Energy Flexibility Realization in Grid-Interactive Buildings for Demand Response: State-of-the-Art Review on Strategies, Resources, Control, and KPIs

Abstract

The increasing penetration of renewable energy into the grid has given rise to an emerging challenge of maintaining the supply–demand balance. Conventional supply-side regulation is now insufficient to maintain this balance, necessitating flexible resources from the demand side to address this challenge. Buildings, as important energy end-use consumers, possess abundant flexible resources and can play a significant role in responding to grid dispatch via demand response. Therefore, grid-interactive buildings (GIBs) have garnered widespread attention. This technology coordinates the scheduling of distributed renewable energies, energy storage, and adjustable loads via advanced control methodologies, leading to the reshaping of building load profiles to enhance grid flexibility. However, the realization of energy flexibility in GIBs has not yet been comprehensively identified in the literature. To narrow the knowledge gap, this review compared GIBs with other technologies of building energy management to highlight the distinct features of GIBs. Additionally, the flexible energy strategies of GIBs were explored, combined with flexible resources within buildings, and the feasible pathways for these strategies were also addressed. Based on the scheduling scenarios in GIBs, the performance characteristics of various control methodologies were compared and analyzed. Finally, an evaluation framework for GIBs was established. This review will facilitate the shift of buildings from traditional energy consumers to flexible resources that actively respond to the grid and provide critical support for the grid stability and reliability.

1. Introduction

The global average surface temperature has increased by 1.1 °C compared with the pre-industrial period, and the atmospheric CO₂ concentration has exceeded 410 ppm, which is mainly attributed to the overuse of fossil fuels [1]. To reduce CO₂ emissions, a number of countries have announced their low-carbon or net-zero carbon roadmaps [2]. As an alternative, conventional fossil fuels are being broadly replaced by renewable energy sources, including roof PV, onshore and offshore wind farms, and solar thermal power plants. The long-term vision predicted by the IEA demonstrates that renewable energy can provide over 30% of the global electricity supply by 2050 [3]. However, due to the nature of instability and intermittency, it is widely considered that the increasing penetration of renewable energy can pose a substantial challenge to grid stability and supply–demand balance [4,5]. Analysis from Villar et al. [6] suggested that effects from the supply side alone cannot address this challenge, and new flexible resources from the demand side should be imperative.

Recently, the notion of buildings as a potential flexible resource on the demand side has emerged as a potential solution to address the imbalance between supply and demand, since they consume 55% of the global power [7]. For decades, numerous efforts have been devoted to developing the technologies of building energy management (BEM). Various technologies have been proposed, including intelligent buildings (IBs), smart buildings, green buildings (SBs), zero-energy buildings (ZEBs), and grid-interactive buildings (GIBs). We conducted a systematic search for research papers published in the BEM field from 2005 to 2025 via the Web of Science (WOS) database. A topic-based search strategy was applied with the keywords such as “intelligent building” or “intelligent buildings”. The search results were further sorted by publication years using the result analysis tool of WOS to illustrate the research trends of various technologies, as shown in Figure 1. Although GIBs exhibit a promising potential, research in this field has very recently begun according to the search in Web of Science, as shown in Figure 1b. The search results also indicate that SBs and GBs have attracted the most research attention, while the number of publications on GIBs is still limited, with only of 329 works in total. This discrepancy highlights the necessity of comprehensive introduction to GIBs.

Table 1. Summary of various building energy management technologies.

BEM	Characteristics	Key Technologies	Grid Interaction
IBs [12]	Focus on equipment automation centralized control	Building automation systems, sensors, controllers, etc. automatically manage HVAC systems and lighting	No grid interaction
SBs [13]	Data-driven optimization initial ability of making decisions	The Internet of Things (IoT) big data analysis elf-learning algorithms collaborative control	Simple interaction with the grid
GBs [14]	CO2 emissions reduction high penetration rate of distributed renewable energy	High-efficiency energy-saving equipment insulation and natural ventilation disturbed renewable energy technologies	Weak grid coordination capability
ZEBs [15]	Energy self-sufficiency distributed energy	Distributed energy energy storage technology dynamic energy balance schedule	Low dependence on the grid and weak grid support
GIBs [7,10]	Real-time interaction with the grid demand response load profile reshapes	Artificial intelligence (AI) big data analysis virtual power plant (VPP) intelligent control systems energy storage system distributed energy real-time load adjustment	Deep integration into the grid and strong grid support

