Energy and AI最新文献_第5页

BatteryTSFM: Generalizable long-horizon degradation prediction across conditions and chemistries via time series foundation models BatteryTSFM：基于时间序列基础模型的跨条件和化学的可推广的长期退化预测

IF 9.6 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Energy and AI

Pub Date : 2025-12-01 DOI: 10.1016/j.egyai.2025.100646

Zimeng Fan , Yuting Guo , Lei Song , Junrong Du , Hongfei Wang , Xuzhi Li

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.

机器学习在预测锂离子电池的健康状态（SOH）方面取得了重大突破。然而，由于使用条件和电化学特性的变化，传统方法在跨域适应、跨数据集泛化和长期预测方面面临着严峻的挑战。受大型语言模型（llm）成功的启发，时间序列基础模型（tsfm）为克服上述问题提供了另一种解决方案。然而，探索TSFMs在不同跨域因子下对电池SOH预测的泛化增强能力的研究仍然不足。为此，提出了一种新的基于tsfm模型的SOH预测框架BatteryTSFM。首先，我们引入了基于主干感知的时间重采样，根据不同tsfm的结构特征动态调整预处理，通过特征缩放实现最优的跨域泛化。其次，将蒙特卡罗dropout集成到自回归推理中，量化多步预测误差。在四个公共数据集中，BatteryTSFM在交叉条件任务中平均降低了35%的RMSE，在交叉化学任务中平均降低了88%，这表明基础模型方法可以为能源系统提供可靠的长期SOH预测。我们还进行了探索性分析，将泛化与微调数据集大小和重新采样粒度联系起来，为部署提供实用指导。

{"title":"BatteryTSFM: Generalizable long-horizon degradation prediction across conditions and chemistries via time series foundation models","authors":"Zimeng Fan , Yuting Guo , Lei Song , Junrong Du , Hongfei Wang , 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}

引用次数: 0

Dual-channel representation learning with wind speed correction factor for enhanced wind power forecasting 带风速修正因子的双通道表示学习增强风电预测

IF 9.6 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Energy and AI

Pub Date : 2025-12-01 DOI: 10.1016/j.egyai.2025.100650

Yanbo Chen , Qintao Du , Tuben Qiang , Liangcheng Cheng , Yongkang She , Zhi Zhang

Wind power forecasting (WPF) accuracy is fundamentally constrained by two critical challenges. First, the high-order nonlinear relationship between wind speed (WS) and power (WP) substantially amplifies errors inherent in numerical weather prediction (NWP) data. Second, conventional models process all input features uniformly, failing to distinguish the dominant role of the primary driving feature from the complementary roles of auxiliary features. To decouple and address these challenges, this paper proposes a novel forecasting method (CFRM-DCM) that integrates a Correction Factor Representation Model (CFRM) and a Dual-Channel Mechanism (DCM). The CFRM is first employed to address the NWP error. It describes the complex correlation and forecasting error between measured WS and NWP WS as correction factors. A generative adversarial network (GAN) is then utilized to learn the distribution of these factors and output a corrected, more accurate WS. This corrected data is then fed into the DCM, a dual-branch architecture designed to enhance complex feature extraction, overcoming the limitations of traditional single-channel structures. The proposed method is validated on four wind farms. Simulation results demonstrate that the CFRM-DCM method achieves significant improvements in WPF accuracy, with error reductions ranging from 3.9 % to 9.4 % across ultra-short-term and short-term timescales. This enhanced WPF performance is directly attributed to the model's ability to first improve WS accuracy, with gains of 8.8 %, 7.6 %, 8.3 %, and 8.8 % for the respective farms.

风电预测（WPF）的准确性从根本上受到两个关键挑战的制约。首先，风速（WS）和功率（WP）之间的高阶非线性关系极大地放大了数值天气预报（NWP）数据固有的误差。其次，传统模型对所有输入特征进行统一处理，未能区分主要驱动特征的主导作用和辅助特征的补充作用。为了解耦和解决这些挑战，本文提出了一种新的预测方法（CFRM-DCM），该方法集成了校正因子表示模型（CFRM）和双通道机制（DCM）。CFRM首先用于解决NWP错误。将实测WS与NWP WS之间的复杂相关关系和预测误差描述为校正因子。然后利用生成对抗网络（GAN）来学习这些因素的分布，并输出一个修正的、更准确的WS。校正后的数据随后被输入DCM， DCM是一种双分支结构，旨在增强复杂特征提取，克服了传统单通道结构的局限性。该方法在四个风电场上进行了验证。仿真结果表明，CFRM-DCM方法可以显著提高WPF精度，在超短期和短期时间尺度上误差降低3.9% ~ 9.4%。这种增强的WPF性能直接归因于模型首先提高WS准确性的能力，分别为各自的农场增加了8.8%,7.6%，8.3%和8.8%。

