Accurate residential energy forecasting is critical for smart grid operation and HVAC efficiency, yet it is inherently constrained by severe data scarcity and complex user heterogeneity. Standard meta-learning approaches often suffer from statistical distribution bias, as the meta-gradient typically overfits to dominant consumption patterns while neglecting minority households. To overcome this limitation, this study proposes a heterogeneity-aware model-agnostic meta-learning framework driven by a deep generative pre-categorization strategy (Cat-MAML). Abandoning heuristic statistical clustering, our approach utilizes a 1D-convolutional time-series variational autoencoder (TS-VAE) and Gaussian mixture model (GMM) soft-clustering to project raw consumption profiles into a continuous latent space, extracting profound behavioral archetypes. By enforcing a balanced task sampling distribution derived from this latent space during meta-training, the framework ensures equitable representation of both dominant and rare user groups. Experiments on a diverse, cross-climatic building energy dataset demonstrate that the proposed heterogeneity-aware framework significantly outperforms both a global baseline and vanilla MAML, achieving a 32.7
Building operation optimization (BOO) increasingly relies on diverse optimization paradigms. However, existing studies often employ inconsistent objectives, constraints, and modeling assumptions, making intrinsic trade-offs difficult to isolate. This study establishes a formulation-consistent benchmarking framework to compare black-box optimization (BBO) and mathematical programming optimization (MPO) for representative building energy systems. Two systems—an air-handling unit and a chiller plant derived from U.S. Department of Energy Commercial Prototype Buildings—are used as benchmark cases. Within a unified BOO formulation, BBO is implemented through iterative input–output evaluations using both high-fidelity EnergyPlus simulations and reduced-order models, whereas MPO employs explicit algebraic formulations solved using mathematical programming techniques. Representative gradient-based, mixed-integer, and population-based algorithms are evaluated under aligned objectives and constraints. The results show that all evaluated methods achieve comparable near-optimal energy performance. However, BBO requires approximately 10–100 times more computation time than MPO and scales less effectively as system complexity increases. Increasing model accuracy improves BBO performance primarily for the simpler air-handling-unit case, while providing limited benefit for the more complex chiller-plant case. These findings indicate that greater model fidelity does not necessarily lead to better optimization outcomes and that computational efficiency, system complexity, and model structure should be considered when selecting optimization methods for building operation.
Currently, the centralized optimization regulation strategy for conventional energy storage systems struggles to meet the operational requirements of centralized-distributed energy storage systems in rural areas. To address this challenge, this study examines a typical village in the Central Shaanxi Plain, Shaanxi Province, China, and integrates the usage patterns of agricultural machinery batteries into energy storage regulation. A peer-to-peer (P2P) trading mechanism for distributed energy storage, based on a combinatorial double auction, is proposed. In the trading environment, a double-layer optimal regulation model for rural centralized-distributed energy storage systems is established using a Stackelberg game approach. A revenue distribution mechanism based on the Shapley value method is introduced to ensure a reasonable distribution of revenue among alliance participants. The results showed that: (i) P2P trading can reduce the annual carbon emissions of centralized-distributed energy storage systems by 10.41
Amid global climate change and the goals of carbon peak and carbon neutrality, photovoltaic-driven air conditioning (PVAC) systems were critical for low-carbon building cooling, but existing controls suffered from poor adaptability, high computation costs, and inadequate real-time performance. This study proposed a hierarchical supervisory control framework integrating the twin delayed deep deterministic policy gradient (TD3) algorithm with large language model (LLM)-based adaptive reward weight adjustment. The TD3 controller performed real-time compressor speed regulation to balance multiple operational objectives. Meanwhile, LLM was used as a high-level module for adaptive reward tuning between control cycles. It adaptively adjusted the reward function weights based on historical operational performance, thereby reducing reliance on manual parameter tuning. Two LLM variants, DeepSeek-R1 and DeepSeek-V3, were employed as supervisory decision agents. The reasoner model DeepSeek-R1 demonstrated strong capability in mechanism-oriented reasoning and optimization exploration, whereas the chat model DeepSeek-V3 provided a lower latency compared to the reasoner model. Results showed that the supervisory mechanism enabled stable convergence of reward weights and improved the overall balance among thermal comfort, energy utilization, and operational cost. By using an adaptive supervisory decision from the LLM, The R1 model achieved a maximum comfort time ratio of 92.75
