Next Article in Journal
A Classification Algorithm for Revenue Range Estimation in Ancillary Service Markets
Previous Article in Journal
A Method for Determining Medium- and Long-Term Renewable Energy Accommodation Capacity Considering Multiple Uncertain Influencing Factors
Previous Article in Special Issue
Discrete-Time Impedance Model-Based Dynamic Analysis and Parameter Design for VSG-Controlled Grid-Forming DG System
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Systematic Review

Systematic Review of Optimization Methodologies for Smart Home Energy Management Systems

by
Abayomi A. Adebiyi
* and
Mathew Habyarimana
*
Department of Electrical Power Engineering, Durban University of Technology, Durban 4001, South Africa
*
Authors to whom correspondence should be addressed.
Energies 2025, 18(19), 5262; https://doi.org/10.3390/en18195262
Submission received: 4 August 2025 / Revised: 19 September 2025 / Accepted: 29 September 2025 / Published: 3 October 2025
(This article belongs to the Special Issue Advanced Application of Mathematical Methods in Energy Systems)

Abstract

Power systems are undergoing a transformative transition as consumers seek greater participation in managing electricity systems. This shift has given rise to the concept of “prosumers,” individuals who both consume and produce electricity, primarily through renewable energy sources. While renewables offer undeniable environmental benefits, they also introduce significant energy management challenges. One major concern is the variability in energy consumption patterns within households, which can lead to inefficiencies. Also, improper energy management can result in economic losses due to unbalanced energy control or inefficient systems. Home Energy Management Systems (HEMSs) have emerged as a promising solution to address these challenges. A well-designed HEMS enables users to achieve greater efficiency in managing their energy consumption, optimizing asset usage while ensuring cost savings and system reliability. This paper presents a comprehensive systematic review of optimization techniques applied to HEMS development between 2019 and 2024, focusing on key technical and computational factors influencing their advancement. The review categorizes optimization techniques into two main groups: conventional methods, emerging techniques, and machine learning methods. By analyzing recent developments, this study provides an integrated perspective on the evolving role of HEMSs in modern power systems, highlighting trends that enhance the efficiency and effectiveness of energy management in smart grids. Unifying taxonomy of HEMSs (2019–2024) and integrating mathematical, heuristic/metaheuristic, and ML/DRL approaches across horizons, controllability, and uncertainty, we assess algorithmic complexity versus tractability, benchmark comparative evidence (cost, PAR, runtime), and highlight deployment gaps (privacy, cybersecurity, AMI/HAN, and explainability), offering a novel synthesis for AI-enabled HEMS.

1. Introduction

The growing demand for electrical energy, rising energy prices, and environmental constraints on conventional energy sources have driven significant changes in the power industry. One major shift is the adoption of smart demand-side management techniques to optimize energy consumption. Studies indicate that in the United States, residential and commercial consumers waste at least 30% of the energy utilized, underscoring the need for more efficient energy practices [1,2]. The advancement of information and communication technologies (ICTs) has further increased the demand for reliable and high-quality power supplies. ICTs are a vital component of smart grid systems, enabling seamless information transfer between systems. This communication capability is instrumental in controlling and coordinating various smart grid technologies to adapt swiftly to fluctuating energy demands. By leveraging these technologies, smart grids can integrate local renewable energy sources, such as wind and solar power, to address environmental challenges, enhance system reliability, and reduce infrastructure costs [3].
ICTs also play a critical role in managing energy demand from consumer premises. They form the backbone of smart home appliances’ energy management systems, a key element of the broader smart grid framework. These systems enable efficient control and scheduling of energy consumption by appliances, reducing energy waste and promoting cost-effective usage. Additionally, smart grids can incorporate distributed renewable energy resources at the local level, further improving energy efficiency and reducing environmental impact. The integration of ICTs within smart grids represents a transformative approach to energy management, combining advanced technologies to create a more sustainable, reliable, and efficient energy ecosystem [4].
The concept of microgrids has emerged as a sustainable solution to environmental challenges by integrating renewable resources such as wind and solar energy with energy generators like microturbines and fuel cells. This approach not only reduces generation costs but also minimizes environmental pollution. These resources are interconnected and interact with higher-level distribution networks, creating an efficient and adaptable energy system [5,6]. The increasing restrictions on fossil fuels, growing energy demand, enhanced living standards, and pressing environmental issues such as global warming have driven significant technological advancements and the adoption of modern energy systems [7]. Smart grids have evolved from the concept of Advanced Metering Infrastructure (AMI), aimed at improving demand management, energy efficiency, and grid reliability. They offer self-repair capabilities and resilience to natural disasters or intentional sabotage [8,9].
Smart grid advancements enable consumers to optimize energy consumption by scheduling appliances based on real-time electricity pricing (RTEP) [10]. This capability promotes energy efficiency and cost savings. Demand-side management (DSM) techniques, long established in power systems, now incorporate demand response (DR) strategies that focus on load shifting and conservation programs [11,12,13]. A primary function of DR techniques is to shift energy usage from peak hours to off-peak hours, reducing stress on the grid and lowering costs [14]. To encourage participation, consumers are offered incentives, fostering greater engagement in energy-efficient practices. Microgrids and smart grids collectively represent a transformative step toward sustainable energy systems, integrating advanced technologies to balance environmental responsibility, cost efficiency, and grid reliability [15,16].
Demand response (DR) programs are designed to shift energy consumption from peak demand periods to off-peak times by offering customers financial incentives [17,18]. This approach helps reduce critical and daily peak demand, alleviating stress on the power grid and enhancing overall energy efficiency. In a smart grid environment, various DR pricing schemes are utilized to encourage participation, including real-time pricing (RTP), time-of-use pricing (TOUP), and critical peak pricing (CPP) [1,19]. These pricing models incentivize customers to modify their energy usage patterns, fostering load management and cost optimization.
This paper provides an in-depth analysis of optimization techniques for smart home energy management systems (HEMSs) [20,21]. Performance evaluation is conducted using normalized outcomes, expressed as relative improvements over baseline strategies in Table 1. In cases where specific numerical results were not provided, qualitative assessments are used to indicate the direction of improvement. The primary performance indicators include cost reduction (↓%), peak-to-average ratio (PAR ↓%), and occupant comfort, which is measured either through comfort scores or the frequency of violations, with lower values reflecting better performance.
Section 2 examines the architecture of HEMSs within the framework of the home area network (HAN) protocol. The HAN facilitates seamless communication between smart appliances and energy management controllers, enabling efficient scheduling and energy allocation. This section highlights the significance of HAN in creating an interconnected and responsive home energy ecosystem. Section 3 focuses on optimization methods for HEMS, reviewing heuristic and machine-learning techniques. Mathematical methods offer precise solutions by formulating energy management as optimization problems, while heuristic techniques provide flexible, computationally efficient approaches for complex scenarios, and machine learning methods forecast energy consumption trends and adjust to user behaviour, improving the accuracy of energy scheduling. These methods are analyzed for their ability to enhance energy efficiency, reduce costs, and optimize appliance operation. Section 4 will present a detailed analysis of algorithmic complexity and optimization challenges in HEMS. The section analyzes problem dimensionality, planning horizons, uncertainty modelling, and computational constraints, highlighting how these factors impact real-time decision-making and the scalability of advanced optimization strategies in practical smart home environments. Section 5 presents a comparative analysis of these optimization techniques, evaluating their strengths, limitations, and practical applications in real-world settings. This analysis underscores the potential of advanced optimization strategies to revolutionize smart home energy management by integrating demand response programs with cutting-edge technologies. The paper concludes by emphasizing the critical role of HEMSs in achieving energy efficiency and sustainability. By adopting DR pricing schemes and advanced optimization techniques, smart homes can significantly contribute to balancing energy demand and improving grid reliability. Future research opportunities in HEMS optimization are identified, focusing on innovative algorithms, enhanced system integration, and the potential for widespread deployment in diverse energy markets.

2. The Architecture of HEMS Within Smart Grids

The integration of household energy management systems (HEMSs) into smart grids is a revolutionary strategy for attaining energy efficiency, diminishing peak demand, and improving grid dependability. This section explores the architecture of HEMSs, emphasizing its essential components, communication protocols, and optimization methods, illustrating how HEMS enables energy consumption monitoring, demand response (DR) integration, and renewable energy management, thereby fostering a sustainable energy environment [22,23]. A smart home is conceptualized as an environment where various appliances are equipped with adequate communication interfaces to exchange information with an energy management system (EMS). As depicted in Figure 1, each appliance exclusively communicates with the EMS, ensuring centralised control without direct intercommunication among appliances.
In the Home Energy Management System (HEMS) configuration, the arrows represent different flows within the setup. The red arrows indicate the flow of electrical power, showing how energy moves from the solar panels through the inverter to household appliances, the electric vehicle, or from the grid when required. The green arrows represent data communication, where power meters send consumption information to the HEMS controller, which then communicates with the power monitor for real-time tracking and optimization, and the black arrows highlight system integration and monitoring links, connecting the power monitor and other components to ensure coordinated operation of the overall system.
The architecture of a household energy management system comprises three primary data communication domains: the internet domain, the home area network (HAN) domain, and the smart meter domain (AMI) [23,24]. The internet domain facilitates consumer interaction with the EMS. Consumers can monitor and control their energy consumption profiles, schedule appliance usage, and optimise power consumption patterns through an in-home display (IHD), which may consist of a computer, tablet, or smartphone. The home area network (HAN) domain connects smart appliances to the EMS, enabling seamless data transfer and real-time control of energy usage. It provides the communication backbone within the home, ensuring efficient management of energy resources [25]. The smart meter domain, also referred to as automatic metering infrastructure (AMI), includes a network of interconnected smart meters installed and monitored by utility companies. This domain enables the bidirectional flow of information, transmitting real-time load data and demand response (DR) signals between smart homes and the power market [26,27]. Leveraging real-time pricing (RTP) signals received 24 h in advance, the EMS schedules electricity usage patterns for the next day, optimizing energy consumption while reducing costs. This structured communication framework ensures the effective integration of demand-side management strategies and enhances energy efficiency within the smart grid ecosystem [28,29,30].
In the context of a smart grid, a HEMS is designed to optimize energy usage in households and manage energy supply efficiently. This includes energy sourced from external providers or self-generated energy from alternative sources such as solar panels or wind turbines. By integrating these components, HEMS enhances energy efficiency and supports the adoption of renewable energy, contributing to sustainable energy management [23,31,32]. The HEMS configuration comprises several key elements:
  • Household Loads: These are categorized into scheduled and unscheduled loads. Scheduled loads include appliances that operate at pre-determined times, such as washing machines and dishwashers, while unscheduled loads are devices used on-demand, such as lighting and televisions. Effective management of these loads helps balance energy demand and reduces peak loads [33].
  • Energy Storage Systems: These include battery storage systems that store excess energy generated from renewable sources or during off-peak grid hours for later use. This ensures energy availability during periods of high demand or low generation [34].
  • Alternative Energy Sources: Solar panels, wind turbines, or other renewable energy systems integrated into HEMSs allow households to generate their power, reducing reliance on the grid and lowering energy costs [35].
  • Grid Connections: A bidirectional connection to the utility grid enables energy exchange. Excess energy generated by the household can be exported to the grid, while energy from the grid can be used during periods of insufficient generation [36,37].
  • Electric Vehicles (EVs): HEMSs integrate EVs as both loads and storage units, utilizing their batteries for energy storage and discharge during grid peaks [38,39,40].
  • Control Systems and Communication Technology: Advanced control systems supported by communication technologies, such as Zigbee, Wi-Fi, or Power Line Communication (PLC), enable real-time monitoring and control of energy usage. These systems interact with smart meters to facilitate demand response (DR) and real-time pricing (RTP) mechanisms, ensuring optimal energy use [41].
The HEMS framework not only empowers households to manage energy efficiently but also plays a vital role in achieving grid stability, integrating renewable energy sources, and advancing smart grid objectives. By incorporating cutting-edge technology, HEMS ensures a sustainable and reliable energy system for the future.