{"title":"Dual-channel representation learning with wind speed correction factor for enhanced wind power forecasting","authors":"Yanbo Chen , Qintao Du , Tuben Qiang , Liangcheng Cheng , Yongkang She , Zhi Zhang","doi":"10.1016/j.egyai.2025.100650","DOIUrl":"10.1016/j.egyai.2025.100650","url":null,"abstract":"<div><div>Wind power forecasting (WPF) accuracy is fundamentally constrained by two critical challenges. First, the high-order nonlinear relationship between wind speed (WS) and power (WP) substantially amplifies errors inherent in numerical weather prediction (NWP) data. Second, conventional models process all input features uniformly, failing to distinguish the dominant role of the primary driving feature from the complementary roles of auxiliary features. To decouple and address these challenges, this paper proposes a novel forecasting method (CFRM-DCM) that integrates a Correction Factor Representation Model (CFRM) and a Dual-Channel Mechanism (DCM). The CFRM is first employed to address the NWP error. It describes the complex correlation and forecasting error between measured WS and NWP WS as correction factors. A generative adversarial network (GAN) is then utilized to learn the distribution of these factors and output a corrected, more accurate WS. This corrected data is then fed into the DCM, a dual-branch architecture designed to enhance complex feature extraction, overcoming the limitations of traditional single-channel structures. The proposed method is validated on four wind farms. Simulation results demonstrate that the CFRM-DCM method achieves significant improvements in WPF accuracy, with error reductions ranging from 3.9 % to 9.4 % across ultra-short-term and short-term timescales. This enhanced WPF performance is directly attributed to the model's ability to first improve WS accuracy, with gains of 8.8 %, 7.6 %, 8.3 %, and 8.8 % for the respective farms.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100650"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145614660","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

Machine learning for photovoltaic single axis tracker fault detection and classification 基于机器学习的光伏单轴跟踪器故障检测与分类

IF 9.6 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Energy and AI

Pub Date : 2025-12-01 DOI: 10.1016/j.egyai.2025.100652

Taos Transue, Marios Theristis, Daniel M. Riley

More than 81% of the annual capacity of utility-scale photovoltaic (PV) power plants in the U.S. use single-axis trackers (SATs) due to SATs delivering 4% in capacity factor on average over fixed-array systems. However, SATs are subject to faults, such as software misconfigurations and mechanical failures, resulting in suboptimal tracking. If left undetected, the overall power yield of the PV power plant is reduced significantly. Minimizing downtime and ensuring efficient operation of SATs requires robust detection and diagnosis mechanisms for SAT faults. We present a machine learning framework for implementing real-time SAT fault detection and classification. Our implementation of the proposed framework reliably identifies measurements taken from a test PV system undergoing emulated SAT faults relative to state-of-the-art algorithms and produces nearly zero false positives on our testing days. Code and data are available at https://pvpmc.sandia.gov/tools.

在美国，超过81%的公用事业规模的光伏（PV）发电厂使用单轴跟踪器（SATs），因为SATs比固定阵列系统平均提供4%的容量系数。然而，sat容易出现故障，例如软件配置错误和机械故障，从而导致次优跟踪。如果不加以检测，光伏电站的总发电量将显著降低。最大限度地减少停机时间并确保SAT高效运行需要强大的SAT故障检测和诊断机制。我们提出了一个实现实时SAT故障检测和分类的机器学习框架。我们所提出的框架的实现可靠地识别了测试PV系统中与最先进算法相关的模拟SAT故障的测量结果，并且在我们的测试日中产生了几乎零误报。代码和数据可在https://pvpmc.sandia.gov/tools上获得。

引用次数: 0

Predictive modelling of slurry viscosity using transfer learning to mitigate uncertainties in pilot-scale lithium-ion battery manufacturing process 使用迁移学习的浆液粘度预测建模，以减轻中试规模锂离子电池制造过程中的不确定性

IF 9.6 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Energy and AI

Pub Date : 2025-12-01 DOI: 10.1016/j.egyai.2025.100661

Mo'ath El-Dalahmeh, Om Siddhapura, Geanina Apachitei, Matthew Capener, James Marco, Adam Todd, Keri Goodwin, Mona Faraji Niri

Reliable prediction of slurry viscosity is essential for consistent electrode coating in lithium-ion battery manufacturing. This study investigates transfer learning to address deployment-phase uncertainty arising from batch-to-batch variability under pilot-scale conditions. Experimental data from two batches are used to model viscosity at 1, 10, and 100 1/s using four inputs: formulation, dispersant type, solid content percent, and solid dispersant percent. Linear baselines (ordinary least squares and ridge) are evaluated alongside tree ensembles and neural architectures, under identical splits and preprocessing. Results show consistent performance gains from transfer learning across all shear rates, with higher R2 and lower MAE and RMSE relative to no-transfer training. Across models, differences in backbone choice are secondary; the transfer step is the principal driver of improvement under the present data regime. Uncertainty is quantified using split-conformal prediction intervals, yielding nominal 90 percent coverage with narrower intervals after transfer learning. Small-data design choices are reported, including balanced splits and sensitivity checks with conservative augmentation used only for analysis. The findings indicate a practical and data-efficient route to viscosity prediction under sequential batches, supporting more robust model deployment in pilot-scale manufacturing.

