Energy and AI最新文献

{"title":"An uncertainty-aware bi-level multitask SqueezeNet for dynamic security assessment in power systems with focus on critical generator identification under small and imbalanced datasets","authors":"Sasan Azad ,&nbsp;Mohammad Taghi Ameli ,&nbsp;Amir Reza Shafieinejad ,&nbsp;Hossein Ameli ,&nbsp;Goran Strbac","doi":"10.1016/j.egyai.2025.100618","DOIUrl":"10.1016/j.egyai.2025.100618","url":null,"abstract":"<div><div>Deep learning (DL)-based methods in pre-fault dynamic security assessment (DSA) have provided significant results, contributing to the safe operation of power systems. However, power systems often suffer from insufficient, small, and imbalanced datasets, which significantly impact the performance of DL-based DSA models. Existing DSA frameworks typically operate as two-class black-box models, assessing only overall system security without providing insights into the causes of insecurity or identifying critical generators (CGs), and they fail to quantify prediction uncertainty. These challenges hinder the implementation of current methods in real-world power systems and reduce operators' confidence in them. To address these issues, this paper proposes an uncertainty-aware bi-level multitask learning framework based on transfer learning and SqueezeNet architecture. The framework assesses system security, identifies CGs during instability, and leverages fine-tuning of a pre-trained SqueezeNet model to facilitate training with limited data. Additionally, evidential deep learning is incorporated to quantify classification uncertainty. Without relying on the complex and challenging data augmentation method, this framework uses a simple technique called optimal classification threshold determination to mitigate the negative impact of imbalanced data on model performance. The optimal threshold is determined by maximizing the area under the receiver operating characteristic (ROC) curve. The application of the proposed method to the IEEE 118-bus system shows its strong performance. These results offer crucial technical insights for the implementation of DL-based DSA in real-world power systems.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100618"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145060520","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":"Enhancing wind speed prediction in wind farms through ordinal classification","authors":"A.M. Gómez-Orellana ,&nbsp;M. Vega-Bayo ,&nbsp;D. Guijo-Rubio ,&nbsp;J. Pérez-Aracil ,&nbsp;V.M. Vargas ,&nbsp;P.A. Gutiérrez ,&nbsp;L. Prieto-Godino ,&nbsp;S. Salcedo-Sanz ,&nbsp;C. Hervás-Martínez","doi":"10.1016/j.egyai.2025.100596","DOIUrl":"10.1016/j.egyai.2025.100596","url":null,"abstract":"<div><div>This paper presents and evaluates two novel ordinal classification methods for wind speed prediction, considering three prediction time-horizons: 1h, 4h, and 8h. To address the problem, wind speed values are discretised into four classes, critical for wind farm management. Each class represents essential information for wind farm production, ranging from very low wind speeds to extreme wind speed events and the corresponding production conditions, facilitating operational decisions for wind farm operators. Ordinal classifiers are more suitable than nominal methods to tackle this problem. The study’s primary objective is to compare recently proposed ordinal classifiers for addressing the challenges of wind speed prediction with a focus on extreme wind conditions, which are responsible for many turbine shutdowns. Hourly wind speed measurements from a Spanish wind farm and predictor variables from the European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA5 Reanalysis) model are used. The proposed methods include an Artificial Neural Network (ANN) model implementing the Cumulative Link Model as an ordinal output function (<span><math><msup><mrow><mtext>MLP-CLM</mtext></mrow><mrow><mtext>O</mtext></mrow></msup></math></span>), which emphasises overall performance, and an ANN model optimised using a soft labelling technique based on triangular distributions (<span><math><msup><mrow><mtext>MLP-T</mtext></mrow><mrow><mtext>O</mtext></mrow></msup></math></span>), which excels at handling extreme class performance. The results demonstrate the superiority of both approaches over other nominal and ordinal methods across performance metrics that account for the unbalanced nature and ordinality of the data. <span><math><msup><mrow><mtext>MLP-CLM</mtext></mrow><mrow><mtext>O</mtext></mrow></msup></math></span> excels in overall and ordinal performance, while <span><math><msup><mrow><mtext>MLP-T</mtext></mrow><mrow><mtext>O</mtext></mrow></msup></math></span> demonstrates superior handling of the extreme class predictions.