Engineering Applications of Artificial Intelligence最新文献_第2页

Transformer-based offline-to-online reinforcement learning for decision-making and control in autonomous driving 基于变压器的离线到在线的自动驾驶决策控制强化学习

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-04-15 Epub Date: 2026-02-13 DOI: 10.1016/j.engappai.2026.114139

Feihong Tan , Ping Lu , Fulin Zhang , Xin Ye , Bo Hu , Xing Shu

Developing robust decision-making and control systems for autonomous driving in complex, dynamic environments involving multi-vehicle interactions at intersections, roundabouts, and merging ramps remains a significant hurdle. In this context, Reinforcement Learning (RL) emerges as a highly promising approach. The primary methods for applying RL, however, present a core dilemma. On one hand, offline RL cannot adapt well to real-world conditions because it learns from a fixed dataset. On the other hand, online RL requires learning through real-world interaction, which is inherently unsafe for driving. To address these issues, this paper proposes a Transformer-based Offline-to-online Reinforcement Learning (TORL) framework. Firstly, the framework's offline learning paradigm integrates a Transformer architecture with a maximum entropy mechanism. This synergistic approach allows the model to capture long-term temporal dependencies for high-performance decision-making and control while ensuring the initial policy is robust and generalizable. Building on this foundation, the framework employs a trifecta of synergistic mechanisms during online fine-tuning, including Human-in-the-Loop (HITL) safe exploration, a hybrid replay buffer, and a mixed data-source learning approach, to simultaneously mitigate performance degradation from distributional shifts and neutralize the critical safety risks of online exploration. Comprehensive experiments conducted in the MetaDrive simulation environment demonstrate that TORL surpasses baseline methods, achieving an absolute increase of approximately 29.4% in normalized return and 46.1% in task success rate, while maintaining a zero-collision record. Furthermore, the framework's real-time feasibility was validated on an experimental autonomous vehicle platform, demonstrating low computational latency suitable for practical deployment. This study demonstrates that the proposed offline-to-online RL paradigm offers a robust and effective solution for developing high-performance decision-making and control systems for autonomous vehicles.

在复杂的动态环境中，包括交叉路口、环形交叉路口和合并坡道上的多车交互，为自动驾驶开发强大的决策和控制系统仍然是一个重大障碍。在这种情况下，强化学习（RL）成为一种非常有前途的方法。然而，应用强化学习的主要方法存在一个核心难题。一方面，离线强化学习不能很好地适应现实世界的条件，因为它从固定的数据集学习。另一方面，在线强化学习需要通过现实世界的互动进行学习，这对驾驶来说本质上是不安全的。为了解决这些问题，本文提出了一种基于变压器的离线到在线强化学习（TORL）框架。首先，该框架的离线学习范式集成了具有最大熵机制的Transformer体系结构。这种协同方法允许模型捕获长期的时间依赖性，以便进行高性能的决策和控制，同时确保初始策略是健壮的和可推广的。在此基础上，该框架在在线微调期间采用了三种协同机制，包括人在环（HITL）安全探索、混合重播缓冲区和混合数据源学习方法，以同时减轻分布转移带来的性能下降，并消除在线探索的关键安全风险。在MetaDrive仿真环境中进行的综合实验表明，TORL超越了基线方法，在保持零碰撞记录的同时，实现了约29.4%的归一化收益率和46.1%的任务成功率的绝对增长。此外，在实验自动驾驶汽车平台上验证了该框架的实时性可行性，证明了适合实际部署的低计算延迟。该研究表明，提出的离线到在线RL范式为开发高性能的自动驾驶车辆决策和控制系统提供了一个鲁棒和有效的解决方案。