3. HEMS Optimization Methodology

Recent years have seen significant research efforts focused on managing home appliances in smart grid environments to reduce electricity costs and optimize energy usage. One of the key challenges in this area is minimizing the peak-to-average ratio, which helps improve grid stability and efficiency. Various studies have explored demand response (DR) strategies as a means to manage household energy consumption effectively [42]. To achieve optimal energy management, researchers have formulated DR problems mathematically and employed machine learning techniques to address the complexities involved. These approaches typically treat home energy management system (HEMS) applications as nonlinear programming problems, aiming to optimize energy usage while considering consumer preferences, electricity pricing, and grid constraints [43]. Integrating advanced optimization techniques has played a crucial role in solving HEMS-related challenges. Machine learning algorithms have been utilized to predict energy demand patterns and dynamically adjust appliance schedules for enhanced efficiency. Additionally, metaheuristic optimization methods have been applied to refine load distribution, further improving cost savings and reducing peak demand [44,45]. Overall, the ongoing advancements in smart grid research are shaping the future of residential energy management, promoting sustainability, and ensuring an intelligent balance between energy supply and demand within households. All the extant works of literature are limited in the comparison of modern heuristic and machine learning techniques application in HEMS optimization.

3.1. Heuristic Optimization Methods for HEMSs

Heuristic optimization techniques iteratively search for practical solutions, making them effective for complex, large-scale problems. They provide fast, scalable, and adaptable alternatives within a reasonable timeframe. Their rising popularity over the past two decades stems from their efficiency and ease of implementation, making them widely used in areas like load scheduling and real-time problem-solving [45]. Heuristic optimization techniques are essential in a Home Energy Management System (HEMS) for effectively scheduling appliance operations and enhancing energy efficiency. These techniques, comprising genetic algorithms, particle swarm optimization, and simulated annealing, yield near-optimal solutions for intricate, multi-objective challenges such as demand response, load balancing, and cost minimization. Heuristic algorithms improve decision-making by incorporating real-time data from smart meters and Non-Intrusive Load Monitoring (NILM) approaches, allowing customers to decrease electricity expenses while ensuring comfort. Their versatility enables them to manage uncertainties in renewable energy production and usage patterns, rendering them essential for intelligent and sustainable energy management in contemporary houses [42].
The study in [46] presents an in-depth review of Non-Intrusive Load Monitoring (NILM) techniques and their applications in Home Energy Management Systems (HEMSs) and Ambient Assisted Living (AAL). NILM methods enable the disaggregation of total power consumption in buildings, identifying the energy usage of individual appliances without requiring additional hardware beyond smart meters. This technology has gained significant interest due to its potential for improving energy efficiency, optimizing demand response programs, and assisting in independent living for elderly individuals. NILM methods rely on different sampling rates to capture steady-state and transient electrical characteristics. Low-sampling methods (below 1 Hz) analyze aggregate power consumption, while high-frequency methods (above 2 kHz) use detailed waveform analysis. Advanced techniques such as machine learning, Hidden Markov Models (HMMs), deep learning, and graph signal processing enhance load identification accuracy. The study shows that NILM can help reduce energy consumption by 0.7–4.5% through user awareness and targeted energy-saving strategies. Furthermore, NILM facilitates load scheduling, helping households shift energy use to off-peak hours, reducing peak demand, and optimizing electricity costs. In an ageing population, AAL technologies are increasingly important for assisting elderly individuals in independent living. NILM plays a crucial role in monitoring appliance usage to infer daily activities. Compared to wearable health monitoring devices, NILM-based AAL systems offer low-cost, non-intrusive monitoring without requiring user intervention. However, privacy concerns arise as energy usage data can infer occupancy patterns.
The paper in [47] examines the utilization of Swarm Intelligence (SI) algorithms for maximizing multiple facets of the Internet of Things (IoT) within Home Energy Management Systems (HEMSs). SI draws inspiration from the collective behaviour of social creatures, like ants, bees, and birds, enabling simple agents to cooperate and identify optimal solutions for intricate issues. The study offers a comprehensive analysis of SI algorithms, such as Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Artificial Bee Colony (ABC), Bacterial Foraging Optimization (BFO), and Butterfly Optimization Algorithm (BOA), examining their capacity to improve wireless sensor networks (WSNs) and other IoT-related applications in Home Energy Management Systems (HEMSs). Wireless Sensor Networks (WSNs) are essential to the Internet of Things (IoT) as they provide real-time data acquisition and communication from dispersed sensor nodes. Nonetheless, WSNs encounter difficulties in energy economy, route optimization, and sensor placement. The research emphasizes how SI algorithms tackle these challenges by optimizing cluster head selection, refining routing protocols, and strengthening node localization. SI-based CH selection methodologies utilize criteria such as residual energy, proximity to the base station, and communication link quality to identify ideal CHs. Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO) are notably proficient in optimizing energy usage and extending network longevity, whereas Artificial Bee Colony (ABC) and hybrid methodologies offer further enhancements in efficiency.
SI approaches enhance WSN routing protocols by optimizing energy-efficient, multipath routing. ACO-based routing protocols effectively identify optimal data transmission routes by mimicking pheromone-driven foraging behaviour, hence assuring equitable load distribution among sensor nodes. PSO-based routing enhances transmission distances and reduces energy usage via adaptive routing algorithms. Hybrid SI methodologies, integrating ACO, PSO, and fuzzy logic, augment routing efficacy by accounting for many elements like energy levels, network architecture, and data traffic conditions. In addition to WSNs, SI applications also encompass other IoT sectors, such as vehicular ad hoc networks (VANETs), smart cities, and industrial IoT. In VANETs, SI algorithms enhance multicast routing by maximizing network connectivity and minimizing latency in dynamic traffic situations. In smart cities, SI-based optimization improves resource allocation, traffic control, and environmental monitoring. The research emphasizes the application of SI in cloud and edge computing systems, where effective resource scheduling and task allocation are essential for optimal performance [47].
Although SI algorithms provide considerable benefits in IoT applications, the research recognizes various limits. Numerous SI-based methodologies exhibit elevated computational complexity, rendering real-time execution difficult in resource-limited settings. Moreover, the majority of SI algorithms depend on parameter tuning, which influences performance and convergence speed, hence requiring additional study to create adaptive SI methods that can dynamically modify parameters according to real-time network conditions. Furthermore, privacy and security issues in SI-based IoT systems necessitate increased focus to avert potential vulnerabilities in data transmission and processing.
The author of [48] focuses on shifting energy usage from peak hours to off-peak times, thereby reducing electricity costs and peak-to-average ratio (PAR). The study classified residential loads into fixed and flexible appliances and proposed an optimization model using two heuristic algorithms: The Genetic Algorithm (GA) and the Harmony Search Algorithm (HSA). These methods are compared against a baseline scenario with no optimization. The study evaluates energy consumption scheduling using real-world time-of-use pricing data. The results indicate that both GA and HSA effectively minimize electricity costs and flatten the demand curve. Specifically, GA reduces electricity costs by 0.9% and PAR by 15%, while HSA achieves reductions of 3.98% and 5.8%, respectively. The study further tests scalability by considering multiple users and different time resolutions, demonstrating that heuristic-based HEMSs can efficiently manage energy consumption across various scenarios. The limitation of this study is that it does not incorporate renewable energy sources or distributed energy resources, which could further enhance energy savings. Additionally, heuristic methods do not always guarantee globally optimal solutions, which may affect performance in more complex energy management scenarios.
The authors of [49] demonstrated effective energy management in a smart grid by integrating demand response systems and renewable energy sources. The research introduces an Appliance Scheduler and Energy Management Controller (ASEMC) utilizing a hybrid heuristic optimization method (HGPDO) that integrates genetic algorithm (GA), particle swarm optimization (PSO), and wind-driven optimization (WDO). The approach seeks to reduce electricity expenses, peak-to-average ratio (PAR), carbon emissions, and waiting time, while enhancing user comfort for thermal, visual, and air quality metrics. The system integrates various renewable energy sources, such as solar, wind, and combined heat and power (CHP), in addition to battery storage and electric vehicle batteries. The suggested strategy is assessed across various price scenarios, revealing substantial decreases in power expenses and peak loads relative to conventional optimization techniques. The findings demonstrate that the proposed HGPDO algorithm surpasses current methods in minimizing costs and emissions while preserving user comfort. Despite its efficiency, the study’s drawbacks are its presumption of optimal operating conditions for renewable energy sources and neglect of such variables as weather variability. Also, the proposed optimization approach exhibits significant computational complexity, rendering real-time implementation difficult, necessitating further study to investigate adaptive algorithms that react dynamically to fluctuating grid circumstances and consumer behaviour.
The authors in [50] presented a novel HEMS that integrates three innovative demand response (DR) optimization approaches aimed at improving energy efficiency in smart homes. Unlike conventional DR programs that rely on pricing mechanisms, the proposed strategies are decision-variable-based, directly targeting peak demand reduction, load synchronization, and demand flattening. The research introduces a new index to evaluate DR effectiveness, balancing DR achievements with the economic impact on homeowners. The developed methods were tested in a benchmark prosumer environment, demonstrating superior performance compared to conventional pricing-based DR mechanisms. Notably, the results indicate that the proposed strategies effectively reduce peak demand and enhance efficiency without significantly increasing electricity costs. Additionally, the study examined the influence of energy storage availability, showing that battery energy storage (BES) plays a crucial role in improving DR effectiveness. Despite these promising results, the study has some limitations. It does not address the load synchronization issue comprehensively, leaving a research gap. Additionally, the effectiveness of different DR strategies remains difficult to compare due to the absence of a universally accepted evaluation index. Lastly, the computational burden of the optimization framework also poses challenges for real-time implementation.
In [51] an efficient scheduling approach for a HEMS using heuristic optimization techniques to optimize energy consumption and reduce electricity costs in smart homes was presented. The study integrates the Genetic Algorithm (GA), Binary Particle Swarm Optimization (BPSO), and Wind Driven Optimization (WDO) to schedule home appliances while considering demand response programs and Time of Use (ToU) pricing schemes. The model is applied to a typical home with multiple appliances, an on-site renewable energy generation system, and battery storage. The results indicate that GA performs best in terms of cost reduction and energy consumption efficiency, while BPSO is superior in reducing the peak-to-average ratio (PAR), which is essential for grid stability. The study identifies the benefits of incorporating renewable energy sources in smart homes, demonstrating that shifting loads to periods of low electricity cost significantly reduces overall energy expenses. Despite its promising results, the study has some limitations. It did not consider the impact of increasing the number of appliances on the efficiency of the proposed algorithms and the model focuses on a single household scenario, which may limit its applicability to larger residential areas. This created a research gap and demonstrated that it is necessary to explore multi-household implementations and enhanced optimization techniques.
The study in [52] presents a meta-heuristic-based HEMS designed to optimize smart appliance scheduling and reduce electricity costs in residential complexes. The proposed approach, named the Gradient-Based Runge–Kutta Optimizer (GRUN), integrates the Runge–Kutta optimizer with a Local Escaping Operator (LEO) from gradient-based optimization techniques. The system effectively manages electricity demand, ensuring that power consumption remains stable before and after scheduling, while also minimizing peak loads and overall costs. The GRUN technique is validated through benchmark testing and compared with well-known optimization algorithms such as artificial ecosystem-based optimization, hunter-prey optimization, and the original Runge–Kutta method. The results demonstrate that the GRUN approach significantly reduces electricity costs, achieving reductions of up to 60% for a single home and 20% for 100 homes. Likewise, peak demand reductions of up to 50% for one home and 25% for multiple homes were observed. The study offers an effective method for household energy management, balancing energy efficiency with cost savings. Despite its advantages, the study has limitations. It assumes ideal conditions for smart appliances and does not fully account for variations in household behaviours.
In [53] an optimal scheduling approach for HEMSs by integrating dynamic pricing and a direct load control (DLC)-based demand response (DR) program were presented. The study employs binary particle swarm optimization (BPSO) and a novel discrete elephant herd optimization (DEHO) technique to optimize load scheduling, minimizing electricity costs and the peak-to-average ratio (PAR). A prototype hardware model is developed to validate the proposed method, demonstrating its practical feasibility. The results indicate that the DEHO algorithm achieves superior performance in reducing energy costs and PAR compared to conventional optimization methods. The optimal scheduling approach improves PAR to 2.504 and results in energy cost savings. The study also highlights the benefits of integrating photovoltaic (PV) systems, which further reduce grid dependency and enhance cost efficiency. Despite the study’s novelty, it has limitations in terms of the assumption of perfect electricity prices and consumer behaviour, which may not always be realistic, and the computational complexity of the DEHO algorithm, which could pose challenges for large-scale real-time applications scheduling accuracy and scalability.
An Enhanced Leader Particle Swarm Optimization (ELPSO) algorithm for optimal scheduling of home appliances under demand response programs in smart grids was present in [54]. The study formulates the appliance scheduling problem as a constrained multi-objective optimization problem with integer decision variables, aiming to minimize electricity bills while maintaining consumer comfort. C optimization techniques often neglect user convenience, but the proposed ELPSO algorithm integrates a mutation scheme to enhance the leader particle, improving search efficiency and avoiding premature convergence. The study evaluates ELPSO against existing metaheuristic algorithms, including basic Particle Swarm Optimization (PSO), Artificial Bee Colony (ABC), and Gravitational Search Algorithm (GSA). The results demonstrate ELPSO’s superiority in optimizing appliance schedules under various tariff structures, including time of use (TOU), real-time pricing (RTP), and incentive-based load curtailment programs. The proposed method significantly reduces electricity bills and peak-to-average load ratios while maintaining a balance between cost savings and user comfort. Despite its effectiveness, the study has some limitations. It assumes accurate predictions of electricity prices and user behaviour, which may not always be realistic.
The article in [55] presents a multi-objective home energy management system (HEMS) that integrates solar energy and electric vehicles, aiming to optimize energy efficiency and cost savings in smart homes. The optimization method used is a combination of the augmented ε-constraint method and lexicographic optimization (AUGMECON-LO). This approach effectively handles multiple conflicting objectives by generating Pareto optimal solutions while ensuring computational efficiency. The proposed HEMS model significantly improves energy management by incorporating vehicle-to-home (V2H) and home-to-grid (H2G) capabilities. Through simulations, it is demonstrated that the system can reduce energy costs by 47.96% and the peak-to-average ratio (PAR) by 55.24%, all while maintaining a low discomfort index (DI). The study evaluates various factors, including battery storage capacity, photovoltaic (PV) system sizing, and uncertainty in renewable energy generation, confirming that the model is adaptable to real-world applications. Despite its strengths, the study has some limitations. It assumes perfect forecast data for solar energy and electricity prices, which may not always be reliable in practice. It does not consider the battery degradation of the storage system and electric vehicle, which could impact long-term efficiency. It is also computationally complex.
The research in [56] presents a heuristic optimization approach for scheduling home appliances in a HEMS. The methodology involves designing a programmable heuristic-based energy management controller (HPEMC) that optimizes appliance scheduling to minimize power costs, reduce carbon emissions, and lower the peak-to-average ratio (PAR). A demand-response model is applied, where appliance operation is strategically adjusted based on dynamic electricity pricing. The research employs a genetic algorithm (GA) to determine optimal scheduling, reducing energy consumption by 25.98% according to simulations. The hybrid microgrid system, which integrates solar photovoltaic (PV) generation and battery storage, is modelled and tested using Simulink SimPowerSystems. The study’s impact is significant in advancing energy efficiency in smart homes. By leveraging demand-side management (DSM) strategies, the system enhances user comfort while reducing reliance on conventional energy sources. The integration of renewable energy resources and automated scheduling contributes to lower electricity costs and peak demand mitigation. The approach aligns with sustainability goals by curbing greenhouse gas emissions. However, the study has limitations. The heuristic-based approach did not fully capture real-time grid fluctuations, and its scalability for larger residential communities remains uncertain. Future improvements could involve incorporating real-time forecasting and adaptive machine-learning techniques to enhance system efficiency and flexibility.
In [57], the study focuses on optimizing HEMSs to address rising energy consumption due to technological advancements. The study employs Ant Colony Optimization (ACO) and a Load Scheduling Algorithm to manage residential loads while considering power quality issues like harmonics. The proposed model ensures a fixed daily energy cost by scheduling loads at different intervals. By reducing costs and enhancing efficiency, this approach supports demand-side management, improving energy sustainability and minimizing the strain on power systems under various operating conditions. Nonetheless, the heuristic-based Ant Colony Optimization (ACO) algorithm did not fully adapt to real-time fluctuations in energy demand and supply, limiting its effectiveness in dynamic environments. Secondly, the model assumes a fixed daily energy consumption cost, which may not account for unpredictable load variations or sudden peak demands. Thirdly, the impact of external factors such as weather conditions and grid stability is not fully considered. Additionally, the study does not extensively explore the scalability of the proposed approach for larger residential communities or its integration with smart grid technologies for enhanced efficiency and flexibility.
The research in [58] employs a Mixed-Integer Linear Programming (MILP) optimization framework to develop an integrated energy management system for residential demand response. The methodology focuses on optimizing the operation of shiftable loads, electric water heaters, air conditioners, and battery storage, including both static and electric vehicle (EV) batteries, within a 24-h planning period. The study incorporates forecasting models—Auto Regressive (AR), Seasonal Auto Regressive Integrated Moving Average (SARIMA), Lasso, Ridge regression, and Random Forest—to predict day-ahead market energy prices, enabling cost-effective energy consumption strategies. The impact of this research is significant, demonstrating that an optimized energy management system allows consumers to lower electricity costs or even generate profits by strategically utilizing batteries and shifting energy consumption to low-price periods. The study highlights the profitability of static batteries, as they enable energy storage during off-peak hours for later use, whereas EV batteries, constrained by departure schedules, offer more limited cost-saving opportunities. The accurate forecasting of hourly electricity prices enhances decision-making in energy consumption and storage. Nevertheless, the research presents certain limitations as it does not account for battery degradation over time, which may impact long-term economic feasibility. The model assumes a single household, limiting its applicability to broader energy systems with multiple users. Further exploration of real-time energy pricing, supply-demand dynamics, and integration of additional renewable sources, such as biomass or hydroelectric energy, would enhance the robustness of the approach, which could also investigate consumer behaviour in response to dynamic tariff adjustments.
In [59], the authors employ a hybridized metaheuristic optimization approach combining Ant Colony Optimization (ACO) and Genetic Algorithm (GA) for cost-efficient demand-side management in smart communities. The methodology involves scheduling household appliances based on price-driven demand response while integrating renewable energy sources such as photovoltaic panels and energy storage systems. The hybrid algorithm enhances ACO by mitigating its tendency to converge at local optima through GA’s mutation and crossover operators, thereby improving solution diversity and convergence efficiency. Simulations were conducted on a smart community of 50 houses, with performance evaluated against alternative scheduling methods, including conventional ACO-based energy management controllers and a mutated ACO model. The research demonstrates significant impacts on energy cost reduction, peak load minimization, and enhanced user comfort. The hybrid approach achieves a 35.4% reduction in peak load demand and a 33.67% reduction in cumulative energy costs. By leveraging renewable energy sources, the system further improves grid independence and sustainability. However, limitations exist. The model assumes predefined appliance operation constraints and does not fully incorporate stochastic user behaviour variations. Additionally, the approach is tested in simulation environments without real-world validation, and the impact of battery degradation over time is not considered. A further study addressing real-time dynamic pricing and multi-user interaction effects to enhance scalability and practical applicability would enhance the research results.

