Energy and AI最新文献_第5页

Assessing long-term electricity market design for ambitious decarbonization targets using multi-agent reinforcement learning 使用多智能体强化学习评估雄心勃勃的脱碳目标的长期电力市场设计

IF 9.6

Energy and AI Pub Date : 2026-01-01 Epub Date: 2025-12-18 DOI: 10.1016/j.egyai.2025.100665

Javier Gonzalez-Ruiz , Carlos Rodriguez-Pardo , Iacopo Savelli , Alice Di Bella , Massimo Tavoni

{"title":"Assessing long-term electricity market design for ambitious decarbonization targets using multi-agent reinforcement learning","authors":"Javier Gonzalez-Ruiz , Carlos Rodriguez-Pardo , Iacopo Savelli , Alice Di Bella , Massimo Tavoni","doi":"10.1016/j.egyai.2025.100665","DOIUrl":"10.1016/j.egyai.2025.100665","url":null,"abstract":"<div><div>Electricity systems are key to transforming today’s society into a carbon-free economy. Long-term electricity market mechanisms, including auctions, support schemes, and other policy instruments, are critical in shaping the electricity generation mix. In light of the need for more advanced tools to support policymakers and other stakeholders in designing, testing, and evaluating long-term markets, this work presents a multi-agent reinforcement learning model capable of capturing the key features of decarbonizing energy systems. Profit-maximizing generation companies make investment decisions in the wholesale electricity market, responding to system needs, competitive dynamics, and policy signals. The model employs independent proximal policy optimization, which was selected for suitability to the decentralized and competitive environment. Nevertheless, given the inherent challenges of independent learning in multi-agent settings, an extensive hyperparameter search ensures that decentralized training yields market outcomes consistent with competitive behavior. The model is applied to a stylized version of the Italian electricity system and tested under varying levels of competition, market designs, and policy scenarios. Results highlight the critical role of market design for decarbonizing the electricity sector and avoiding price volatility. The proposed framework allows assessing long-term electricity markets in which multiple policy and market mechanisms interact simultaneously, with market participants responding and adapting to decarbonization pathways.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100665"},"PeriodicalIF":9.6,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926797","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}

引用次数: 0

TRACE: Time series representation learning with contrastive embeddings for anomaly detection in photovoltaic systems TRACE：光伏系统异常检测的对比嵌入时间序列表示学习

IF 9.6

Energy and AI Pub Date : 2026-01-01 Epub Date: 2025-12-18 DOI: 10.1016/j.egyai.2025.100670

Chandana Priya Nivarthi, Zhixin Huang, Christian Gruhl, Bernhard Sick

{"title":"TRACE: Time series representation learning with contrastive embeddings for anomaly detection in photovoltaic systems","authors":"Chandana Priya Nivarthi, Zhixin Huang, Christian Gruhl, Bernhard Sick","doi":"10.1016/j.egyai.2025.100670","DOIUrl":"10.1016/j.egyai.2025.100670","url":null,"abstract":"<div><div>Reliable anomaly detection in photovoltaic (PV) inverters is critical for ensuring operational efficiency and reducing maintenance costs in renewable energy systems. We introduce TRACE (<strong>T</strong>ime series <strong>R</strong>epresentation learning with <strong>A</strong>utoencoder-based <strong>C</strong>ontrastive <strong>E</strong>mbeddings), a self-supervised contrastive learning framework for multivariate time series anomaly detection in PV systems. TRACE employs a two-stage architecture: autoencoder-based representation learning with interchangeable backbones followed by contrastive training through a Siamese network. The framework generates semantically coherent augmentations by perturbing autoencoder reconstructions and applies three negative mining strategies to create challenging contrastive pairs. Comprehensive experiments on a real-world PV inverter dataset and two industrial benchmarks demonstrate TRACE’s superiority. Autoencoder-based augmentations deliver a 21.3% relative improvement in mean F1 (0.616 vs. 0.508) over traditional perturbation methods, with TransformerAE emerging as the optimal backbone architecture. While negative sampling strategies show dataset-specific advantages, their impact remains secondary to encoder capacity. TRACE with TransformerAE and reconstruction-error negatives consistently outperforms fourteen state-of-the-art time series anomaly detection methods, achieving highest F1 scores on all the three datasets while maintaining exceptional precision up to 0.99. Visualization analysis confirms TRACE’s capacity for early fault detection up to three days before failure and interpretable embedding separation. The framework addresses the fundamental challenge of label scarcity in industrial monitoring through self-supervised learning, providing a practical and transparent solution for predictive maintenance in PV systems and broader industrial applications.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100670"},"PeriodicalIF":9.6,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926792","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}

