Abstract

Charging infrastructure is seen as the catalyst for electric vehicle (EV) uptake, which is essential to reach Net Zero targets. However, access to public charging and deployment of charging infrastructure in the areas most needed are key to increase EV ownership. This paper presents a framework based on K-means clustering and spatial analytics for the equitable deployment planning of EV charging infrastructure within South East Wales—a geographical region in the United Kingdom with low public charger deployment. Results show that there is a public EV charging divide between rural and urban communities in the region and, thus, a greater effort is needed to bridge the gap between these communities.

Abstract

Predicting and optimizing the performance of proton exchange membrane fuel cells (PEMFCs) poses significant challenges because of the large number of design variables and their intricate interdependencies over the full operating voltage range. In this study, with the power of an artificial neural network (ANN), an ANN-based framework is proposed to perform prediction and optimization of PEMFC performance efficiently. A comprehensive dataset is produced using a validated multiphysics model, enabling robust training of ML models. An ANN achieves high predictive accuracy (R² > 0.999). Results show that with this framework, predictions agree well with experimental data, with predictive error below 8.1%. Furthermore, the framework enables quick identification of optimal design variables for multi-objective optimization. Overall, this work highlights an ANN-based research paradigm for designing efficient, application-tailored fuel cell systems, thereby reducing experiments and shortening design cycles.

Abstract

Integrating Large Language Models (LLMs) into home energy management systems promises personalized, autonomous energy scheduling. However, reliance on cloud-based LLMs introduces severe privacy vulnerabilities regarding household data and critical latency bottlenecks for control loops. While deploying edge LLMs offers a localized solution, their human-computer interaction (HCI) capabilities in continuous, dynamic physical environments remain insufficiently evaluated. This work proposes a novel evaluation framework bridging this gap. We develop a context-aware scenario generator to create physically grounded HCI testbenches, and introduce an event-driven “LLM-as-a-Judge” sandbox that merges subjective semantic scoring with objective, solver-derived metrics. Case studies evaluate various sub-12B edge models, quantifying their intent comprehension and physical modeling boundaries. This research provides essential insights into the capability limits of edge intelligence, laying a foundation for their reliable deployment in privacy-sensitive energy systems.

Abstract

Residential modular buildings require Heating, Ventilation, and Air Conditioning (HVAC) control strategies that can be deployed rapidly while maintaining reliable performance under dynamic operating conditions. This need is intensified by rising cooling demand, stricter Indoor Air Quality (IAQ) requirements, and growing interest in on-site Photovoltaic (PV) systems for net-zero-oriented operation. This study proposes a plug-and-play multi-objective HVAC control framework for PV-integrated residential modular buildings based on a Multi-Objective Markov Decision Process (MOMDP) and Envelope Q-Learning (EQL). A residential modular case study is evaluated in whole-year simulation. EQL outperformed scalarized Deep Reinforcement Learning (DRL) and Multi-Objective Model Predictive Control (MOMPC) across all tested preference settings. Under the balanced preference, annual HVAC grid import was reduced to 261 kWh, versus 767 kWh for Scalarized DRL and 1028 kWh for MOMPC. Under the energy preference, grid import fell further to 150 kWh. In the balanced case, EQL also improved cumulative energy, comfort, and IAQ rewards by 61.4%, 58.6%, and 21.1% over Scalarized DRL, and by 71.6%, 81.0%, and 23.0% over MOMPC. These results show strong potential for robust, scalable, and net-zero-oriented HVAC control in PV-integrated modular buildings.

Abstract

Blockages in air-conditioning water pipe networks can increase hydraulic resistance, reduce energy distribution efficiency, and threaten the safe operation of air-conditioning systems. However, existing studies mainly focus on leakage localization in pipe networks, while research on blockage localization remains limited. To address the difficulty of obtaining sufficient fault data from real systems, this study proposes a model-data fusion-driven method for blockage localization in air-conditioning water pipe networks. In the proposed framework, measurement data from the supervisory control and data acquisition system are first used to calibrate the hydraulic model through impedance identification using a genetic algorithm. The calibrated model is then employed to simulate multiple blockage scenarios and generate a dataset for training a multilayer perceptron model. During diagnosis, the pressure variation rates of sensors before and after faults are used as input features to predict the most likely blockage components, and a candidate blockage component set based on the Top-k predicted probabilities is introduced to improve localization reliability. A case study shows that the deviations between simulated and measured values of the calibrated hydraulic model are within 1%. The trained multilayer perceptron model achieves a blockage localization accuracy of 90.52%. When the candidate set contains the two components with the highest predicted probabilities, the coverage of the actual blockage component increases by 6.02%. Expanding the candidate set to the three components with the highest predicted probabilities further improves the coverage by 1.76%. These results demonstrate that the proposed model-data fusion framework can effectively localize blockage components and support intelligent operation and maintenance of air-conditioning systems.