3.2. Machine Learning Optimization Methods for HEMS

Artificial Intelligence (AI) is a broad field that enables machines to perform tasks more innovatively and in a human-like manner by replicating human behaviour and decision-making processes. It allows machines to analyze, interpret, and respond to data dynamically [60]. A key subfield of AI is Machine Learning (ML), which focuses on developing systems that can learn from experience, improving their performance over time. In the case of machines, experience is derived from data, which ML algorithms use to make decisions and predictions. These algorithms become more accurate as they process new data, refining their outputs continuously [61]. ML is generally categorized into three types: Supervised Learning, Unsupervised Learning, and Reinforcement Learning. Supervised Learning involves training a model using labelled data, enabling it to make predictions based on past examples. Unsupervised Learning, on the other hand, deals with unlabelled data, where the algorithm identifies patterns and relationships without predefined categories. Reinforcement Learning is based on a reward system, where an agent interacts with an environment, learns from feedback, and optimizes actions to maximize rewards. These three ML approaches, as illustrated in Figure 2, form the foundation of modern AI applications, driving advancements in automation, decision-making, and data analysis across various industries [62].
In [63], the study focuses on reducing standby power consumption in home appliances using an Intelligent Home Energy Management System (i-HEMS) based on supervised learning ML techniques. The methodology involves deploying i-HEMS in a standard home environment, integrating IoT sensors, power usage monitoring, and behavioural pattern algorithms. The system uses Long Short-Term Memory (LSTM) networks to predict occupancy and optimize power scheduling. Experiments were conducted over six days, comparing energy consumption before and after implementing i-HEMS. The results show a 20% reduction in overall energy consumption, decreasing from 13,062 Wh to 10,434 Wh, with 9060 Wh attributed to home appliances and 1374 Wh to i-HEMS operations. The impact of the study demonstrates significant energy savings through intelligent control, contributing to energy efficiency and cost reduction in residential settings. However, limitations include the need for further validation in diverse home environments and potential adaptation challenges for different user behaviours. Future studies could focus on expanding experimental validation and optimizing control algorithms to enhance adaptability. This study highlights the potential of AI-driven energy management systems in reducing household energy waste and improving sustainability.
The authors in [64] proposed a supervised learning strategy to minimize HEMS energy costs by optimally scheduling the operation of an Energy Storage System (ESS) and an Electric Vehicle (EV) utilizing historical data on energy prices, consumption, solar irradiation, and EV availability to solve a Mixed-Integer Linear Programming (MILP) model, generating optimal decisions for ESS and EV operations. These optimal decisions, represented as state-action pairs, are then used to train Deep Neural Networks (DNNs) that learn how to predict the best actions in real time based on current conditions. The DNNs then act as decision-makers, using real-time data to forecast the optimal charge and discharge actions for the ESS and EV. This method is validated through simulations and compared with a multi-agent deep deterministic policy gradient (MADDPG) algorithm and a forecasting-based approach, showing better cost savings and adaptability. The study addresses key research gaps in HEMS, particularly the limitations of traditional and reinforcement learning methods that struggle with uncertainty and real-time adaptability. It also fills a gap in supervised learning research, which has previously focused only on single devices like HVACs or ESS, by incorporating EVs into the scheduling process. This comprehensive study uses supervised learning for real-time control of both ESS and EV in a smart home setting. However, a clear research gap in the study is the limited consideration of broader household appliance integration and dynamic user behaviour modelling within the HEMS framework. While the study focuses on optimizing the operations of ESS and EV using supervised learning, it assumes flexible appliance usage without explicitly incorporating smart appliance scheduling, user comfort constraints, or behaviour-driven energy patterns. Future studies could expand this work by integrating multiple controllable appliances (e.g., HVAC systems, washing machines, dishwashers) into the optimization model and learning framework. Additionally, the current supervised learning model relies on historical data to train the DNNs, which may not fully capture real-time behavioural shifts or unexpected appliance usage. Future research could explore online learning or hybrid learning models that combine supervised learning with real-time adaptation techniques. Another gap lies in the scalability to community-level or multi-home energy management systems, where interactions between households, shared renewable resources, and peer-to-peer energy trading could introduce new challenges. The paper does not explicitly address cybersecurity and data privacy concerns, which are critical in AI-driven energy systems and represent an important direction for future research.
The study in [65] introduces a supervised-learning-based approach for hour-ahead demand response (DR) in HEMSs to minimize daily energy costs without altering residents’ appliance usage behaviour. Instead of directly controlling home appliances, the proposed method manages Energy Storage Systems (ESSs) and Renewable Energy Systems (RESs) based on learned user behaviours. The methodology involves training deep neural networks (DNNs) using historical data optimized through Mixed-Integer Linear Programming (MILP). Once trained, these networks predict near-optimal scheduling actions for ESS and RES in real-time. The study compares its MILP-based supervised learning approach with two alternative strategies: a multi-agent deep deterministic policy gradient (MADDPG)-based strategy and a forecast-based MILP strategy. The impact of the study is evident in its improved energy cost savings and system efficiency. Simulations across three real-world households show that the MILP-based supervised learning strategy outperforms the other approaches, achieving up to 82.8% of the cost savings achieved by an optimal MILP solver. The approach effectively balances energy use while maintaining user comfort. However, the study’s limitations include reduced performance in unpredictable user behaviour scenarios and reliance on historical data, which may affect adaptability. Future research could improve the model’s generalizability and integrate cooperative strategies for multiple households.
The study in [66] explores a real-time energy management framework for smart homes incorporating energy storage systems (ESSs) and electric vehicles (EVs). The methodology is based on a supervised-learning approach using deep neural networks (DNNs) trained on optimal energy scheduling results generated by a mixed-integer linear programming (MILP) solver. The framework also includes two additional AI-driven strategies, multi-agent deep deterministic policy gradient (MADDPG) and a forecasting-based method, for comparative performance analysis. The supervised-learning method significantly improves real-time scheduling accuracy, overcoming forecasting errors associated with traditional optimization techniques. The impact of the study is demonstrated through simulations in two smart home scenarios, where the proposed system achieved up to 28.96% cost reduction while maintaining user convenience. The limitations include the need for robust generalization in unpredictable conditions such as grid outages, hardware failures, or spontaneous EV usage. Additionally, real-time communication network delays and the performance of power electronic inverters were not accounted for, which could affect practical implementation. Future work aims to enhance system adaptability, integrate hardware-in-the-loop simulations, and scale the approach to large-scale microgrids and EV charging networks.
The study in [67] presents a machine learning (ML)-based framework designed to optimize household energy consumption by leveraging predictive analytics. Traditional energy management approaches focus on optimizing power system efficiency for energy suppliers, whereas this research prioritizes consumer benefits by encouraging behavioural changes based on dynamic energy pricing. The methodology involves integrating data from smart meters, renewable energy generation, and dynamic pricing structures to develop an ML-based recommendation system. The system predicts optimal energy usage times and provides personalized notifications to users, suggesting either increased or reduced consumption based on upcoming cost fluctuations. Simulations were conducted in MATLAB to analyze the performance of different ML algorithms using varying input data and model characteristics. The impact of this research is significant, as it enhances consumer decision-making in energy usage, reduces peak loads on the grid, and supports the integration of renewable energy sources. The model promotes energy efficiency and cost savings by influencing user behaviour through intelligent recommendations. Its implementation in Home Energy Management Systems (HEMSs) or smart meters offers a scalable solution for widespread adoption. However, limitations include the dependency on accurate and real-time data from smart meters, variations in user response to recommendations, and potential challenges in integrating the model with existing household appliances. Future work could explore real-world deployment, refine the ML model for greater adaptability, and incorporate additional parameters such as weather conditions and appliance-specific energy consumption patterns.
The study in [68] proposed a deep reinforcement learning-based home energy management system (DRL-HEMS) to optimize household energy consumption. The system incorporates various household appliances, an energy storage system, a photovoltaic system, and an electric vehicle. The DRL-HEMS aims to minimize electricity costs while considering resident comfort and distribution transformer loading, a novel aspect compared to existing works. The optimization is achieved using a deep Q-network (DQN) algorithm, which learns optimal appliance scheduling from real-world electricity pricing, transformer load data, and renewable energy generation. The system successfully reduces electricity bills by scheduling appliance operations based on dynamic pricing and renewable energy availability. Additionally, it minimizes resident discomfort by integrating a generalized reward function that considers electricity costs, dissatisfaction costs, and the loss of life of the distribution transformer. The model is evaluated using real-world datasets, demonstrating that it outperforms traditional model-based approaches in cost savings and computational efficiency. A key limitation of the study is that deep reinforcement learning methods require extensive training and may not generalize well to unseen environments. Furthermore, the model operates as a black-box system, making interpretability and transparency a challenge for real-world implementation.
The research in [69] proposes a HEMS integrating solar photovoltaic (PV) energy with an artificial intelligence-based control system using the Levenberg–Marquardt (LM) algorithm. The methodology involves designing an Internet of Home Energy Management System (IoHEMS) that optimizes energy usage through a smart plug system, IoT functionality, and machine learning-based energy harvesting. Data is collected from solar panels and electrical appliances, processed through an artificial neural network, and used to predict energy consumption and optimize appliance switching to reduce peak-hour electricity tariffs. The model was tested using MATLAB/Simulink (R2016a), achieving high accuracy with a regression coefficient of 0.9999 and minimal mean squared error. The impact of this research lies in its ability to transform conventional homes into smart energy-efficient homes, reducing energy costs and improving grid-to-battery efficiency. The IoHEMS allows real-time monitoring and control through a mobile application, enhancing consumer participation in energy management. Additionally, it contributes to sustainable energy practices by integrating renewable sources. However, the study has limitations, including scalability challenges for large-scale implementation and computational complexity associated with AI-based optimization. Future improvements could include expanding the system’s capabilities to predict dynamic energy patterns for larger residential communities.