引用次数: 0

Towards secure federated learning for energy forecasting under adversarial attacks 面向对抗性攻击下能源预测的安全联邦学习

IF 9.6

Energy and AI Pub Date : 2026-01-01 Epub Date: 2026-01-17 DOI: 10.1016/j.egyai.2026.100680

Jonas Sievers , Krupali Kumbhani , Thomas Blank , Frank Simon , Andreas Mauthe

{"title":"Towards secure federated learning for energy forecasting under adversarial attacks","authors":"Jonas Sievers , Krupali Kumbhani , Thomas Blank , Frank Simon , Andreas Mauthe","doi":"10.1016/j.egyai.2026.100680","DOIUrl":"10.1016/j.egyai.2026.100680","url":null,"abstract":"<div><div>Federated learning is increasingly used in energy forecasting, enabling buildings to collaboratively predict load, photovoltaic generation, and prosumption while preserving data privacy. However, this collaborative nature introduces new vulnerabilities, as manipulations by a single participant can propagate across the network. Such attacks can undermine grid balancing, limit flexibility provision, and reduce trust in decentralized energy systems. This work presents a comprehensive study of adversarial threats and defenses in federated energy forecasting. We compare structured manipulations generated with Generative Adversarial Networks against simple random perturbations in two attack scenarios: (i) data poisoning, where corrupted training data degrade global accuracy, and (ii) backdoors, where hidden triggers distort predictions in targeted time windows. Our experiments show that poisoning can increase global forecasting errors by up to 131%, while backdoors raise local errors by up to 48%. In both cases, Generative Adversarial Network-based attacks are consistently more effective than random perturbations, with backdoors proving especially challenging to detect due to their localized effect. To mitigate these threats, we evaluate four defense strategies: weighted aggregation, larger participant clusters, local retraining, and their coordinated integration into a secure framework. Results demonstrate that these defenses substantially reduce the impact of attacks, and in some cases even improve baseline accuracy, thereby enhancing the resilience of federated energy forecasting against adversarial manipulation.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100680"},"PeriodicalIF":9.6,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146022833","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}

引用次数: 0

A graph neural network enhanced decision transformer for efficient optimization in dynamic smart charging environments 一种基于图神经网络的决策变压器，用于动态智能充电环境的高效优化

IF 9.6

Energy and AI Pub Date : 2026-01-01 Epub Date: 2026-01-16 DOI: 10.1016/j.egyai.2026.100679

Stavros Orfanoudakis, Nanda Kishor Panda, Peter Palensky, Pedro P. Vergara

{"title":"A graph neural network enhanced decision transformer for efficient optimization in dynamic smart charging environments","authors":"Stavros Orfanoudakis, Nanda Kishor Panda, Peter Palensky, Pedro P. Vergara","doi":"10.1016/j.egyai.2026.100679","DOIUrl":"10.1016/j.egyai.2026.100679","url":null,"abstract":"<div><div>Electric-vehicle smart charging requires quick decision-making under uncertainty while enforcing strict electricity grid and user requirements. Mathematical optimization becomes too slow at scale, while online reinforcement learning struggles with sparse rewards and safety. This paper proposes GNN-DT, a topology-aware Decision Transformer that combines graph neural network embeddings with sequence modeling to learn charging policies from offline trajectories. The method operates over variable numbers of vehicles and chargers without retraining. Evaluated on realistic smart charging scenarios, GNN-DT achieves near-optimal performance, reaching rewards within 5 percent of an oracle solver while using up to 10<span><math><mo>×</mo></math></span> fewer training trajectories than baseline methods. It consistently outperforms online and offline reinforcement learning approaches and generalizes to unseen fleet sizes and network topologies. Inference runs in milliseconds, making the approach suitable for real-time deployment in large-scale charging systems.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100679"},"PeriodicalIF":9.6,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146022834","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}

