Systematic Review of Advanced Optimization Techniques and Multi-Asset Integration in Home Energy Management Systems

Abstract

Home Energy Management Systems (HEMS) are increasingly positioned at the center of residential flexibility, particularly as homes integrate photovoltaics, battery storage, electric vehicles, and responsive loads. This systematic review examines recent advances in optimization and multi-asset coordination for HEMS. Searches were conducted in Scopus, Web of Science, IEEE Xplore, and ScienceDirect for studies published between 2020 and 2025; after screening and eligibility assessment, 90 studies were included. The findings indicates that deterministic optimization remains well suited to structured scheduling problems, whereas metaheuristic, hybrid, and learning-based methods are better able to address nonlinearity, uncertainty, and real-time adaptation. Across the reviewed literature, multi-asset integration generally improves cost, peak demand, self-consumption, and, in some cases, user comfort and emissions. Yet the field remains dominated by simulation-based validation. Future progress of HEMS will depend on real-world validation, interoperable system design, explainable control, and stronger alignment with user behavior, communication constraints, and regulatory frameworks.

1. Introduction

The increasing global energy demand and the urgent challenge of climate change have compelled all sectors to prioritize energy efficiency improvements in both industrial and residential contexts. The building sector is a key area, accounting for approximately 30% of global final energy consumption and a similar share of energy-related greenhouse gas emissions [1]. As urbanization continues, buildings are expected to consume more energy, which increases the need for efficient energy management systems. [2].

In response to this challenge, the concept of “smart buildings” has emerged, driven by recent breakthroughs in Internet of Things technologies, advanced sensors for data acquisition, and sophisticated analytics platforms. These innovations enable the development of intelligent environments that not only enhance occupant comfort and well-being but also achieve significant energy savings [3]. At the core of this framework are Building Energy Management Systems (BEMS) and Home Energy Management Systems (HEMS), the main difference between HEMS and BEMS is the scale of the environment they manage. While both systems are designed to monitor and optimize energy usage, HEMS is built for individual households, while BEMS is designed for commercial or industrial buildings [4].

Figure 1 illustrates the fundamental design of a HEMS architecture, featuring an optimization controller that regulates energy flow between local renewable sources (PV/wind), battery storage, and the grid. By leveraging cloud data and smart metering, the system automates load scheduling and peak shaving to minimize operational costs. It effectively balances user comfort with economic objectives by optimizing demand response efficiency and integrating smart appliances and electric vehicles as active nodes within the power grid.

Figure 1. HEMS architecture.

The integration of distributed energy resources, including rooftop photovoltaic systems and battery energy storage systems, has transformed buildings from passive consumers into active prosumers capable of generating, storing, and intelligently managing their own energy flows. This transition brings both major opportunities for energy savings and substantial operational complexity [5].

Traditional rule-based control systems are inadequate for managing such complex energy ecosystems. Consequently, artificial intelligence, combined with advanced optimization algorithms, is increasingly recognized as essential. Modern AI-enabled systems transcend simple sensing and actuation; they incorporate forecasting capabilities, self-adaptation, and continuous learning from operational data. These intelligent systems can predict occupant behavior, anticipate energy demand patterns, optimize appliance scheduling, and dynamically respond to grid signals and electricity price fluctuations [6].

This paper presents a systematic review of optimization Algorithms applied to HEMS. Our objectives are:

• To analyze and compare the major Optimization approaches in HEMS (supervised learning, deep learning, reinforcement learning) currently employed in HEMS applications.
• Evaluate Optimization Algorithms: Conduct a detailed inquiry into diverse optimization algorithms for resolving multi-objective trade-offs within HEMS and BEMS.
• Examine how individual optimization strategies are scaled to Integrated Home and Building Energy Management Systems.

Related Work

The existing literature on HEMS and integrated building energy management has evolved from foundational architectural taxonomies toward highly specialized, algorithm-centric evaluations. Reviews, such as [7], primarily emphasize the conceptual breadth of demand response frameworks and the communication infrastructures necessary for smart technology integration, offering foundational taxonomies for HEMS architectures. In contrast, more recent surveys by [8,9] adopt a more technical focus on optimization methodologies, with the latter utilizing a rigorous systematic review process to categorize heuristic, meta-heuristic, and mathematical programming approaches. While [10] provide a comprehensive survey of research challenges encompassing load forecasting and scheduling, reviews like [11] offer a narrow but high-depth examination of DRL as a pivotal tool for real-time adaptive control. Furthermore, literature extends from isolated home units toward larger management ecosystems; for instance, ref. [12] emphasize AI-driven sustainable management within property and building-scale frameworks, whereas [13,14] analyzed emerging lens of digital twins and real-time power system reliability. Similarly, while earlier works focus on static scheduling, recent studies increasingly show the need to addressing uncertainty and multi-objective trade-offs between cost and comfort [8,10]. Previous studies have examined these aspects separately, which shows the need for a more unified review.

In contrast, this review offers a detailed study, carefully evaluating diverse optimization algorithms that help address the multi-objective trade-offs inherent to HEMS, such as the balance between cost-efficiency and user comfort. These individual optimization strategies are then scaled to Integrated Home and Building Energy Management Systems, shifting the focus from isolated units to collective, grid-interactive architectures. Finally, the review links these models to real applications and their limits, identifying the empirical barriers that influence the transition from simulated models to practical deployment. This review helps fill that gap by bringing these points together.

2. Methodology

This article presents a systematic review of Advanced Optimization and Multi-Asset Integration in Home Energy Management Systems. The methodology was designed to identify relevant studies, apply clear selection criteria, and compare the findings in a consistent way. The review process included four main steps: literature search, screening, eligibility assessment, and final synthesis.

2.1. Search Strategy

A comprehensive search was executed across four major academic databases to ensure broad coverage of the relevant literature: Scopus, Web of Science, IEEE Xplore, and ScienceDirect. The search period was strictly defined as 1 January 2020, to 26 December 2025, to focus on the most recent advancements in the field.

The search employed multiple query combinations using Boolean operators to capture the intersection of key research areas:

• Core HEMS and Optimization/AI: (“home energy management” OR “HEMS” OR “residential energy” OR “smart home energy”) AND (optimization OR “machine learning” OR “deep learning” OR “reinforcement learning” OR “artificial intelligence”)
• Specific Applications: Queries focused on key BEMS functions, including load forecasting, demand response, metaheuristic algorithms (e.g., Genetic Algorithm, Particle Swarm), advanced AI techniques (e.g., Reinforcement Learning, Deep Q-learning), integration of renewable energy (e.g., Photovoltaic, Battery Storage), and electric vehicle integration (e.g., V2H, V2G).

