Due to the complex and dynamic structure of building energy systems, improving energy efficiency in smart buildings while ensuring occupant comfort has proven difficult. This paper proposes a hybrid framework which combines machine learning, physics-based thermal model of the building and optimization for intelligent building energy management. The proposed framework does not just use data-driven or physics-based approaches but rather integrates the complementary benefits to enhance energy prediction and control performance. The dataset for ASHRAE Great Energy Predictor III was used for evaluating the framework. After data preprocessing, the data was randomly split into three sets for training (70
This paper explores how machine learning and multi agent systems can be used in self healing protection of a smart distribution grid. This approach can increase the robustness and dependability of the power grid even in situations like high renewable energy penetration and complex energy sources scenario. Despite significant advances, existing self-healing protection schemes still face challenges in achieving scalable, autonomous fault detection and coordinated service restoration in active distribution networks. To address these limitations, this paper proposes an autonomous AI-based self-healing framework that integrates supervised and unsupervised machine learning with a distributed Multi-Agent System (MAS) for coordinated fault detection, isolation, and service restoration. The overall control detects the fault using Machine learning techniques, isolate the fault and restores the fault using multi agent systems. The framework employs supervised machine learning models for known fault detection and an autoencoder-based anomaly detection model for identifying previously unseen faults. A nine-bus system is used as the active distribution system under study here and it is used to validate the proposed control. Experimental results demonstrate that the proposed autoencoder achieved 94.5
Positive Energy Districts (PEDs) are increasingly framed through both expert-driven policy narratives and data-driven analyses of emerging project discourse. However, the extent to which different knowledge extraction approaches produce conceptually meaningful representations of the PED concept remains insufficiently examined. This study examines thematic patterns in two representative English-language corpora using expert-guided and machine-derived topic modeling with a three-way analytical design. A Pseudo-document-Guided LDA (PG-LDA) model was applied to structured case descriptions from the PED-DB (66 documents after preprocessing), while unsupervised Latent Dirichlet Allocation (LDA) was applied to an unstructured corpus of 35 documents retained from 44 identified PED project sources (approximately 149,420 words). To provide a comparative reference point for interpreting differences attributable to modeling approach and corpus type, an additional unsupervised control LDA (K = 7) was trained on the same structured corpus as PG-LDA. Cross-corpus semantic alignment was evaluated using sentence-embedding cosine similarity (SBERT; all-MiniLM-L6-v2), which accommodates the near-zero lexical overlap between expert-defined bigram topics (PG-LDA) and corpus-derived unigram topics (LDA). Results indicate moderate semantic alignment across most thematic dimensions (SBERT cosine similarity range: 0.348–0.535), with the cross-model comparison (PG-LDA vs. control LDA; mean SBERT = 0.453) and the cross-corpus comparison (control LDA vs. unstructured LDA; mean SBERT = 0.462) showing broadly comparable alignment scores. Stakeholder integration showed consistently weak alignment across all model comparisons, while financial feasibility and mobility integration remained comparatively underrepresented in both corpora. These asymmetries suggest differences in how PED knowledge is produced, communicated, and institutionalized across expert-curated and project-communication contexts. The findings suggest that expert-guided modeling supports conceptual consolidation of policy-relevant themes, whereas machine-driven modeling reveals emergent narratives and contextual experimentation dynamics. By integrating deductive and inductive analytical perspectives, the study contributes to a more reflexive understanding of PED discourse and supports the development of more balanced knowledge frameworks with broader qualitative and institutional evidence.
The energy consumption estimation and remaining driving range estimation are important aspects of electric vehicles for intelligent transportation systems, battery management systems, and for reducing range anxiety. Typical deterministic models, however, do not account for the uncertainty associated with dynamic driving behaviour, traffic variations, environmental conditions, and battery characteristics, and therefore have limited prediction ability. This study implements a comparative analysis of several uncertainty-aware machine learning models to estimate energy consumption and driving range of EVs using real-world driving data. The developed methods include SVR Residual Bootstrap, RF Variance Estimation, SVR Conformal Prediction, SVR Adaptive Conformal Prediction, LightGBM Quantile Regression, LightGBM Conformal Quantile Regression (CQR), and LightGBM Adaptive Conformal Quantile Regression (ACQR). The models are evaluated using deterministic metrics including RMSE, MAE, R2, and uncertainty-specific metrics including PICP and MPIW. The experimental results showed that accuracy of predictions, uncertainty coverage reliability, and interval sharpness were not equally good for the frameworks evaluated, resulting in significant trade-offs. SVR Residual Bootstrap had the best prediction performance in a deterministic prediction with RMSE = 2.358, MAE = 1.742, and R2 = 0.893. RF Variance and LightGBM CQR had higher uncertainty coverage but with wider intervals, while LightGBM ACQR had better balance between uncertainty coverage reliability and interval efficiency. Among all implemented approaches, the results of the SVR Conformal approach showed consistent and stable uncertainty estimation with reliability of the coverage (PICP = 0.95) and moderate interval width (MPIW = 15), making it suitable for real-world EV applications. The results underscore the significance of incorporating uncertainty quantification in EV prediction systems to ensure reliable and dependable sustainable transportation applications.