summarizes the characteristics of the above-mentioned technologies of building energy management and the interaction between the buildings and the grid. As shown in Table 1, the smart buildings focused on the automated operation of the HVAC system, introducing Internet of Things (IoT) technology to achieve data-driven real-time optimization of energy consumption [8]. In contrast, green buildings focused more on the increasing penetration of renewable energy and CO₂ emissions reduction [9]. On the basis of green buildings, zero-energy buildings are further proposed to fully utilize distributed renewable energy for power supply, with almost no interaction with the grid. Given the high penetration rate of renewable energy in the modern power system, the grid-interactive buildings (GIBs) have received significant attention due to their flexibility. GIBs can dynamically regulate their load profile in response to grid dispatching signals, utilizing intelligent dispatching technology, distributed renewable energy, flexible loads, and energy storage systems [7]. This technology drives the transformation of buildings from traditional energy consumption terminals into flexible resources that actively respond to the grid. Therefore, GIBs can enhance the stability of the grid and promote the consumption of renewable energy and the reduction of CO₂ emissions [10,11]. Despite the significant potential of grid-interactive buildings (GIBs) to enhance grid stability and facilitate renewable energy integration, several challenges can hinder their widespread implementation. These challenges include high initial investment costs, technical complexities in system integration and control, cybersecurity risks associated with grid interactions, and the requirement for reliable and robust communication infrastructure. Moreover, user acceptance and behavioral uncertainties present additional obstacles to the effective deployment of GIBs. Addressing these challenges is essential to fully realize the potential of GIBs in future energy systems, which necessitates a comprehensive understanding of GIBs.

However, current research still lacks a systematic summary of the flexible resources of GIBs, and the critical path of dispatching the flexible resources remains unclear. In order to narrow the knowledge gap, this paper firstly presents three strategies for building energy flexibility and then identifies the corresponding flexible resources in buildings to support the various strategies. The optimized control methodologies for dispatching the flexible resources have also been reviewed. Moreover, for the different control methodologies, a comprehensive evaluation method has been proposed from five dimensions, including independence, stability, energy efficiency, flexibility, and environmental friendliness. This review can promote the large-scale application of grid-interactive buildings in the new power system.

2. Building Energy Flexibility

The core of building energy flexibility lies in coordinated scheduling of various flexible resources in buildings according to the dispatching signal of the grid, while meeting the daily needs of occupants [16]. As shown in Figure 2, after receiving grid dispatching signals at the interactive layer, the GIB employs its intelligent decision-making system to select an appropriate energy flexibility strategy. Then, the scheduling instructions are issued to various flexible resources via the information network to execute the flexibility strategies. Finally, the adjustment results are fed back to the interaction layer to achieve real-time grid response and support its stability.

2.1. Strategies for Building Energy Flexibility

According to the dispatching purpose, response speed, and duration, energy flexibility strategies mainly include three types: load covering, load shifting, and load shedding. Figure 3 shows the schematic diagrams of three energy flexibility strategies. In Figure 3a–c, the black dashed line represents the original load curve of the building before regulation, and the blue line is the actual building load curve after implementing corresponding flexibility strategies.

(1)

Load covering: As shown in Figure 3a, the power generation from on-site distributed renewable energy (e.g., PV and wind turbine) is used to cover a portion or the entirety of building energy demand during peak hours. Thereby, the primary objective of this strategy is to curtail peak power demand on the grid and enhance the grid reliability, which persists for a duration of 2 to 4 h during the peak period.

(2)

Load shifting: This strategy refers to the use of flexible resources such as energy storage and shiftable loads to achieve peak shaving and valley filling of the load profile of buildings, as shown in Figure 3b. The implementation of this strategy is commonly guided by electricity price signals or incentive mechanisms, which realizes 4 to 6 h of power mitigation.

(3)

Load shedding: This strategy refers to rapidly cutting off the load in response to the emergency regulation demands of the grid, thereby leading to a swift reduction in the power consumption of buildings. Its duration usually only lasts a few minutes or seconds, which can help the power system recover from emergencies. In recent applications, this strategy has also been employed as a virtual energy storage capacity by the integration of an intermittent run of HVAC system and building thermal inertia.

The topic-based search was conducted in the Web of Science database, with combined keywords such as “grid-interactive building” and “load shift”. The results indicate the distribution of the three strategies is almost equal, as shown in Figure 3d. This rank can be attributed to two factors. One is that the complexity of building energy management includes distributed renewable energy, energy storage, and demand response. Conventional PID controllers, originally designed for process control, exhibit fundamental limitations when applied to the nonlinear, multivariable, and stochastic building energy systems. The other factor is that the emerging control methodologies, such as data-driven control and agent-based control, are still under investigation to validate their robustness and reliability.

2.2. The Flexible Resources of Buildings

In buildings, the flexible resources that support the above-mentioned energy flexibility strategies mainly include distributed renewable energy, energy storage systems, shiftable loads, and interruptible loads [17]. Notably, as demonstrated in Figure 2, the energy storage system can support all three strategies, playing a significant role in building energy flexibility. Distributed renewable energy, shiftable load, and curtailable load can support load covering, load shifting, and load shedding, respectively.