2.2. Inclusion and Exclusion Criteria

Studies were included when they met the following conditions:

(1)

they focused on HEMS or closely related residential energy management systems;

(2)

they examined optimization methods, artificial intelligence techniques, machine learning models, deep learning models, reinforcement learning approaches, metaheuristic algorithms, or hybrid control methods;

(3)

they addressed one or more important HEMS functions, such as appliance scheduling, demand response, renewable energy integration, storage management, electric vehicle coordination, or multi-objective energy optimization; and

(4)

they provided sufficient technical detail on the proposed method, model, or evaluation results.

Studies were excluded when they were not peer-reviewed, were duplicated across databases, lacked enough technical detail, or fell outside the scope of residential energy management. Papers were also excluded when their focus did not apply to HEMS operation, optimization, or intelligent control.

2.3. Study Selection Process

Figure 2 outlines the multi-stage study selection process.

Figure 2. PRISMA Flow Diagram of the Study Selection and Quality Assessment Process.

2.4. Data Extraction

For each included study, the main information was extracted and recorded in a structured manner. The extracted data included the publication year, the main research objective, the optimization or AI method used, the HEMS components considered, the type of energy resources integrated, the decision variables or control targets, and the main performance indicators reported by the authors.

Special attention was given to studies that addressed multi-objective optimization and multi-asset coordination. These studies were examined in terms of how they managed trade-offs between electricity cost, peak-to-average ratio, user comfort, renewable energy use, storage operation, and electric vehicle charging.

2.5. Quality Assessment

A quality assessment was performed to improve the reliability of the final synthesis. Each study was examined using four main criteria: clarity of the research objective, appropriateness of the methodology, quality of the data or simulation setup, and rigor of the analysis.

2.6. Data Synthesis

The synthesis focused on identifying the strengths and limitations of each approach, the types of uncertainty addressed, the main optimization objectives, and the practical challenges reported in the literature.

2.7. Scope and Limitations of the Review Method

This review focused on recent peer-reviewed studies with sufficient technical content. As a result, some relevant works may not have been included if they were outside the selected databases, outside the defined period, lacked sufficient methodological detail, or were not retrieved in full text.

2.8. Protocol and Registration

This systematic review was not registered.

3. Optimization Approaches in HEMS

The management of electrical energy in residential settings has undergone a notable transformation, driven by the dual needs to enhance energy efficiency and integrate distributed energy resources [15]. The evolution of HEMS is fundamentally linked to advances in sophisticated optimization and AI techniques [16]. Studies often relied on classical, deterministic optimization methods, which have progressively given way to more dynamic, intelligent, and adaptive strategies capable of handling the complexities and uncertainties of modern smart home environments [3]. This section provides a structured overview of optimization approaches in HEMS as shown in Figure 3, categorizing the key techniques and discussing their application to optimizing energy consumption, cost, and user comfort.

Figure 3. Classification of optimization approaches in HEMS.

3.1. Metaheuristics

Metaheuristic algorithms have been widely adopted in HEMS for their effectiveness in solving complex, non-linear optimization problems without requiring a precise mathematical formulation of the entire system. Approaches such as GAs and PSO are commonly employed to derive near-optimal solutions for the scheduling of household appliances, the management of energy storage, and the coordination with DR initiatives [17]. For instance, previous studies have shown the efficacy of employing GAs to reduce electricity costs while preserving user comfort by optimizing the operational schedules of smart appliances, HVAC systems, and EV charging [18]. In a related context, PSO has been successfully applied to regulate renewable energy resources, refining the energy transfer among PV systems, battery storage, and the grid to reduce costs and peak demands [19]. Other significant metaheuristic methodologies include the HGSO for multi-objective cost-peak optimization, as well as algorithms such as the VOA for demand management [20]. These techniques work well in large search spaces to reconcile competing objectives, including minimizing energy costs, reducing the PAR, and maximizing occupant comfort [21].

3.2. Deterministic and Mathematical Optimization

Deterministic optimization methods, including MILP and Convex Optimization, underpin many scheduling challenges in HEMS [22]. These techniques exhibit substantial efficacy in scenarios that can be formulated as a definite objective function subject to linear constraints [23]. MILP has been widely employed to determine optimal day-ahead scheduling for domestic appliances, energy storage systems, and Distributed Energy Resources to minimize electricity expenditures under time-of-use pricing structures [24]. The strength of these mathematical programming techniques lies in their ability to guarantee the optimality of the solution within the bounds of a defined model [25]. In contrast, convex optimization provides a coherent framework for managing energy in real time, yielding effective and resilient solutions to challenges such as economic dispatch and load scheduling [3]. Despite their strengths, these methodologies can be computationally demanding. They may struggle to address the nonlinearities and uncertainties pervasive in practical home energy systems, often necessitating their integration into hybrid frameworks [26].

3.3. AI-Driven and Data-Driven Approaches

The development of smart devices, coupled with the availability of detailed energy analytics, has significantly driven the integration of AI and ML within HEMS [27]. ML frameworks, particularly those for deep learning and reinforcement learning, have made HEMS more adaptive and flexible in energy governance [28]. ANNs are frequently employed to predict energy consumption and generation from renewable energy sources, thereby providing essential data for optimization algorithms [29]. More refined frameworks employ DRL to enable autonomous decision-making in real-time settings [30]. DRL agents can learn optimal energy management strategies by interacting with the domestic environment, adapting to fluctuating energy tariffs, climatic variations, and user preferences without requiring direct programming [25].

Multi-agent systems (MAS) are now seen as an effective approach for distributed HEMS coordination because they distribute decision-making across households, devices, and grid layers rather than concentrating control in a centralized optimizer [31]. Hierarchical and supervisor-based MAS architectures enable coordinated real-time management of PV units, storage systems, EVs, and flexible appliances while preserving local autonomy and operational scalability [31,32]. Recent studies further show that learning-based MAS, particularly DDQN- and MAPPO-driven schemes, improve scheduling quality, convergence speed, and operating cost relative to conventional rule-based control [32,33]. In parallel, decentralized negotiation methods based on game theory and ADMM strengthen congestion management and peer-level coordination without requiring full central information exchange [34]. When combined with stochastic–robust optimization, MAS also enhance resilience and self-healing capability under uncertainty in renewable-rich smart distribution environments [35].