Data-driven methods can extract valuable information from building smart meter data. This article presents COF-Tool, a supervised methodological framework and open-source tool with trained models for classifying building categories and heating types, and for disaggregating heating electricity use directly from hourly electric smart meter data of buildings. The tool is trained and validated on real energy use measurements from Norwegian buildings with at least one year of data and consists of two modules: the classification module based on a random forest classifier, trained on data from 6403 buildings across 12 building categories, and the disaggregation module based on a Categorical Boosting (CB) regressor, trained and validated on data from 323 buildings across 9 building categories. To address the scarcity of labelled electric-heating data, the disaggregation module employs a cross-domain training approach where smart electricity and district heating measurements from buildings with district heating (DH) serve as a proxy for all-electric buildings. Results show that the cross-domain approach generalises to school and apartment buildings but is only tested on DH buildings for other categories. The effect of including classification probabilities and extended features in the disaggregation module is also evaluated. Classification performance is reduced compared to the original version, due to an increased number of residential units in the training set and common confusion between apartments and houses, as well as a reduced number of schools in the training set to avoid information leakage, as these buildings are reserved for the disaggregation training set. COF-Tool is publicly available as a GitHub repository and a web application, providing an accessible interface for extracting building information and heating electricity estimates from hourly smart meter data.
The rapid convergence of transportation electrification and urban energy systems introduces complex interdependencies among humans, electric vehicles, charging infrastructures, road networks, and distribution grids. This paper proposes a unified deep learning–based optimization framework for coordinated operation and resource allocation across the human–vehicle–charger–road–grid ecosystem. Unlike existing formulations relying on static optimization or distributionally robust approximations, the proposed method adopts a dynamic multi-agent deep reinforcement learning paradigm that jointly captures the spatiotemporal evolution of energy flow, traffic conditions, and user behaviors. Each agent−representing a distinct domain component−learns adaptive decision policies through cooperative reward sharing and environment embedding, enabling scalable coordination without centralized supervision. The mathematical formulation integrates a multi-term objective that minimizes total energy cost, travel delay, and grid imbalance, subject to nonlinear power–traffic coupling constraints. A hybrid convolutional–recurrent neural architecture encodes spatial topology and temporal dependencies, while federated gradient aggregation preserves cross-agent learning stability. Case studies based on a synthetic urban energy–mobility testbed demonstrate that the proposed framework reduces total system cost by 21.7
Abstract Decentralised energy flexibility in electricity systems faces a range of non-technical barriers that constrain widespread implementation. This study focuses on flexibility from residential heat pumps (HP), especially in combination with thermally activated building systems (TABS). It provides a comparative analysis of regulatory, financial and stakeholder-related barriers in Austria, Belgium, Canada, Denmark, Germany and Spain. Besides these types of barriers, the cross-country analytical framework is structured around specific flexibility use cases and utilisation mechanisms, which are schemes and market structures through which end-users’ flexibility can be activated. The analysis is based on expert consultations and a systematic review of scientific literature, offering insights into the multi-dimensional nature of the identified barriers. The findings highlight a significant disconnect between the technological availability of flexibility from residential heating systems and implementation. The main identified barriers are perceived high initial costs in combination with uncertain return on investments, insufficient awareness among end-users and professionals, reinforced by the insufficiently adopted regulatory setting. Insufficient regulatory consideration was identified, particularly for TABS, shortcomings in current energy policy frameworks were observed.
Energy systems are being reshaped by decarbonisation, decentralisation, electrification, sector coupling, digitalisation, and rising demands for resilience. These developments are changing energy systems from relatively centralised infrastructures into distributed cyber-physical systems that foster increasingly intelligent energy ecosystems. Almost a decade after the process of establishing the journal Energy Informatics began in 2017, the field has reached a level of thematic breadth, methodological maturity, and community identity that makes it timely to clarify its intellectual foundations. Energy Informatics has developed from an emerging research area concerned with digital applications in energy systems into a broader scientific and engineering discipline concerned with the information foundations of digital energy transformation. It is not defined by individual technologies such as data analytics, artificial intelligence, optimisation, smart meters, or digital twins, although these remain important parts of the field. It is concerned with how energy related information is represented, acquired, managed, exchanged, governed, analysed, and used to support the planning, operation, coordination, optimisation, and transformation of energy systems and energy ecosystems. Energy Informatics therefore provides the information centred perspective needed to connect energy engineering, information systems, computer science, software engineering, data science, artificial intelligence, cyber physical systems, energy economics, and sustainability transitions.