在锂离子电池制造中，浆液粘度的可靠预测对于保证电极涂层的一致性至关重要。本研究探讨了迁移学习，以解决在中试规模条件下由批对批可变性引起的部署阶段不确定性。从两个批次的实验数据被用来模拟粘度在1,10,100 1/s使用四个输入：配方，分散剂类型，固体含量百分比和固体分散剂百分比。在相同的分割和预处理下，线性基线（普通最小二乘和脊线）与树集成和神经架构一起进行评估。结果显示，在所有剪切速率下，迁移学习的性能都有一致的提高，相对于无迁移训练，具有更高的R2和更低的MAE和RMSE。在各个模型中，骨干网选择的差异是次要的；在目前的数据制度下，转移步骤是改进的主要推动力。不确定性使用分裂-适形预测区间进行量化，在迁移学习后，以更窄的区间产生名义上90%的覆盖率。报告了小数据设计选择，包括平衡分裂和敏感性检查，保守增强仅用于分析。研究结果表明，在连续批次下，粘度预测是一种实用且数据高效的方法，支持在中试规模生产中更稳健的模型部署。

{"title":"Predictive modelling of slurry viscosity using transfer learning to mitigate uncertainties in pilot-scale lithium-ion battery manufacturing process","authors":"Mo'ath El-Dalahmeh, Om Siddhapura, Geanina Apachitei, Matthew Capener, James Marco, Adam Todd, Keri Goodwin, Mona Faraji Niri","doi":"10.1016/j.egyai.2025.100661","DOIUrl":"10.1016/j.egyai.2025.100661","url":null,"abstract":"<div><div>Reliable prediction of slurry viscosity is essential for consistent electrode coating in lithium-ion battery manufacturing. This study investigates transfer learning to address deployment-phase uncertainty arising from batch-to-batch variability under pilot-scale conditions. Experimental data from two batches are used to model viscosity at 1, 10, and 100 1/s using four inputs: formulation, dispersant type, solid content percent, and solid dispersant percent. Linear baselines (ordinary least squares and ridge) are evaluated alongside tree ensembles and neural architectures, under identical splits and preprocessing. Results show consistent performance gains from transfer learning across all shear rates, with higher R2 and lower MAE and RMSE relative to no-transfer training. Across models, differences in backbone choice are secondary; the transfer step is the principal driver of improvement under the present data regime. Uncertainty is quantified using split-conformal prediction intervals, yielding nominal 90 percent coverage with narrower intervals after transfer learning. Small-data design choices are reported, including balanced splits and sensitivity checks with conservative augmentation used only for analysis. The findings indicate a practical and data-efficient route to viscosity prediction under sequential batches, supporting more robust model deployment in pilot-scale manufacturing.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100661"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145680992","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 novel method for multivariate short-term offshore wind forecasting via time–frequency clustering and inverted attention 基于时频聚类和反向注意的多变量短期海上风预报新方法

IF 9.6 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Energy and AI

Pub Date : 2025-12-01 DOI: 10.1016/j.egyai.2025.100654

Keqi Chen , Tianshuai Pei , Lina Yang , Thomas Wu , Yunxuan Dong

Short-term offshore wind power forecasting is crucial for stable power system operations. However, accurate forecasting is hindered by multivariate interactions that generate multi-scale repetitive patterns and obscure cross-variate correlations. In this paper, we propose a hierarchical framework, the Time–Frequency Clustering Inverted Transformer, for multivariate offshore wind power forecasting. First, a Time–Frequency Clustering component applies Wavelet Packet Decomposition to each series and quantifies sub-series similarity by overall activity and evolutionary trend, grouping repetitive patterns into structured clusters. Second, an inverted Transformer captures multivariate correlations within clusters by treating time points of individual sub-series as variate tokens, enabling self-attention to focus on multivariate correlations rather than temporal dependencies. On two real-world offshore wind datasets (horizons 8–48 h), our proposed framework reduces MSE/MAE by 14.11% and outperforms 12 recognised baselines (e.g., PatchTST, TimesNet), with the advantage persisting even when the TFC component is applied to the baselines. Moreover, our method demonstrates remarkable generalisability on three public datasets (Solar-Energy, Traffic, and ECL), reducing MSE/MAE by 7.36%. These results indicate that associating repetitive patterns with attention to cross-variate structure materially improves multivariate offshore wind power forecasting.

海上风电短期预测对电力系统的稳定运行至关重要。然而，产生多尺度重复模式和模糊的交叉变量相关性的多变量相互作用阻碍了准确的预测。在本文中，我们提出了一种分层框架，即时频聚类逆变器，用于多元海上风电预测。首先，时频聚类组件对每个序列应用小波包分解，并根据整体活动和进化趋势量化子序列的相似性，将重复模式分组到结构化的聚类中。其次，倒置的Transformer通过将单个子序列的时间点作为变量标记来捕获集群内的多变量相关性，从而使自关注集中于多变量相关性而不是时间依赖性。在两个真实的海上风电数据集（8-48小时）上，我们提出的框架将MSE/MAE降低了14.11%，优于12个公认的基线（例如PatchTST， TimesNet），即使将TFC组件应用于基线，其优势也会持续存在。此外，我们的方法在三个公共数据集（Solar-Energy， Traffic和ECL）上显示出显著的通用性，将MSE/MAE降低了7.36%。这些结果表明，将重复模式与对交叉变量结构的关注联系起来，可以有效地改善多元海上风电预测。