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100596"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144908071","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":"Opening the AI black-box: Symbolic regression with Kolmogorov–Arnold Networks for advanced energy applications","authors":"Nataly R. Panczyk,&nbsp;Omer F. Erdem,&nbsp;Majdi I. Radaideh","doi":"10.1016/j.egyai.2025.100595","DOIUrl":"10.1016/j.egyai.2025.100595","url":null,"abstract":"<div><div>While most modern machine learning methods offer speed and accuracy, few promise interpretability or explainability—two key features necessary for highly sensitive industries, like medicine, finance, and engineering. Using eight datasets representative of one especially sensitive industry, nuclear power, this work compares a traditional feedforward neural network (FNN) to a Kolmogorov–Arnold Network (KAN). We consider not only model performance and accuracy, but also interpretability through model architecture and explainability through a post-hoc SHapley Additive exPlanations (SHAP) analysis, a game-theory-based feature importance method. In terms of accuracy, we find KANs and FNNs comparable across all datasets when output dimensionality is limited. KANs, which transform into symbolic equations after training, yield perfectly interpretable models, while FNNs remain black-boxes. Finally, using the post-hoc explainability results from Kernel SHAP, we find that KANs learn real, physical relations from experimental data, while FNNs simply produce statistically accurate results. Overall, this analysis finds KANs a promising alternative to traditional machine learning methods, particularly in applications requiring both accuracy <em>and</em> comprehensibility.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100595"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145045870","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":"Enhancing energy control stability under extreme conditions by integrating weather forecasts into ARLEM","authors":"Vahid M. Nik","doi":"10.1016/j.egyai.2025.100617","DOIUrl":"10.1016/j.egyai.2025.100617","url":null,"abstract":"<div><div>Increasing the stability of energy management systems is crucial, especially when facing extreme weather events. A common approach to prevent sudden oscillations is the implementation of predictive models; however, this often requires more complex models and greater computational power. In this work, weather forecasts are integrated into an approach based on Adaptive Reinforcement Learning for Energy Management (ARLEM). Since ARLEM is an online model-free reward-based RL method, it does not need any form of (predictive) modelling and weather forecasts serve as an additional source of information about the agent’s environment. The impacts of incorporating weather forecasts into ARLEM are investigated for a typical urban neighborhood in Stockholm during a winter with two cold waves and considering 17 future climate scenarios for the period of 2040–2069. Multiple adaptive policy schemes are assessed considering four forecast horizons of 3, 6, 12, and 24 h. Results show that integrating weather forecasts into decision-making can significantly reduce fluctuations in the control system, leading to more stable energy management. Moreover, it enhances energy efficiency by reducing both average and peak demands, particularly during extreme weather events. Overall, this contributes to improving the climate resilience of energy systems.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100617"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145060519","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":"Imputing the long-term missing heating load data using a generative network","authors":"Mengbo Yu,&nbsp;Alexander Neubauer,&nbsp;Pedram Babakhani,&nbsp;Stefan Brandt,&nbsp;Martin Kriegel","doi":"10.1016/j.egyai.2025.100637","DOIUrl":"10.1016/j.egyai.2025.100637","url":null,"abstract":"<div><div>Accurately filling in missing heating data is essential for ensuring data quality in applications such as energy management optimization and building efficiency analysis. Traditional machine learning methods use historical heating data as an input feature to predict the following missing data. However, when the duration of missing data is long, previous estimated values are inevitably used for further imputation, leading to error accumulation and a growing deviation from true values. To overcome this problem, this paper proposes a generative network that can fill missing data solely based on weather and temporal data, without using previous imputed values for further imputation. Our method outperformed the state of the art such as Seq2seq and Transformer, achieving relative normalized root mean square error (NRMSE) reductions of 1.65<!--> <!-->% to 41.38<!--> <!-->%, 0.30<!--> <!-->% to 66.43<!--> <!-->%, and 14.84<!--> <!-->% to 50.22<!--> <!-->% across three different data sources. In addition, with our proposed method, the effect of selecting different weather variables on model performance, and the benefits of transfer learning under limited data were also demonstrated. The relative NRMSE reduction is between 3.88<!--> <!-->% to 15.85<!--> <!-->% in cold months and from 7.49<!--> <!-->% to 12.29<!--> <!-->% in warm months when applying transfer learning.