3.4. Hybrid Algorithms

Recent research has shown that no single method is best in all cases, prompting a shift toward the development of integrated strategies that combine the strengths of different optimization and artificial intelligence systems [36]. A prevalent tactic entails the conjunction of machine learning-based forecasting frameworks with optimization algorithms; for instance, employing an artificial neural network to project solar energy generation, subsequently utilizing this prognostication within a mixed-integer linear programming or genetic algorithm model to optimize energy resource scheduling [37]. An additional trend is the integration of metaheuristic algorithms with deterministic strategies, exemplified by using a genetic algorithm to explore a large solution space, followed by a linear programming solver to refine the optimal solution in a local context [20].

Table 1 provides a comparison of algorithm classes and HEMS performance, distinguishing between deterministic, heuristic, and data-driven approaches. Deep Learning and ANN algorithms, specifically kCNN-LSTM and MLP-ANN, dominate the forecasting domain, delivering superior accuracy in load and energy prediction compared to traditional regression models. Conversely, the control and scheduling layer is split between deterministic and stochastic optimization. While MILP achieves global optimality for appliance scheduling under strict technical constraints, it faces significant computational bottlenecks as system dimensionality increases. In contrast, metaheuristic approaches like LMARO and HGSO offer enhanced scalability for multi-dwelling environments, achieving cost reductions of up to 39% and peak-load shavings of approximately 55%. Furthermore, DRL variants, such as DDPG+MASAC, provide a robust model-free alternative for real-time control, though they necessitate intensive data training and stability mechanisms to reach near-optimal performance.

Table 1. Comparison of Optimization and AI Techniques for HEMS.

Study/Context	Algorithm Family & Specific Methods	HEMS-Relevant Task	Main Performance Outcomes (vs. Baselines)	Notable Strengths	Limitations/Caveats	Citations
Multi-objective DSM in smart homes	Metaheuristic + MCDM: HGSO + VIKOR ranking	Appliance scheduling under TOU and adaptive pricing	>80% and 76% cost savings (TOU, adaptive) and 55% peak reduction vs. baseline	Very high reported savings; multi-criteria decision framework; compared with literature methods. Offline search (no learning); computational burden depends on HGSO population/iterations but suitable for single-home case; flexible across multiple pricing schemes and explicit inconvenience index via VIKOR ranking	Extremely large savings may depend on optimistic baselines; scalability and online deployment not explored	[20]
Smart-home cost/peak optimization	Metaheuristics: ARO and LMARO; compared with NGO, GWO on CEC benchmarks	Multiple-knapsack HEMS scheduling to minimise cost and PAR	LMARO cuts single-home cost by 22% vs. 8% with ARO; for 10 homes, 39% vs. 24%; PAR reduced up to 25–30%	Time-window comfort; keeps demand under threshold to limit peaks and grid stress. Offline heuristic search (MATLAB); no explicit training phase; computational load scales with number of rabbits and devices but manageable for the tested sizes; flexible for different household counts and price scenarios via knapsack formulation	No real-time hardware tests; comfort only indirectly modeled; communication and cyber issues ignored	[21]
MILP for smart-home DR peaks	MILP solved with CPLEX	Household appliance scheduling to improve load factor and avoid new peaks	Improves load factor and lowers costs for both customers and utility in homogeneous/heterogeneous groups	Clear trade-off analysis for LF vs. cost; rigorous handling of appliance/PEV constraints and income groups. Deterministic solver (no learning); guarantees global optimum for modeled problem; supports customer hourly preference windows and detailed appliance/PEV constraints; tractable for the two studied communities though MILP complexity grows with scale	Uncertainty only in start times via Monte Carlo; not designed for high-speed real-time control	[23]
Rolling-horizon neighbourhood EMS	MILP with rolling horizon	Real-time community EMS with PV/ESS fairness	Achieves cost reduction, load equalisation, emissions reduction while ensuring fair PV/ESS usage	Incorporates fairness metrics for PV/ESS usage among households; Repeated MILP over short horizons limits computational load and enables near-real-time control for studied neighborhood; flexible to updated PV forecasts and changing participation via rolling horizon	Results from one case study; broader scalability and user-behavior effects not fully explored	[24]
ANN-based smart home management	Supervised ML: multi-layer perceptron ANN	Smart-home demand prediction to support automated control	Training accuracy 99.96%, validation 99.99% on apartment dataset; better than several compared models	Very high predictive accuracy; demonstrates ANN advantages on larger datasets Trained on Google Colab T4 GPU; each epoch ~645 s for 12,595 iterations and 15 epochs total; once trained, inference is lightweight for deployment; flexible to different apartments given retraining on local data	Focus on single-building dataset; generalization to diverse buildings and climates not yet proven	[27]
DRL EMS in microgrid	Deep RL: DQN, SARSA, Double DQN, REINFORCE, actor–critic, PPO/PPO++, A3C/A3C++; A3C++ adds replay + semi-deterministic training	Optimal scheduling of generation, storage and flexible demand	A3C++ reaches ~50% of GA optimum profit and 24% higher profit than retailer baseline	Systematic comparison of 7 DRL algorithms; shows big convergence differences; near-optimal with no explicit model, high-dimensional Markov decision process with experience replay and semi-deterministic training	Data-hungry, tuning-sensitive; stability gains rely on replay buffer size and learning-rate settings	[29]
Two-timescale Volt-VAR control	DRL: DDPG for slow devices + MASAC for smart inverters	Distribution-level Volt-VAR control under PV uncertainty	Two-timescale DRL achieves near-optimal voltage regulation and reduced operating cost vs. baseline VVCs	Model-free, handles fast inverter dynamics; hierarchical agent design. Training uses ~20 weeks of historical data and converges in ~2 h on CPU+GPU; once trained, per-hour decision time is lower than GUROBI-based optimization; model-free, handles complex network without full topology/parameter knowledge; flexible to changing PV patterns through retraining	Needs large historical/simulated data; hyperparameter tuning (e.g., MASAC temperature, DDPG noise) non-trivial	[30]
Residential energy prediction comparison	ANN vs. SVM and decision tree on 300 buildings	Annual building energy use prediction	ANN outperforms SVM and DT in R2 and error metrics; DT worst	Shows ANN scalability to big datasets and superiority over SVM in this context. Once trained, prediction is fast and suitable for online EMS; flexibility to incorporate many features and long histories; computational load dominated by offline training	Focus on offline prediction; not integrated with dynamic HEMS control	[38]
EA review for engineering	Evolutionary algorithms: GA, GP, DE, ES, EP and many variants	Generic optimization of engineering/control problems	GA dominates in publications and patents; DE and GP also widely used	Very broad algorithmic toolbox; many specialised GA/DE variants (e.g., linkage-learning GA, memory-based DE)	Evidence heavily skewed to GA; most results on test functions, not building/HEMS problems	[39]
Building energy forecast (IIT Bombay)	Hybrid deep learning: k-means + CNN + LSTM (kCNN-LSTM)	Short- to mid-term building energy demand forecasting	kCNN-LSTM outperforms k-means variants of state-of-the-art models on real building data	Offline training of CNN–LSTM (moderate–high compute) but fast online inference; explicitly captures spatio-temporal dependencies via clustering + convolution + recurrence, improving flexibility for complex buildings	Single building case; no closed-loop HEMS control; computationally heavy	[40]
eDemand forecast	LSTM + (ISCOA-LSTM)	Forecasting for energy management (short/mid/long-term)	Lower MAE, MAPE, MSE, RMSE, Theil stats than several benchmark models	Evolutionary optimizer tunes LSTM hyperparameters offline (computationally heavier than plain training but yields robust models); once tuned, LSTM forecasts are efficient; flexible across multiple forecast horizons on real building data	Extra optimization layer adds complexity; focused on one academic building	[41]

Overall, the reviewed studies show that no single optimization approach is universally optimal for HEMS. Mathematical programming methods such as MILP remain highly effective for small- to medium-scale problems requiring strict constraint satisfaction and exact scheduling. Metaheuristic techniques provide greater flexibility and scalability in complex multi-objective problems, although they generally yield near-optimal rather than guaranteed optimal solutions. AI-based methods, particularly deep reinforcement learning, are increasingly valuable for real-time and adaptive control, but their effectiveness depends on data availability, training effort, and tuning stability. Consequently, recent HEMS research increasingly favors hybrid frameworks that combine forecasting, optimization, and adaptive control to balance accuracy, computational tractability, and operational flexibility.

4. Multi-Objective Optimization and Trade-Offs in Smart Homes/Buildings

Energy management problems in smart homes and buildings are inherently multi-objective. Typical BEMS and smart-building models balance more than two competing goals.

Figure 4 illustrates most common goals in HEMS such energy cost, peak load and PAR, greenhouse gas emissions, occupant comfort/discomfort, self-sufficiency, and asset health.

Figure 4. Multi Objective Optimization Goals In Hems.

4.1. Comparative Analysis of Multi-Objective Trade-Offs

The studies consistently demonstrates that the nature and magnitude of trade-offs are highly dependent on the objectives included in the optimization model and the context of the study (e.g., residential vs. industrial, operational vs. design). A direct comparison of key studies highlights these differences, as summarized in Table 2.

Table 2. Examples Of Quantitative Trade-Offs Across Multi-Objective Energy Management Studies.

Study Context	Scale	Objectives	Trade-Offs	Citations
HEMS with PV+BESS (China; historical weather-based PV)	Single home with PV and battery; three operational strategies	Tri-objective cost, PAR, comfort, new strategy + tri-objective model	Approximately 39.8% bill reduction, 50.4% peak reduction, and a 1.6-fold increase in comfort index compared with the base case	[42]
Residential home (Morocco-like TOU/CPP tariffs; no local generation)	Single home without DER; only appliance time-shifting	Cost, PAR, comfort; Pareto + lexicographic	Under TOU: cost reduction of 28% and PAR reduction of 49%; under CPP: cost reduction of 48.9% and PAR reduction of 47.9%, with acceptable waiting time	[43]
Smart cities DSM in Portugal (residential, commercial, industrial; varying RES mix)	City-level, multi-building; grid-only loads with sector-specific appliances	Cost, CO2, discomfort; NSGA-II	In the industrial case, the multi-objective approach preserves 46–90% of the cost savings while providing 30–195% better emissions performance and up to 447% better PAR reduction compared with cost-only optimization	[44]
Single smart home (unspecified region; PV+WT+EES; bi-directional trading)	Small-scale, one dwelling with PV, wind, storage; various RER sizes, charge rates, selling tariffs	Cost & PAR; IBBO vs. GWO, WOA	Both objectives kept near optimum; IBBO more robust and lower average objective than GWO/WOA	[45]
Single smart home (Malaysia-like solar profile; PV, storage, EMS)	One EMS-controlled home with PV and battery; no EVs	Cost per satisfaction vs. satisfaction (%S); MGWASA vs. NSGA-II, MOBPSO, MOABC, MOEA	MGWASA achieved 97% satisfaction with competitive or lower cost per satisfaction unit than the benchmark methods	[46]
Collective residential building (Europe; PV, EVs, BESS)	Multi-apartment smart building; shared PV and BESS, multiple EVs	Cost and peak; Pascoletti–Serafini scalarization	Approximately 35.6% cost reduction and 45.5% peak reduction, with different Pareto points reflecting operator and user priorities	[47]
Smart home with PV, ESS, EVs (Iran) under flexibility limits	Single, DER- home; PV+storage+EV, explicit flexibility budget	Energy payment, user satisfaction, self-sufficiency; lexicographic	Re-ordering priorities changes energy payment, satisfaction, and self-sufficiency by 27.4%, 100%, and 56.6%, respectively	[48]
Generic residential buildings (HEMS-DR, region not fixed; with PV, EVs, BESS, full appliance mix)	Single LV feeder with many homes; rich DER portfolio (PV, BESS, EVs, flexible and fixed loads)	Cost, peak, transformer LoL; multi-objective	31% cost reduction, 18% peak reduction, and 28% LoL improvement	[49]

Table 2 summarizes the quantitative outcomes of multi-objective optimization across diverse smart home and building contexts. It serves to illustrate two primary findings: first, the magnitude of performance gains achievable when optimizing core objectives (cost, PAR, comfort); and second, the quantifiable nature of the trade-offs that emerge when expanding the objective set to include environmental impact, self-sufficiency, and asset health. The data confirms that multi-objective modeling is essential for achieving balanced solutions, as improvements in one dimension are frequently negotiated against another. These trade-offs presented in the table will be further analyzed in the subsequent section.

4.1.1. Cost, Comfort, and PAR Trade-Offs

Studies focusing on the core objectives of cost, comfort, and PAR show that significant economic and grid-stability improvements are achievable, provided comfort is explicitly constrained. For instance, Ref. [43] using a hybrid PSO–branch-and-bound formulation, reported substantial reductions in electricity bills (28–48.9%) and PAR (47.9–49.3%) under TOU and CPP tariffs, while ensuring user waiting-time constraints were met. In contrast, a study by [47] on collective residential buildings, which utilized scalarization and multi-objective mixed-binary linear programming, achieved a 35.56% cost reduction and a 45.52% peak-load reduction. The difference in cost savings (up to 13.34%) suggests that the choice of optimization technique and the specific tariff structure (CPP in the former case) can significantly influence the economic outcome. Furthermore, the inclusion of storage (PV–BESS) in the HEMS model, as demonstrated by [42], not only cut bills by 39.81% but also improved comfort by a factor of 1.6, illustrating that the introduction of flexible assets can mitigate the traditional cost-comfort trade-off.

4.1.2. Expanding the Scope: Emissions and Discomfort

When environmental impact is introduced as a third objective, the trade-off space shifts, often requiring a deliberate sacrifice of some cost savings to achieve greater reductions in emissions and discomfort [44], in a multi-environment DSM framework using NSGA-II, quantified this shift, reporting meaningful cost savings (up to ~20% in industrial settings) while simultaneously reducing CO₂ and discomfort. This contrasts with cost-only optimization, where the multi-objective approach, while retaining 46–90% of the cost savings, yielded 30–195% better emissions and up to 447% better PAR reduction. Similarly, Ref. [50] showed that microgrid-level tri-objective optimization, coupling smart homes with demand response, successfully reduced operating costs, emissions, and PAR while maintaining customer comfort. This finding is critical, as it challenges the assumption that environmental goals might increase user discomfort, provided flexibility is modeled explicitly, and user preferences are respected.

4.1.3. Design vs. Operation: Long-Term Trade-Offs

Design-oriented multi-objective optimization extends the trade-off analysis to the building envelope and HVAC systems, where the potential for energy and emission gains is significantly larger than in operational scheduling alone. [51], through simulation-based NSGA-II retrofits, achieved massive reductions in primary energy (up to 60%) and GHG emissions (60%), alongside significant improvements in visual discomfort (83%) and thermal dissatisfaction (65%). However, they caution that these efficient solutions can entail design compromises (e.g., window/insulation selections that affect daylighting), highlighting a long-term trade-off between energy efficiency and specific comfort/IAQ constraints. This is supported by [52], who found that HVAC choices, responsible for approximately 67% of building energy use, dominate the trade-offs in design parameters, underscoring the need for careful negotiation of these competing objectives early in the design phase.

4.1.4. Beyond Core Objectives: Self-Sufficiency and Asset Health

Emerging multi-objective formulations are incorporating objectives that reflect broader system goals. Ref. [48] explicitly modeled the trade-off between autonomy and cost by optimizing energy payment, user satisfaction, and self-sufficiency. Their results showed that re-ordering priorities in a lexicographic approach could change payment, satisfaction, and self-sufficiency by 27.4%, 100%, and 56.6%, respectively, making the cost of autonomy quantifiable. Furthermore, at the distribution level, Ref. [49] demonstrated that embedding transformer LoL into the objective function led to co-optimization, resulting in a 31% cost reduction, 18% peak reduction, and 28% LoL improvement. This illustrates that customer-centric objectives (cost, peak) and operator-centric objectives (asset health) can be successfully integrated and optimized together.

4.2. Algorithmic Trends and Methodological Comparison

Recent literature indicates a strong predominance of evolutionary and metaheuristic algorithms in multi-objective energy management for smart homes and buildings.

As illustrated in Table 3 Algorithms such as NSGA-II, MOABC, MOPSO, MOEA, MOALO, MOAOA, IBBO, and more specialized frameworks like MGWASA are extensively adopted because of their natural ability to approximate Pareto fronts in complex trade-off environments involving cost, comfort, PAR, and emissions. Their robustness and capacity to generate diverse non-dominated solutions make them well-suited for the heterogeneous and constraint-rich scenarios typical of HEMS and building performance optimization.

Table 3. Approaches to Trade-Off Management and Negotiation.

Approaches	Role in Trade-Offs	Insight
Pareto front (NSGA-II, MOPSO, MOEA, IBBO, MOAOA, MGWASA)	Generate a set of non-dominated solutions showing explicit cost & comfort & PAR or cost & emission trade-offs [42,45,46,52,53,54,55,56].	Users/decision-makers choose a preferred point
Scalarization (weighted sum, Pascoletti–Serafini, lexicographic)	Convert multi-objective to single-objective with weights or priority ordering, implicitly encoding preferences [43,47,48,51].	Changing weights/priorities can change cost, satisfaction, self-sufficiency by 27–100% [48]
Fuzzy or satisfaction indices	Map comfort/emissions into normalized indices; optimize cost per unit satisfaction or satisfaction under constraints [42,46,49,50,53]	MGWASA maximizes satisfaction (97%) at low cost per satisfaction unit [46]

Overall, the reviewed studies show that multi-objective optimization is essential in HEMS because it captures the unavoidable trade-offs among cost, PAR, comfort, environmental performance, self-sufficiency, and asset health. The evidence indicates that substantial gains can be achieved in core operational objectives, but these gains become increasingly dependent on user and system priorities as additional objectives are introduced. Therefore, the value of multi-objective optimization lies not in eliminating trade-offs, but in making them explicit and manageable within a balanced decision framework.

5. Integrated Hems: Multi-Asset Coordination and Energy Market Participation

5.1. Multi-Asset Integration and Prosumer Operation

Recent studies show that the coordinated integration of renewable generation (mainly PV), multi-type storage systems, EVs, and DR can simultaneously reduce operational costs, peak demand, and emissions while improving reliability and, in many cases, user comfort. Recent studies highlight the benefits of coupling these components under optimized control, enabling households and energy communities to act as active participants in energy markets.

Figure 5 illustrates the topology of a modern smart residential environment, often referred to as a prosumer node. This architecture relies on a central HEMS controller to interface via IoT with distributed energy resources, specifically PV, Battery Storage (BS), and EV, and controllable Smart Appliances.

Figure 5. Typical Architecture BEMS and Component.

As illustrated in Table 4 A large body of recent work incorporates rooftop PV, stationary batteries, and EVs acting not only as loads but as flexible storage assets through V2H, V2G operation.

Table 4. Integration Strategies and Impacts.

Integration Strategy in BEMS/HEMS	Main Integrated Resources	Control/Optimization Approach	Key Performance Impacts Reported	Citations
Multi-objective SHEM with PV–wind–EES	PV, wind, electrical energy storage, flexible/fixed loads	Improved biogeography-based optimization	Daily bill and PAR reduced; effective peak-time discharge and energy selling to grid	[45]
Optimal appliance scheduling with PV–battery	PV, battery energy storage, flexible/deferrable loads	Mixed-integer programming in MATLAB R2015a	Peak load was reduced by 61.5%, and cost was reduced by 71.4% vs. no PV storage	[57]
Rule-based PV–BES peak shaving in commercial building BEMS	Grid-connected PV, battery storage, DR on building loads	Rule-based control + GA for parameter optimization	Energy use was reduced by up to 20.9%; peak demand was reduced by up to 49.8%	[58]
Robust HEMS with PV–battery–EV and DR	PV, battery, EV charging, deferrable loads, TOU/RTP tariffs	Two-stage robust optimization	Cost savings up to 53.8% (TOU) and 35.5% (RTP) vs. no PV-storage	[59]
Microgrid-based HEMS with DSM	PV/wind, storage, household loads under TOU/flexible tariffs	Binary particle swarm optimization	Reduced peak load, peak-to-average ratio (PAR), and electricity cost; trade-off with renewable export	[60]
Life-cycle aware PV–battery–EV scheduling	PV, stationary battery, EV, DR, grid interaction	PSO-based multi-objective scheduling and ESS sizing	Cost was reduced by 28.9% (summer) and 17.7% (winter); losses were reduced by approximately 25%; EV battery life was nearly doubled; CO2 emissions were reduced by up to 59.7%	[61]
Community HEMS with PV–BES–EV peak shaving	Community microgrid, PV, BES, EVs, clustered loads	PSO + rule-based dynamic demand/feed-in limits	Peak shaving was improved by approximately 7% vs. earlier methods; community load profile became flatter	[62]
PV–battery DSM with hybrid metaheuristics	Grid, PV, battery, residential loads, RTP	Hybrid Genetic–Ant Colony Optimization	Cost was reduced by up to 59%; CO2 emissions were reduced by approximately 57%; peak load was reduced by up to 25.7% vs. unscheduled operation	[63]
Multi-home HEMS with PV–BESS–EV	Multiple smart homes, PV, BESS, EVs, RTP	Mixed-integer linear programming	For multiple homes, cost was reduced by 46.4% vs. no PV/BESS/EV	[64]
Hierarchical domestic BEMS with PV–BES–EV (V2G)	PV, BES, EVs with V2G, multiple households	Hierarchical coordination framework	Significant peak reduction on distribution grid; economic benefits vs. prediction-based control	[65]
Forecast-driven EMS with PV–home battery–EV (V2H)	PV, home battery, EV (V2H), HVAC, flexible loads	LSTM-HHO forecasting + predictive EMS	PV self-consumption was increased by 18.6%; grid reliance was reduced by 25%; household cost was reduced by 17.3%	[66]
DR-based HEMS with RES + BSS	PV/other RES, battery storage, smart loads, RTP DR	Hybrid ACO + TLBO scheduler	With grid+RES+BSS: bill was reduced by 28.8%, PAR was reduced by 41.5%, CO2 emissions were reduced by 21.9%; comfort was improved	[67]
Source–load–storage coordination in intelligent buildings	PV, ESS, EVs, HVAC, LED, DR, user comfort	Two-stage multi-objective (cost vs. satisfaction)	Cost was reduced by 12.9%; user satisfaction was maintained at 0.95 or higher; grid volatility was reduced by 7.6%	[68]
PV–BESS BEMS with DR using PSO	PV, BESS, building loads, DR (peak clipping, valley filling)	PSO-based EMS vs. linear programming	Daily cost was reduced by up to 15.3%; smoother SoC; reduced grid power fluctuations	[69]
Packetized energy management with PV–battery	PV, battery, packetized residential loads, QoS constraints	Hybrid DE + BPSO	Lower total cost while preserving QoS and improving PV self-sufficiency	[70]
Day-ahead DR-integrated EMS for off-grid hybrid system	Diesel + multiple RES/storage, DR in remote community	Probabilistic fuzzy DR + day-ahead scheduling	Higher RES utilization, reduced diesel use and improved reliability	[71]
Heuristic programmable HEMS with PV–ESS–EV	PV, ESS, EVs, smart appliances, DR	Hybrid GA-PSO HPEMC controller	Cost was reduced by 25.6%, PAR was reduced by 37.0%, and CO2 emissions were reduced by 24.0% vs. no scheduling	[72]

These integrated systems enhance peak shaving, reduce grid reliance, and improve energy autonomy. Also, several models extend beyond single-battery configurations by integrating multiple storage technologies combined with grid trading, enabling households to operate as prosumers with optimized day-ahead energy decisions. These systems achieve meaningful operational cost reductions and greenhouse-gas mitigation, though not all storage technologies or renewable sources (e.g., wind) are economically viable in all settings.

5.2. Energy Market Participation in HEMS/BEMS: Markets, Mechanisms, and Enabling Technologies

The coordinated integration of PV generation, battery storage, EVs, and demand response does not only improve internal household scheduling; it also enables HEMS-equipped homes to operate as active prosumers capable of participating in local and grid-interactive energy markets. Accordingly, beyond technical coordination, recent literature increasingly examines how HEMS support peer-to-peer trading, aggregator-mediated services, and flexibility exchange through advanced market-clearing and digital transaction mechanisms.

5.2.1. Types of Energy Markets and HEMS Roles

According to the reviewed studies, HEMS do not operate within a single market structure; instead, their function changes with the institutional design of the local market. Ref. [73] distinguish three dominant models, peer-to-peer (P2P), community self-consumption, and transactive energy, and this classification is reinforced by [74], who show that each model implies different rules for coordination, pricing, and grid interaction. In comparative terms, P2P markets give HEMS the highest decision autonomy because households can act as active trading agents, which explains why [73,75,76] emphasize data exchange, local optimization, and bid preparation. By contrast, Ref. [77] present community trading as more coordinated and operationally stable, whereas transactive designs are more grid-responsive, prioritizing dynamic balancing over purely bilateral exchange.

5.2.2. Trading and Market-Clearing Mechanisms

Trading performance depends more on choosing a mechanism that fits the market context. Ref. [78] frame blockchain-enabled P2P trading around secure decentralized settlement, while [77] show that pool trading is better suited to coordinated local communities with flexible loads, photovoltaics, and storage. Ref. [79] demonstrate that auction-based mechanisms become particularly valuable when heterogeneous assets, such as second-life EV batteries, must be allocated competitively. Ref. [80] adds that game-theoretic pricing improves responsiveness by tying price formation to the supply-demand ratio, whereas [81] indicate that learning-based control can improve exchange decisions under dynamic conditions. Comparatively, auctions strengthen price discovery, pools reduce coordination complexity, and intelligent control improves adaptability under uncertainty.

5.2.3. Role of AI, Blockchain, and IoT

Recent studies present AI and blockchain as tools that work together. Ref. [76] identifies forecasting and scheduling as the central AI contributions to P2P smart homes, while [82] shows that AI-enhanced IoT infrastructures improve building-level monitoring and control. Ref. [81] further demonstrate that deep reinforcement learning can reduce costs by about 19% while preserving prosumer comfort, indicating that AI is most valuable when market participation requires adaptive decision-making. Refs. [83,84] move a step further by integrating prediction, blockchain, and HEMS coordination into unified platforms. In contrast, Refs. [75,78,85,86] show that blockchain mainly contributes trust, automation, and tamper-resistant settlement, although scalability and implementation overhead remain unresolved.

5.2.4. Multi-Agent Coordination in Local Energy Markets

Local energy markets (LEMs) increasingly rely on distributed agents, including households, microgrids, and network operators, to enable energy trading while preserving grid security [31,34,35]. Recent studies show that this coordination is progressively supported by multi-agent and optimization-based frameworks that jointly address market negotiation, congestion management, and system resilience [34,35]. In smart distribution networks with multiple microgrids, for example, zone agents, microgrid operators, and the distribution system operator (DSO) are coordinated through a two-stage multi-agent system (MAS) to manage both energy dispatch and self-healing functions [35]. Similarly, decentralized hierarchical MAS architectures distinguish monitoring agents from control agents, thereby allowing prosumers to participate in decision-making while maintaining grid-level operational objectives [31]. At the household level, fully decentralized LEM models often assign a market or negotiation agent for game-theoretic trading and a congestion agent based on ADMM to handle network constraints [34]. This dual-agent structure is particularly relevant because it preserves the regulatory separation between trading and grid operation, avoids reliance on a central coordinator, and has demonstrated effective congestion management in IEEE test systems [34]. Under uncertainty, hybrid stochastic-robust MAS frameworks further strengthen local energy markets by coordinating energy management and self-healing against variability in load, electricity prices, renewable energy sources, electric vehicles, and agent availability, while reducing operating costs, network losses, and voltage deviations [35].

Overall, the reviewed studies show that integrated HEMS create the greatest value when renewable generation, storage, EVs, and flexible demand are coordinated rather than optimized in isolation. This integration consistently improves cost efficiency, peak and PAR mitigation, self-consumption, emissions performance, and, in some cases, user comfort and battery longevity. As asset portfolios expand, however, the control problem becomes more complex and increasingly requires multi-objective, robust, or hybrid optimization frameworks. Beyond operational benefits, this coordinated flexibility also enables households and communities to participate more effectively in local energy markets, shifting HEMS from passive load schedulers to active prosumer platforms.

6. Real-World Applications and Limitations of Integrated HEMS

HEMS are moving from theory to pilots and small-scale deployments. Research shows clear economic and technical benefits, but also substantial gaps around scalability, user behavior, ICT infrastructure, and regulatory fit.

Figure 6 illustrates where HEMS are being applied in practice, categorizing the implementation into three main domains: Real Households & Communities, which involve standard grid-connected residential setups; Remote/Off-Grid Communities, where systems manage isolated microgrids and battery storage; and Hardware-in-the-Loop/Laboratory Setups, utilized for testing and algorithm validation.

Figure 6. Classification of HEMS application domains.

Table 5 synthesizes findings from recent simulated and real-world HEMS studies. Most HEMS research remains simulation-based, which enables extensive testing across diverse scenarios and supports the evaluation of advanced optimization methods, but often overlooks hardware constraints, communication delays, and behavioral uncertainties. Emerging field and real-time prototypes confirm that energy savings are achievable in practice, although usually at lower levels than those reported in simulation, while also underscoring the importance of user convenience, reliability, and interoperability as major research priorities. Advanced and robust optimization methods, such as two-stage robust optimization and reptile search algorithms, report the highest theoretical performance, with cost reductions of up to 60% [59], particularly under uncertainty in PV generation and EV charging. Heuristic and metaheuristic approaches, including PSO, GA, and MMGO, provide a strong compromise between flexibility and performance; although their cost savings are typically more moderate, ranging from 12% to 42%, they are especially effective in multi-objective settings and can reduce PAR by up to 75% [67]. MILP-based formulations remain a strong benchmark because of their transparency, interpretability, and suitability for reliable day-ahead scheduling, even if they are generally less flexible than adaptive or learning-based approaches.

Table 5. Real World Application Overview.

HEMS Focus	Real-World/Simulation Context	Main Application	Outcomes	Citations
PV + ESS + EV + DR (robust/det.)	MATLAB-based simulation with TOU and RTP tariffs	Cost minimization under ToU/RTP, battery tech comparison	Up to ~60% cost savings vs. no PV+ESS; +5–7% better than prior work	[59]
An Optimization Based Power Usage Scheduling Strategy Using PV, BS for DSM in Smart Grid	Simulation of DSM under real-time pricing	HGACO metaheuristic to co-schedule grid, PV, and battery; MKP to avoid rebound peaks	Up to 59.1% cost, 57.4% CO2, and 25.7% peak-load reduction vs. unscheduled use	[63]
Smart home DSM with RES+BSS	Simulation of SEMC-based DSM with RES and battery	User-comfort-aware DSM under RTP	21–42% bill cut; up to 41.5% PAR & ~28% CO2 reduction	[67]
Heuristic HEMS (PV+ESS+EV)	Simulation of HPEMC for a smart home with PV, ESS, EVs	Cost, PAR, CO2, comfort	25.55% cost, 36.98% PAR, 24% CO2 reduction	[72]
Hybrid backup (PV, WT, H2, BESS, EV/V2H)	Case study home in Liaoning (China) with PV, WT, HSS, BESS, EVs	AI-optimized HEMS with DR (0–40%) and hybrid storage/V2H	2.34% operating-cost cut but very low LPSP (high reliability)	[87]
HEMS with RES, ESS, grid buy/sell	Simulation of HEMS with grid, PV, ESS, and electricity selling	PSO/BPSO to minimize cost and PAR; allows buy-low/sell-high operation of ESS	Energy cost cut to 19.7% of baseline; extra 4.3–8.5% savings by higher ESS power; PAR improvement	[88]
IoT real-time cluster HEMS	Real-time IoT implementation with smart plugs and hardware prototype	Hardware smart plugs + MMGO	11–12% cost and up to 74.68% PAR reduction	[89]
Prosumer multi-storage EMS	Simulation using real consumption data; day-ahead planning via MILP	Li-ion/lead-acid + PV + EV + DR	18.26% (Li-ion) & 14.88% (lead-acid) cost cuts; large GHG drop	[90]