{"title":"A novel method for multivariate short-term offshore wind forecasting via time–frequency clustering and inverted attention","authors":"Keqi Chen , Tianshuai Pei , Lina Yang , Thomas Wu , Yunxuan Dong","doi":"10.1016/j.egyai.2025.100654","DOIUrl":"10.1016/j.egyai.2025.100654","url":null,"abstract":"<div><div>Short-term offshore wind power forecasting is crucial for stable power system operations. However, accurate forecasting is hindered by multivariate interactions that generate multi-scale repetitive patterns and obscure cross-variate correlations. In this paper, we propose a hierarchical framework, the Time–Frequency Clustering Inverted Transformer, for multivariate offshore wind power forecasting. First, a Time–Frequency Clustering component applies Wavelet Packet Decomposition to each series and quantifies sub-series similarity by overall activity and evolutionary trend, grouping repetitive patterns into structured clusters. Second, an inverted Transformer captures multivariate correlations within clusters by treating time points of individual sub-series as variate tokens, enabling self-attention to focus on multivariate correlations rather than temporal dependencies. On two real-world offshore wind datasets (horizons 8–48 h), our proposed framework reduces MSE/MAE by 14.11% and outperforms 12 recognised baselines (e.g., PatchTST, TimesNet), with the advantage persisting even when the TFC component is applied to the baselines. Moreover, our method demonstrates remarkable generalisability on three public datasets (Solar-Energy, Traffic, and ECL), reducing MSE/MAE by 7.36%. These results indicate that associating repetitive patterns with attention to cross-variate structure materially improves multivariate offshore wind power forecasting.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100654"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145680875","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

Fast and generalisable parameter-embedded neural operators for lithium-ion battery simulation 用于锂离子电池仿真的快速、通用参数嵌入神经算子

IF 9.6 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Energy and AI

Pub Date : 2025-12-01 DOI: 10.1016/j.egyai.2025.100647

Amir Ali Panahi , Daniel Luder , Billy Wu , Gregory Offer , Dirk Uwe Sauer , Weihan Li

Digital twins of lithium-ion batteries are increasingly used to enable predictive monitoring, control, and design at system scale. Increasing their capabilities involves improving their physical fidelity while maintaining sub-millisecond computational speed. In this work, we introduce machine learning surrogates that learn physical dynamics. Specifically, we benchmark three operator-learning surrogates for the Single Particle Model (SPM): Deep Operator Networks (DeepONets), Fourier Neural Operators (FNOs) and a newly proposed parameter-embedded Fourier Neural Operator (PE-FNO), which conditions each spectral layer on particle radius and solid-phase diffusivity. We extend the comparison to classical machine-learning baselines by including U-Nets. Models are trained on simulated trajectories spanning four current families (constant, triangular, pulse-train, and Gaussian-random-field) and a full range of State-of-Charge (SOC) (0 % to 100 %). DeepONet accurately replicates constant-current behaviour but struggles with more dynamic loads. The basic FNO maintains mesh invariance and keeps concentration errors below 1 %, with voltage mean-absolute errors under 1.7 mV across all load types. Introducing parameter embedding marginally increases error but enables generalisation to varying radii and diffusivities. PE-FNO executes approximately 200 times faster than a 16-thread SPM solver. Consequently, PE-FNO’s capabilities in inverse tasks are explored in a parameter estimation task with Bayesian optimisation, recovering anode and cathode diffusivities with 1.14 % and 8.4 % mean absolute percentage error, respectively, and 0.5918 percentage points higher error in comparison with classical methods. These results pave the way for neural operators to meet the accuracy, speed and parametric flexibility demands of real-time battery management, design-of-experiments and large-scale inference. PE-FNO outperforms conventional neural surrogates, offering a practical path towards high-speed and high-fidelity electrochemical digital twins.

锂离子电池的数字孪生体越来越多地用于系统规模的预测监测、控制和设计。提高它们的能力需要提高它们的物理保真度，同时保持亚毫秒的计算速度。在这项工作中，我们引入了学习物理动力学的机器学习代理。具体来说，我们对单粒子模型（SPM）的三种算子学习替代方法进行了基准测试：深度算子网络（DeepONets）、傅立叶神经算子（FNOs）和新提出的参数嵌入傅立叶神经算子（PE-FNO），该算子根据粒子半径和固相扩散率来约束每个光谱层。我们通过包括U-Nets将比较扩展到经典的机器学习基线。模型在模拟轨迹上进行训练，这些轨迹跨越四个电流族（常量、三角形、脉冲序列和高斯随机场）和全范围的充电状态（SOC）（0%至100%）。DeepONet精确地复制了恒流行为，但在处理更多动态负载时遇到了困难。基本FNO保持网格不变性，将浓度误差保持在1%以下，所有负载类型的电压平均绝对误差低于1.7 mV。引入参数嵌入会略微增加误差，但可以泛化到不同的半径和扩散系数。PE-FNO的执行速度大约是16线程SPM求解器的200倍。因此，在贝叶斯优化参数估计任务中探索了PE-FNO在逆任务中的能力，回收阳极和阴极扩散系数的平均绝对百分比误差分别为1.14%和8.4%，与经典方法相比误差提高了0.5918个百分点。这些结果为神经算子满足实时电池管理、实验设计和大规模推理的准确性、速度和参数灵活性需求铺平了道路。PE-FNO优于传统的神经替代物，为实现高速高保真的电化学数字孪生提供了切实可行的途径。

{"title":"Fast and generalisable parameter-embedded neural operators for lithium-ion battery simulation","authors":"Amir Ali Panahi , Daniel Luder , Billy Wu , Gregory Offer , Dirk Uwe Sauer , Weihan Li","doi":"10.1016/j.egyai.2025.100647","DOIUrl":"10.1016/j.egyai.2025.100647","url":null,"abstract":"<div><div>Digital twins of lithium-ion batteries are increasingly used to enable predictive monitoring, control, and design at system scale. Increasing their capabilities involves improving their physical fidelity while maintaining sub-millisecond computational speed. In this work, we introduce machine learning surrogates that learn physical dynamics. Specifically, we benchmark three operator-learning surrogates for the Single Particle Model (SPM): Deep Operator Networks (DeepONets), Fourier Neural Operators (FNOs) and a newly proposed parameter-embedded Fourier Neural Operator (PE-FNO), which conditions each spectral layer on particle radius and solid-phase diffusivity. We extend the comparison to classical machine-learning baselines by including U-Nets. Models are trained on simulated trajectories spanning four current families (constant, triangular, pulse-train, and Gaussian-random-field) and a full range of State-of-Charge (SOC) (0% to 100%). DeepONet accurately replicates constant-current behaviour but struggles with more dynamic loads. The basic FNO maintains mesh invariance and keeps concentration errors below 1%, with voltage mean-absolute errors under 1.7mV across all load types. Introducing parameter embedding marginally increases error but enables generalisation to varying radii and diffusivities. PE-FNO executes approximately 200 times faster than a 16-thread SPM solver. Consequently, PE-FNO’s capabilities in inverse tasks are explored in a parameter estimation task with Bayesian optimisation, recovering anode and cathode diffusivities with 1.14% and 8.4% mean absolute percentage error, respectively, and 0.5918 percentage points higher error in comparison with classical methods. These results pave the way for neural operators to meet the accuracy, speed and parametric flexibility demands of real-time battery management, design-of-experiments and large-scale inference. PE-FNO outperforms conventional neural surrogates, offering a practical path towards high-speed and high-fidelity electrochemical digital twins.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100647"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145680996","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

Deep reinforcement learning for joint dispatch of battery storage and gas turbines in renewable-powered microgrids 可再生微电网中电池储能与燃气轮机联合调度的深度强化学习

IF 9.6 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Energy and AI

Pub Date : 2025-12-01 DOI: 10.1016/j.egyai.2025.100653

Manuel Sage , Khalil Al Handawi , Yaoyao Fiona Zhao

This study introduces a novel deep reinforcement learning (DRL) framework for the joint dispatch of Gas Turbines (GTs) and Battery Energy Storage Systems (BESs) in microgrids that face the variability of renewable energy sources and demands. BESs can store surplus renewable energy for nearly instantaneous use, while GTs offer sustained energy output over longer periods, offering complementary benefits. Previous studies often oversimplified GT operations, neglecting critical factors such as ramp-up times and increased degradation from frequent starts. This research addresses these gaps by proposing an advanced modeling framework that accurately captures the dynamic interaction between GTs and BESs, including GT ramp-up times and maintenance costs associated with operational cycles. Through extensive case studies involving diverse microgrid configurations, we demonstrate that DRL effectively learns dispatch policies directly from historical data, outperforming traditional optimization techniques. Deploying DRL to our framework yields more realistic dispatch policies, reducing GT maintenance costs by avoiding frequent starts. The proposed framework has significant potential to improve energy management strategies and to streamline the planning of hybrid energy systems. To encourage further research, we have released our codebase to the public, enabling the scientific community to build upon our findings.

本研究引入了一种新的深度强化学习（DRL）框架，用于面对可再生能源和需求变化的微电网中燃气轮机（gt）和电池储能系统（BESs）的联合调度。BESs可以储存多余的可再生能源，几乎可以即时使用，而GTs可以在更长的时间内提供持续的能源输出，提供互补的好处。以前的研究往往过于简化了GT操作，忽略了一些关键因素，如加速时间和频繁启动导致的性能下降。本研究通过提出一种先进的建模框架来解决这些差距，该框架准确地捕获了GT和BESs之间的动态交互，包括GT的启动时间和与操作周期相关的维护成本。通过涉及不同微电网配置的广泛案例研究，我们证明了DRL直接从历史数据中有效地学习调度策略，优于传统的优化技术。将DRL部署到我们的框架中可以产生更现实的调度策略，通过避免频繁启动来降低GT维护成本。提出的框架在改善能源管理战略和简化混合能源系统规划方面具有重大潜力。为了鼓励进一步的研究，我们向公众发布了我们的代码库，使科学界能够在我们的发现的基础上进行构建。

{"title":"Deep reinforcement learning for joint dispatch of battery storage and gas turbines in renewable-powered microgrids","authors":"Manuel Sage , Khalil Al Handawi , Yaoyao Fiona Zhao","doi":"10.1016/j.egyai.2025.100653","DOIUrl":"10.1016/j.egyai.2025.100653","url":null,"abstract":"<div><div>This study introduces a novel deep reinforcement learning (DRL) framework for the joint dispatch of Gas Turbines (GTs) and Battery Energy Storage Systems (BESs) in microgrids that face the variability of renewable energy sources and demands. BESs can store surplus renewable energy for nearly instantaneous use, while GTs offer sustained energy output over longer periods, offering complementary benefits. Previous studies often oversimplified GT operations, neglecting critical factors such as ramp-up times and increased degradation from frequent starts. This research addresses these gaps by proposing an advanced modeling framework that accurately captures the dynamic interaction between GTs and BESs, including GT ramp-up times and maintenance costs associated with operational cycles. Through extensive case studies involving diverse microgrid configurations, we demonstrate that DRL effectively learns dispatch policies directly from historical data, outperforming traditional optimization techniques. Deploying DRL to our framework yields more realistic dispatch policies, reducing GT maintenance costs by avoiding frequent starts. The proposed framework has significant potential to improve energy management strategies and to streamline the planning of hybrid energy systems. To encourage further research, we have released our codebase to the public, enabling the scientific community to build upon our findings.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100653"},"PeriodicalIF":9.6,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145680874","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 intelligent multi-layer framework for attack conduction, detection and reconstruction in the smart grid with renewable energies 一种面向可再生能源智能电网攻击传导、检测与重构的多层智能框架

IF 9.6 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Energy and AI

Pub Date : 2025-11-19 DOI: 10.1016/j.egyai.2025.100636

Mostafa Mohammadpourfard , Yang Weng , Mahdi Eghbali , Manohar Chamana , Suhas Pol

The rapid proliferation of renewable energy sources (RESs) has enhanced operational flexibility but intensified cybersecurity concerns in modern power systems. In this work, we investigate how attackers can exploit the increased variability introduced by RESs to orchestrate false data injection attacks (FDIAs). First, we propose a targeted attack strategy based on Jensen–Shannon divergence (JSD) and the Kolmogorov–Smirnov (KS) test. This two-stage procedure identifies measurements that exhibit minimal distributional shifts after RES integration. False data are then injected into these stable measurements, blending seamlessly into the expanded measurement space and increasing attack stealth. Second, we develop a customized hybrid deep learning model, combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) units to capture both spatial correlations and temporal dynamics in power system measurements. This design explicitly addresses concept drift arising from fluctuating load and generation profiles, ensuring persistent detection accuracy. Third, we integrate an Autoencoder (AE)–based reconstruction mechanism to repair compromised measurements upon detection, mitigating denial-of-service (DoS) scenarios that could result from discarding suspect data. Our evaluations on the IEEE 14-bus and 118-bus systems, using real-world load profiles, confirm that the JSD–KS approach boosts attack stealth while the CNN–LSTM–AE pipeline achieves robust detection and recovery. Our experiments on the IEEE 14-bus and 118-bus systems demonstrate F1-score gains of up to 3% over the strongest CLSTM baseline under traditional FDIA scenarios, and up to 13% under our intelligent FDIA, while also reducing AE reconstruction RMSE by approximately 6%–7%. This integrated strategy offers a multi-layered defense against evolving cyber threats in renewable-rich smart grids.

可再生能源（RESs）的快速扩散提高了现代电力系统的运行灵活性，但也加剧了对网络安全的担忧。在这项工作中，我们研究了攻击者如何利用RESs引入的增加的可变性来编排虚假数据注入攻击（FDIAs）。首先，我们提出了一种基于Jensen-Shannon散度（JSD）和Kolmogorov-Smirnov （KS）检验的针对性攻击策略。这个两阶段的程序确定了RES整合后表现出最小分布变化的测量值。然后将虚假数据注入这些稳定的测量中，无缝地融合到扩展的测量空间中，并增加攻击的隐蔽性。其次，我们开发了一个定制的混合深度学习模型，将卷积神经网络（cnn）和长短期记忆（LSTM）单元结合起来，以捕获电力系统测量中的空间相关性和时间动态。这种设计明确地解决了由波动负载和发电剖面引起的概念漂移，确保了持久的检测精度。第三，我们集成了基于自动编码器（AE）的重建机制，以在检测时修复受损的测量，减轻因丢弃可疑数据而导致的拒绝服务（DoS）场景。我们对IEEE 14总线和118总线系统的评估，使用真实的负载配置文件，证实了JSD-KS方法提高了攻击的隐身性，而CNN-LSTM-AE管道实现了强大的检测和恢复。我们在IEEE 14总线和111总线系统上的实验表明，在传统FDIA场景下，f1得分比最强CLSTM基线提高了3%，在我们的智能FDIA场景下提高了13%，同时也将声发射重建RMSE降低了大约6%-7%。这种综合战略为可再生能源丰富的智能电网提供了多层次的防御，以应对不断发展的网络威胁。

{"title":"An intelligent multi-layer framework for attack conduction, detection and reconstruction in the smart grid with renewable energies","authors":"Mostafa Mohammadpourfard , Yang Weng , Mahdi Eghbali , Manohar Chamana , Suhas Pol","doi":"10.1016/j.egyai.2025.100636","DOIUrl":"10.1016/j.egyai.2025.100636","url":null,"abstract":"<div><div>The rapid proliferation of renewable energy sources (RESs) has enhanced operational flexibility but intensified cybersecurity concerns in modern power systems. In this work, we investigate how attackers can exploit the increased variability introduced by RESs to orchestrate false data injection attacks (FDIAs). First, we propose a targeted attack strategy based on Jensen–Shannon divergence (JSD) and the Kolmogorov–Smirnov (KS) test. This two-stage procedure identifies measurements that exhibit minimal distributional shifts after RES integration. False data are then injected into these stable measurements, blending seamlessly into the expanded measurement space and increasing attack stealth. Second, we develop a customized hybrid deep learning model, combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) units to capture both spatial correlations and temporal dynamics in power system measurements. This design explicitly addresses concept drift arising from fluctuating load and generation profiles, ensuring persistent detection accuracy. Third, we integrate an Autoencoder (AE)–based reconstruction mechanism to repair compromised measurements upon detection, mitigating denial-of-service (DoS) scenarios that could result from discarding suspect data. Our evaluations on the IEEE 14-bus and 118-bus systems, using real-world load profiles, confirm that the JSD–KS approach boosts attack stealth while the CNN–LSTM–AE pipeline achieves robust detection and recovery. Our experiments on the IEEE 14-bus and 118-bus systems demonstrate F1-score gains of up to 3% over the strongest CLSTM baseline under traditional FDIA scenarios, and up to 13% under our intelligent FDIA, while also reducing AE reconstruction RMSE by approximately 6%–7%. This integrated strategy offers a multi-layered defense against evolving cyber threats in renewable-rich smart grids.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"23 ","pages":"Article 100636"},"PeriodicalIF":9.6,"publicationDate":"2025-11-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145739285","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

Autonomous wind turbine performance curve modeling based on SCADA data: A vision intelligence powered method 基于SCADA数据的自主风力机性能曲线建模：一种视觉智能驱动方法

IF 9.6 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Energy and AI

Pub Date : 2025-11-07 DOI: 10.1016/j.egyai.2025.100645

Chufan Wu , Luoxiao Yang , Zijun Zhang

A central question in efficient wind farm big-data analytics is how to design an algorithm for autonomously extracting performance curves of wind turbines based on data collected via wind farm supervisory control and data acquisition (SCADA) systems. This paper investigates this question systematically, focusing on a challenging setting: the end-to-end autonomous analytics for directly generating mathematical functions of wind turbine performance curves from raw SCADA data. We propose a vision generative modeling (VGM) paradigm for autonomous development of wind turbine performance curve models. We discover that, compared with prevalently discussed numerical fitting-based performance curve modeling (NFM) methods, VGM directly working on raw data without any data preprocessing and model parameter tuning offers more generalizable and accurate results in deriving performance curves as well as their mathematical forms. The success of VGM is achieved by three computational steps developed in this study. By comparing with a set of state-of-the-art NFM benchmarks in multiple performance curve modeling tasks, we observe that VGM consistently performs more advantageously by achieving a 75.1% accuracy improvement in wind power curve modeling with insufficient SCADA data and an 84.3% improvement in modeling the rotor speed curve based on faulty field data. This work presents a milestone in autonomous wind turbine SCADA data analytics, which possesses a great potential of spanning to autonomous analytics of measured data of other industrial systems.

高效风电场大数据分析的一个核心问题是，如何设计一种算法，根据风电场监控和数据采集（SCADA）系统收集的数据，自动提取风力涡轮机的性能曲线。本文系统地研究了这个问题，重点研究了一个具有挑战性的设置：从原始SCADA数据直接生成风力涡轮机性能曲线的数学函数的端到端自主分析。我们提出了一种视觉生成建模（VGM）范式，用于风力发电机性能曲线模型的自主开发。我们发现，与目前广泛讨论的基于数值拟合的性能曲线建模（NFM）方法相比，VGM直接处理原始数据而不进行任何数据预处理和模型参数调优，在导出性能曲线及其数学形式方面提供了更通用和准确的结果。VGM的成功是通过本研究开发的三个计算步骤实现的。通过在多个性能曲线建模任务中与一组最先进的NFM基准进行比较，我们观察到VGM始终表现出更大的优势，在SCADA数据不足的情况下，风电曲线建模精度提高了75.1%，在基于故障现场数据的转子转速曲线建模精度提高了84.3%。这项工作是风力发电机SCADA数据自主分析的一个里程碑，具有跨越其他工业系统测量数据自主分析的巨大潜力。

{"title":"Autonomous wind turbine performance curve modeling based on SCADA data: A vision intelligence powered method","authors":"Chufan Wu , Luoxiao Yang , Zijun Zhang","doi":"10.1016/j.egyai.2025.100645","DOIUrl":"10.1016/j.egyai.2025.100645","url":null,"abstract":"<div><div>A central question in efficient wind farm big-data analytics is how to design an algorithm for autonomously extracting performance curves of wind turbines based on data collected via wind farm supervisory control and data acquisition (SCADA) systems. This paper investigates this question systematically, focusing on a challenging setting: the end-to-end autonomous analytics for directly generating mathematical functions of wind turbine performance curves from raw SCADA data. We propose a vision generative modeling (VGM) paradigm for autonomous development of wind turbine performance curve models. We discover that, compared with prevalently discussed numerical fitting-based performance curve modeling (NFM) methods, VGM directly working on raw data without any data preprocessing and model parameter tuning offers more generalizable and accurate results in deriving performance curves as well as their mathematical forms. The success of VGM is achieved by three computational steps developed in this study. By comparing with a set of state-of-the-art NFM benchmarks in multiple performance curve modeling tasks, we observe that VGM consistently performs more advantageously by achieving a 75.1% accuracy improvement in wind power curve modeling with insufficient SCADA data and an 84.3% improvement in modeling the rotor speed curve based on faulty field data. This work presents a milestone in autonomous wind turbine SCADA data analytics, which possesses a great potential of spanning to autonomous analytics of measured data of other industrial systems.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100645"},"PeriodicalIF":9.6,"publicationDate":"2025-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145519753","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

Explainable artificial intelligence relates perovskite luminescence images to current-voltage metrics 可解释的人工智能将钙钛矿发光图像与电流-电压指标联系起来

IF 9.6 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Energy and AI

Pub Date : 2025-11-03 DOI: 10.1016/j.egyai.2025.100640

Andrew Glaws , Jackson W. Schall , Andrew Ballen , Amy E. Louks , Kristopher O. Davis , Axel F. Palmstrom , Juliette Ugirumurera , Dana B. Kern

As the demand for low-cost, high-efficiency solar energy technologies grows, metal halide perovskite (MHP) solar cells have emerged as a promising candidate for next-generation photovoltaics due to their high power conversion efficiencies. However, their poor durability and issues with manufacturing consistency remain significant barriers to commercialization. In this work, we develop deep learning models to support materials characterization and provide insight into features and processes influencing performance. The models are trained using transfer learning of a pretrained model to predict relevant current–voltage (IV) metrics based on different combinations of input electroluminescence (EL) and photoluminescence (PL) images of MHP devices. We examine which image types are most informative in accurately predicting different IV metrics. Additionally, we use explainable artificial intelligence (XAI) techniques to provide insights into specific spatial features in the devices that drive differences in performance. We find that stabilized luminescence images (e.g. those collected after biasing the devices for at least 1 min) are better for predicting metrics of open-circuit voltage (by PL) and short-circuit current (by PL with EL), but that predicting fill factor and overall power output may use the time-evolution of EL images. Based on attribution masks generated by integrated gradients for each device performance metric, we further suggest different loss mechanisms associated with categories of large and small spatial defects. Overall, this case study highlights the potential applicability of XAI methodology for streamlining MHP device analysis and accelerating detailed understanding of the relationships between spatial defects and impacts on performance.

随着对低成本、高效率太阳能技术的需求不断增长，金属卤化物钙钛矿（MHP）太阳能电池因其高功率转换效率而成为下一代光伏电池的有希望的候选者。然而，它们的耐久性差和制造一致性问题仍然是商业化的重大障碍。在这项工作中，我们开发了深度学习模型来支持材料表征，并提供对影响性能的特征和过程的洞察。使用预训练模型的迁移学习来训练模型，以基于MHP器件的输入电致发光（EL）和光致发光（PL）图像的不同组合来预测相关的电流-电压（IV）指标。我们研究了哪些图像类型在准确预测不同的静脉注射指标方面最具信息性。此外，我们使用可解释的人工智能（XAI）技术来提供对设备中驱动性能差异的特定空间特征的见解。我们发现稳定的发光图像（例如，在器件偏置至少1分钟后收集的图像）更适合预测开路电压（通过PL）和短路电流（通过带有EL的PL）的指标，但预测填充因子和总功率输出可能使用EL图像的时间演变。基于每个器件性能指标的集成梯度生成的归因掩模，我们进一步提出了与大小空间缺陷类别相关的不同损耗机制。总的来说，这个案例研究强调了XAI方法在简化MHP设备分析和加速空间缺陷与性能影响之间关系的详细理解方面的潜在适用性。

{"title":"Explainable artificial intelligence relates perovskite luminescence images to current-voltage metrics","authors":"Andrew Glaws , Jackson W. Schall , Andrew Ballen , Amy E. Louks , Kristopher O. Davis , Axel F. Palmstrom , Juliette Ugirumurera , Dana B. Kern","doi":"10.1016/j.egyai.2025.100640","DOIUrl":"10.1016/j.egyai.2025.100640","url":null,"abstract":"<div><div>As the demand for low-cost, high-efficiency solar energy technologies grows, metal halide perovskite (MHP) solar cells have emerged as a promising candidate for next-generation photovoltaics due to their high power conversion efficiencies. However, their poor durability and issues with manufacturing consistency remain significant barriers to commercialization. In this work, we develop deep learning models to support materials characterization and provide insight into features and processes influencing performance. The models are trained using transfer learning of a pretrained model to predict relevant current–voltage (IV) metrics based on different combinations of input electroluminescence (EL) and photoluminescence (PL) images of MHP devices. We examine which image types are most informative in accurately predicting different IV metrics. Additionally, we use explainable artificial intelligence (XAI) techniques to provide insights into specific spatial features in the devices that drive differences in performance. We find that stabilized luminescence images (e.g. those collected after biasing the devices for at least 1 min) are better for predicting metrics of open-circuit voltage (by PL) and short-circuit current (by PL with EL), but that predicting fill factor and overall power output may use the time-evolution of EL images. Based on attribution masks generated by integrated gradients for each device performance metric, we further suggest different loss mechanisms associated with categories of large and small spatial defects. Overall, this case study highlights the potential applicability of XAI methodology for streamlining MHP device analysis and accelerating detailed understanding of the relationships between spatial defects and impacts on performance.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"22 ","pages":"Article 100640"},"PeriodicalIF":9.6,"publicationDate":"2025-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145465940","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