2.2.1. Distributed Renewable Energy

Distributed renewable energy, such as rooftop photovoltaics, micro-wind turbines, and geothermal heat pumps, has been widely integrated into buildings to meet occupants’ various energy demands, including electricity, heat, and cooling. Energy demand during peak hours can be significantly reduced through load covering. However, renewable energy such as wind and solar power is intermittent and fluctuating. Therefore, the match between the building energy demand and renewable energy is the key to implementing the strategy of load covering. At the building level, the combination of distributed renewable energy and an energy storage system is the main approach to addressing the aforementioned challenge [18]. Kang et al. [19] proposed a capacity optimization analysis framework and cost/carbon efficiency assessment for a microgrid system. The optimized design of the battery storage system capacity can reduce the cost by up to 21.5% or cut carbon emissions by 38.4%. With the popularization of electric vehicles, many scholars also regard them as important distributed energy storage devices for buildings. Liu et al. [20] proposed a hierarchical control strategy for the orderly management of electric vehicle charging in an interactive energy system with renewable sources. The investigation revealed that the control strategy can enhance the matching capability and flexibility of the building energy, minimizing reliance on the power grid. Considering the slow charging and discharging rate and low capacity of electric vehicles, Robledo et al. [21] studied the integration of hydrogen energy vehicles with BIPV in the vehicle-to-grid mode, which can reduce the power demand from the grid by about 71% per year. Moreover, hydrogen energy vehicles can be used as hydrogen storage systems to supply energy to the grid during peak load periods. In terms of energy storage capacity matching, although building energy demand and distributed energy are matched on an annual scale, there is a severe imbalance between local power generation and building energy consumption on minute or even hour scales. This imbalance results in frequent interaction between buildings and the grid. Consequently, the design of energy storage capacity for distributed energy still needs to be refined through minute-level simulation of energy storage operation to enhance the dynamic regulation performance of grid-interactive buildings. In addition, the accurate prediction of building loads, meteorological conditions, and renewable energy output by using historical data and deep learning also contributes to the implementation of load covering strategies [22,23,24].

2.2.2. Energy Storage Systems

Energy storage systems are the core of realizing building energy flexibility, which effectively supports all the flexibility strategies. According to the form of energy storage, building energy storage mainly includes heat storage, cold storage, and electric energy storage. Heating and cooling are basic needs of residents, accounting for approximately 60% of building energy consumption. Heat and cold storage technologies can effectively reduce the peak load of buildings. Off-peak electricity can be used to freeze water or heat molten salts for energy storage, which will be released during peak hours to achieve load shifting. Qiang et al. [25] proposed a control strategy based on the cooling load forecasting model to address the issue of poor energy performance caused by the mismatch between the cold storage capacity and the cooling demands of buildings. This model can predict future cooling demands and employ off-peak electricity for cold storage, achieving cost savings of up to 41.9%. Liu et al. [26] proposed an energy system including an electric chiller, an electric heater, and storage tanks. In comparison with the system without energy storage, the deployment of the tanks can improve cost-effectiveness by 16.5% and reduce CO₂ emissions by 30.9%, indicating that the energy storage system can improve building flexibility and meet the load shifting requirements. Considering the thermal inertia of buildings, researchers have proposed direct load control (DLC) to implement the periodic on–off cycling of HVAC systems, thereby optimizing the air conditioning loads. Wei et al. [27] comprehensively considered the thermal comfort requirements and peak shaving demands of different users and proposed an optimal scheduling strategy for DR based on DLC. The simulation results show that during the DR period, the load regulation of this strategy exceeded the target by approximately 6.65%, effectively achieving the peak shaving goal while ensuring thermal comfort. In addition, phase change materials are widely used in building heat storage due to their highly efficient heat storage capacity. Jin et al. [28] combined phase change heat storage with heat pump technology to transfer the heat load during the peak power period to the valley power period. Their comparisons of different operation strategies revealed that the night-heating strategy notably flattened the building load profiles. Additionally, electrical energy storage can be further integrated with district heating networks. In the integration, electricity is used to power heat pumps for thermal energy production, which is subsequently fed to centralized heating systems. This approach not only expands the role of GIBs from consumers to “prosumers”, but also establishes an essential connection between electrical and thermal energy systems. Muntean et al. [29] demonstrated this concept by effectively utilizing surplus renewable electricity to power heat pumps. They found that this approach significantly reduced reliance on conventional fossil fuels, improved overall energy efficiency, and increased the renewable energy share. Based on the cross-system integration, GIBs can evolve from merely participating in electrical demand response to serving as flexible regulators across both electrical and thermal networks, substantially enhancing their roles in multi-energy systems.

In addition to heat and cold storage, battery energy storage has also played a significant role in promoting the flexibility of buildings. Current research interests focus on determining battery capacity compatible with distributed renewable energy and its healthy operation management. Symeonidou et al. [30] developed a nonlinear method to manage the energy generated by PV, the energy stored, and the energy purchased from the grid based on the life cycle cost optimization of battery energy storage (BES) systems. This method explicitly considers building energy system requirements, including energy costs, initial investment in batteries, and arbitrage by selling excess electricity to the public grid. Das et al. [31] established two distinct energy management strategies: load tracking and battery cycle charging. With the non-dominated sorting genetic algorithm II (NSGA-II), they optimized the energy management of lead–acid batteries and lithium batteries with different strategies, respectively. The comparative analysis results revealed that the PV-battery system adopting the cycle charging strategy has a lower energy cost and enhances battery operational health.