Abstract

Currently, geopolitical conflicts are escalating, with key energy channels such as the Strait of Hormuz facing disruptions in navigation and sharp fluctuations in energy prices. Geopolitical risks have become a core variable threatening the stability, smoothness, and efficiency of global energy supply. Traditional energy scenario analysis methods suffer from lagging responses and cannot timely provide support for energy system risk response. This paper focuses on the potential risks to energy supply under geopolitical risks, introduces artificial intelligence technology to construct an analytical model in energy scenario analysis to achieve rapid construction and efficient analysis of scenarios, and centers on designing an energy scenario framework from the three dimensions of “supply, efficiency, smoothness”. Integrating geopolitical conflict-related information, energy supply data, international oil price data, shipping data, and market transaction data, an integrated scenario analysis model is built relying on artificial intelligence technology, replacing the cumbersome traditional manual modeling and calculation processes to quickly construct three core scenarios: the supply guarantee scenario, the efficiency guarantee scenario, and the smoothness guarantee scenario. Through the model, the stability, economic efficiency, and circulation efficiency of energy supply under different scenarios are quickly quantified and deduced, and the applicable scenarios and implementation priorities of each scenario are clarified. The research results show that introducing artificial intelligence technology into energy scenario analysis can effectively simplify the modeling process, improve the speed and accuracy of scenario analysis, and quickly respond to the dynamic changes of geopolitical risks; the three-dimensional scenario of ” supply, efficiency, smoothness” can achieve the multi-objective balance of the energy system under geopolitical risks. The research conclusions can provide accurate decision-making reference and technical support for energy enterprises to formulate risk response strategies and for governments to improve energy security policies.

Abstract

Residential heating electrification is central to the
UK’s 2050 net-zero target but poses significant
challenges to local distribution networks. This study
develops a data-driven framework to assess the optimal
heating electrification pathways and the associated
network reinforcement. The model combes half-hourly
heat demand profiles at LSOA level with allocated
primary substation capacity to capture network
reinforcements requirements with spatial and temporal
data. Two electrification scenarios for 2035 and 2050 are
assessed under network-constrained and network
unconstrained conditions. Results of worst scenarios
indicate that achieving 2050 targets may require up to
fourfold increases in local network capacity, while heat
pump adoption can reduce this by around half. The
framework enables grid-aware assessment of local
electrification pathways and supports more informed
network planning and policy decisions.

Abstract

Accurate identification of high consequence areas (HCAs) is critical for oil and gas pipeline safety management, because the final grading depends on whether building exposure evidence within the pipeline buffer zone remains reliable from image interpretation to engineering rule evaluation. This paper presents a complete workflow that converts building recognition results from unmanned aerial vehicle (UAV) corridor imagery into structured engineering evidence for rule-based HCA grading. Building masks predicted by a deep segmentation model are restored to geographic coordinates, vectorized into footprint polygons, and intersected with the pipeline buffer. Each footprint is then combined with digital surface model (DSM) data to estimate building height, floor count, household number, and population. Point-of-interest (POI) records are spatially filtered and classified into hazardous-facility and sensitive-place triggers through keyword matching. The fused evidence vector, including building count, estimated households, population density, and POI triggers, is evaluated by a configurable rule operator that outputs both an HCA grade and its decision reason. Two decision-oriented reliability measures, evidence margin and grade-stability index, are further introduced to quantify how close a segment is to a rule boundary and whether plausible evidence perturbations would change the final grade. The proposed method provides a traceable pathway from UAV imagery to auditable HCA grading, supporting differentiated inspection planning in pipeline integrity management.

Abstract

In existing buildings, the efficiency of air-conditioning water systems is frequently constrained by flaws in de-sign, installation, and control, resulting in substantial en-ergy waste and poor operational performance. This re-search centers on the severe hydraulic imbalance in the main branches of such systems within an industrial sys-tem. It usually causes inadequate cooling capacity in the coils located upstream of AHUs. Additionally, the lack of permanent sensors complicates on-site hydraulic balance adjustments. To resolve these challenges, a hydraulic balancing method based on simulation technology is pro-posed. Specifically, a physics-based simulation model is constructed in MWorks using the Modelica language. Validated against historical operational data, the model exhibits relative errors of less than 3% for flow rate and 6% for pressure drop, thereby showing excellent con-sistency with the actual hydraulic characteristics of the system. A PID-based control strategy is integrated into the simulation model to calculate target valve openings for key branches. These computed values offer direct, quantitative guidance for on-site manual adjustments, eliminating the need for a permanent sensor network. Moreover, AHUs continue to operate normally even when the pump frequency is reduced by 2 Hz, which indi-cates enhanced overall system performance and energy efficiency. This study presents a practical, low-cost digital twin approach for retrofitting existing HVAC systems, underscoring the effectiveness of Modelica-based physi-cal modeling in the analysis and control of intelligent en-ergy systems.

Abstract

Fault diagnosis is essential for safe thermal power system operation. Large language models (LLMs) can process complex industrial logs and provide diagnostic insights, making them a promising approach for fault diagnosis. However, conventional LLMs often produce physical hallucinations and logical inconsistencies, since physical knowledge and fault-specific patterns are injected together in a single training stage. This paper proposes a physics-augmented curriculum learning (PACL) method that uses a two-stage curriculum to Integrate physical laws into diagnostic reasoning. The method includes two steps: physics-field chain-of-thought-driven knowledge extraction and diagnostic alignment to fine-tune the model. Multi-source documents from Hanglian thermal power plant in Hangzhou, China are utilized, including fault cases, equipment regulations, and technical manuals. Evaluations on representative fault scenarios show that the proposed PACL achieves an average diagnostic score of 92.0, outperforming SFT (47.8), Mixed-FT (53.5), and R-PACL (60.8), demonstrating expert-level reasoning in complex scenarios.