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100637"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145416081","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 robust hybrid machine learning framework for short-term load forecasting: integrating multi-linear regression, long short-term memory, and feed-forward neural networks for enhanced accuracy and efficiency","authors":"Fareeduddin Mohammed,&nbsp;Ameni Boumaiza,&nbsp;Antonio Sanfilippo,&nbsp;Daniel Perez-Astudillo,&nbsp;Dunia Bachour","doi":"10.1016/j.egyai.2025.100625","DOIUrl":"10.1016/j.egyai.2025.100625","url":null,"abstract":"<div><div>Efficient energy management and grid stability strongly rely on accurate Short-Term Load Forecasting (STLF). Existing forecasting models, unfortunately, are often inaccurate and computationally demanding. To overcome these challenges, a novel hybrid model, combining both linear regression and machine learning techniques, is proposed in this study. The hybrid model, MLR-LSTM-FFNN, captures both temporal and non-linear dependencies in load data by integrating multi-linear regression (MLR) with long short-term memory (LSTM) networks and feed-forward neural networks (FFNN). Using datasets from Qatar, with 5 min, 15 min, 30 min, and 1 h time intervals and from Panama City with a 1 h interval, experiments were conducted to thoroughly test the robustness of the model. The results showed that the MLR-LSTM-FFNN hybrid model outperformed the baseline and state-of-the-art hybrid models for each of the datasets, in terms of lower RMSE, MAE, and MAPE values along with a faster training time. This superior performance across different datasets underscores the model’s scalability and reliability as an STLF approach, providing a practical solution to energy demand prediction tasks. The improvement in short-term forecasting accuracy provides utilities with a practical tool to optimize demand-side management, reduce operational costs, and enhance grid reliability.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100625"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145219589","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":"Machine-learning-accelerated screening of multi-element doped CuSb catalysts for enhanced C2+ selectivity in CO2 electroreduction","authors":"Xin Cheng ,&nbsp;Hang Wang ,&nbsp;Xun Zhu ,&nbsp;Yang Wang ,&nbsp;Qian Fu","doi":"10.1016/j.egyai.2025.100613","DOIUrl":"10.1016/j.egyai.2025.100613","url":null,"abstract":"<div><div>Electrochemical CO₂ reduction (CO₂RR) to value-added fuels and chemicals offers a promising route toward carbon neutrality. However, developing efficient and selective catalysts for the generation of multi-carbon (C₂₊) products remains a significant challenge. In this work, we propose a combined density functional theory (DFT) and machine learning (ML) approach to systematically screen CuSb-based catalysts with varying surface Sb atomic fractions and <span><span>non-metal dopants</span><svg><path></path></svg></span> (O, N, S, Se, and P) on the Cu₂Sb(100) surface for CO₂RR. Approximately 200 representative adsorption configurations were randomly selected for DFT calculations, which were then used to train a predictive ML model. This model enables high-accuracy predictions of the adsorption energies of key intermediates (*CO and *H) for the remaining uncalculated configurations. By integrating the K-means clustering analysis and the optimal adsorption energy selection criteria based on the Sabatier principle, the candidate configuration with the best potential for C₂₊ product formation was identified: O-doped CuSb with a surface Sb atomic fraction of 3/12. Mechanistic studies further reveal that O doping significantly strengthens *CO adsorption while suppressing *H adsorption by modulating the electronic structure, thereby lowering the CO₂RR energy barrier and improving the thermodynamic selectivity toward C₂₊ products. This work not only elucidates the synergistic effect of surface Sb atomic fraction and <span><span>non-metal dopants</span><svg><path></path></svg></span> on CO₂RR activity, but also establishes a scalable ML prediction and screening framework, providing theoretical support and methodological pathways for the design of high-performance CuSb-based catalysts.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100613"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145096104","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":"Reproducibility of machine learning-based fault detection and diagnosis for HVAC systems in buildings: An empirical study","authors":"Adil Mukhtar ,&nbsp;Michael Hadwiger ,&nbsp;Franz Wotawa ,&nbsp;Gerald Schweiger","doi":"10.1016/j.egyai.2025.100658","DOIUrl":"10.1016/j.egyai.2025.100658","url":null,"abstract":"<div><div>Reproducibility is a cornerstone of credible scientific research. The topic gained prominence in fields such as psychology, medicine and artificial intelligence where concerns about non-replicable results sparked ongoing discussions about research practices. However, its status within machine learning for building systems is underexamined. Therefore, this work contributes to closing this gap by analyzing the reproducibility of machine learning-based fault detection and diagnosis studies published over the past decade. We found that nearly all articles are not reproducible due to insufficient disclosure across key dimensions of reproducibility. Notably, 72% of the articles do not specify whether the dataset used is public, proprietary, or commercially available. Only two papers share a link to their code, one of which was broken. Two-thirds of the publications were authored exclusively by academic researchers, yet no significant differences in reproducibility were observed compared to publications with industry-affiliated authors. These findings highlight the need for targeted interventions, including reproducibility guidelines, training for researchers, and policies by journals and conferences that promote transparency and reproducibility.