{"title":"Transformer-based offline-to-online reinforcement learning for decision-making and control in autonomous driving","authors":"Feihong Tan , Ping Lu , Fulin Zhang , Xin Ye , Bo Hu , Xing Shu","doi":"10.1016/j.engappai.2026.114139","DOIUrl":"10.1016/j.engappai.2026.114139","url":null,"abstract":"<div><div>Developing robust decision-making and control systems for autonomous driving in complex, dynamic environments involving multi-vehicle interactions at intersections, roundabouts, and merging ramps remains a significant hurdle. In this context, Reinforcement Learning (RL) emerges as a highly promising approach. The primary methods for applying RL, however, present a core dilemma. On one hand, offline RL cannot adapt well to real-world conditions because it learns from a fixed dataset. On the other hand, online RL requires learning through real-world interaction, which is inherently unsafe for driving. To address these issues, this paper proposes a Transformer-based Offline-to-online Reinforcement Learning (TORL) framework. Firstly, the framework's offline learning paradigm integrates a Transformer architecture with a maximum entropy mechanism. This synergistic approach allows the model to capture long-term temporal dependencies for high-performance decision-making and control while ensuring the initial policy is robust and generalizable. Building on this foundation, the framework employs a trifecta of synergistic mechanisms during online fine-tuning, including Human-in-the-Loop (HITL) safe exploration, a hybrid replay buffer, and a mixed data-source learning approach, to simultaneously mitigate performance degradation from distributional shifts and neutralize the critical safety risks of online exploration. Comprehensive experiments conducted in the MetaDrive simulation environment demonstrate that TORL surpasses baseline methods, achieving an absolute increase of approximately 29.4% in normalized return and 46.1% in task success rate, while maintaining a zero-collision record. Furthermore, the framework's real-time feasibility was validated on an experimental autonomous vehicle platform, demonstrating low computational latency suitable for practical deployment. This study demonstrates that the proposed offline-to-online RL paradigm offers a robust and effective solution for developing high-performance decision-making and control systems for autonomous vehicles.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"170 ","pages":"Article 114139"},"PeriodicalIF":8.0,"publicationDate":"2026-04-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146161831","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A quantum group decision-making model for patient-capital project selection integrating cumulative prospect theory under linear Diophantine fuzzy uncertainty 线性丢番图模糊不确定性下结合累积前景理论的耐心资本项目选择量子群决策模型

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-04-15 Epub Date: 2026-02-14 DOI: 10.1016/j.engappai.2026.114169

Wen Li , Fanfan Mu , Zhiliang Ren , Obaid Ur Rehman

The effective selection of patient-capital projects is crucial for promoting China’s transition toward low-carbon development, technological innovation, and sustainable value creation. However, such decision-making processes typically involve multiple experts whose assessments are affected by individual risk preferences and inter-expert opinion interference—factors that are seldom modeled simultaneously yet jointly exert significant influence in existing studies. To address these limitations, this study develops an integrated multi-criteria quantum group decision-making framework for patient-capital project selection, which combines linear Diophantine fuzzy sets (LDFSs), cumulative prospect theory (CPT), quantum probability theory (QPT), and the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS). First, a linear Diophantine fuzzy Deng entropy (LDFDE) method is proposed to characterize the information uncertainty inherent in LDFSs. Second, a personalized criterion-weighting approach that integrates LDFDE with the classical entropy weighting method is developed to reflect both objective data variability and subjective expert preferences. Third, an integrated quantum CPT-TOPSIS model is constructed, in which the CPT-TOPSIS component captures behavioral biases and risk attitudes, while the QPT-based aggregation component models opinion interference to achieve more consistent and robust group decisions. Finally, a real-world case study on patient-capital project selection demonstrates the practicality and effectiveness of the proposed framework. We also report sensitivity and comparative analyses to validate that the proposed approach enhances ranking stability and reliability. In summary, the findings underscore the potential of the model as a reliable decision-support tool for complex, uncertain, and psychologically driven investment scenarios.

耐心资本项目的有效选择对于推动中国向低碳发展、技术创新和可持续价值创造转型至关重要。然而，这样的决策过程通常涉及多位专家，他们的评估受到个人风险偏好和专家间意见干扰的影响——这些因素很少同时建模，但在现有研究中却共同产生重大影响。为了解决这些限制，本研究开发了一个集成的多准则量子群体决策框架，用于患者资本项目选择，该框架结合了线性丢phantine模糊集（LDFSs）、累积前景理论（CPT）、量子概率论（QPT）和理想解相似性排序偏好技术（TOPSIS）。首先，提出了一种线性丢芬图模糊邓熵（LDFDE）方法来表征LDFSs固有的信息不确定性。其次，将LDFDE与经典熵权法相结合，提出了一种个性化的标准加权方法，以反映客观数据的可变性和主观专家的偏好。第三，构建了一个集成的量子CPT-TOPSIS模型，其中CPT-TOPSIS组件捕获行为偏差和风险态度，而基于qpt的聚合组件建模意见干扰，以实现更一致和鲁棒的群体决策。最后，对耐心资本项目选择的实际案例研究证明了所提出框架的实用性和有效性。我们还报告了敏感性和比较分析，以验证所提出的方法提高了排名的稳定性和可靠性。总之，研究结果强调了该模型作为复杂、不确定和心理驱动的投资场景的可靠决策支持工具的潜力。