Long-term solar power forecasting at month- and year-ahead horizons is indispensable for grid investment planning, capacity adequacy assessment, and renewable energy policy in high-irradiance arid regions — yet deep learning architectures have not been systematically evaluated at these horizons across geographically diverse sites. This study addresses that gap by comparing Long Short-Term Memory (LSTM) networks and Temporal Fusion Transformers (TFT) for month-ahead (730-hour) and year-ahead (8,670-hour) photovoltaic (PV) power forecasting using three years of real on-site sensor data from ten grid-tied school PV systems ( 130 kW each) spanning seven governorates of Oman. Thirty-five cross-location experiments evaluate generalisation to climatically unseen sites, providing a substantially more demanding benchmark than standard temporal train–test splits. Results reveal a fundamental accuracy–generalisability trade-off: Single-site LSTM achieves high month-ahead accuracy (best RMSE = 43.0 kW, R² = 0.989) at specific training-test pairs (Mudhaibi-trained on Yunqul). but degrades severely on transfer, while TFT trained on merged multi-site data outperforms LSTM by up to 57
Abstract This study explores the ethical, social and interpretive aspects underlying the application of optimization to real-world decision-making, with special emphasis on energy-related applications. Based on ten years of interdisciplinary experience across academia, industry, and education, this research investigates how abstract optimization models gain meaning in practice within energy informatics applications such as electric vehicle charging, battery systems, energy system planning, and sustainable logistics. In this work, an autoethnographic approach is used to show that modeling is not only a technical activity, but also a highly interpretive process shaped by values such as efficiency, fairness, and accountability, as well as by the need to balance formal simplicity with complex stakeholder realities. Building on these insights the study proposes a reflexive framework that conceptualizes optimization as a meaning-making practice, with the objective of capturing how model outputs are constructed, interpreted, and negotiated across different contexts. The findings show that optimization models do not generate purely objective solutions. At the same time their outputs acquire meaning and authority through processes of framing, storytelling, and interaction with stakeholders. Ethical tensions, such as those between technical elegance and social responsibility, or between abstraction and contextual complexity, emerge as central features of modeling practice and are often managed through informal negotiation and compromise. This paper advances the understanding of optimization as a socially embedded practice, by introducing a structured framework grounded in empirical and reflexive analysis. It contributes to ongoing discussions in operations research and energy informatics by providing a conceptual lens for interpreting the human dimensions of modeling, and by supporting more reflective, context-aware, and ethically informed approaches to decision support.
Abstract When maximum power point tracking (MPPT) is implemented in practical photovoltaic systems, rapid variations in irradiance and temperature, as well as partial shading conditions, can create multiple local maxima in the power–voltage curve and degrade the performance of conventional tracking methods. In this study, a fuzzy logic–based MPPT controller is adopted and its scaling factors and membership-function spread parameters are optimized using an improved Harris Hawks Optimization (I-HHO) algorithm to avoid subjective parameter tuning. In addition, two practical modifications are introduced: a soft dead zone to reduce steady-state ripple near the maximum power point and a lightweight global scanning mechanism with adaptive resolution to improve tracking under severe irradiance changes and partial shading. The proposed method is evaluated under several operating scenarios, including irradiance and temperature variations and partial shading, and compared with the conventional P&O and INC methods. The results show that the multi-scenario objective function improves from 0.085267 in the baseline fuzzy controller to 0.071317 in the optimized version (about 16.36% improvement). In addition, the ripple in the temperature-step scenario decreases from 16.254% to 9.072%, while the settling time under partial shading improves from 0.078 s to 0.066 s. These results indicate that the proposed approach provides more stable tracking and improved dynamic performance for MPPT in practical operating conditions.