3.2.1. Supervised Learning Optimization Methods for HEMS

This section presents a synthesis of supervised learning optimization methods for HEMS, highlighting their capacity to enhance forecasting precision, computational efficiency, and control robustness in smart energy ecosystems. The study in [66] presents a real-time home energy management system (HEMS) that aims to minimize residential electricity costs by optimally managing an energy storage system (ESS) and electric vehicle (EV). Unlike traditional optimization or deep reinforcement learning (DRL) methods, which face limitations in handling uncertainty and computational demands, the study introduces a supervised-learning-based strategy. This approach enhances scheduling stability and scalability by using historical MILP-generated optimal decisions to train deep neural networks (DNNs), which then make real-time control decisions without relying on complex forecasts. The model was evaluated against a DRL method (MADDPG) and a forecast-based MILP model, showing that it achieves near-optimal cost performance with lower computational overhead and improved response time. While effective, the model assumes ideal conditions, such as perfect communication and no battery degradation, which may limit its real-world applicability. The model’s success is dependent on the availability and quality of expert-labelled training data. The study also addresses a key research gap by incorporating bidirectional EV scheduling into the HEMS framework, an area often overlooked in existing supervised learning models. Using real-world datasets including UK residential load profiles, European solar data, Spanish electricity prices, and EV usage patterns, the simulation confirms the model’s practicality and robustness. However, future work should consider real-world uncertainties, battery lifecycle effects, and hybrid control strategies. The study highlights the growing potential of supervised learning to provide fast, reliable, and cost-effective energy management in smart home environments with increasing integration of distributed energy resources.
The study in [70] proposes a supervised learning optimization method for Home Energy Management Systems (HEMSs) using a Quantum Support Vector Machine (QSVM). This model is trained to forecast short-term household energy consumption by learning from the AMPds2 dataset, which contains detailed energy usage data collected from 21 different household appliances over two years. The methodology begins with resampling and preprocessing the data using MinMax scaling to standardize the range. The data is then transformed into quantum states and processed through quantum circuits using Hadamard and CNOT gates to map features into a high-dimensional quantum space. A quantum kernel is employed to calculate similarities between data points, and the final forecasting is performed using a traditional Support Vector Machine trained with this quantum kernel. The study compares the QSVM’s performance with deep learning models such as LSTM and RNN and shows that the QSVM outperforms them with a prediction accuracy of 97.36% using 2 qubits, alongside lower RMSE and MAE values. However, a key research gap lies in the limited evaluation of the model’s generalizability across different datasets and environments. The QSVM model is tested only on the AMPds2 dataset, which may limit its applicability in diverse real-world scenarios. Future research can expand by applying QSVM to varied and larger datasets, investigating hybrid models that combine QSVM with other deep learning techniques, and evaluating its robustness against fluctuating external variables like user behaviour, appliance scheduling, and environmental changes to ensure more adaptable and scalable energy management solutions.
The article in [71] provides a comprehensive review of supervised learning methods for PV power forecasting in home energy management systems (HEMSs), focusing particularly on applying various supervised deep learning architectures, including multilayer perceptrons (MLPs), recurrent neural networks (RNNs), convolutional neural networks (CNNs), and graph neural networks (GNNs), to model and forecast PV power output. These models are trained using historical time-series data of solar irradiance, temperature, humidity, and PV output, with appropriate preprocessing and hyperparameter tuning. MLPs are effective for nonlinear regression tasks, RNNs and their variants like LSTM and GRU are utilized for capturing temporal dependencies, CNNs extract local temporal patterns, and GNNs handle spatiotemporal dependencies using graph-based representations of data across multiple locations. The evaluation of these models is conducted using standardized datasets such as Solar-6 and NSRDB, and RMSE, MAE, and MAPE performance metrics. Despite the success of these approaches, a notable research gap remains in their limited performance under ultra-long-term forecasting conditions. Most models struggle to maintain accuracy when predicting far into the future due to increased variability and computational complexity. Furthermore, the integration of user behaviour, weather uncertainty, and hybrid architectures remains underexplored. Future research can focus on improving long-term prediction reliability by enhancing temporal feature extraction, combining deep learning with physical modelling, and integrating real-time adaptive learning strategies to handle unexpected changes in PV output patterns. Addressing these challenges will lead to more robust and practical HEMS applications for dynamic grid environments.
The article in [72] explores supervised learning optimization methods within HEMS by categorizing their applications across various architectural frameworks such as centralized, decentralized, distributed, and hierarchical models. The methodology primarily involves applying supervised learning techniques like support vector machines (SVMs), regression models, and deep neural networks (DNNs) for forecasting, control, and scheduling tasks in key HEMS components, including energy management information systems (EMISs), energy storage systems (ESSs), and demand-side management systems (DSMSs). These supervised learning models leverage labelled datasets to optimize load scheduling, energy storage utilization, and demand response while ensuring data-driven decision-making in energy flow management. Within centralized frameworks, SVM and regression-based methods provide global optimization, while decentralized systems benefit from modular supervised algorithms with localized control. Despite these advancements, a critical research gap remains in the scalability and interoperability of these supervised models under dynamic grid conditions. Existing studies often evaluate models in isolated environments with limited scalability considerations, overlooking challenges like data heterogeneity, latency, and coordination across distributed nodes. Moreover, most implementations lack robust mechanisms for integrating real-time user behaviour and ensuring cybersecurity in large-scale applications. Further research is needed to develop scalable, adaptive supervised learning models that can generalize across diverse grid topologies and incorporate real-time feedback from heterogeneous energy sources and consumer behaviours, ensuring resilience and efficiency in future smart grid environments.
In [73], the authors present a Forecast-based Home Energy Management System Optimization (FHEMSO) model that applies supervised learning methods for energy optimization in smart homes. The methodology involves an ensemble learning approach integrating three time-series forecasting techniques of ARMA, ARIMA, and SARIMA, whose outputs are combined and enhanced using an Artificial Neural Network (ANN). This hybrid model improves prediction accuracy for household energy demands, enabling dynamic and efficient energy distribution. The system leverages the Kaggle Smart Home Dataset for simulation, aiming to reduce energy waste, stabilize grid operations, and promote sustainable urban energy practices. Despite its promising results, showing a performance improvement of 6% to 21% over existing models, the study primarily emphasizes forecasting accuracy while overlooking a quantitative evaluation of the energy distribution phase. This represents a notable research gap, as future studies could enhance the FHEMSO framework by incorporating advanced optimization algorithms such as Genetic Algorithms (GAs) or Particle Swarm Optimization (PSO) to refine distribution strategies. These improvements could not only elevate the model’s operational efficiency but also offer a more comprehensive solution to grid load management and renewable energy integration in smart city infrastructures. The proposed methodology thus lays a solid foundation for future intelligent energy management systems, while leaving room for deeper optimization-focused exploration.
The reviewed studies employ diverse supervised learning paradigms from MILP-trained deep neural networks for real-time ESS and EV coordination, to quantum-enhanced SVMs leveraging high-dimensional kernel spaces for short-term load forecasting. Advanced architectures such as CNNs, RNNs, GNNs, and ensemble methods integrating ARMA, ARIMA, SARIMA, and ANN further exemplify the methodological breadth applied to PV power prediction and demand scheduling. Despite high accuracy and reduced latency, limitations persist regarding generalization, long-horizon forecasting, and integration of user behaviour, uncertainty, and dynamic grid variables. Scalability, interoperability, and security in distributed environments also remain underexplored. Collectively, these approaches underscore the transformative role of supervised learning in HEMS optimization, while emphasizing the need for hybrid models, adaptive learning mechanisms, and real-world validations to bridge existing gaps in smart grid deployment scenarios.