Since urban residents heavily rely on rail transit on a daily basis, controlling airborne infectious disease transmission in railway trains is critical for public health. In recent years, 222 nm far-ultraviolet C (UVC) lamps have emerged as a promising solution for disinfection of bioaerosols in occupied environments with enhanced safety. However, there is a lack of experimental data in railway train compartments, especially regarding the effectiveness of far-UVC against bioaerosols that are suspended in the air and that settle on handrails. Therefore, this study experimentally and numerically evaluated the effectiveness of far-UVC disinfection of bioaerosols suspended in air and deposited on surfaces using a simplified, scaled railway carriage model. The experimental data show that, under a high ventilation rate of 107 air changes per hour (ACH), the far-UVC lamp still provided supplementary reduction of viable airborne bioaerosols, achieving a disinfection efficiency of 23
Buildings consume large amounts of energy for heating and cooling, while peak electricity demand places significant stress on the power grid. This paper presents a reduced-order co-simulation framework and load-oriented supervisory control strategy for a geothermal-integrated thermally anisotropic building envelope with a ground loop (TABE+GL). In TABE+GL, a hydronic loop embedded in the building envelope is directly coupled with a geothermal ground loop, allowing for bidirectional heat exchange between the envelope, the ground, and the indoor environment. A hybrid co-simulation framework was established by coupling a reduced-order resistor–capacitor (RC) thermal network model with EnergyPlus augmented with GHEDesigner modules. The RC model generated feasible heat flux options under three operating modes, and EnergyPlus predicted sensible loads, energy use, and pump energy demand. At each simulation step, a supervisory control algorithm selected the optimal loop configuration and duty factor that maximizes useful TABE geothermal utilization without exceeding the predicted sensible load, thereby avoiding overheating or cooling. Case studies were conducted for Los Angeles, California, Charleston, South Carolina, and Denver, Colorado. Results showed that the proposed framework reduced HVAC electricity consumption by 43
Artificial intelligence (AI) has transformed the field of building energy and environment over recent decades. It enables substantial improvements in energy efficiency, cost-effective operation, and carbon emission mitigation in buildings. Despite these advances, the applications of AI in real-world engineering still face barriers in interpretability, physics-consistency, and generalization. These limitations lead to a performance gap between objective function of AI model and the targets of final application. To bridge the performance gap of AI applications in buildings, this perspective paper focuses on how domain knowledge can be integrated throughout the AI workflow in building energy and environmental applications. Furthermore, this paper proposes a framework of whole-process AI in building applications as a fundamental perspective of integrating AI and domain knowledge. The framework emphasizes that domain knowledge can be systematically embedded throughout the entire AI application process, including data preparation, model structure, model training, and performance evaluation. Rather than viewing AI as a substitute for domain knowledge, this study argues that domain knowledge is critical to application-oriented AI solutions. In particular, the novel technologies of large language models and physics-informed machine learning are highlighted as promising pathways to integrate data and physics in AI applications. This study promotes a synergistic integration of AI technologies and domain knowledge, enabling further improvements on whole-process application performance of AI in the building domains.
Distributed battery energy storage systems (BESS) have become increasingly popular to reduce utility costs, to mitigate the mismatch between uncontrollable renewable generation and electricity demand, and therefore to improve grid stability. However, determining the optimal BESS capacity remains a complex challenge, especially for large-scale industrial parks with distributed renewable energy systems. Existing approaches typically rely on Mixed-Integer Linear Programming approaches and nonlinear economic battery models, which create significant bottlenecks for computational efficiency and financial evaluation. To tackle these problems, this paper proposed a computationally efficient Linear Programming optimization framework and developed an off-the-shelf software tool for optimal BESS sizing in large-scale industrial parks. A key contribution of this work is the development of a linearized battery life-cycle cost model that embeds self-degradation rates directly into the Capital Recovery Factor, avoiding the computational burden of non-linear degradation models while ensuring economic accuracy. A comprehensive case study is conducted at a real-world industrial park located in Changsha, China, to verify the performance of the developed software and investigate the impact of multiple factors on the optimal BESS capacity. The results demonstrate that the developed Linear Programming framework identifies the maximum profitability bound for BESS more accurately than rule-based control framework and operates with significantly higher computational efficiency than Mixed-Integer Linear Programming models. Furthermore, the synergistic effect between the BESS and photovoltaic systems is also identified, revealing that their combined deployment yields a payback period of 4.56 years, thereby providing valuable quantitative insights for the optimal design of distributed energy systems.