Beyond the mentioned storage method above, compressed air energy storage (CAES) and gravity energy storage (GES) are increasingly recognized as viable alternatives for large-scale energy storage applications. CAES converts electrical energy into high-pressure air stored in underground reservoirs, which is later expanded through turbines to generate electricity during periods of high demand. It offers advantages including large storage capacity, extended service life, and the absence of chemical pollution, making it well-suited for supporting building-integrated photovoltaic or wind systems for long-duration storage, ranging from 4 to 24 h. However, the application CAES is constrained by geographical limitations, since it requires suitable underground caverns. GES stores energy by using electricity to lift heavy objects to elevated positions and generates power via gravitational descent to drive generators. This technology offers advantages including rapid response, ultra-long cycle life, and low life cycle costs. Modular GES units can be deployed in building basements or on rooftops, enhancing their integration flexibility. For instance, Bowoto et al. [32] proposed a hybrid system combining solar photovoltaics with gravity energy storage.

2.2.3. Shiftable Loads and Interruptible Loads

The shiftable loads in the buildings mainly include dishwashers, washing machines, electric vehicles, etc. The core of the dispatching strategy for this type of load lies in fully leveraging its long-term load shift capacity by proactively delaying its start-up during peak hours, thereby effectively shifting load demand and achieving peak shaving and valley filling. Figure 4 shows the average flexible time window of shiftable loads. As can be seen, the shiftable horizon of all these loads exceeds 5 h, enabling effective peak shaving and valley filling of the load profile of buildings.

As the deep integration of distributed renewable energy into the power system intensifies grid instability, it poses a considerable challenge to the dynamic regulation of shiftable loads. Ning et al. [34] proposed a bi-layer optimal scheduling model for virtual power plants (VPPs), considering source–load coordination via the regulation of shiftable loads. This model offers a low-cost flexible framework for high-volatility renewable energy systems, more effectively enhancing peak shaving capacity and improving the overall economic performance. Liu et al. [35] proposed a building-integrated photovoltaic (BIPV) energy management strategy based on the optimal schedule of the grid. This strategy can improve the grid stability and energy economy of BIPV and building energy storage systems by load shifting and BIPV power management. On the other hand, the uncertainties in user behavior patterns are also a key factor affecting the regulation of shiftable load. Yan et al. [36] proposed a dynamic pricing and multi-objective optimization strategy based on peak–valley price difference to guide users to charge electric vehicles during off-peak hours. This strategy achieves the effective regulation of shiftable load and reduces the peak–valley difference between power supply and demand. Tifoura et al. [37] employed a real-time scheduling algorithm to adjust the operation of shiftable load to balance solar generation and building demand. The test in a solar house demonstrated that the optimal operation of shiftable load can boost renewable energy utilization and self-consumption rate without affecting user comfort.

Interruptible loads in buildings mainly include elevators, air conditioners, lighting, etc. During peak hours or emergencies of the grid, these loads can rapidly shut down or curtail power demand, enabling a swift response to the grid dispatch signal. The uncertainties in user behavior patterns have a significant impact on the regulation of interruptible loads. Yang et al. [38] proposed an incentive-based interruptible load scheduling scheme to reduce the peak load of the grid and operation costs. The results show that the scheme fully meets the constraints of interrupted loading time and minimal daily load reduction. Based on the dynamic economic compensation mechanism, Duan et al. [39] developed the optimal operation strategy of an electrical integrated energy system using incentive-based demand response. They conducted a comparative analysis of cases under different circumstances. The comparisons revealed that the piecewise incentive mechanism was effective for motivating users’ participation in demand response and lowering system costs. In addition to the above-mentioned simulation studies, a large-scale field test was carried out in Shanghai by switching off some elevators and lights in commercial buildings during peak hours, achieving a load reduction of 500 kW in a single building.

3. Control Methodologies for Building Energy Flexibility

Section 2 outlines the three core strategies for flexible energy and the adjustable resources in buildings. However, the realization of the potential of these resources and the implementation of the strategies depends on precise optimization scheduling methods. These resources exhibit significant differences in response speed, control objectives, and operational constraints. Appropriate control methodologies are therefore needed to balance energy efficiency, comfort, and grid response requirements. Section 3 focuses on control methodologies, which compare their applicability and limitations. The topic-based search in Web of Science illustrated that the primary control methodologies employed in GIBs include PID control, model-based predictive control (MPC), data-driven control, and agent-based control. Figure 5 shows the distribution of these control methodologies. The control methodologies ranked by academic interests are MPC (76.19%), data-driven control (14.29%), agent-based control (7.14%), and PID-based control (2.38%), highlighting the predominant focus on MPC in current research trends.

3.1. Rule-Based PID Control

Rule-based PID control dynamically adjusts control parameters through proportional, integral, and derivative operations on the feedback deviations to achieve stable control of building equipment (e.g., HVAC systems) [40]. This method delivers a characteristic of rapid response and instantaneous regulation of energy system operating states without complex algorithms or optimization, resulting in its prevalent application in building energy management [41]. Figure 6 shows the schematic of the PID control. Chojecki et al. [42] proposed a PID-based energy-saving control method for intelligent building heating systems. Multiple environmental state factors were considered in the tuning of PID parameters to meet building comfort requirements while significantly reducing total energy consumption. However, PID control inherently lacks the capability to predict future trends of the system, especially in systems with significant delays and large inertia. Its independent application often leads to poor control performance [43]. When building heat loads fluctuate drastically or in extreme climates, control lag or overshoot may occur, which compromises indoor environmental comfort and causes energy wastage. Thus, rule-based PID control is not suitable for building energy systems that have complex dynamic characteristics and uncertainties [44].