引用次数: 0

Dual-path feature extraction with MoE optimization for ultra-short-term wind power forecasting 基于MoE优化的双路径特征提取超短期风电预测

IF 9.6

Energy and AI Pub Date : 2026-01-01 Epub Date: 2026-01-11 DOI: 10.1016/j.egyai.2026.100684

Yongsheng Wang , Xuehui Wang , Guangchen Liu , Shijie Guan , Zhe Zhang , Hailong Li

{"title":"Dual-path feature extraction with MoE optimization for ultra-short-term wind power forecasting","authors":"Yongsheng Wang , Xuehui Wang , Guangchen Liu , Shijie Guan , Zhe Zhang , Hailong Li","doi":"10.1016/j.egyai.2026.100684","DOIUrl":"10.1016/j.egyai.2026.100684","url":null,"abstract":"<div><div>Accurate ultra-short-term wind power forecasting is critical for maintaining the stability and efficiency of modern power systems, yet it remains challenging due to the high volatility and nonlinear dynamics of wind energy. This study proposes a hybrid forecasting framework that integrates deep learning-based temporal modeling, fluctuation-aware feature engineering, and expert-driven hyperparameter optimization. A convolutional attention network is designed to capture both local temporal patterns and long-term dependencies in wind power time series, while financial technical indicators are incorporated to enhance the representation of short-term fluctuation characteristics. To further improve forecasting accuracy and robustness, a mixture of experts strategy is employed to jointly optimize model hyperparameters and indicator construction. The final power forecasting is performed using a gradient boosting decision tree model with strong generalization capability. Experimental evaluations conducted on multiple real-world wind farm datasets demonstrate that the proposed framework consistently outperforms state of the art machine learning and deep learning approaches in terms of forecasting accuracy and efficiency. For example, on Dataset 1, the proposed method achieves a 5.03% reduction in mean square error compared with the strongest deep learning baseline. The results indicate that the proposed approach effectively captures both rapid fluctuations and underlying temporal trends, providing a reliable and practical solution for ultra-short-term wind power forecasting in complex operational environments.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100684"},"PeriodicalIF":9.6,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977681","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}

引用次数: 0

Evolving power system operator rules for real-time congestion management 基于实时拥塞管理的电力系统算子规则演化

IF 9.6

Energy and AI Pub Date : 2026-01-01 Epub Date: 2026-01-06 DOI: 10.1016/j.egyai.2025.100672

Ferinar Moaidi , Ricardo J. Bessa

{"title":"Evolving power system operator rules for real-time congestion management","authors":"Ferinar Moaidi , Ricardo J. Bessa","doi":"10.1016/j.egyai.2025.100672","DOIUrl":"10.1016/j.egyai.2025.100672","url":null,"abstract":"<div><div>The growing integration of renewable energy sources and the widespread electrification of the energy demand have significantly reduced the capacity margin of the electrical grid. This demands a more flexible approach to grid operation, for instance, combining real-time topology optimization and redispatching. Traditional expert-driven decision-making rules may become insufficient to manage the increasing complexity of real-time grid operations and derive remedial actions under the N-1 contingency. This work proposes a novel hybrid AI framework for power grid topology control that integrates genetic network programming (GNP), reinforcement learning, and decision trees. A new variant of GNP is introduced that is capable of evolving the decision-making rules by learning from data in a reinforcement learning framework. The graph-based evolutionary structure of GNP and decision trees enables transparent, traceable reasoning. The proposed method outperforms both a baseline expert system and a state-of-the-art deep reinforcement learning agent on the IEEE 118-bus system, achieving up to an 28% improvement in a key performance metric used in the Learning to Run a Power Network (L2RPN) competition.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100672"},"PeriodicalIF":9.6,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926799","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}

引用次数: 0

Edge-cloud artificial intelligence digital twin thermal modeling for rotating sintered core heat pipes 旋转烧结芯热管的边缘云人工智能数字孪生热建模

IF 9.6

Energy and AI Pub Date : 2026-01-01 Epub Date: 2025-12-13 DOI: 10.1016/j.egyai.2025.100666