Rapid urbanization has exacerbated the urban heat island effect, significantly increasing cooling energy demand. Daytime radiative cooling (RC) offers a promising zero-energy passive cooling strategy; however, urban-scale evaluations of RC potential and the specific influence of urban morphology remain scarce. This study integrates a physical RC model with the local climate zone (LCZ) framework to simulate 10,549 buildings across 85 LCZ models in Shenzhen, China. To improve spatial generalization, we developed rapid machine learning assessment models evaluated via 5-fold spatial cross-validation. Additionally, ensemble SHapley Additive exPlanations (SHAP) analysis was utilized to quantify the impacts of meteorological and morphological parameters. The results demonstrate that CatBoost and XGBoost are the optimal algorithms for predicting roof and facade RC energy savings, achieving average test-set R2 values of 0.888 and 0.854, respectively. SHAP analysis revealed that meteorological conditions (primarily solar irradiance and wind speed) dictate performance, contributing 66.0
The neglect of subjective factors such as cost consideration and accessibility in occupant behavior models leads to unrealistic device usage simulations, which ultimately undermines prediction accuracy. To address this gap, this paper proposes a Multi-Device Occupant Behavior (MDOB) model that integrates the Observe–Orient–Decide–Act (OODA) loop framework. The model systematically divides the behavioral process into four stages: in the Observe and Orient stages, both objective and subjective data are collected simultaneously, and key influencing factors are identified through correlation analysis; in the Decide stage, a multi-device coordinated behavioral decision-making mechanism is constructed by combining fuzzy theory and survival analysis; in the Act stage, the behavior is executed, and dynamic feedback is implemented. Through this closed-loop modeling framework, the model achieves a more accurate characterization of multi-device coordinated adjusting behaviors. Validation results show that the proposed method achieves higher accuracy in predicting actual energy consumption and thermal comfort levels. Furthermore, simulation results based on the MDOB model indicate that, compared to standard assumption scenarios, the model not only improves thermal comfort by 4.74
In hot and humid climates, the application of conventional radiant cooling systems is often limited by condensation risk and insufficient cooling capacity. To address these limitations, a decoupled radiant cooling (DRC) technology has been developed, in which a new type of radiant cooling panel is adopted. The new panel, titled as DRC panel, uses an air layer that is sealed by an infrared-transparent membrane to separate its radiant cooling surface from the air-contact surface. This design enables high radiant cooling capacity while minimizing the risk of condensation. However, existing research mainly focuses on the thermal performance analysis of DRC systems using steady-state models. There is a lack of research on their dynamic behavior under varying external boundary conditions and internal load disturbances. This study addresses the research gap in the dynamic thermal performance analysis of DRC systems by developing a quasi-2D dynamic model using a resistance-capacitance (RC) thermal network, calibrated through a non-dominated sorting genetic algorithm II (NSGA-II). The model was validated against experimental data, achieving mean absolute errors (MAE) and root mean square errors (RMSE) not exceeding 0.28 °C and 0.34 °C for both the membrane surface temperature and the radiant cooling surface temperature. The model also showed reasonable agreement in predicting the indoor air temperature and outlet water temperature. A simulation platform based on the developed model was constructed for a typical office conditioned by a DRC system, demonstrating its effectiveness as a tool for analyzing the dynamic thermal environment. The results highlight the model’s ability to provide accurate predictions, laying a solid foundation for future studies on energy consumption forecasting and operation control method development for DRC systems.