3.2.2. Unsupervised Learning Optimization Methods for HEMS

Unsupervised learning optimization techniques for Home Energy Management Systems (HEMSs) provide a robust alternative to supervised methods by obviating the necessity for labelled training data. These methods independently discern patterns, clusters, and anomalies in energy usage without prior categorization, facilitating adaptive and scalable energy management across multiple scenarios. This is especially beneficial in practical situations where labelled datasets are limited or dynamic.
The authors in [74] discuss the application of unsupervised learning optimization methods for HEMS, emphasizing their utility in scenarios where labelled datasets are unavailable or impractical to obtain. Unsupervised techniques such as clustering, dimensionality reduction, and anomaly detection are employed to enhance the intelligence and autonomy of HEMS. Among the most commonly applied methodologies are k-means clustering, hierarchical clustering, and self-organizing maps (SOMs), which are used to segment consumer load profiles, identify user behaviour patterns, and support dynamic demand-side management. These methods enable the grouping of similar energy consumption patterns and help in load forecasting, appliance usage profiling, and identification of abnormal consumption behaviour. Dimensionality reduction techniques such as Principal Component Analysis (PCA) and Independent Component Analysis (ICA) are also utilized to reduce the complexity of high-dimensional energy datasets, allowing for more efficient processing and feature selection. These unsupervised approaches contribute significantly to real-time energy optimization by uncovering hidden structures and patterns in the data without prior labelling. However, a notable research gap exists in the limited deployment of these techniques within real-time control systems and their integration with external contextual variables such as occupancy, weather fluctuations, and grid status. Most existing studies emphasize static or offline data analysis, lacking mechanisms for continuous learning or adaptability to evolving consumption behaviour and external influences. Future research should focus on developing real-time unsupervised learning frameworks that can integrate heterogeneous data sources, respond to temporal dynamics, and interact with other machine learning paradigms for robust and scalable energy management solutions in smart grid-enabled homes.

3.2.3. Reinforcement Learning Optimization Methods for HEMS

The authors in [75] presented a structured survey of recent applications, first covering value-based (tabular Q-learning/SARSA and deep variants like DQN, DDQN, and the rarely used dueling-DQN), policy-based (TRPO, PPO), and actor–critic families (A2C/A3C, DDPG, TD3, SAC). The study categorized HEMS studies into five optimization domains—HVAC/water heating, EV–storage–renewables coordination, other appliance scheduling, demand response, and peer-to-peer trading—across objectives of cost, comfort, and load balancing and across residential, commercial, and academic building types. Methodologically, the survey synthesized articles from the last five years, mapped each to its application class, objective, and algorithm choice, and summarized algorithm prevalence using tables (e.g., DQN/DDQN for discrete control; PPO/TRPO for policy learning; DDPG/TD3/SAC as dominant actor–critic methods). To standardize evaluation, the authors proposed four RL performance metrics for HEMS, saturation reward (R∞), variance at saturation (σ∞), exploitation risk (Rmin), and convergence rate (C) with practical estimation guidance. A key research gap is translation to practice: only 12% of surveyed studies reported real-world deployments, with many simplifying discretizations and model-free simulations, indicating limited external validity. Future work could emphasize safety-aware, sample-efficient, and transferable RL with standardized benchmarks using the proposed metrics, tighter integration with physical constraints (e.g., MPC–RL hybrids), and multiagent coordination for grid-interaction. The study in [76] presented a foundational review of home energy management systems (HEMS) that foregrounded smart technologies and intelligent controllers; building on that base, the study present a reinforcement learning (RL) optimization pipeline in which real-time HEMS control is cast as a Markov decision process with a carefully engineered state (controllable, exogenous, and temporal variables), action space (including appliance scheduling and ES/EV active–reactive set-points), and a tri-objective reward trading off electricity cost, comfort, and power-factor deviation. Methodologically, a model-free proximal policy optimization (PPO) actor–critic is trained and evaluated against DQN, DDPG, and an idealized full-information MILP benchmark, using 15-min resolution data with stochastic appliance, PV, EV, and thermal dynamics; the PPO agent converges faster and generalizes better under uncertainty. Critically, the action sets coordinate active and reactive power of ES/EV converters to compensate household reactive demand, which materially improves grid-facing power factor. Empirically, the PPO controller reduces bills by 31.5% and lifts the average power factor from 0.44 to 0.90, approaching the MILP upper bound while retaining online feasibility. Most HEMS studies seldom optimize active and reactive power jointly in real time, and reactive-power management in residences, which remains underexplored, especially when EVs are unavailable for compensation. Future work should therefore pursue safety-aware, sample-efficient PPO variants with standardized benchmarks and robustness to EV absence to secure power-factor compliance without sacrificing consumer comfort.
The authors in reference [77] presented a two-stage reinforcement learning optimization for HEMS in which a soft actor–critic (SAC) agent first schedules plug-in EV charging/discharging under real-time pricing and uncertainty, and the resulting EV trajectory is then passed to a mixed-integer linear program (MILP) that co-schedules household appliances and a BESS to minimize the electricity bill. Methodologically, the EV problem is cast as a model-free MDP with state comprising SOC and price, a continuous action space spanning charge, discharge, and idle within power limits, and a reward that penalizes unmet departure SOC while minimizing normalized cost; arrival/departure times and initial SOC are sampled from truncated Gaussians, and γ = 0.99 emphasizes future returns. The HEMS MILP enforces net-power non-negativity, uninterruptible appliance operation, and BESS SOC dynamics while optimizing hourly schedules. Evaluated over a 24-h horizon with ComEd RTP data and a Nissan Leaf across four seasonal days, SAC achieved the highest average rewards and the lowest charging cost versus PPO, DDPG, TD3 and disorderly scheduling, albeit with longer training time. The study integrates model-free RL for EV C/D with simultaneous appliance/BESS scheduling under uncertainty, SAC has been used for EVs, but not within a full HEMS optimization, leaving scalability, user-preference modelling, battery-degradation costs, energy trading, and real-world validation underexplored.
The authors in [78] introduce EdgeHEM, an edge reinforcement learning optimization framework for home energy management (HEM) tailored to the computational and storage constraints of distributed edge devices. The methodology combines two innovative components. First, a dynamic sparse learning strategy with topology evolution is employed to reduce memory usage while maintaining network learning capability. Second, a compressed federated learning mechanism with gradient approximation is proposed to leverage cached transitions across multiple edge devices, thereby improving scalability and distributed training efficiency. These methods are validated on a hardware testbed with real-world datasets, demonstrating their ability to enhance both computational efficiency and practicality in edge-based HEM scenarios. Despite these contributions, a notable research gap remains. The proposed framework primarily addresses memory constraints and distributed learning but pays limited attention to other key aspects of home energy management, such as multi-objective optimization incorporating user comfort, grid interaction dynamics, and varying pricing mechanisms. Also, the study is constrained to controlled testbed environments, leaving open questions about its robustness in highly heterogeneous real-world smart grid ecosystems with diverse device types and communication uncertainties. Future research could therefore extend the approach to broader optimization objectives and cross-domain scalability, while also exploring integration with real-time demand response and interoperability standards.

4. Algorithmic Complexity and Optimization Challenges in HEMS

Designing effective Home Energy Management Systems (HEMSs) requires careful attention to algorithmic complexity that directly influences optimal decision-making and real-time operation. These challenges emerge from the challenges and scale of the problem, as well as the need to deliver timely solutions using limited computational resources. Among the most significant issues are problem size, planning horizon and time resolution, and the treatment of uncertainty. The problem size in HEMSs is strongly linked to the number of controllable devices, the resolution of the time intervals, and the length of the planning horizon [79]. As more devices are incorporated and the system integrates renewable energy sources and electric vehicles (EVs), the problem grows exponentially. This increase is further exacerbated by the need to model uncertainty, particularly in loads, renewable generation, and EV availability. Stochastic programming approaches often represent uncertainty through large scenario sets, which can significantly increase computational demands [79]. To address this, scenario reduction techniques have been proposed to reduce the number of scenarios while maintaining representativeness, thereby achieving a balance between solution accuracy and computational feasibility [43,80,81].
The selection of the planning horizon and time resolution also plays a critical role. Longer horizons and finer resolutions increase the number of time steps and the problem size, thus requiring greater computational effort [82]. Studies have demonstrated how varying horizons (from 1 h to 24 h) and resolutions (from one minute to one hour) can influence system performance and computational tractability [83,84,85,86,87,88,89].
Uncertainty remains a central challenge in HEMS design [90]. Deterministic models often fail to capture the variability of market prices, user loads, renewable energy outputs, and EV behaviours. Stochastic programming and robust optimization are widely used to address these challenges [91,92,93]. While stochastic methods rely on probability distributions and require scenario reduction to limit algorithmic complexity, robust optimization focuses on worst-case bounds, offering resilience without dependence on precise probability distributions [87,94]. However, both methods can be computationally intensive, particularly when real-time solutions are required.
Algorithmic complexity in HEMS stem from the increasing dimensionality of the problem, the trade-offs between planning horizon and resolution, and the effective treatment of uncertainty. Current approaches, such as scenario reduction and robust optimization have alleviated some of these issues, but achieving scalable, real-time operation remains an open research problem. Future work should prioritize hybrid methods capable of balancing solution quality, computational efficiency, and adaptability to emerging technologies such as bidirectional EVs and distributed storage. These advancements are essential for enabling reliable, cost-effective, and intelligent energy management in modern residential settings.

5. Comparative Analysis of Optimization Techniques for HEMS

The diverse optimization methodologies reviewed in Section 3, encompassing heuristic, metaheuristic, and machine learning (ML)-based techniques, reveal complementary strengths and limitations when applied to smart HEMS. Heuristic and metaheuristic methods such as Genetic Algorithms (GAs), Particle Swarm Optimization (PSO), Harmony Search Algorithm (HSA), and hybridized variants have demonstrated their ability to handle multi-objective problems involving appliance scheduling, peak-to-average ratio (PAR) reduction, and electricity cost minimization. Their ease of implementation and low computational complexity make them attractive for real-time HEMS scenarios with constrained processing resources. However, their performance is often sensitive to parameter tuning and can degrade when faced with stochastic and highly dynamic energy environments. These methods typically do not guarantee globally optimal solutions, limiting their scalability in larger multi-household settings.
Machine learning-based techniques, particularly supervised learning, deep reinforcement learning (DRL), and unsupervised clustering methods, have shown marked improvements in adaptability, prediction accuracy, and decision-making robustness. Supervised learning methods leverage historical data to optimize the scheduling of energy storage systems (ESSs), electric vehicles (EVs), and photovoltaic (PV) generation while reducing reliance on complex real-time forecasting. DRL approaches, such as Deep Q-Networks (DQN), Proximal Policy Optimization (PPO), and Soft Actor–critic (SAC), excel at learning optimal policies through continuous interaction with the environment, making them highly suited for uncertain, dynamic settings. Nevertheless, ML-based methods demand substantial datasets and computational resources for model training, and their “black box” nature raises concerns regarding interpretability and user trust.
A key observation emerging from this analysis is that hybrid frameworks combining heuristic search techniques with ML-driven forecasting and control can mitigate the drawbacks of each method while enhancing overall performance. In scenario generation using GA integrated with DRL-based real-time scheduling has been shown to accelerate convergence while preserving adaptability. Furthermore, the incorporation of these optimization strategies with demand response (DR) programs strengthens grid interactivity by shifting and shaping loads according to grid constraints and renewable energy availability. Such integration is vital for supporting higher penetrations of distributed energy resources and facilitating coordinated peer-to-peer energy trading among households. Table 2 provides a consolidated comparison of heuristic, ML-based, and hybrid optimization methods, emphasizing their relative strengths, weaknesses, and real-world applications.
This analysis underscores that the future of HEMS optimization lies in hybridized frameworks that fuse the robust search capabilities of heuristics with the learning and prediction strengths of ML-based techniques. Such frameworks, when coupled with advanced DR mechanisms, can deliver cost-effective, resilient, and user-centric energy management. Moving forward, research must address critical gaps in scalability, real-world validation, cybersecurity, and integration of user behaviour to realize next-generation, grid-interactive HEMSs capable of driving the energy transition in residential sectors.