Jialan Liu , Chi Ma , Wenhui Zhou , Mingming Li , Jialong He , Giovanni Totis , Chunlei Hua , Liang Wang , Gangwei Cui , Ruijuan Xue , Zhi Tan , Jun Yang , Kuo Liu , Yuansheng Zhou , Jianqiang Zhou , Shengbin Weng

{"title":"Edge-cloud artificial intelligence digital twin thermal modeling for rotating sintered core heat pipes","authors":"Jialan Liu , Chi Ma , Wenhui Zhou , Mingming Li , Jialong He , Giovanni Totis , Chunlei Hua , Liang Wang , Gangwei Cui , Ruijuan Xue , Zhi Tan , Jun Yang , Kuo Liu , Yuansheng Zhou , Jianqiang Zhou , Shengbin Weng","doi":"10.1016/j.egyai.2025.100666","DOIUrl":"10.1016/j.egyai.2025.100666","url":null,"abstract":"<div><div>The sintered core heat pipe is widely used in precision thermal control and aerospace systems due to its high heat transfer performance. However, conventional computational fluid dynamics approaches for predicting its thermal behavior are computationally expensive and inflexible under varying operating conditions, while experimental methods are time-consuming and costly. To address the above challenges, in this study, a digital twin-based predictive thermal modeling framework is presented for sintered core heat pipes under rotational conditions, implemented within an edge-cloud artificial intelligence architecture. A fully parameterized physical model is developed on the Simulink platform using the SIMSCAPE Fluids module, enabling dynamic simulations of phase transitions and temperature responses. Validation against experimental data shows prediction errors within ±5 %. Simulation and experimental datasets are integrated to train three models-physics-informed neural network, Transformer, and light gradient boosting machine-evaluated under steady and transient thermal conditions. The physics-informed neural network achieves the lowest mean absolute error of 0.85 °C in high thermal inertia cases, while the Transformer attains the best steady-state accuracy with a root mean square error of 0.58 °C and inference latency of 150 ms after Turing Tensor R-Engine deployment. Docker-based deployment enables real-time edge inference, with the Transformer achieving an optimal balance of accuracy, memory footprint (36 MB), and response speed. The proposed framework offers a practical and scalable approach for accurate thermal prediction in advanced thermal management applications.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100666"},"PeriodicalIF":9.6,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145791891","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}

引用次数: 0

Bubble segmentation and heat flux estimation in nucleate boiling using a resource-efficient deep learning framework 基于资源高效深度学习框架的核沸腾气泡分割和热流估计

IF 9.6

Energy and AI Pub Date : 2026-01-01 Epub Date: 2025-12-13 DOI: 10.1016/j.egyai.2025.100668

UngJin Na , Bruno Pinheiro Serrao , JunYoung Seo , Youngjoon Suh , Juliana Pacheco Duarte , Yoonjin Won , HangJin Jo

{"title":"Bubble segmentation and heat flux estimation in nucleate boiling using a resource-efficient deep learning framework","authors":"UngJin Na , Bruno Pinheiro Serrao , JunYoung Seo , Youngjoon Suh , Juliana Pacheco Duarte , Yoonjin Won , HangJin Jo","doi":"10.1016/j.egyai.2025.100668","DOIUrl":"10.1016/j.egyai.2025.100668","url":null,"abstract":"<div><div>Boiling is a critical heat transfer mechanism in high-power-density systems driven by bubble ebullition cycles that efficiently manage thermal loads. However, the simultaneous challenges of the millisecond-scale lifecycle of strongly deformable bubbles and the heavy computational burden of processing gigabytes of high-speed photographic data demand immense resources with meticulous parameter tuning for conventional image processing techniques. Herein, we present an end-to-end resource-efficient deep learning framework that overcomes the diagnostic bottlenecks of bubble dynamics visualization and delivers quantitative insights into these processes using a Mask R-CNN model trained on a synthetically augmented boiling dataset. The model extracts spatiotemporal bubble characteristics by segmenting bubble contours across varying heat fluxes ranging from 100 to 1000 kW/m², tracking unique bubbles, and detecting bubble departure from the surface, enabling the estimation of bubble-level contributions to macroscopic cooling performance. Our resource-efficient approach can be rapidly retuned for new boiling experiments with low manual effort, independent of the setup.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100668"},"PeriodicalIF":9.6,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926798","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}