Buildings are interconnected cyber-physical systems (CPS) that couple with occupancy, heating, ventilation, and air-conditioning (HVAC), distributed energy resources (DERs), and power grids. This increasing complexity poses significant challenges for scalable modeling, control, and optimization. Existing approaches either rely on physics-based models with limited scalability, data-driven methods with weak physical consistency, or simplified representations that cannot fully capture system dynamics, restricting their applicability in real-world scenarios. To address these limitations, this study presents BESTOpt, a modular physics-informed machine learning (PIML)-based runtime environment for unified modeling, control, and optimization of interconnected occupancy–building–HVAC–DER–grid systems. The framework introduces a hierarchical structure (cluster–domain–system/building–component) and a standardized state–action–disturbance–observation data typology, enabling scalable coordination and integration of heterogeneous subsystems. Three case studies demonstrate the capabilities of the proposed framework. First, BESTOpt is compared with EnergyPlus, long short-term memory (LSTM), and a reduced-order 3R2C model to evaluate computational performance, prediction accuracy, physical consistency, and scalability from 1 to 200 buildings. Second, a 30-building residential cluster simulation evaluates 12 demand-response and retrofit strategies, showing that PV–battery integration can reduce peak-hour grid import by up to 97.8
Accurate building energy forecasting is essential for efficient energy management and sustainable building operation. While recent studies suggest that large language models (LLMs) exhibit strong potential for time-series forecasting, their application to building energy prediction remains limited by high computational cost, inefficient parameter utilization, and inadequate modeling of multivariate dependencies. To overcome these challenges, this paper proposes MaPL-LLM, a lightweight LLM-based forecasting framework that integrates multivariate prompt fusion and temporal patching. MaPL-LLM adopts a frozen LLaMA-1B backbone and introduces two complementary modules: (1) a multivariate prompt-based embedding module that encodes temporal context and statistical characteristics into structured textual prompts, and (2) a multivariate patching-based numerical embedding module that captures local temporal patterns and cross-variable interactions. Only lightweight input transformation and output projection layers are trained, significantly improving computational efficiency while maintaining stable performance. Extensive experiments on multiple building types from the BDG2 dataset demonstrate that MaPL-LLM consistently outperforms state-of-the-art methods, including TimeXer, PatchTST, and TFDFNet. For a 24-step forecasting horizon, MaPL-LLM achieves an MAE of 0.166, an MSE of 0.072, and an R2 of 0.946, reducing forecasting error by over 10
Accurate assessment of rooftop photovoltaic (PV) potential is critical for urban energy transitions, yet complex rooftop morphologies and fragmented obstacles often compromise estimation accuracy. Most existing approaches focus on selecting module tilt and azimuth to improve per-module yield based on meteorological conditions, while overlooking how mounting parameters affect geometric feasibility, particularly packing density and layout efficiency on irregular rooftops. To address this gap, this study proposes a Deep Learning-based Morphology-Aware Layout (DL-MAL) framework for obstacle-aware PV layout planning on complex rooftops. First, a DeepLabV3+ model with an Xception backbone was trained on a custom three-class high-resolution imagery dataset to extract pixel-level rooftop availability, excluding static obstacles and defining realistic installation zones. Then, an adaptive layout algorithm (ALA) evaluates predefined azimuth–tilt configurations while dynamically adjusting inter-row spacing to avoid shading, generating feasible layouts and corresponding capacity estimates. The reported best-performing configurations are identified within the predefined discrete azimuth–tilt search space under the adopted no-inter-row-shading constraint. Case studies demonstrate that the framework achieves reliable rooftop availability segmentation and significantly improves assessment accuracy, reducing the mean absolute relative error from 25.3
Urban microclimate exerts a significant influence on building indoor thermal performance, yet the coupled assessment of outdoor–indoor conditions, and the predictive evaluation of urban interventions prior to implementation, remain computationally and methodologically demanding. This study develops an automated, user-oriented workflow to evaluate how neighbourhood-scale urban interventions influence both outdoor heat exposure and indoor thermal comfort. The proposed parametric pipeline links open-source Geographic Information Systems (GIS) data, Grasshopper-based algorithmic modelling, the Vertical City Weather Generator (VCWG), and EnergyPlus Weather file morphing for building energy simulation. The workflow automates urban data extraction, parametric model generation, remote microclimate simulation, and climate-file preparation, allowing seasonal assessments without requiring Python programming expertise. Its capabilities are compared with Urban Weather Generator (UWG) and ENVI-met, revealing that, among the analysed tools, VCWG provides a balanced combination of sensitivity to radiative properties and vegetation changes, long-term simulation capacity, and operational computation times. The workflow is applied to a mid-20th century residential neighbourhood in Seville, Spain, under three scenarios: baseline, higher-reflectance surfaces, and a combined reflectance and vegetation strategy. Results illustrate the pipeline’s capacity to capture multi-scale thermal effects: outdoor air temperature reductions of up to −3 °C (mean summer decrease of −0.84 °C) translated into a mean indoor temperature reduction of −1.1 °C and a 14