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100658"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145680885","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":"Kinetic-model identification in metal-hydride reactions using neural network autoencoder surrogate models","authors":"André Martins Neves ,&nbsp;Willi Großmann ,&nbsp;Julián Atílio Puskiel ,&nbsp;Jan Warfsmann ,&nbsp;Vahid Reza Hosseini ,&nbsp;Maximilian Passing ,&nbsp;Thomas Carraro ,&nbsp;Thomas Klassen ,&nbsp;Oliver Niggemann ,&nbsp;Julian Jepsen","doi":"10.1016/j.egyai.2025.100659","DOIUrl":"10.1016/j.egyai.2025.100659","url":null,"abstract":"<div><div>Solid-state hydrides can reversibly absorb and desorb H₂ under comparatively mild temperature and pressure conditions, making them promising candidates for H₂ storage in renewable energy applications. The underlying gas-solid interactions are complex and involve multiple intermediary steps. Because they occur in series, by fitting experimental data employing several proposed models, it is possible to identify the rate-limiting step of the reaction, driving the development of new catalysts and the design of H₂-storage systems. The corresponding state-of-the-art method for model identification is the reduced-time method (RTM), which is time-consuming and often yields inconclusive results. To overcome these limitations and to facilitate automatization, this work proposes a framework with 12 unsupervised neural networks (NNs) which are trained using simulated curves from selected kinetic models. These networks are applied to a dataset of 144 experimental kinetic curves of an AB₂ hydride-forming alloy as a blueprint material. Each NN attempts to reconstruct the input data, and the model with the lowest reconstruction loss is selected. The machine learning algorithm achieved a match of 97% and 91% for the absorption/desorption curves compared to the benchmark. Both reactions follow predominantly the Avrami-Erofeyev model with exponents (<span><math><mi>n</mi></math></span>) between 0.8 and 0.9. The kinetic constants (<span><math><mi>k</mi></math></span>) derived from the assigned model are used to simulate kinetic curves, showing excellent agreement with experimental data and RTM results. The proposed method provides an advantageous approach that can be applied to most gas-solid or even solid-solid reactions.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100659"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145680877","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":"BatteryTSFM: Generalizable long-horizon degradation prediction across conditions and chemistries via time series foundation models","authors":"Zimeng Fan ,&nbsp;Yuting Guo ,&nbsp;Lei Song ,&nbsp;Junrong Du ,&nbsp;Hongfei Wang ,&nbsp;Xuzhi Li","doi":"10.1016/j.egyai.2025.100646","DOIUrl":"10.1016/j.egyai.2025.100646","url":null,"abstract":"<div><div>Machine learning has demonstrated remarkable breakthroughs in predicting the state of health (SOH) for lithium-ion batteries. However, conventional methods face critical challenges in cross-domain adaptation, inter-dataset generalization, and long-horizon forecasting due to variations in usage conditions and electrochemical characteristics. Inspired by the success of large language models (LLMs), time-series foundation models (TSFMs) offer an alternative solution to overcome the issues above. Nevertheless, studies to explore the generalization enhancement capability of TSFMs for battery SOH forecasting under different cross domain factors remain insufficient. Therefore, a novel TSFMs based framework named BatteryTSFM is proposed for SOH forecasting. First, we introduce backbone-aware temporal resampling that dynamically adapts preprocessing to structural characteristics of diverse TSFMs, enabling optimal cross-domain generalization through feature scaling. Second, Monte Carlo dropout is integrated into autoregressive inference to quantify the multi-step prediction errors. Across four public datasets, BatteryTSFM reduces RMSE by an average of 35% in cross-condition tasks and 88% in cross-chemistry tasks, indicating that foundation-model methods can deliver reliable long-horizon SOH forecasts for energy systems. We also conduct exploratory analyses that link generalization to fine-tuning dataset size and resampling granularity, yielding practical guidance for deployment.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100646"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145680995","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}