6. Conclusions

This study has comprehensively reviewed the landscape of optimization methodologies for smart HEMS, highlighting their potential to transform residential energy management and support the broader transition toward sustainable, grid-interactive power systems. The paper reviewed conventional, metaheuristic approaches, machine learning (ML) techniques, and hybrid frameworks that combine the strengths of both paradigms. By analyzing their algorithmic structures, computational demands, and real-world applicability, we have developed a holistic understanding of the opportunities and limitations that characterize the current state of HEMS optimization. Heuristic and metaheuristic algorithms, such as Genetic Algorithms (GAs), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), and their numerous derivatives, remain popular for appliance scheduling, peak-to-average ratio (PAR) reduction, and demand response (DR) integration. Their relative simplicity, low computational cost, and suitability for small-scale and single-household settings make them a viable choice for resource-constrained environments. However, these methods are often hindered by sensitivity to parameter tuning, limited adaptability to stochastic conditions, and a tendency to converge to local optima in high-dimensional search spaces. These drawbacks are particularly concerning in the context of real-time HEMS decision-making, where system states evolve rapidly due to renewable energy fluctuations, dynamic tariffs, and unpredictable user behaviour.
Machine learning-based methods, particularly deep reinforcement learning (DRL), supervised learning, and clustering-based unsupervised approaches, offer significant advantages in adaptability, scalability, and prediction accuracy. DRL methods such as Deep Q-Networks (DQN), Proximal Policy Optimization (PPO), and Soft Actor–critic (SAC) have demonstrated the ability to learn optimal control policies from interactions with the environment, making them highly effective for dynamic energy management tasks. A key insight from this study is the growing relevance of hybrid optimization frameworks that combine heuristic search and ML-based prediction and control. By leveraging the strengths of each paradigm, these integrated approaches can address computational scalability while improving adaptability to uncertainty. Hybrid solutions also enable multi-objective optimization, balancing cost savings, user comfort, and system reliability, while effectively integrating distributed energy resources (DERs) such as EVs, energy storage systems (ESSs), and rooftop PV. The synergy between hybrid optimization and demand response programs holds particular promise for enhancing grid stability by dynamically aligning household consumption with utility-level constraints and renewable generation profiles.
However, several challenges must be addressed before the full potential of advanced HEMS optimization can be realized. Algorithmic complexity remains a significant barrier, as increasing problem dimensionality and the need for fine time resolutions often lead to prohibitive computational demands. Uncertainty in renewable energy availability, user behaviour, and market conditions continues to limit the robustness of existing methods. Furthermore, many studies rely heavily on simulations with idealized conditions and lack real-world validation, raising questions about scalability and external validity. Issues of cybersecurity, data privacy, and user acceptance are also becoming increasingly critical as HEMS evolve into interconnected components of the broader smart grid ecosystem.
Looking ahead, future research must focus on developing hybrid, adaptive, and interpretable optimization frameworks capable of learning and adapting to evolving conditions in real-time while operating within the computational constraints of embedded residential systems. The incorporation of explainable AI (XAI) techniques will be vital to increase user trust and regulatory compliance. Standardized evaluation benchmarks, incorporating realistic uncertainty modelling and hardware-in-the-loop testing, will accelerate progress by enabling more rigorous cross-study comparisons. Additionally, scaling HEMS optimization beyond individual households to community- and neighbourhood-level energy management will be essential to unlock the full benefits of distributed energy resources and peer-to-peer energy trading.
The integration of advanced optimisation techniques in HEMS represents a cornerstone of future smart grids, with the potential to significantly enhance energy efficiency, reduce operational costs, and enable higher penetrations of renewable energy. While substantial progress has been made in the development of heuristic, ML-based, and hybrid methods, overcoming algorithmic complexity, uncertainty, and real-world implementation barriers remains crucial. Addressing these challenges will lay the foundation for resilient, user-centric, and sustainable residential energy management systems capable of supporting the global transition toward a cleaner and more reliable energy future.
The original contributions of this paper are as follows:
  • A unified taxonomy of HEMS methods (2019–2024) spanning mathematical, heuristic/metaheuristic, and ML/DRL approaches, systematically mapped to decision horizons, controllability, and uncertainty modeling, with distinctions between real-time and day-ahead scheduling.
  • An algorithmic-complexity perspective synthesizing how horizon length, time resolution, and scenario growth affect tractability and what this implies for practical on-device controllers, an aspect rarely addressed in prior reviews.
  • A comparative evidence table across studies reporting cost savings, PAR reduction, comfort penalties, and runtime, enabling like-for-like benchmarking beyond prior narrative surveys.
  • Deployment considerations integrating privacy, cybersecurity, AMI/HAN constraints, and explainability, contextualized to AI-enabled HEMS, an often-overlooked dimension in the literature.

6.1. Practical Implications

The findings provide actionable insights for multiple stakeholders. For policymakers, the unified taxonomy and benchmarking framework inform regulatory guidelines and performance standards for HEMS evaluation. For utilities, the algorithmic complexity perspective highlights the feasibility of deploying optimization methods at scale while balancing flexibility and computational limits. For technology developers and the industry, the emphasis on cybersecurity, explainability, and interoperability underscores the design requirements for next-generation HEMS platforms that must operate reliably within heterogeneous smart grid environments.

6.2. Future Research Directions

Building on these insights, several research priorities are identified:
  • Adaptive Hybrid Optimization: Develop lightweight frameworks that dynamically switch between heuristic and learning-based strategies according to system state and uncertainty.
  • Explainable and Trustworthy AI: Advance explainable reinforcement learning and interpretable models to enhance transparency, trust, and regulatory compliance.
  • Standardized Benchmarks and Validation: Create open datasets and hardware-in-the-loop testbeds for reproducible, realistic cross-study comparisons.
  • Scalable Community-Level HEMS: Extend optimization beyond single households to neighbourhood-level coordination, integrating shared DERs and peer-to-peer trading.
  • Privacy-Preserving and Secure Architectures: Employ federated learning and robust cybersecurity techniques to safeguard user data while enabling efficient control.
In summary, the integration of advanced optimization techniques into HEMS is a cornerstone for future smart grids. By bridging methodological advances with deployment realities, this paper offers a structured foundation for developing scalable, secure, and user-centric energy management systems that can accelerate the global transition to cleaner and more resilient energy futures.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/en18195262/s1, Figure S1: Adapted from a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram, customized for our study.

Author Contributions

Conceptualization, A.A.A.; methodology, A.A.A.; formal analysis, A.A.A. and M.H.; data curation, A.A.A. and M.H.; writing—original draft preparation, A.A.A.; writing—review and editing, A.A.A. and M.H. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

The original contributions presented in the study are included in the article/Supplementary Materials, further inquiries can be directed to the corresponding author.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
AMIAdvanced Metering Infrastructure
BESSBattery Energy Storage System
CPPCritical Peak Pricing
DRDemand Response
DERsDistributed Energy Resources
DRLDeep Reinforcement Learning
DQNDeep Q-Networks
EVElectric Vehicle
XAIExplainable AI
HANHome Area Network
HEMSHome Energy Management System
MLMachine Learning
PARPeak-to-Average Ratio
PPOProximal Policy Optimization
ICTsInformation and Communication Technologies
RTEPReal-time Electricity Pricing
RTPReal-time Pricing
SACSoft Actor–critic
TOUPTime-of-Use Pricing
NILMNon-Intrusive Load Monitoring