{"title":"A quantum group decision-making model for patient-capital project selection integrating cumulative prospect theory under linear Diophantine fuzzy uncertainty","authors":"Wen Li , Fanfan Mu , Zhiliang Ren , Obaid Ur Rehman","doi":"10.1016/j.engappai.2026.114169","DOIUrl":"10.1016/j.engappai.2026.114169","url":null,"abstract":"<div><div>The effective selection of patient-capital projects is crucial for promoting China’s transition toward low-carbon development, technological innovation, and sustainable value creation. However, such decision-making processes typically involve multiple experts whose assessments are affected by individual risk preferences and inter-expert opinion interference—factors that are seldom modeled simultaneously yet jointly exert significant influence in existing studies. To address these limitations, this study develops an integrated multi-criteria quantum group decision-making framework for patient-capital project selection, which combines linear Diophantine fuzzy sets (LDFSs), cumulative prospect theory (CPT), quantum probability theory (QPT), and the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS). First, a linear Diophantine fuzzy Deng entropy (LDFDE) method is proposed to characterize the information uncertainty inherent in LDFSs. Second, a personalized criterion-weighting approach that integrates LDFDE with the classical entropy weighting method is developed to reflect both objective data variability and subjective expert preferences. Third, an integrated quantum CPT-TOPSIS model is constructed, in which the CPT-TOPSIS component captures behavioral biases and risk attitudes, while the QPT-based aggregation component models opinion interference to achieve more consistent and robust group decisions. Finally, a real-world case study on patient-capital project selection demonstrates the practicality and effectiveness of the proposed framework. We also report sensitivity and comparative analyses to validate that the proposed approach enhances ranking stability and reliability. In summary, the findings underscore the potential of the model as a reliable decision-support tool for complex, uncertain, and psychologically driven investment scenarios.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"170 ","pages":"Article 114169"},"PeriodicalIF":8.0,"publicationDate":"2026-04-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146193014","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Theory-guided data-driven based on the learning curve for fracturing performance prediction 基于学习曲线的理论导向数据驱动压裂动态预测

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-04-15 Epub Date: 2026-02-13 DOI: 10.1016/j.engappai.2026.114163

Yunjin Wang , Leyi Zheng , Gong Chen , Jianlong Zhang , Hao Bai , Hanxuan Song , Tingxue Jiang , Fujian Zhou

Accurate and robust prediction of fracturing performance is essential for optimizing fracturing strategies. Here, a fracturing learning curve is proposed based on the fracturing characteristics in Gimsar shale oil, and is used as a theoretical guide to build a theory-guided data-driven (TgDD) model to predict the fracturing performance. The fracturing learning curve is further decomposed into dimensionless trends and local fluctuations. Convolutional neural network (CNN) and gated recurrent unit (GRU) were combined to build a CNN-GRU to predict the dimensionless trend. Using adaptive boosting (AdaBoost) integrated random forest (RF) to build an AdaBoost-RF to predict the local fluctuations. The results show that dimensionless trend has time series characteristics. CNN-GRU can extract and select the features, and its prediction ability is 28.1 % and 12.9 % higher than that of CNN and GRU. AdaBoost-RF can dynamically adjust the weights, and its prediction ability is about 37% higher than that of the RF. TgDD is more sensitive to engineering parameters. Relative to the direct prediction, the prediction accuracy of the TgDD is improved by 47.6 %. There are two main reasons for the higher prediction accuracy of TgDD. One is that the dimensionless trend belongs to the time series data, for which the established CNN-GRU model has an extremely strong prediction ability. The second is that the fluctuation amplitude of local fluctuations is reduced, which improves the data quality. The engineering parameters of the newly fractured wells were optimized using TgDD, and its estimated ultimate recovery was improved from 0.4847 to 0.4917.

准确、可靠的压裂性能预测对于优化压裂策略至关重要。基于Gimsar页岩油的压裂特征，提出了压裂学习曲线，并将其作为理论指导，建立理论指导数据驱动（TgDD）模型，预测压裂性能。压裂学习曲线进一步分解为无量纲趋势和局部波动。将卷积神经网络（CNN）和门控循环单元（GRU）相结合，构建了CNN-GRU模型，用于预测无量纲趋势。利用自适应增强（AdaBoost）集成随机森林（RF）构建AdaBoost-RF预测局部波动。结果表明，无量纲趋势具有时间序列特征。CNN-GRU能够对特征进行提取和选择，预测能力分别比CNN和GRU高28.1%和12.9%。AdaBoost-RF可以动态调整权重，其预测能力比RF提高约37%。TgDD对工程参数更为敏感。与直接预报相比，TgDD的预报精度提高了47.6%。TgDD预测精度较高的原因主要有两点。一是无量纲趋势属于时间序列数据，对此所建立的CNN-GRU模型具有极强的预测能力。二是降低了局部波动的波动幅度，提高了数据质量。利用TgDD对新压裂井的工程参数进行了优化，最终采收率由0.4847提高到0.4917。