3.2. Model-Based Predictive Control

One advantage of model predictive control (MPC) is its ability to predict the future dynamic behavior of a system based on established models. It determines optimal control actions through objective function optimization, dynamically evaluates alternative strategies via simulation, and implements the solution within a rolling predictive horizon [45]. Figure 7 shows the schematic of MPC and rolling optimization. In the scenarios of GIBs, MPC is expected to outperform rule-based PID control via the prediction of systems’ dynamic behavior and rolling optimization of the response. This approach can prevent overshoot or control lag, thereby maintaining indoor comfort while yielding significant cost and energy savings. Pedersen et al. [46] investigated the performance of a scenario-based MPC approach for residential building heating. In comparison with PID control, the MPC-based approach reduces system operating costs by 13.1–16.1%. Reynolds et al. [47] combined day-ahead optimization and MPC to optimize building energy management. Simulation demonstrated that this optimization reduced energy consumption by approximately 25% compared to the baseline strategy. With the introduction of a time-of-use (TOU) tariff, the optimization strategy successfully shifted loads to off-peak price periods, reducing energy costs by around 27%.

Although MPC exhibits a promising potential in energy consumption reduction and cost minimization, it still remains under-exploited in real application. High implementation costs and computational complexity severely prevent MPC applications in medium-sized buildings [48]. Additionally, the performance of MPC depends on the quality of established models, the accuracy of predicted data, and disturbance effects of the system [49].

3.3. Multi-Scale Model-Based Predictive Control

Multi-scale model predictive control considers system characteristics across different temporal and spatial scales. Time-scale coordination incorporates short-term equipment responses and long-term energy demands, divided into day-ahead scheduling, intra-day scheduling, and real-time scheduling. Each schedule differs in data input characteristics, control target settings, and time scales, as shown in Figure 8. Geng et al. [50] proposed a bi-level MPC strategy for the run of an island energy hub. A day-ahead scheduling is optimized at the higher level, while a real-time scheduling is tracked in the lower level. This strategy is expected to improve integrated energy system reliability while reducing operating costs. Wang et al. [51] used a multi-timescale MPC for managing hydrogen utilization to meet various demands of buildings. In the day-ahead scheduling, seasonal balance between renewable energy and load was considered, while rolling optimization is employed to identify the predicted errors of renewable energy in the intraday scheduling. Finally, in the real-time scheduling, the chance-constrained method was adopted to address supply–demand balance in the short-term. In addition to multi-timescale optimization, multi-spatial scale control was also implemented for co-optimization of energy consumption and equipment operation in different building zones [52].

3.4. Adaptive Model-Based Predictive Control

With the development of adaptive model predictive control (AMPC), new solutions have emerged for variability and uncertainty issues in GIBs. AMPC dynamically adjusts model parameters and strategies via real-time system monitoring. This method can effectively respond to building uncertainties, including structural changes, usage patterns, and climatic impacts, to enhance prediction accuracy and control performance [53]. Figure 9 shows the schematics of adaptive model predictive control. Dou et al. [54] addressed the source–load uncertainties in flexible distribution networks through adaptive adjustments in prediction model design and control parameters. Although AMPC can cope with complex and dynamically varying building environments, its algorithm is complex and computationally expensive. To address this limitation, simplified models were established to reduce computational burden (e.g., combining CFD simulation with real-time data) [55]. In GIBs, AMBC can regulate flexible loads such as air conditioners and lighting to achieve power demand response and reduce peak–valley differences. Additionally, this method can also improve building energy flexibility by balancing indoor comfort and energy savings [56].

3.5. Data-Driven Predictive Control

Due to the complex and stochastic nature of the building energy system, the establishment of a control-oriented model is still a challenge. As a potential alternative, data-driven predictive control can be performed without models and mainly depend on intelligent algorithms, including neural networks, reinforcement learning, genetic algorithms, and particle swarm optimization. It utilizes a large amount of historical and real-time data to learn operating rules of building energy systems and predicts future trends of the building energy system. According to the learning and prediction, this data-driven method can achieve efficient decision-making. Bao et al. [57] optimized microgrid energy management based on deep reinforcement learning, enhancing adaptability to renewable energy power fluctuations and reducing system costs. Zhang et al. [58] used communication neural networks (CommNet) as intelligent agents trained by proximal policy optimization (PPO) to achieve optimal scheduling strategies for distributed power sources. Krishnan et al. [59] applied neural network-based ADHDP to the scheduling of residential building energy. This method predicts energy demand using historical data for the optimal scheduling. Most of the above studies focus on training neural network models with reinforcement learning algorithms. Since reinforcement learning is characterized by varying underlying data distribution during training, it suffers from poor convergence and sensitivity to numerous hyperparameters. Thus, there still exists a great potential for data-driven predictive control in terms of both theoretical development and practical application.