References

  1. Molla, T.; Khan, B.; Singh, P. A comprehensive analysis of smart home energy management system optimization techniques. J. Auton. Intell. 2018, 1, 15–21. [Google Scholar] [CrossRef]
  2. Davarzani, S.; Pisica, I.; Taylor, G.A.; Munisami, K.J. Residential demand response strategies and applications in active distribution network management. Renew. Sustain. Energy Rev. 2021, 138, 110567. [Google Scholar] [CrossRef]
  3. Khan, B.; Getachew, H.; Alhelou, H.H. Components of the smart-grid system. In Solving Urban Infrastructure Problems Using Smart City Technologies; Vacca, J.R., Ed.; Elsevier: Amsterdam Netherlands, 2021; pp. 385–397. [Google Scholar]
  4. Boopathy, P.; Liyanage, M.; Deepa, N.; Velavali, M.; Reddy, S.; Maddikunta, P.K.R.; Khare, N.; Gadekallu, T.R.; Hwang, W.J.; Pham, Q.V. Deep learning for intelligent demand response and smart grids: A comprehensive survey. Comput. Sci. Rev. 2024, 51, 100617. [Google Scholar] [CrossRef]
  5. Salam, I.U.; Yousif, M.; Numan, M.; Billah, M. Addressing the challenge of climate change: The role of microgrids in fostering a sustainable future—A comprehensive review. Renew. Energy Focus 2024, 48, 100538. [Google Scholar] [CrossRef]
  6. Saeed, M.H.; Fangzong, W.; Kalwar, B.A.; Iqbal, S. A review on microgrids’ challenges & perspectives. IEEE Access 2021, 9, 166502–166517. [Google Scholar] [CrossRef]
  7. Adebiyi, A.A.; Moloi, K. Renewable energy source utilization progress in South Africa: A review. Energies 2024, 17, 3487. [Google Scholar] [CrossRef]
  8. Mohassel, R.R.; Fung, A.; Mohammadi, F.; Raahemifar, K. A survey on advanced metering infrastructure. Int. J. Electr. Power Energy Syst. 2014, 63, 473–484. [Google Scholar] [CrossRef]
  9. Wali, M.; Channi, H.K. (Eds.) Smart Meter Infrastructure for Distributed Renewable Power. In AI Approaches to Smart and Sustainable Power Systems; IGI Global: Hershey, PA, USA, 2024; Volume 1, pp. 81–99. [Google Scholar]
  10. Tepe, İ.F.; Irmak, E. Optimizing real-time demand response in smart homes through fuzzy-based energy management and control system. Electr. Eng. 2024, 107, 2121–2145. [Google Scholar] [CrossRef]
  11. Palensky, P.; Dietrich, D. Demand side management: Demand response, intelligent energy systems, and smart loads. IEEE Trans. Ind. Inform. 2011, 7, 381–388. [Google Scholar] [CrossRef]
  12. Sarker, E.; Halder, P.; Seyedmahmoudian, M.; Jamei, E.; Horan, B.; Mekhilef, S.; Stojcevski, A. Progress on the demand side management in smart grid and optimization approaches. Int. J. Energy Res. 2021, 45, 36–64. [Google Scholar] [CrossRef]
  13. Mishra, A.; Singh, S. Demand-side flexibility management: State of the art. In Renewable Energy Integration in Utility Grids; Academic Press: Cambridge, MA, USA, 2025; pp. 321–333. [Google Scholar]
  14. Nebey, A.H. Recent advancement in demand side energy management system for optimal energy utilization. Energy Rep. 2024, 11, 5422–5435. [Google Scholar] [CrossRef]
  15. Khaleel, M.; Yusupov, Z.; Alfalh, B.; Guneser, M.T.; Nassar, Y.; El-Khozondar, H. Impact of smart grid technologies on sustainable urban development. Int. J. Electr. Eng. Sustain. 2024, 2, 62–82. [Google Scholar]
  16. Zheng, Z.; Shafique, M.; Luo, X.; Wang, S. A systematic review towards integrative energy management of smart grids and urban energy systems. Renew. Sustain. Energy Rev. 2024, 189, 114023. [Google Scholar] [CrossRef]
  17. Binz Varghese, K. Baseline Load Estimation of Residential Customers for Incentive-Based Demand Response Programs. Master’s Thesis, Universitat Politècnica de Catalunya, Barcelona, Spain, 2024; p. 86. [Google Scholar]
  18. Stanelyte, D.; Radziukyniene, N.; Radziukynas, V. Overview of demand-response services: A review. Energies 2022, 15, 1659. [Google Scholar] [CrossRef]
  19. Wanjala, J.; Langa, H.M.; Walingo, T. Cost Minimization in a Smart Grid using a Genetic Algorithm. In Proceedings of the 2024 International Conference on Science, Engineering and Business for Driving Sustainable Development Goals (SEB4SDG), Omu-Aran, Nigeria, 2–4 April 2024; IEEE: Piscataway, NJ, USA, 2024; pp. 1–6. [Google Scholar]
  20. Jo, H.C.; Park, H.A.; Kwon, S.Y.; Cho, K.H. Home Energy Management Systems (HEMSs) with Optimal Energy Management of Home Appliances Using IoT. Energies 2024, 17, 3009. [Google Scholar] [CrossRef]
  21. Raza, A.; Jingzhao, L.; Ghadi, Y.; Adnan, M.; Ali, M. Smart home energy management systems: Research challenges and survey. Alex. Eng. J. 2024, 92, 117–170. [Google Scholar] [CrossRef]
  22. Youssef, H.; Kamel, S.; Hassan, M.H. Smart home energy management and power trading optimization using an enhanced manta ray foraging optimization. Sci. Rep. 2023, 13, 22163. [Google Scholar] [CrossRef]
  23. Han, B.; Zahraoui, Y.; Mubin, M.; Mekhilef, S.; Seyedmahmoudian, M.; Stojcevski, A. Home energy management systems: A review of the concept, architecture, and scheduling strategies. IEEE Access 2023, 11, 19999–20025. [Google Scholar] [CrossRef]
  24. Balakrishnan, R.; Geetha, V. Review on home energy management system. Mater. Today Proc. 2021, 47, 144–150. [Google Scholar] [CrossRef]
  25. Kuzlu, M.; Pipattanasomporn, M.; Rahman, S. Communication network requirements for major smart grid applications in HAN, NAN and WAN. Comput. Netw. 2024, 67, 74–88. [Google Scholar] [CrossRef]
  26. Li, Y.; Qin, D.; Poor, H.V.; Wang, Y. Introducing edge intelligence to smart meters via federated split learning. Nat. Commun. 2024, 15, 1–14. [Google Scholar] [CrossRef]
  27. Abdullah, A.A.; El-Den, B.M.; Abo-Al-Ez, K.M.; Hassan, T.M. Security management for an advanced metering infrastructure (AMI) system of smart electrical grids. Appl. Sci. 2023, 13, 8990. [Google Scholar] [CrossRef]
  28. Said, D. A survey on information communication technologies in modern demand-side management for smart grids: Challenges, solutions, and opportunities. IEEE Eng. Manag. Rev. 2022, 51, 76–107. [Google Scholar] [CrossRef]
  29. Panda, S.; Mohanty, S.; Rout, P.K.; Sahu, B.K.; Bajaj, M.; Zawbaa, H.M.; Kamel, S. Residential Demand Side Management model, optimization and future perspective: A review. Energy Rep. 2022, 8, 3727–3766. [Google Scholar] [CrossRef]
  30. Mahapatra, B.; Nayyar, A. Home energy management system (HEMS): Concept, architecture, infrastructure, challenges and energy management schemes. Energy Syst. 2022, 13, 643–669. [Google Scholar] [CrossRef]
  31. Missaoui, R.; Joumaa, H.; Ploix, S.; Bacha, S. Managing energy smart homes according to energy prices: Analysis of a building energy management system. Energy Build. 2014, 71, 155–167. [Google Scholar] [CrossRef]
  32. Al-Ali, A.R.; El-Hag, A.; Bahadiri, M.; Harbaji, M.; El Haj, Y.A. Smart home renewable energy management system. Energy Procedia 2011, 12, 120–126. [Google Scholar] [CrossRef]
  33. Honarmand, M.E.; Hosseinnezhad, V.; Hayes, B.; Shafie-Khah, M.; Siano, P. An overview of demand response: From its origins to the smart energy community. IEEE Access 2021, 9, 96851–96876. [Google Scholar] [CrossRef]
  34. Mohammad, A.; Zuhaib, M.; Ashraf, I. An optimal home energy management system with integration of renewable energy and energy storage with home to grid capability. Int. J. Energy Res. 2022, 46, 8352–8366. [Google Scholar] [CrossRef]
  35. Ali, A.O.; Elmarghany, M.R.; Abdelsalam, M.M.; Sabry, M.N.; Hamed, A.M. Closed-loop home energy management system with renewable energy sources in a smart grid: A comprehensive review. J. Energy Storage 2022, 50, 104609. [Google Scholar] [CrossRef]
  36. Espe, E.; Potdar, V.; Chang, E. Prosumer communities and relationships in smart grids: A literature review, evolution and future directions. Energies 2018, 11, 2528. [Google Scholar] [CrossRef]
  37. Shivam, K.; Tzou, J.C.; Wu, S.C. A multi-objective predictive energy management strategy for residential grid-connected PV-battery hybrid systems based on machine learning technique. Energy Convers. Manag. 2021, 237, 114103. [Google Scholar] [CrossRef]
  38. Mohammad, A.; Zuhaib, M.; Ashraf, I.; Alsultan, M.; Ahmad, S.; Sarwar, A.; Abdollahian, M. Integration of electric vehicles and energy storage system in home energy management system with home to grid capability. Energies 2021, 14, 8557. [Google Scholar] [CrossRef]
  39. Umoh, V.; Adebiyi, A.; Moloi, K. Hosting Capacity Assessment of South African Residential Low-Voltage Networks for Electric Vehicle Charging. Engineering 2023, 4, 1965–1980. [Google Scholar] [CrossRef]
  40. Umoh, V.B.; Adebivi, A.A.; Moloi, K. Electric Vehicles Hosting Capacity Assessment for a South African Low Voltage Distribution Network. In Proceedings of the IEEE 2023 3rd International Conference on Electronic and Electrical Engineering and Intelligent System (ICE3IS), Yogyakarta, Indonesia, 9–10 August 2023; pp. 53–57. [Google Scholar]
  41. Abrahamsen, F.E.; Ai, Y.; Cheffena, M. Communication technologies for smart grid: A comprehensive survey. Sensors 2021, 21, 8087. [Google Scholar] [CrossRef] [PubMed]
  42. Amer, A.; Shaban, K.; Gaouda, A.; Massoud, A. Home energy management system embedded with a multi-objective demand response optimization model to benefit customers and operators. Energies 2021, 14, 257. [Google Scholar] [CrossRef]
  43. Gomes, I.; Bot, K.; Ruano, M.G.; Ruano, A. Recent techniques used in home energy management systems: A review. Energies 2022, 15, 2866. [Google Scholar] [CrossRef]
  44. Jha, A.V.; Appasani, B.; Gupta, D.K.; Ramavath, S.; Khan, M.S. Machine Learning and Deep Learning Approaches for Energy Management in Smart Grid 3.0. In Smart Grid 3.0: Computational and Communication Technologies; Springer: Berlin/Heidelberg, Germany, 2023; pp. 121–151. [Google Scholar] [CrossRef]
  45. Manzoor, A.; Javaid, N.; Ullah, I.; Abdul, W.; Almogren, A.; Alamri, A. An intelligent hybrid heuristic scheme for smart metering-based demand side management in smart homes. Energies 2017, 10, 1258. [Google Scholar] [CrossRef]
  46. Ruano, A.; Hernandez, A.; Ureña, J.; Ruano, M.; Garcia, J. NILM techniques for intelligent home energy management and ambient assisted living: A review. Energies 2019, 12, 2203. [Google Scholar] [CrossRef]
  47. Sun, W.; Tang, M.; Zhang, L.; Huo, Z.; Shu, L. A survey of using swarm intelligence algorithms in IoT. Sensors 2020, 20, 1420. [Google Scholar] [CrossRef]
  48. Hussain, H.M.; Nardelli, P.H.J. A Heuristic-based Home Energy Management System for Demand Response. In Proceedings of the 2020 IEEE Conference on Industrial Cyberphysical Systems (ICPS), Tampere, Finland, 10–12 June 2020; pp. 285–290. [Google Scholar]
  49. Rehman, A.U.; Hafeez, G.; Albogamy, F.R.; Wadud, Z.; Ali, F.; Khan, I.; Rukh, G.; Khan, S. An efficient energy management in smart grid considering demand response program and renewable energy sources. IEEE Access 2021, 9, 148821–148844. [Google Scholar] [CrossRef]
  50. Tostado-Véliz, M.; Arévalo, P.; Kamel, S.; Zawbaa, H.M.; Jurado, F. Home energy management system considering effective demand response strategies and uncertainties. Energy Rep. 2022, 8, 5256–5271. [Google Scholar] [CrossRef]
  51. Mahmood, Z.; Cheng, B.; Butt, N.A.; Rehman, G.U.; Zubair, M.; Badshah, A.; Aslam, M. Efficient scheduling of home energy management controller (HEMC) using heuristic optimization techniques. Sustainability 2023, 15, 1378. [Google Scholar] [CrossRef]
  52. Youssef, H.; Kamel, S.; Hassan, M.H. Meta-heuristic-based home energy management system for optimizing smart appliance scheduling and electricity cost reduction in residential complexes. Neural Comput. Appl. 2024, 36, 23077–23102. [Google Scholar] [CrossRef]
  53. Priolkar, J.; Sreeraj, E.S. Optimal scheduling and demand response implementation for home energy management. Int. J. Electr. Comput. Eng. 2024, 14, 1352–1368. [Google Scholar] [CrossRef]
  54. Rezaee Jordehi, A. Enhanced leader particle swarm optimisation (ELPSO): A new algorithm for optimal scheduling of home appliances in demand response programs. Artif. Intell. Rev. 2020, 53, 2043–2073. [Google Scholar] [CrossRef]
  55. Huy, T.H.B.; Dinh, H.T.; Kim, D. Multi-objective framework for a home energy management system with the integration of solar energy and an electric vehicle using an augmented ε-constraint method and lexicographic optimization. Sustain. Cities Soc. 2023, 88, 104289. [Google Scholar] [CrossRef]
  56. Youssef, H.; Kamel, S.; Hassan, M.H. Smart home energy management with long-term memory artificial rabbit algorithm for cost and load optimization. Clust. Comput. 2025, 28, 1–28. [Google Scholar] [CrossRef]