Building energy consumption accounts for a significant portion of global energy use, with heating, ventilation, and air conditioning (HVAC) systems representing one of the largest contributors to building energy demand. Accurate short-term HVAC load forecasting is therefore essential for efficient building operation, demand-response strategies, and intelligent energy-management systems.This study investigates the impact of temporal feature representation on one-hour-ahead HVAC load forecasting using real operational data from the PLEIAData building dataset. Three forecasting models representing both machine learning and deep learning paradigms—Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU)—are evaluated under progressively enriched temporal feature configurations. The experiments systematically compare baseline environmental variables, autoregressive load features, rolling statistical representations, and extended temporal feature sets designed to capture recent HVAC operational dynamics.The results demonstrate that models relying only on contemporaneous environmental variables exhibit limited predictive capability, whereas incorporating temporal load-history information substantially improves forecasting accuracy. Extended temporal representations based on autoregressive and rolling statistical features consistently enhance model performance across all forecasting architectures. Among the evaluated configurations, XGBoost combined with the extended temporal feature representation achieves one of the best and most stable forecasting performances with an R2 value of approximately 0.799. Additional ablation and interpretability analyses further show that recent temporal load information is the primary contributor to forecasting performance, while extending the feature space with increasingly complex rolling statistics and larger temporal representations provides only marginal additional gains. Overall, the study highlights the importance of temporal feature design, robustness-oriented evaluation, and interpretable forecasting analysis in HVAC load prediction applications.
Abstract The aging of wind energy assets and the expiration of government subsidies make decisions on continued operation, repowering, and decommissioning increasingly critical for wind farm operators. Existing studies provide valuable insights into wind energy investment under uncertainty, but often provide limited consideration of turbine-level heterogeneity when analyzing decisions from a wind farm perspective. Following Design Science Research, this study proposes a Decision Support System (DSS) for wind farm operations that integrates several methodological components with distinct roles. Spatial and regulatory feasibility assessment identifies whether turbines and sites are eligible for repowering. Stochastic process simulation generates uncertain paths of electricity prices, turbine lifetimes, energy yields, and costs. Least Squares Monte Carlo-based real options valuation evaluates repowering timing for eligible turbines, and Monte Carlo-based net present value analysis assesses decommissioning timing for turbines without repowering eligibility. Building on these turbine-level valuations, the DSS applies a rule-based clustering approach to translate individual decision timings into batch-level coordinated strategies at the farm scale. The artifact is evaluated using a real wind farm in central Germany. The results indicate that, under the given assumptions, repowering-eligible turbines exhibit a pronounced early exercise window, whereas non-eligible turbines tend to be decommissioned at the end of their lifetimes. At the farm level, the system identifies three decision scenarios and compares their expected total returns. The batch-execution strategy, in which actions are carried out at each cluster’s timing, achieves the highest expected return in the case study. This study provides a reusable information-systems-based methodological approach and case-based evidence for coordinating repowering and decommissioning decisions in wind farms. The findings suggest that the proposed DSS can translate complex uncertainties into executable farm-level strategies and improve transparency in strategic decision-making.
The equilibrium of the energy financial market design for new energy trading models and the analysis of the energy financial market have become central challenges as electricity systems are rapidly transforming due to the integration of renewable energy. The expansion of renewable energy sources has created a structural dislocation between the physical dimensions of energy dispatch and financial risk management. Current energy trading models treat the physical and financial aspects as two independent optimisation problems and overlook the joint strategic behaviour that arises when participants operate with both operational exposure and financial hedge positions simultaneously. To address this gap, a co-optimised framework called the Dual-Layer Energy Trading Model (DLETM) is proposed in this paper, integrating both physical energy exchange and financial market participation. The physical layer controls the day-ahead flow of energy between renewable generators, prosumers, storage operators and retailers whereas the financial layer allows forward contracts, green energy certificates (GECs) and risk hedging instruments. The interconnection between layers is enforced by a Cross-Layer Volume Consistency Constraint (CLVCC) and coordinated by a Dual-Layer Pricing Signal (DLPS). Market equilibrium is formulated as a Generalized Nash Equilibrium Problem (GNEP) and reformulated as a Variational Inequality (VI). The existence of at least one Variational Equilibrium is established under standard convexity and compactness assumptions [1]. An Augmented Lagrangian Decomposition with Primal-Dual Coupling (ALD-PDC) algorithm is introduced for decentralised computation. Simulation experiments on a stylised market with 12 participants with 40
Addressing the issues of insufficient precision and difficulty in identifying small targets in deep learning-based aerial image power grid facility segmentation methods, this paper proposes an improved framework integrating multi-scale features and boundary awareness based on SAM. A dual-branch architecture combining ResNet-34 and ViT is constructed within the Image Encoder, balancing global structure capture and local detail extraction through channel-wise concatenation. An adaptive prompting mechanism is established leveraging the Feature Pyramid Network (FPN) and Adaptive Prompt Generator (APG), enhancing the automation and accuracy of power grid segmentation. A Boundary Awareness Module (BAM) and Channel-Spatial Hybrid Attention (CBAM) are embedded in the Mask Decoder, paired with a cross-entropy-edge hybrid loss function that strengthens the boundary segmentation accuracy of small targets by constraining both boundary precision and recall. Ablation studies and comparative experiments with different algorithms validate the effectiveness and advancement of the proposed enhanced SAM model for aerial image power grid segmentation tasks.