3.6. Agent-Based Control

Due to the complexity of building energy systems, they can be divided into several subsystems as individual agents according to their functionalities. The optimization of the building energy system can be achieved through the cooperation and communication of the agents. Thus, agent-based control (ABC) is proposed for the scheduling of building energy systems, utilizing multiple intelligent agents to enable distributed decisions and optimize building energy management. The multi-agent systems can be designed as hierarchical structures. In such a structure, a coordinating agent can send optimized parameters setpoints to sub-agents, while the sub-agents are responsible for adjusting the corresponding equipment to meet these setpoints [60]. Multi-agent systems can achieve coordinated management of building photovoltaic systems and energy storage. Raju et al. [61] find that building systems using agent-based control can dynamically adjust the charging and discharging strategies of BES. This adjustment is based on real-time light intensity and household electricity load forecasts. Ultimately, this method can reduce grid electricity consumption by 15% and increase solar energy utilization by 10% to 12%. Despite their advantages, agent-based building control systems face challenges such as difficulty in coordinating and optimizing multiple subsystems and inflexibility in optimization multiple objectives. To solve this problem, Michailidis et al. [62] proposed a hybrid control approach that combines RL, MPC, and other algorithms to enhance coordination capabilities. At the same time, agent-based control has limited adaptability to environmental dynamics (such as sudden changes in lighting and load fluctuations), and coordinating the interests of multiple agents is complex. Without a global optimization mechanism, this may lead to local optima.

Table 2. Comparative summary of advantages and disadvantages of several optimization scheduling methods.

Optimizing Scheduling Methods	Advantages	Drawbacks	Applications
Rule-based PID control	Easy-to-understand principle high reliability wide range of applications	Difficulty in relying on accurate models complex parameterization lack of ability to predict future trends	PID control can be applied to hot water tank management and HVAC systems, determining their behavior based on preset logic (such as schedules, weather conditions, etc.) to achieve more efficient energy management.
Model predictive control	Strong ability to consider system constraints good adaptation to model accuracy predictive capabilities	Higher computational complexity model dependency problem parameter tuning is difficult	MPC has a significant effect on climate control and cost–benefit analysis.
Multi-scale modelling predictive control	cope with systems with multiple temporal and spatial scales improved control accuracy	Model construction and multi-scale coordination strategies are designed to be complex involving extensive calculations	Multi-scale MPC not only meets building energy consumption management requirements but also achieves decarbonization goals.
Adaptive model predictive control	Adaptation to time-varying system characteristics automatic tracking of system changes	Adaptive control mechanisms require meticulous design	AMPC can be applied to heat pump-assisted solar hot water systems, which can effectively respond to grid fluctuations, provide flexible energy management, and optimize hot water supply through predictive control.
Data-driven predictive control	Strong ability to model complex systems self-learning and self-adaptation access to rich data resources	High data quality requirements poor model interpretability high training costs	Data-driven predictive control enables accurate prediction of temperature, humidity, and energy consumption, improving energy efficiency and user comfort.
Agent-based Control	Handles increased system complexity beyond traditional methods Provides operational flexibility & adaptability	High deployment costs and complexity Difficult multi-subsystem coordination	Agent-based control technology can balance energy efficiency and indoor comfort in buildings by dynamically adjusting equipment settings to reduce energy consumption.

systematically summarizes the advantages and disadvantages of several optimal scheduling strategies.

4. Evaluation Indicators and Methodology

The development of GIBs requires the implementation of their key functions in both new and existing buildings. The KPIs are essential to scientifically evaluate the performance of GIBs and guide their progress. Previous studies mainly analyzed the dynamic interaction between buildings and the grid from the aspect of the supply–demand balance [63]. Based on the features of GIBs, this review provides a methodology to quantify their KPIs from five dimensions: independence, stability, energy efficiency, flexibility, and environmental impact, as summarized in

Table 3. Key performance indicators and their definitions for GIBs.