  57. Keerthana, K.; Snekha, S.; Nandan, G.; Ramalingam, S.P.; Shanmugam, P.K.; Raglend, I.J. Minimising Energy Consumption Cost Using Ant Colony Optimization with Power Quality Considerations Under HEMS Environment. In Proceedings of the 2023 Innovations in Power and Advanced Computing Technologies (i-PACT), Kuala Lumpur, Malaysia, 8–10 December 2023; pp. 1–8. [Google Scholar]
  58. Van Groesen, S. Optimal Demand Response Management for HEMS Using Real-Time Pricing. Bachelor Dissertation, Erasmus University Rotterdam, Rotterdam, The Netherlands, 2024; p. 95. [Google Scholar]
  59. Silva, B.N.; Khan, M.; Wijesinghe, R.E.; Wijenayake, U. Meta-heuristic optimization based cost efficient demand-side management for sustainable smart communities. Energy Build. 2024, 303, 113599. [Google Scholar] [CrossRef]
  60. Sarker, I.H. Machine learning: Algorithms, real-world applications and research directions. SN Comput. Sci. 2021, 2, 160. [Google Scholar] [CrossRef]
  61. Ali, S.; Abuhmed, T.; El-Sappagh, S.; Muhammad, K.; Alonso-Moral, J.M.; Confalonieri, R.; Guidotti, R.; Del Ser, J.; Díaz-Rodríguez, N.; Herrera, F. Explainable Artificial Intelligence (XAI): What we know and what is left to attain Trustworthy Artificial Intelligence. Inf. Fusion 2023, 99, 101805. [Google Scholar] [CrossRef]
  62. Nutakki, M.; Mandava, S. Review on optimization techniques and role of Artificial Intelligence in home energy management systems. Eng. Appl. Artif. Intell. 2023, 119, 100918. [Google Scholar] [CrossRef]
  63. Park, B.; Kwon, S.H.; Oh, B. Standby Power Reduction of Home Appliance by the i-HEMS System Using Supervised Learning Techniques. Energies 2024, 17, 2404. [Google Scholar] [CrossRef]
  64. Huy, T.H.B.; Dinh, H.T.; Vo, D.N.; Kim, D. Real-time energy scheduling for home energy management systems with an energy storage system and electric vehicle based on a supervised-learning-based strategy. Energy Convers. Manag. 2023, 292, 117340. [Google Scholar] [CrossRef]
  65. Dinh, H.T.; Lee, K.H.; Kim, D. Supervised-learning-based hour-ahead demand response for a behaviour-based home energy management system approximating MILP optimization. Appl. Energy 2022, 321, 119382. [Google Scholar] [CrossRef]
  66. Kermansaravi, A.; Refaat, S.S.; Trabelsi, M.; Vahedi, H. AI-based energy management strategies for electric vehicles: Challenges and future directions. Energy Rep. 2025, 13, 5535–5550. [Google Scholar] [CrossRef]
  67. Powroźnik, P.; Szcześniak, P. Predictive Analytics for Energy Efficiency: Leveraging Machine Learning to Optimize Household Energy Consumption. Energies 2024, 17, 5866. [Google Scholar] [CrossRef]
  68. Amer, A.A.; Shaban, K.; Massoud, A.M. DRL-HEMS: Deep reinforcement learning agent for demand response in home energy management systems considering customers and operators perspectives. IEEE Trans. Smart Grid 2022, 14, 239–250. [Google Scholar] [CrossRef]
  69. Rokonuzzaman, M.; Rahman, S.; Hannan, M.A.; Mishu, M.K.; Tan, W.S.; Rahman, K.S.; Pasupuleti, J.; Amin, N. Levenberg-Marquardt algorithm-based solar PV energy integrated internet of home energy management system. Appl. Energy 2025, 378, 124407. [Google Scholar] [CrossRef]
  70. Nutakki, M.; Koduru, S.; Mandava, S. Quantum support vector machine for forecasting house energy consumption: A comparative study with deep learning models. J. Cloud Comput. 2024, 13, 105. [Google Scholar] [CrossRef]
  71. Yu, J.; Li, X.; Yang, L.; Li, L.; Huang, Z.; Shen, K.; Yang, X.; Yang, X.; Xu, Z.; Zhang, D.; et al. Deep learning models for PV power forecasting. Energies 2024, 17, 3973. [Google Scholar] [CrossRef]
  72. Lee, S.; Seon, J.; Hwang, B.; Kim, S.; Sun, Y.; Kim, J. Recent trends and issues of energy management systems using machine learning. Energies 2024, 17, 624. [Google Scholar] [CrossRef]
  73. Reddy, K.H.K.; Behera, R.K.; Singh, R.K.B. An Ensemble Learning based Energy Forecasting Model: A Sustainable Home Energy Management System for Smart City. Procedia Comput. Sci. 2025, 258, 1316–1325. [Google Scholar] [CrossRef]
  74. Alfaverh, K.; Fawaier, M.; Szamel, L. Optimizing home energy flows and battery management with supervised and unsupervised learning in renewable systems. Electronics 2025, 14, 1166. [Google Scholar] [CrossRef]
  75. Al-Ani, O.; Das, S. Reinforcement learning: Theory and applications in HEMS. Energies 2022, 15, 6392. [Google Scholar] [CrossRef]
  76. Aldahmashi, J.; Ma, X. Real-time energy management in smart homes through deep reinforcement learning. IEEE Access 2024, 12, 43155–43172. [Google Scholar] [CrossRef]
  77. Mansour, S.H.; Azzam, S.M.; Hasanien, H.M.; Tostado-Véliz, M.; Alkuhayli, A.; Jurado, F. Deep reinforcement learning-based plug-in electric vehicle charging/discharging scheduling in a home energy management system. Energy 2025, 316, 134420. [Google Scholar] [CrossRef]
  78. Li, Y.; Chen, X.; Wang, Y. EdgeHEM: Sparse Federated Reinforcement Learning for Home Energy Management at the Edge. IEEE Trans.Smart Grid 2025, 92, 1–13. [Google Scholar] [CrossRef]
  79. Birge, J.R.; Louveaux, F. Introduction to Stochastic Programming; Springer Science & Business Media: New York, NY, USA, 2011; ISBN 1461402379. [Google Scholar]
  80. Duan, J.; Li, Y.; Li, Y.; Yang, P.; Zeng, Z. Statistical Feasibility Robust Optimization with Polyhedron Uncertainty Set for Hydrogen-Data Center Microgrid Operations. IEEE Trans. Autom. Sci. Eng. 2024, 22, 9884–9902. [Google Scholar] [CrossRef]
  81. Germscheid, S.H.; Mitsos, A.; Dahmen, M. Scenario Reduction Methods for Risk-Averse Demand Response Scheduling Under Price Uncertainty. Comput. Aided Chem. Eng. 2024, 53, 3223–3228. [Google Scholar] [CrossRef]
  82. Makhadmeh, S.N.; Al-Betar, M.A.; Alyasseri, Z.A.A.; Abasi, A.K.; Khader, A.T.; Damaševičius, R.; Mohammed, M.A.; Abdulkareem, K.H. Smart Home Battery for the Multi-Objective Power Scheduling Problem in a Smart Home Using Grey Wolf Optimizer. Electronics 2021, 10, 447. [Google Scholar] [CrossRef]
  83. Hussain, A.; Musilek, P. Energy management of buildings with energy storage and solar photovoltaic: A diversity in experience approach for deep reinforcement learning agents. Energy AI 2024, 15, 100313. [Google Scholar] [CrossRef]
  84. Ahrarinouri, M.; Rastegar, M.; Seifi, A.R. Multiagent Reinforcement Learning for Energy Management in Residential Buildings. IEEE Trans. Ind. Inform. 2021, 17, 659–666. [Google Scholar] [CrossRef]
  85. Diyan, M.; Silva, B.N.; Han, K.A. Multi-Objective Approach for Optimal Energy Management in Smart Home Using the Reinforcement Learning. Sensors 2020, 20, 3450. [Google Scholar] [CrossRef]
  86. Gomes, I.L.R.; Ruano, M.G.; Ruano, A.E. MILP-based model predictive control for home energy management systems: A real case study in Algarve, Portugal. Energy Build. 2023, 281, 112774. [Google Scholar] [CrossRef]
  87. Ikram, A.I.; Ullah, A.; Datta, D.; Islam, A.; Ahmed, T. Optimizing energy consumption in smart homes: Load scheduling approaches. IET Power Electron. 2024, 17, 2656–2668. [Google Scholar] [CrossRef]
  88. Yang, Y.; Wang, S. Resilient residential energy management with vehicle-to-home and photovoltaic uncertainty. Int. J. Electr. Power Energy Syst. 2021, 132, 107206. [Google Scholar] [CrossRef]
  89. Akbari, S.; Martins, J.; Camarinha-Matos, L.M.; Petrone, G. A two-stage probabilistic flexibility management model for aggregated residential buildings. Energy Build. 2025, 332, 115404. [Google Scholar] [CrossRef]
  90. Avordeh, T.K.; Gyamfi, S. Optimizing residential demand response in Ghana through iterative techniques and home appliance trend analysis. Heliyon 2024, 10, e25807. [Google Scholar] [CrossRef]
  91. Tesfatsion, L. Locational marginal pricing: A fundamental reconsideration. IEEE Open Access J. Power Energy 2024, 11, 104–116. [Google Scholar] [CrossRef]
  92. Li, C.; Grossmann, I.E. A review of stochastic programming methods for optimization of process systems under uncertainty. Front. Chem. Eng. 2021, 2, 622241. [Google Scholar] [CrossRef]
  93. Fariman, S.K.; Danesh, K.; Pourtalebiyan, M.; Fakhri, Z.; Motallebi, A.; Fozooni, A. A robust optimization model for multi-objective blood supply chain network considering scenario analysis under uncertainty: A multi-objective approach. Sci. Rep. 2024, 14, 9452. [Google Scholar] [CrossRef]
  94. Wang, K.; Aydemir, M.; Jacquillat, A. Scenario-based robust optimization for two-stage decision making under binary uncertainty. Inf. J. Optim. 2024, 6, 84–117. [Google Scholar] [CrossRef]
Figure 1. HEMS configuration.
Figure 1. HEMS configuration.
Energies 18 05262 g001
Figure 2. ML method classification.
Figure 2. ML method classification.
Energies 18 05262 g002
Table 1. Benchmark Summary of Anchor Studies.
Table 1. Benchmark Summary of Anchor Studies.
Ref.ObjectiveDataset/Case StudyScenario Uncertainty ConsideredNormalized Outcomes
Sarker et al. (2021) [12]Review of DSM in smart grids; optimization approaches for DRSurvey of global DSM/DR worksStochastic renewable generation, demand variabilityCost ↓ (reported), PAR ↓ (✓), Comfort ↑ (discussed qualitatively), Runtime—not benchmarked
Mishra & Singh (2025) [13]Demand-side flexibility management, state-of-the-artConceptual/aggregated studiesLoad forecast uncertainty, DR participationCost ↓ (review evidence), PAR ↓ (not quantified), Comfort trade-offs noted, Runtime—not reported
Nebey (2024) [14]Demand-side EMS for optimal utilizationResidential & microgrid case studiesUncertain load profilesCost ↓ (✓), PAR ↓ (✓), Comfort ↑ (fewer violations), Runtime moderate
Zheng et al. (2024) [16]Integrative smart grid + urban energy managementMulti-system reviews (urban grids)High variability of demand + urban supplyCost ↓ (conceptual evidence), PAR—not focus, Comfort N/A, Runtime—not reported
Binz Varghese (2024) [17]Baseline load estimation for DRResidential customer data (Barcelona)Baseline uncertainty (consumer profiles)Cost ↓ (via incentives), PAR ↓ (indirect), Comfort—trade-offs discussed, Runtime—not benchmarked
Stanelyte et al. (2022) [18]Review of DR servicesUtility DR programs (Europe, US)Customer participation uncertaintyCost ↓ (✓), PAR ↓ (conceptual), Comfort—N/A, Runtime—N/A
Wanjala et al. (2024) [19]GA-based smart grid cost minimizationIEEE test systemsUncertain load and pricingCost ↓ 10–20%, PAR ↓ (✓), Comfort N/A, Runtime ↑ (due to GA iterations)
Jo et al. (2024) [20]IoT-based HEMS for appliancesResidential IoT dataset (Korea)Appliance usage uncertaintyCost ↓ 12–15%, PAR ↓ (✓), Comfort ↑ (less violations), Runtime—moderate
Table 2. Comparative Analysis of Optimization Methods for HEMSs.
Table 2. Comparative Analysis of Optimization Methods for HEMSs.
Optimization MethodStrengthsLimitationsReal-World Applications
Heuristic & Metaheuristic Low computational cost, straightforward implementation, effective for multi-objective schedulingSensitive to parameter tuning, prone to local optima, limited adaptability in dynamic or stochastic environmentsAppliance scheduling, PAR reduction, single-household DR programs
Machine Learning (Supervised, DRL, Unsupervised)High adaptability, data-driven forecasting, capable of handling uncertainty and complex energy patternsRequires large datasets and significant training time, high computational complexity, interpretability challengesESS and EV optimization, PV power forecasting, adaptive DR participation
Hybrid (Heuristic + ML)Leverages complementary strengths, improved convergence and robustness, scalable to multi-home systemsIncreased design complexity, potential for high resource requirementsCommunity-level DR, peer-to-peer energy trading, integrated RES and storage optimization
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Adebiyi, A.A.; Habyarimana, M. Systematic Review of Optimization Methodologies for Smart Home Energy Management Systems. Energies 2025, 18, 5262. https://doi.org/10.3390/en18195262

AMA Style

Adebiyi AA, Habyarimana M. Systematic Review of Optimization Methodologies for Smart Home Energy Management Systems. Energies. 2025; 18(19):5262. https://doi.org/10.3390/en18195262

Chicago/Turabian Style

Adebiyi, Abayomi A., and Mathew Habyarimana. 2025. "Systematic Review of Optimization Methodologies for Smart Home Energy Management Systems" Energies 18, no. 19: 5262. https://doi.org/10.3390/en18195262

APA Style

Adebiyi, A. A., & Habyarimana, M. (2025). Systematic Review of Optimization Methodologies for Smart Home Energy Management Systems. Energies, 18(19), 5262. https://doi.org/10.3390/en18195262

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop