Energy and AI最新文献

{"title":"Reinforcement learning-based energy management for hybrid electric vehicles: A comprehensive up-to-date review on methods, challenges, and research gaps","authors":"Mohamed Nadir Boukoberine ,&nbsp;Muhammad Fahad Zia ,&nbsp;Tarek Berghout ,&nbsp;Mohamed Benbouzid","doi":"10.1016/j.egyai.2025.100514","DOIUrl":"10.1016/j.egyai.2025.100514","url":null,"abstract":"<div><div>Reinforcement learning is widely used for control applications and has also been successfully implemented for efficient energy management within hybrid electric vehicles. Reinforcement learning algorithms offer various advantages, including fast convergence, broad applicability, stability, and robustness, particularly with the integration of deep and transfer learning. This paper provides a comprehensive understanding of reinforcement learning principles and a critical review of various reinforcement learning methods, states, actions, and rewards used to optimize the energy management performance of hybrid electric vehicles. Furthermore, the advantages and limitations of these algorithms are also discussed. This review reveals that deep reinforcement learning techniques show superior performance in handling complex energy management tasks thanks to their ability to learn from high-dimensional state spaces. Nevertheless, their implementation faces notable obstacles, including computational complexity and generalization across diverse driving conditions. Finally, key research directions for future work and challenges are highlighted in the domain of reinforcement-learning-based hybrid electric vehicle energy management.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"21 ","pages":"Article 100514"},"PeriodicalIF":9.6,"publicationDate":"2025-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144070771","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Leveraging neural networks to optimize heliostat field aiming strategies in Concentrating Solar Power Tower plants","authors":"Antonio Alcántara ,&nbsp;Pablo Diaz-Cachinero ,&nbsp;Alberto Sánchez-González ,&nbsp;Carlos Ruiz","doi":"10.1016/j.egyai.2025.100520","DOIUrl":"10.1016/j.egyai.2025.100520","url":null,"abstract":"<div><div>Concentrating Solar Power Tower (CSPT) plants rely on heliostat fields to focus sunlight onto a central receiver. Although simple aiming strategies, such as directing all heliostats to the receiver’s equator, can maximize energy collection, they often result in uneven flux distributions that cause hotspots, thermal stresses, and reduced receiver lifetimes. This paper presents a novel, data-driven approach that combines constraint learning, neural network-based surrogates, and mathematical optimization to address these challenges. The methodology learns complex heliostat-to-receiver flux interactions from simulation data and embeds the resulting surrogate model in a tractable optimization framework. By maximizing a tailored quality score that balances energy collection with flux uniformity, the approach produces smoothly distributed flux profiles and mitigates excessive thermal peaks. An iterative refinement process, guided by a trust region strategy and progressive data sampling, ensures continual improvement of the surrogate model by exploring new solution spaces at each iteration. Results from a real CSPT case study show that the proposed approach outperforms conventional heuristic methods, delivering flatter flux distributions with nearly a 10% reduction in peak values and safer thermal conditions (reflected by up to a 50% decrease in deviations from safe concentration distributions), without significantly compromising overall energy capture.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"21 ","pages":"Article 100520"},"PeriodicalIF":9.6,"publicationDate":"2025-05-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143949077","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Federated reinforcement learning for sustainable and cost-efficient energy management","authors":"J. Sievers ,&nbsp;P. Henrich ,&nbsp;M. Beichter ,&nbsp;R. Mikut ,&nbsp;V. Hagenmeyer ,&nbsp;T. Blank ,&nbsp;F. Simon","doi":"10.1016/j.egyai.2025.100521","DOIUrl":"10.1016/j.egyai.2025.100521","url":null,"abstract":"<div><div>Integrating renewable energy sources into the electricity grid introduces volatility and complexity, requiring advanced energy management systems. By optimizing the charging and discharging behavior of a building’s battery system, reinforcement learning effectively provides flexibility, managing volatile energy demand, dynamic pricing, and photovoltaic output to maximize rewards. However, the effectiveness of reinforcement learning is often hindered by limited access to training data due to privacy concerns, unstable training processes, and challenges in generalizing to different household conditions. In this study, we propose a novel federated framework for reinforcement learning in energy management systems. By enabling local model training on private data and aggregating only model parameters on a global server, this approach not only preserves privacy but also improves model generalization and robustness under varying household conditions, while decreasing electricity costs and emissions per building. For a comprehensive benchmark, we compare standard reinforcement learning with our federated approach and include mixed integer programming and rule-based systems. Among the reinforcement learning methods, deep deterministic policy gradient performed best on the Ausgrid dataset, with federated learning reducing costs by 5.01<!--> <!-->% and emissions by 4.60<!--> <!-->%. Federated learning also improved zero-shot performance for unseen buildings, reducing costs by 5.11<!--> <!-->% and emissions by 5.55<!--> <!-->%. Thus, our findings highlight the potential of federated reinforcement learning to enhance energy management systems by balancing privacy, sustainability, and efficiency.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"21 ","pages":"Article 100521"},"PeriodicalIF":9.6,"publicationDate":"2025-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144072497","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A holistic multimodal approach for real-time anomaly detection and classification in large-scale photovoltaic plants","authors":"Zoubir Barraz ,&nbsp;Imane Sebari ,&nbsp;Hicham Oufettoul ,&nbsp;Kenza Ait el kadi ,&nbsp;Nassim Lamrini ,&nbsp;Ibtihal Ait Abdelmoula","doi":"10.1016/j.egyai.2025.100525","DOIUrl":"10.1016/j.egyai.2025.100525","url":null,"abstract":"<div><div>This paper presents a holistic multimodal approach for real-time anomaly detection and classification in large-scale photovoltaic plants. The approach encompasses segmentation, geolocation, and classification of individual photovoltaic modules. A fine-tuned Yolov7 model was trained for the individual module’s segmentation of both modalities; RGB and IR images. The localization of individual solar panels relies on photogrammetric measurements to facilitate maintenance operations. The localization process also links extracted images of the same panel using their geographical coordinates and preprocesses them for the multimodal model input. The study also focuses on optimizing pre-trained models using Bayesian search to improve and fine-tune them with our dataset. The dataset was collected from different systems and technologies within our research platform. It has been curated into 1841 images and classified into five anomaly classes. Grad-CAM, an explainable AI tool, is utilized to compare the use of multimodality to a single modality. Finally, for real-time optimization, the ONNX format was used to optimize the model further for deployment in real-time. The improved ConvNext-Tiny model performed well in both modalities, with 99 % precision, recall, and F1-score for binary classification and 85 % for multi-class classification. In terms of latency, the segmentation models have an inference time of 14 ms and 12 ms for RGB and IR images and 24 ms for detection and classification. The proposed holistic approach includes a built-in feedback loop to ensure the model’s robustness against domain shifts in the production environment.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"21 ","pages":"Article 100525"},"PeriodicalIF":9.6,"publicationDate":"2025-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143943269","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Comparative of control strategies on electrical vehicle fleet charging management strategies under uncertainties","authors":"Zhewei Zhang ,&nbsp;Rémy Rigo-Mariani ,&nbsp;Nouredine Hadjsaid","doi":"10.1016/j.egyai.2025.100522","DOIUrl":"10.1016/j.egyai.2025.100522","url":null,"abstract":"<div><div>The growing penetration of Electric Vehicles (EVs) in transportation brings challenges to power distribution systems due to uncertain usage patterns and increased peak loads. Effective EV fleet charging management strategies are needed to minimize network impacts, such as peak charging power. While existing studies have addressed uncertainties in future arrivals, they often overlook the uncertainties in user-provided inputs of current ongoing charging EVs, such as estimated departure time and energy demand. This paper analyzes the impact of these uncertainties and evaluates three management strategies: a baseline Model Predictive Control (MPC), a data-hybrid MPC, and a fully data-driven Deep Reinforcement Learning (DRL) approach. For data-hybrid MPC, we adopted a diffusion model to handle user input uncertainties and a Gaussian Mixture Model for modeling arrival/departure scenarios. Additionally, the DRL method is based on a Partially Observable Markov Decision Process (POMDP) to manage uncertainty and employs a Convolutional Neural Network (CNN) for feature extraction. Robustness tests under different user uncertainty levels show that the data hybrid MPC performs better on the baseline MPC by 20 %, while the DRL-based method achieves around 10 % improvement.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"21 ","pages":"Article 100522"},"PeriodicalIF":9.6,"publicationDate":"2025-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143937679","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Rescheduling costs and adaptive asymmetric errors guided closed-loop prediction of power loads in mine integrated energy systems","authors":"Qi Miao ,&nbsp;Xiaoyan Sun ,&nbsp;Chen Ma ,&nbsp;Yong Zhang ,&nbsp;Dunwei Gong","doi":"10.1016/j.egyai.2025.100516","DOIUrl":"10.1016/j.egyai.2025.100516","url":null,"abstract":"<div><div>The development of an integrated energy system for mining that efficiently recycles multiple resources is a crucial strategy for achieving dual carbon reduction targets in the mining sector. Precise load forecasting is fundamental to ensuring the safe and efficient scheduling of this system. However, existing studies often overlook the coupling between load forecasting and scheduling results, treating them independently, which frequently leads to high rescheduling costs due to forecasting errors. To address this issue, we propose a closed-loop load forecasting algorithm that incorporates rescheduling costs and asymmetric errors. We first proposed a data generation and model construction strategy by using real load, predicted load, and rescheduling costs to capture the relationship between load forecasting and rescheduling costs. Considering the different impacts of under-forecasting and over-forecasting on scheduling results, the rescheduling cost model is further integrated with asymmetric prediction errors to define the loss function of the Bi-LSTM based forecasting model. Additionally, an optimization strategy for self-tuning asymmetric prediction error fusion coefficients is designed to ensure the accuracy of load forecasting. The proposed algorithm is applied to the power load forecasting of an integrated energy system in a coal mine in Shanxi. The results demonstrate the effectiveness of the algorithm in reducing system rescheduling costs while ensuring forecasting accuracy, highlighting its potential application in power load forecasting for mine integrated energy systems.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"21 ","pages":"Article 100516"},"PeriodicalIF":9.6,"publicationDate":"2025-05-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144070770","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A review of machine learning techniques for building electrical energy consumption prediction","authors":"Yuyao Chen ,&nbsp;Wei Gong ,&nbsp;Christian Obrecht ,&nbsp;Frédéric Kuznik","doi":"10.1016/j.egyai.2025.100518","DOIUrl":"10.1016/j.egyai.2025.100518","url":null,"abstract":"<div><div>The ongoing energy transition, essential for mitigating global warming, stands to benefit significantly from advances in building energy consumption prediction. With the rise of big data, data-driven models have become increasingly effective in forecasting, with machine learning emerging as the most efficient method for constructing these predictive models. While previous reviews have typically listed various machine learning models for energy consumption prediction, they have often lacked a theoretical perspective explaining why certain models are suitable for different aspects of this domain. In contrast, this review introduces machine learning techniques based on their application phases, covering preprocessing techniques such as feature selection, extraction, and clustering, as well as state-of-the-art predictive models. We provide a comparative theoretical analysis of various models, examining their strengths, weaknesses, and suitability for different forecasting tasks. Additionally, we discuss spatial–temporal considerations in energy consumption forecasting, including the role of Graph Neural Networks and multitask learning. Furthermore, we address a significant challenge in the field, the difficulty of accurately predicting high-fluctuation electricity consumption, and propose potential solutions to tackle this issue.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"21 ","pages":"Article 100518"},"PeriodicalIF":9.6,"publicationDate":"2025-05-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143917537","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Revolutionizing clean energy labs: Robotic imitation learning for efficient fabrication AI-powered electrical units assembly platform","authors":"Xi Xu ,&nbsp;Yijun Gu ,&nbsp;Tianyi Zhang ,&nbsp;Jiwen Yu ,&nbsp;Stephen Skinner","doi":"10.1016/j.egyai.2025.100517","DOIUrl":"10.1016/j.egyai.2025.100517","url":null,"abstract":"<div><div>The energy industry, now in an era of digitization driven by computational design, is gradually moving towards automating the entire process from computational prediction to device assembly, aiming to minimize the reliance on time-consuming, manual trial-and-error validation. In this study, guided by computational density functional theory (DFT) predictions, a humanoid robotic arm, based on artificial intelligence (AI), was creatively utilized to assemble clean energy devices, solid oxide fuel cells (SOFCs). The material <figure><img></figure> (LBSF) was DFT-predicted to have high oxygen reduction reactions (ORRs) ability, suitable for the cathode in SOFCs compared to the conventional <figure><img></figure> (LSF). The material was made into ink then passed to the assembly platform with AI-driven robotics. AI-driven robotics was employed with an imitation learning method to effectively learn skills directly from human demonstrations, thereby alleviating researchers from labor-intensive tasks. We demonstrate our approach for autonomous SOFCs fabrication. For easy platform usage in the future, Large Language Models (LLMs) were incorporated to understand human commands. Visual information was captured by an RGBD camera to identify and locate the cathode painting spot. An imitation learning framework was then applied to learn the painting path from human operations and can be generalized to different conditions. The auto-fabricated single cells with the DFT-predicted LBSF cathode were tested and achieved a power density of 966 <span><math><mrow><mi>m</mi><mi>W</mi><mo>/</mo><mi>c</mi><msup><mrow><mi>m</mi></mrow><mrow><mn>2</mn></mrow></msup></mrow></math></span> at 700 <span><math><mrow><mo>°</mo><mi>C</mi></mrow></math></span>, more than double the performance of LSF. By integrating computational design with an AI-driven assembly platform, this study marks an initial step towards an AI-driven material lab, exponentially accelerating material design in the near future. The platform can also help disabled researchers achieve their ideas through the behavior cloning approach.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"21 ","pages":"Article 100517"},"PeriodicalIF":9.6,"publicationDate":"2025-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143917637","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep learning-based dual monitoring system for power forecasting and fault detection in nuclear power applications","authors":"Mingzhe Lyu ,&nbsp;Helin Gong ,&nbsp;Zhang Chen ,&nbsp;Jiangyu Wang ,&nbsp;Mingxiao Zhong ,&nbsp;Zhiyong Wang ,&nbsp;Qing Li ,&nbsp;Zefei Pan","doi":"10.1016/j.egyai.2025.100515","DOIUrl":"10.1016/j.egyai.2025.100515","url":null,"abstract":"<div><div>Monitoring key parameters in nuclear power plant control rooms is critical, as human errors can result in severe safety and operational consequences. This study proposes a hybrid framework for power prediction and fault detection that integrates multi-head self-attention mechanisms with long short-term memory networks, combined with a dual-monitoring system. The framework is evaluated using real-time data from two pressurized water reactor units (Units 5 and 6) under four realistic operational scenarios. In the most informative case, the model achieves a 56.6% reduction in root mean square error and a 36.8% reduction in mean absolute error, with a coefficient of determination (<span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span>) of 0.9924—significantly outperforming the next-best benchmark. For fault diagnosis, the dual-monitoring system reduces the false negative rate to 18.73% and improves recall to 81.27%, demonstrating strong anomaly detection under complex conditions. By combining short-term fluctuation sensitivity with long-term trend stability, the proposed approach offers a robust and generalizable solution for intelligent monitoring. These findings advance the development of artificial intelligence–enhanced systems for secure and efficient operation of critical energy infrastructure.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100515"},"PeriodicalIF":9.6,"publicationDate":"2025-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143904192","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Bayesian optimized LSTM-based sensor fault diagnosis of organic Rankine cycle system","authors":"Qiyao Zuo ,&nbsp;Pengcheng Liu ,&nbsp;Weijia Meng ,&nbsp;Xianyu Zeng ,&nbsp;Hua Li ,&nbsp;Xuan Wang ,&nbsp;Hua Tian ,&nbsp;Gequn Shu","doi":"10.1016/j.egyai.2025.100519","DOIUrl":"10.1016/j.egyai.2025.100519","url":null,"abstract":"<div><div>As the energy crisis intensifies, the organic Rankine cycle (ORC) is increasingly employed for efficient recovery of low-temperature waste heat. The operation of the ORC system necessitates the use of numerous sensors to monitor its status. Over time, these sensors may become faulty, rendering accurate and timely diagnosis is critical for proper and safe functioning of the ORC system. Currently, there is a lack of rapid diagnostic methods for sensor faults in ORC systems. This study establishes an ORC test bench utilizing cyclopentane as the working fluid. Experimental data incorporating induced faults from the ORC test bench is employed to train machine learning-based models for sensor fault diagnosis. The test results indicate that the diagnostic model developed in this study can accurately diagnose various sensor faults in the ORC system, thereby ensuring its safe operation. Notably, the method based on Bayesian-optimized long short-term memory network (BO-LSTM) achieved the highest diagnostic accuracy, reaching up to 95.92 %.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"21 ","pages":"Article 100519"},"PeriodicalIF":9.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143931340","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}