EvaluationDimensions	Supported Key Performance Indicators	Define	Formula
	Probability of grid-less interaction	Probability that a building or system will not interact with the grid within a certain period	$P_{GNI}={\displaystyle \frac{T_{NI}}{T_{total}}}$ PGNI: probability of no grid interaction; TNI: no grid interaction time; Ttotal: total system time.
Independence	Mismatch compensation factor	Coefficient indicates the amount of compensation required by the mismatch between load and generation	$F_{mc}={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^N\left|P_{load,i}-P_{res,i}\right|dt}}{{\displaystyle {\sum }_{i=1}^N{P}_{load,i}}}}$ Fmc: mismatch compensation factor (range 0–1); Pload,i: building area at the i-th moment Pres,i: renewable generation output at time i
	Microgrid synergy	Evaluation of the response capacity of microgrid to the grid	$C_{grid}=1-{\displaystyle \frac{{\displaystyle {\sum }_{i=1}^N\left|P_{net,i}-P_{ref,i}\right|}}{2{\displaystyle {\sum }_{i=1}^N\left|P_{net,i}\right|}}}$ Pnet,i: net ineractive power at time i; Pref,i: grid dispatch reference power at time i.
	Grid interactivity	Ratio of real-time net power interaction to the annual maximum interaction for a given time period	$R_{inst}(t)={\displaystyle \frac{\left|P_{net}(t)\right|}{P_{net,max}^{year}}}$ Rinst(t): real-time interactivity ratio at time t; Pnet(t): net interactive power at time t; $P_{net,max}^{year}$: annual maximum net interactive power.
Stability	Capacity factor	Ratio of average annual power interaction between the building and the grid to the nominal design capacity of the grid	$R_{annual}={\displaystyle \frac{\frac{1}{N}{\displaystyle {\sum }_{i=1}^N\left|P_{net,i}\right|}}{P_{grid.nominal}}}$ Rannual: annual average interactive capacity ratio; Pnet,i: net exchange power at time i; Pgrid,nominal: grid nominal design capacity; N: annual number of time points.
	Recovery rate	Ratio of peak switching to rated design capacity of the grid	${\eta }_{rec}={\displaystyle \frac{P_{peak}}{P_{grid}}}$ ηrec: recovery rate; Ppeak: battery swap power during peak hours Pgrid: grid-rated design capacity.
	Total energy consumption	Total amount of all forms of energy consumed by buildings in a given period of time	$EB={\displaystyle \sum D_{b,s,t}}={\displaystyle \sum {TA}_{b,s,t}}·{EI}_{b,s,t}$ EB: total building energy consumption; Db,s,t: energy consumption of a specific building type; TAb,s,t: activity level of a specific technology in buildings EIb,s,t: energy intensity of a specific technology in buildings.
Efficiency	Peak power	Maximum electrical load reached by the building or system in a short period of time	$P_{max}=max(P_{net,1},\dots ,P_{net,N})$ Pmax: maximum value in the net load sequence; Pnet,i.: net load at time i.
	Energy conservation	Percentage reduction in energy consumption compared to the baseline situation	$Esr={\displaystyle \frac{E_{base}-E_{actual}}{E_{base}}}$ Esr: energy saving rate; Ebase: baseline energy consumption Eactual: actual energy consumption.
Flexibility	Demand response potential	Measurement of achieving aggressive renewable energy goals while maintaining grid reliability	$DRP={\displaystyle {\sum }_{i=1}^N\left(P_{flex,i}^{max}-P_{flex,i}^{min}\right)\Delta t}$ DRP: aggregate demand response potential; $P_{flex,i}^{max}$: maximum upward dispatchable power; $P_{flex,i}^{min}$: minimum downward dispatchable power.
	Energy storage utilization efficiency	Measurement of the efficiency of energy storage systems actually stored and released energy	$h_{ess}={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^N\left|P_{dis,i}\right|\Delta t}}{{\displaystyle {\sum }_{i=1}^N\left|P_{ch,i}\right|\Delta t}}}$ hess: energy storage round-trip efficiency; Pch,i: charging power at time i; Pdis,i: discharging power at time i.
	Adjustment speed	Speed of system response to grid load changes	$u_{i,j}=\left\{\begin{array}{ll}{\displaystyle \frac{P_{Ei,j}-P_{Si,j}}{T_{Ei,j}-T_{Si,j}}}& \mathrm{upward} \mathrm{adjustment}\\ {\displaystyle \frac{P_{Si,j}-P_{Ei,j}}{T_{Ei,j}-T_{Si,j}}}& \mathrm{downgrade}\end{array}\right.$ ui,j: ramp rate; PEi,j: end-of-interval output power; Psi,j: start-of-interval output power; TEi,j: end time; Tsi,j: start time.
	Storage capacity	Available storage capacity for storage technologies integrated into the smart grid	$C_{avail}={\displaystyle \frac{1}{N}}{\displaystyle {\sum }_{i=1}^N\left(E_{max}-E_{res,i}\right)}$ Cavail: average usable stored energy; Emax: energy storage system maximum capacity; Eres,i: reserved capacity at time interval i N: total time steps.
Environmental Impact	carbon footprint	Refers to the emissions of greenhouse gases generated in a process	$CF={\displaystyle \sum ({AD}_i\times {EF}_i)}$ CF: carbon footprint; ADi: activity data for activity type i; EFi: emission factor for activity type i.
	Renewable energy ratio	Percentage of renewable energy used in buildings	$\lambda ={\displaystyle \frac{{\displaystyle \sum E_i}}{E_b}}$ λ: renewable energy share of total energy consumption; Ei: annual renewable energy utilization; Eb: annual total operational energy consumption of building project.

.

Independence: The independence indicators measure the degree of GIBs dependence on the grid for energy supply. They are characterized by self-supply capacity and grid-interaction frequency, with the ambition of increasing the penetration of distributed renewable energy and avoiding high frequency of grid–building interaction. Three quantitative indicators were proposed, i.e., probability of grid-less interaction, mismatch compensation factor, and microgrid synergy. The indicator of probability of grid-less interaction measures a building’s capability to satisfy its energy demand solely through on-site power generation and energy storage [64]. The mismatch compensation factor is a coefficient that indicates the amount of compensation required by the mismatch between load and generation [65]. Microgrid synergy is proposed to evaluate the response capacity of microgrid to the grid.

Stability: The stability indicators for GIBs quantify power fluctuations during building–grid interaction, focusing on interaction frequency and amplitude. These indicators are considered to limit the rapid interaction between the grid and buildings in a short term, preventing the grid instability. Thus, the temporal characteristics of building–grid interaction frequency and amplitude become critical constraints, which affects the stability of both the grid and building energy system [66].