引用次数: 0

An analytical solutions-enhanced neural network for high-accuracy modeling: A case study on gas turbine 用于高精度建模的解析解增强神经网络：以燃气轮机为例

IF 9.6

Energy and AI Pub Date : 2026-01-01 Epub Date: 2025-12-22 DOI: 10.1016/j.egyai.2025.100674

Zepeng Han , Wei Han , Wenjing Ma , Jichao Li , Jun Sui

{"title":"An analytical solutions-enhanced neural network for high-accuracy modeling: A case study on gas turbine","authors":"Zepeng Han , Wei Han , Wenjing Ma , Jichao Li , Jun Sui","doi":"10.1016/j.egyai.2025.100674","DOIUrl":"10.1016/j.egyai.2025.100674","url":null,"abstract":"<div><div>Physics-informed neural networks have gained wide application due to their data efficiency and generation ability. Nevertheless, their application is restricted in practical engineering problems where complex physical mechanisms make it difficult to formulate or solve the corresponding partial differential equations. To solve the above problems, Analytical solutions are obtained by finite-step algebraic computation. Based on these solutions, an analytical solutions-enhanced physical model is trained using a multi-layer perceptron to replace conventional partial differential equations. By incorporating weighted loss functions into the multi-layer perceptron, an analytical solutions-enhanced neural network is developed to address complex problems in practical engineering applications. Taking gas turbines in the energy sector as a case, the proposed model is compared with conventional prediction models. Results indicate that the proposed model achieves a mean square error of 0.0606, a mean absolute error of 0.0867 on the test set, corresponding to maximum reductions of 99.17 % and 90.98 % relative to conventional prediction models, respectively. Additionally, sample size analyses indicate that the proposed model can reduce the sample size by at least 35.45 %. The analytical solutions-enhanced neural network established in this study provides a novel approach to integrating physical models and neural networks, thereby expending the application of physics-informed neural networks to real-word engineering problems.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100674"},"PeriodicalIF":9.6,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926795","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}

引用次数: 0

Thermal conductivity prediction of BN composites based on Enhanced Co-ANN combined with physical attention mechanisms 基于增强Co-ANN结合物理注意机制的BN复合材料导热系数预测

IF 9.6

Energy and AI Pub Date : 2026-01-01 Epub Date: 2025-12-10 DOI: 10.1016/j.egyai.2025.100663

Chen Liu , Tong Li , Yuandong Guo , Guiping Lin

{"title":"Thermal conductivity prediction of BN composites based on Enhanced Co-ANN combined with physical attention mechanisms","authors":"Chen Liu , Tong Li , Yuandong Guo , Guiping Lin","doi":"10.1016/j.egyai.2025.100663","DOIUrl":"10.1016/j.egyai.2025.100663","url":null,"abstract":"<div><div>This study proposes an enhanced co-training artificial neural network (Enhanced Co-ANN) guided by the physical attention mechanisms for the effective thermal conductivity prediction of high filler volume fraction polymer/BN composites. The thermal conductivity of BN composites is influenced by multiple factors, including the morphology of the fillers, interface thermal resistance, and experimental noise. This model tackles complex physical processes by integrating a customized multi-head physical attention layer to emphasize key features, along with a physics-constrained loss function to ensure prediction consistency. A collaborative training strategy based on curriculum learning and consistency discrimination is adopted. The model is optimized using 3174 labeled experimental datasets and 50,000 unlabeled data generated from physical models. Weight distribution is systematically designed across three core levels: model architecture, loss function, and training strategy. This approach differs from traditional parameter weight adjustments, as it emphasizes key features, especially volume fraction (vf), and balances different learning objectives through a physically guided mechanism and dynamic training strategies. Attention visualization indicates that the model adaptively focuses on the volume fraction of the packing material and the interface effect, verifying the effectiveness of the physically guided design. Six groups of samples with different packing volume fractions were made for testing and validation. This model has high accuracy (R² = 0.982; MAE = 0.045 W/m K) and is extremely consistent with physical laws. This network framework provides a method with broad application prospects for the rapid calculation, screening, and efficient design of high-performance polymer/BN thermal conductive materials.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100663"},"PeriodicalIF":9.6,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145718921","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}

引用次数: 0