{"title":"Theory-guided data-driven based on the learning curve for fracturing performance prediction","authors":"Yunjin Wang , Leyi Zheng , Gong Chen , Jianlong Zhang , Hao Bai , Hanxuan Song , Tingxue Jiang , Fujian Zhou","doi":"10.1016/j.engappai.2026.114163","DOIUrl":"10.1016/j.engappai.2026.114163","url":null,"abstract":"<div><div>Accurate and robust prediction of fracturing performance is essential for optimizing fracturing strategies. Here, a fracturing learning curve is proposed based on the fracturing characteristics in Gimsar shale oil, and is used as a theoretical guide to build a theory-guided data-driven (TgDD) model to predict the fracturing performance. The fracturing learning curve is further decomposed into dimensionless trends and local fluctuations. Convolutional neural network (CNN) and gated recurrent unit (GRU) were combined to build a CNN-GRU to predict the dimensionless trend. Using adaptive boosting (AdaBoost) integrated random forest (RF) to build an AdaBoost-RF to predict the local fluctuations. The results show that dimensionless trend has time series characteristics. CNN-GRU can extract and select the features, and its prediction ability is 28.1 % and 12.9 % higher than that of CNN and GRU. AdaBoost-RF can dynamically adjust the weights, and its prediction ability is about 37% higher than that of the RF. TgDD is more sensitive to engineering parameters. Relative to the direct prediction, the prediction accuracy of the TgDD is improved by 47.6 %. There are two main reasons for the higher prediction accuracy of TgDD. One is that the dimensionless trend belongs to the time series data, for which the established CNN-GRU model has an extremely strong prediction ability. The second is that the fluctuation amplitude of local fluctuations is reduced, which improves the data quality. The engineering parameters of the newly fractured wells were optimized using TgDD, and its estimated ultimate recovery was improved from 0.4847 to 0.4917.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"170 ","pages":"Article 114163"},"PeriodicalIF":8.0,"publicationDate":"2026-04-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146193284","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Multimodal emotion recognition from complete modality to missing modality based on text, audio, and visual: A review 基于文本、音频和视觉的多模态情感识别：从完整情态到缺失情态

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-04-15 Epub Date: 2026-02-13 DOI: 10.1016/j.engappai.2026.114127

Huihui Li , Huiqi Han , Chunlin Xu , Tongbao Chen , Xiaoyong Liu , Guihua Wen

Multimodal Emotion Recognition (MER), as a key component of affective computing, significantly improves the accuracy and robustness of emotion recognition by integrating multiple modalities such as text, audio, and visual information. However, most existing studies are based on the assumption of data integrity, while missing modality data is inevitable in practical applications, which poses new challenges to MER. This paper, for the first time, conducts a comprehensive and systematic review of MER methods from complete modality to missing modality, covering the analysis of common datasets, feature extraction techniques, information fusion mechanisms, and the latest methods. In particular, we elaborate on the construction methods of missing modality data and conduct a comprehensive comparison of MER methods under both complete and missing modalities. Furthermore, we summarize the common evaluation metrics in the field of MER, deeply discuss the core challenges currently faced, and prospect the future research directions. This review aims to provide researchers with a comprehensive understanding of the state of MER technology, thereby offering directional suggestions for subsequent research.

多模态情感识别（MER）作为情感计算的重要组成部分，通过整合文本、音频和视觉等多模态信息，显著提高了情感识别的准确性和鲁棒性。然而，现有的研究大多是建立在数据完整性假设的基础上，而在实际应用中，模态数据的缺失是不可避免的，这对MER提出了新的挑战。本文首次从完整模态到缺失模态对MER方法进行了全面系统的综述，涵盖了常用数据集的分析、特征提取技术、信息融合机制以及最新方法。我们特别阐述了缺失模态数据的构建方法，并对完整模态和缺失模态下的MER方法进行了全面比较。在此基础上，总结了市场营销领域常用的评价指标，深入探讨了当前面临的核心挑战，并对未来的研究方向进行了展望。本文旨在让研究者对MER技术的现状有一个全面的了解，从而为后续的研究提供方向性的建议。