The reviewed studies further indicate that system performance depends not only on the optimization method, but also on the degree of asset integration within the HEMS architecture. Systems combining PV, battery energy storage, EVs, and demand response generally achieve higher resilience and operational efficiency than single-asset configurations. In particular, hybrid storage systems integrating BESS and V2H exhibit strong reliability performance; for example, Ref. [87] reported a very low loss of power supply probability, highlighting the evolution of HEMS from cost-saving tools toward resilience-oriented infrastructures. Beyond economic gains, integrated HEMS also deliver meaningful environmental benefits, including CO₂ emission reductions of up to 28% [67] and substantial greenhouse gas mitigation when lithium-ion storage is used [90]. A key finding is the persistent gap between simulation and field deployment: hardware-validated studies, such as the IoT-based residential cluster in [89], report more modest savings of 11–12%, yet provide important evidence that HEMS can be deployed successfully in existing dwellings through smart plugs and mobile applications.

Overall, the analysis suggests that the most effective HEMS systems combine the integration of multiple types of storage, such as photovoltaics, energy storage systems (ESS), and electric vehicles (EVs), with hybrid or uncertainty-aware optimization frameworks. However, from a practical deployment perspective, user-friendly and interoperable IoT platforms currently appear to offer the most realistic path to adoption in the short term.

7. Discussion and Future Directions

The studies included in this systematic review suggest a clear shift in HEMS research, from relatively static and single-objective scheduling models toward more dynamic frameworks that coordinate multiple assets under competing operational goals. This evolution is not merely algorithmic. It reflects the fact that residential energy systems have become more complex, particularly with the growing presence of rooftop PV, stationary batteries, EV, and DR. Under these conditions, cost minimization can no longer be treated as the unique decision criterion. Instead, it must be evaluated alongside peak reduction, user comfort, emissions, self-consumption, and, increasingly, asset health. Within this broader decision space, deterministic mathematical programming still performs well when the problem structure is explicit and the constraints are clearly defined. By contrast, metaheuristic and learning-based approaches appear more attractive when uncertainty, nonlinearity, and adaptive control requirements become more pronounced.

A central finding of this review is that the effectiveness of a given optimization method is largely context-dependent. In other words, the literature does not support the idea of a universally superior algorithmic family. MILP-based formulations remain especially valuable where transparency, strict constraint handling, and reproducible day-ahead scheduling are required. Their main strength lies in analytical clarity and predictable behavior, which is still important in residential settings where safety, comfort bounds, and device limits must be enforced explicitly. Metaheuristic methods, on the other hand, offer greater flexibility in non-convex and multi-objective search spaces, although this flexibility often comes at the expense of guaranteed optimality. Learning-based methods, particularly DRL and hybrid AI frameworks, seem promising for adaptive decision-making under variable tariffs, renewable intermittency, and changing user behavior. Even so, their reported performance depends heavily on training quality, data availability, reward formulation, and tuning stability. Taken together, the evidence points less toward methodological replacement than toward methodological complementarity, with different approaches serving different operational needs.

Another important observation concerns the role of coordinated multi-asset control. Across the reviewed studies, the benefits of HEMS become more substantial when photovoltaics, stationary storage, electric vehicles, and flexible loads are managed jointly rather than in isolation. This coordinated operation tends to improve not only energy cost and peak-to-average ratio, but also self-consumption, resilience, and, in some cases, battery longevity and user satisfaction. At the same time, these gains should be interpreted with caution. Most of the studies remain simulation-based, often under favorable assumptions regarding forecasts, user compliance, communication reliability, and equipment response. Real deployments must contend with a different set of constraints, including latency, interoperability limitations, behavioral variability, hardware degradation, and regulatory boundaries. This explains why field-reported savings are often more modest than those obtained in optimization studies.

The review also highlights several implications for future work. First, more attention should be given to validation under realistic operating conditions, including pilot deployments, hardware-in-the-loop platforms, and interoperable IoT test environments. Second, future HEMS models would benefit from a more explicit treatment of behavioral uncertainty, heterogeneous comfort preferences, and component degradation, as these factors directly affect practical performance. Third, the field still lacks shared evaluation protocols, reference datasets, and standardized metrics, which makes cross-study comparison difficult and sometimes misleading. Greater methodological consistency would improve the interpretability of published results. Finally, as HEMS move toward more active prosumer participation, future research will need to address community-level coordination, multi-agent control, cybersecurity, and local energy market participation within realistic technical and regulatory limits. These issues are no longer peripheral. They are becoming central to whether advanced HEMS methods can move from promising designs to credible large-scale deployment.

8. Conclusions

This systematic review shows that HEMS have evolved beyond appliance-level scheduling and now operate as integrated control frameworks capable of coordinating flexible demand, distributed renewable generation, stationary storage, and electric vehicles. The evidence reviewed suggests that deterministic optimization remains well suited to structured planning problems with clearly defined technical constraints. By contrast, metaheuristic, hybrid, and learning-based methods appear more adaptable when uncertainty, nonlinear system behavior, and competing objectives must be handled simultaneously. In practice, however, no single approach emerges as uniformly superior. Algorithmic suitability depends on the scale of the decision problem, the quality of available information, and the operational requirements of the HEMS itself.

The reviewed studies also indicate that combining photovoltaics, battery energy storage, electric vehicles, and demand response can improve both economic and operational performance. Yet these gains should not be treated as universal, since they remain strongly shaped by modeling assumptions, tariff structures, control design, and local implementation conditions. A further point deserves emphasis: despite the breadth of recent methodological progress, the literature is still dominated by simulation-based validation, while long-term field evidence remains comparatively scarce.

Taken together, the studies synthesized here provide a structured view of recent HEMS research by linking optimization methods, multi-objective trade-offs, and multi-asset coordination within a single analytical perspective. This review therefore helps clarify the transition from conventional HEMS toward smarter, more adaptive, and grid-interactive residential energy systems, while also pointing to a clear research priority: stronger real-world validation and closer alignment between algorithm design and deployment constraints.