Decentralized air conditioners offer significant potential for electricity market regulation. However, the compressor lock mechanism, often overlooked by existing models, causes units to reject switching commands, leading to prediction errors. To address this, this study proposes a high-precision aggregate power prediction method based on an improved state-queuing model. The framework integrates a second-order equivalent thermal parameter model with k-means clustering and introduces an explicit timer to accurately characterize delayed restart behaviors caused by compressor lock protection mechanism. Validation using GridLAB-D simulations covers regular operation and varying demand response events (0.5 h, 1 h, 2 h). Results demonstrate the method’s superiority. In regular operation, it maintains a mean relative error of 3.37
In urban buildings, indoor pollutants disperse outward through windows and may reenter other rooms, causing inter-unit dispersion. To address public concern regarding cross-infection risk from infectious individuals in self-isolation during window ventilation, this study quantified inter-unit dispersion of indoor respiratory aerosols in two-dimensional ideal street canyons with aspect ratios (AR = H/W = 0.5–3). Indoor-outdoor coupled airflow under single-sided ventilation conditions was simulated using the RNG k-ε turbulence model. Room air change rate (ACH) was determined via the tracer gas decay method. Dispersion of indoor respiratory aerosols in street canyons and multi-storey buildings was simulated using the tracer gas technique and quantified by reentry ratio (Rk). Infection probability (P) was estimated by the Wells-Riley equation. Numerical results shows that the clockwise main vortex in the street canyon dominates pollutant dispersion. The adjacent room of the source room along the main vortex exhibited the highest Rk values. The maximum observed Rk across all ARs was 5.9
Building simulation has evolved from manual calculations to a sophisticated interdisciplinary field essential for designing energy-efficient, resilient, and healthy built environments. However, current engineering practice faces interconnected challenges—including decarbonization, grid integration, climate resilience, and energy equity—that are redefining the scope of building simulation beyond traditional annual energy compliance. Addressing these demands requires a fundamental paradigm shift. This perspective proposes a multifaceted paradigm shift for next-generation building simulation, where technical advancements and model developments are strictly framed by critical engineering problems and validated against real-world scenarios to ensure applicability and reliability. Under this new paradigm, the field is undergoing a fundamental transformation that simultaneously expands its temporal, spatial, and physical horizons. Next-generation building simulation is moving beyond static annual evaluations to span temporal scales from sub-hourly grid interactions to multi-decadal climate adaptation. In parallel, it is extending spatial boundaries from individual buildings to urban and national clusters to capture mobility dynamics and microclimates, while deepening physical representation to resolve complex mechanisms in novel materials and human-centric indoor environments. By integrating AI and domain knowledge, this new paradigm enables these capabilities, transforming building simulation into a dynamic decision-support platform for co-creating a sustainable, resilient, and decarbonized future.
Extreme heat driven by climate change increasingly challenges building energy performance and indoor thermal comfort by intensifying cooling demand. Although typical and extreme weather-year datasets are widely used to estimate future energy consumption, limited attention has been paid to how buildings physically respond to extreme heat. This study applies a three-parameter change-point regression to daily cooling energy under extreme warm-year conditions to characterize the cooling energy–temperature response through the change-point temperature (β1), temperature sensitivity (β2), and base load (β3), while evaluating cooling peak demand and indoor thermal conditions. Residential and office prototypes were simulated across four Korean climate zones under SSP5-8.5 for present, mid-century, and late-century climates using Typical Downscaled Year and Extreme Warm Year weather datasets. In the extreme warm-year baseline for the Seoul medium office, β3 increased by 112.7
The risk of indoor airborne transmission among co-located individuals is generally non-uniform, which remains a critical challenge for public health modelling. Thus, we present CompARE, an integrated risk assessment framework for indoor airborne disease transmission that reveals a striking bimodal distribution of infection risk driven by airflow dynamics and human behaviour. Combining computational fluid dynamics (CFD), machine learning (ML), and agent-based modelling (ABM), our model captures the complex interplay between aerosol transport, human mobility, and environmental context. Based on a prototypical childcare centre, our approach quantifies how incorporation of ABM can unveil significantly different infection risk profiles across agents, with a more than two-fold change in risk of infection between the individuals with the lowest and highest risks in more than 90