A blockchain-powered demand response system is suggested to mitigate the minimal consumer engagement and insufficient transparency in existing demand response (DR) programmers. The framework is deployed on Hyperledger Fabric, within which to automate settlements and secure data exchange between stakeholders the smart contracts are implemented. Simulation in MATLAB is employed to analyze the strategy, which demonstrates that the peak demand decreases by 15
This paper presents a dynamic, privacy-preserving context-aware geofencing system designed to support indoor climate data access and control in IoT-enabled buildings. The approach addresses two interrelated challenges: safeguarding sensitive indoor climate data in compliance with the General Data Protection Regulation’s (GDPR) data minimization and purpose limitation requirements, and enabling responsive environmental control based on user presence without relying on continuous tracking. During installation, each IoT device is registered via a mobile application that automatically collects spatial metadata, including GPS coordinates, which are clustered and aggregated into convex hulls representing building footprints. These geofences govern both data access and control logic, with evaluations occurring locally on users’ mobile devices to preserve privacy. The system architecture integrates a Java-Spring-MongoDB backend with an Angular-based dashboard and supports real-time visualization using Leaflet, OpenStreetMap, and Three.js. Evaluations at two case study sites, a mid-sized Danish university building and a large Malaysian library, demonstrate accurate spatial modeling, responsive access control enforcement, and successful actuation of climate control systems based on geofence crossings. Results show that anticipatory thermal preconditioning and energy-saving setbacks can be triggered reliably by geofence events, confirming the viability of location-based automation as a privacy-aware control mechanism for smart buildings.
Large Language Models-based AI Agents (LLM Agents) are capable of replicating human-like intelligent behaviors such as reasoning, planning, decision-making and executing actions across various environments. Recent studies have demonstrated the effectiveness of applying LLM Agents to building energy systems, enhancing automation, streamlining information processing, and supporting decision-making processes while reducing the need for manual intervention and domain-specific expertise. However, the fundamental challenge of how physical building systems are properly textualized so that LLMs can process them remains largely unaddressed. This paper analyzes how various LLM applications in the building and energy domains represent and textualize their physical system targets, based on a preliminary review of recent literature. The study reveals that most current applications rely on custom, simple, unstructured text-based representations. In contrast, a number of existing works have adopted ontology-based representations, which introduce formal semantic graphs that can help integrate heterogeneous information. Building on this observation, the paper highlights ontology-based approaches as a promising direction for enhancing LLM-building interactions.
The management of electrical infrastructure maintenance has evolved significantly with the integration of artificial intelligence and machine learning technologies. This paper presents a comprehensive hybrid framework combining rule-based defect prioritization with machine learning-based failure prediction for electrical substation maintenance optimization. Using historical defect data spanning 2023–2024 from Malaysia’s Pantai Timur region encompassing Kelantan, Terengganu, and Pahang states, we developed a two-phase system: (1) a transparent multi-criteria rule-based engine for immediate defect triage, and (2) ensemble machine learning models for monthly failure forecasting. The rule-based component processes switchgear defects using business logic derived from equipment type, defect location, and severity indicators. For predictive capabilities, we compared XGBoost and Random Forest classifiers on severely imbalanced data (88.40
Energy-intensive manufacturing processes, such as induction melting, suffer from limited visibility into operational variability, hindering efforts to optimize energy use and productivity. Existing studies primarily focus on structural improvements or single-variable analysis, offering little support for dynamic, data-driven decision-making. This paper addresses this critical gap by proposing a novel framework that integrates full-cycle segmentation, unsupervised clustering, and multi-criteria decision-making (MCDM) to systematically discover, evaluate, and benchmark operational patterns in industrial melting. Unlike prior work that isolates temperature or energy profiles, the proposed method fuses high-frequency power, weight, and temperature signals to identify melting cycles using correlation-informed segmentation. K-means clustering is applied to full-process feature vectors, enabling interpretable groupings based on melting rate and energy-specific consumption. These clusters are then evaluated using an ensemble of MCDM techniques (SAW, TOPSIS, VIKOR) across both performance and operational dimensions. The framework is validated on a 16-month dataset from a Danish foundry, comprising over 3,400 melting cycles. Results reveal five distinct operational patterns and show that standardizing operations toward the top-performing cluster can reduce energy consumption by 9.3