Efficiency: The efficiency indicators for GIBs mainly assess building energy utilization efficiency and optimization potential. These indicators are specifically categorized into total energy consumption, peak power demand, and load balancing coefficient [67]. The refinement of energy consumption evaluation methodologies provides a critical technical foundation for building energy optimization.

Flexibility: The concept of flexibility for GIBs was defined as the capability of a building to regulate its demand and on-site power supply according to local climate, occupant comfort, and the grid requirements [68]. Flexible resources and advanced control methods enable buildings to transform from passive energy consumers into participants that actively support the supply–demand balance of the grid. The key to this transformation lies in the ability to dynamically adjust building loads according to the real-time dispatching signals of the grid. This dynamic response significantly enhances the adaptability and stability of the grid, effectively mitigates peak–valley load differences, and reduces CO₂ emissions [69]. The supported indicators involve demand response potential, energy storage utilization efficiency, adjustment speed, and storage capacity [70,71]. The core functionality of these indicators lies in measuring buildings’ response capability and regulation capacity.

Environmental impact: The environmental indicators for GIBs quantify the environmental impact of building energy management and operation. Typical environmental indicators include carbon footprint and renewable energy proportion, reflecting buildings’ sustainability performance. Carbon footprint quantifies the total direct and indirect CO₂ emissions throughout an entire cycle of a building. The renewable energy ratio indicates the penetration of renewable energy used in building systems [72].

5. Discussion

Since the 1990s, various BEM technologies have been proposed to address energy and environmental challenges across different stages. Currently, the proposed GIBs aim to assist the grid in tackling the challenge of maintaining supply–demand balance, also offering the advantages of improving the penetration of renewable energy and reducing CO₂ emissions. Energy flexibility is the primary feature of GIBs that distinguishes them from other BEM technologies, namely, a capacity of dynamically adjusting their load profile in response to grid dispatching signals. However, based on our review, the realization of energy flexibility still faces the following challenges and needs further investigation.

For flexible resources, energy storage plays a significant role in realizing energy flexibility. Although the storage capacity is enhanced by the growing number of EVs, thermal energy storage (TES) is expected to be a dominant energy storage form in future building applications. This is attributed to the fact that heating and cooling loads amount to approximately 60% of building energy consumption. Additionally, based on this review, both load covering and load shifting require long-term energy storage, typically ranging from 2 h to 8 h. According to the techno-economic evaluation of various energy storage technologies [73], TES is a more cost-effective alternative than BES when the storage duration time exceeds 2.3 h. Moreover, novel thermal mediums are currently being investigated, including phase change materials and molten salts, which have the potential to further improve TES performance.

Control methodologies are another key factor in supporting building energy flexibility. Our review has revealed that MPC is predominantly adopted in current studies. This is attributed to the complex nature of buildings, characterized by nonlinearity, multivariable optimization, and stochastic behaviors. However, most existing MPC methods primarily depend on reduced linear models such as RC models for indoor temperature prediction, which probably limits their control performance in practical application. The date-driven method can serve as a complementary method to address the challenge associated with the MPC method. Nevertheless, due to the high demand for training data and opaque nature of the black-box mechanism, the data-driven method is still controversial and necessitates further investigation and improvement. Moreover, present studies of MPC and data-driven methods are mainly conducted in labs, lacking practical verification. Partial optimization of models or data training processes require substantial computational resources, constraining their practical application. Future work should focus on developing computationally efficient scheduling algorithms in combination with advanced AI algorithms and strengthening real-time decision-making capabilities.

Additionally, most studies in this review only consider the implementation of GIB in a single residential building. The load profiles of non-residential buildings are different from those of residential buildings in the horizon of a day, and this distinction can be combined to enhance energy flexibility and grid response capacity. Therefore, future research should consider the implementation of GIB across different types of buildings. Additionally, it is also essential to explore incentive mechanisms and standardized policies to promote the deployment of GIBs in both existing and new buildings.

6. Conclusions

GIBs, as an emerging building energy management technology, are critical to the transition of buildings from energy consumers to flexible resources on the demand side. By implementing flexible energy strategies, GIBs effectively enhance the supply–demand balance and reliability of the grid, simultaneously increasing renewable energy penetration rates and reducing carbon emissions. This paper systematically reviewed the evolution of building energy management technologies, highlighting the distinct characteristics of GIBs. The primary building energy flexibility strategies and flexible resources were identified, including distributed renewable energy, energy storage systems, shiftable loads, and interruptible loads. The critical roles of energy storage systems were revealed. Through the coordination with other flexible resources, the energy storage system can effectively support load reduction, load shift, and load interruption. Furthermore, the pathways of realizing the flexible strategies and their temporal characteristics were identified. Considering uncertainties in energy sources, loads, and occupant behaviors, the paper summarized the scheduling methods for building flexible resources, i.e., PID-based control, MPC and its variants, and agent-based control. Compared to traditional PID control, MPC and agent-based control significantly enhance the dynamic response capability of buildings and energy efficiency, demonstrating that they are more suitable for building energy scheduling. Finally, an evaluation framework for GIBs was established, including KPIs from five dimensions: independence, stability, energy efficiency, flexibility, and environmental impact.