{"title":"Multimodal emotion recognition from complete modality to missing modality based on text, audio, and visual: A review","authors":"Huihui Li , Huiqi Han , Chunlin Xu , Tongbao Chen , Xiaoyong Liu , Guihua Wen","doi":"10.1016/j.engappai.2026.114127","DOIUrl":"10.1016/j.engappai.2026.114127","url":null,"abstract":"<div><div>Multimodal Emotion Recognition (MER), as a key component of affective computing, significantly improves the accuracy and robustness of emotion recognition by integrating multiple modalities such as text, audio, and visual information. However, most existing studies are based on the assumption of data integrity, while missing modality data is inevitable in practical applications, which poses new challenges to MER. This paper, for the first time, conducts a comprehensive and systematic review of MER methods from complete modality to missing modality, covering the analysis of common datasets, feature extraction techniques, information fusion mechanisms, and the latest methods. In particular, we elaborate on the construction methods of missing modality data and conduct a comprehensive comparison of MER methods under both complete and missing modalities. Furthermore, we summarize the common evaluation metrics in the field of MER, deeply discuss the core challenges currently faced, and prospect the future research directions. This review aims to provide researchers with a comprehensive understanding of the state of MER technology, thereby offering directional suggestions for subsequent research.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"170 ","pages":"Article 114127"},"PeriodicalIF":8.0,"publicationDate":"2026-04-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146161878","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Forecast-enhanced bilevel real-time pricing for microgrids via hybrid-action reinforcement learning 基于混合行动强化学习的微电网预测增强双层实时定价

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-04-15 Epub Date: 2026-02-14 DOI: 10.1016/j.engappai.2026.114195

Jingqi Wang , Yan Gao , Youmeng He

The integration of distributed energy resources into microgrids faces many complex challenges, including renewable intermittency, hybrid decision-making, and hierarchical coordination. This paper presents a forecast-enhanced bilevel real-time pricing framework using a hybrid-action deep reinforcement learning (DRL) algorithm with Gumbel-Softmax reparameterization. The framework manages both discrete generator commitment and continuous pricing decisions through integrated optimization. Our approach integrates Long Short-Term Memory (LSTM) forecasting to enhance proactive scheduling, while coordinating microgrid agents through a bilevel optimization architecture. The main innovations include: a hybrid-action DRL algorithm integrating Gumbel-Softmax reparameterization for joint discrete–continuous optimization; LSTM-based renewable forecasting integrated into state representation. Our DRL approach shows enhanced system performance with improved constraint satisfaction and operational efficiency, offering a practical solution for complex hybrid-action energy optimization problems.

分布式能源整合到微电网中面临着许多复杂的挑战，包括可再生能源间歇性、混合决策和分层协调。本文提出了一种基于Gumbel-Softmax再参数化的混合动作深度强化学习（DRL）算法的预测增强双层实时定价框架。该框架通过集成优化管理离散发电机承诺和连续定价决策。我们的方法集成了长短期记忆（LSTM）预测来增强主动调度，同时通过双层优化架构协调微电网代理。主要创新包括：一种结合Gumbel-Softmax再参数化的混合动作DRL算法，用于联合离散-连续优化；基于lstm的可再生预测与状态表示相结合。我们的DRL方法提高了系统性能，提高了约束满意度和运行效率，为复杂的混合作用能量优化问题提供了实用的解决方案。

{"title":"Forecast-enhanced bilevel real-time pricing for microgrids via hybrid-action reinforcement learning","authors":"Jingqi Wang , Yan Gao , Youmeng He","doi":"10.1016/j.engappai.2026.114195","DOIUrl":"10.1016/j.engappai.2026.114195","url":null,"abstract":"<div><div>The integration of distributed energy resources into microgrids faces many complex challenges, including renewable intermittency, hybrid decision-making, and hierarchical coordination. This paper presents a forecast-enhanced bilevel real-time pricing framework using a hybrid-action deep reinforcement learning (DRL) algorithm with Gumbel-Softmax reparameterization. The framework manages both discrete generator commitment and continuous pricing decisions through integrated optimization. Our approach integrates Long Short-Term Memory (LSTM) forecasting to enhance proactive scheduling, while coordinating microgrid agents through a bilevel optimization architecture. The main innovations include: a hybrid-action DRL algorithm integrating Gumbel-Softmax reparameterization for joint discrete–continuous optimization; LSTM-based renewable forecasting integrated into state representation. Our DRL approach shows enhanced system performance with improved constraint satisfaction and operational efficiency, offering a practical solution for complex hybrid-action energy optimization problems.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"170 ","pages":"Article 114195"},"PeriodicalIF":8.0,"publicationDate":"2026-04-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146193015","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Maximal margin hyper-ellipsoid support vector machine for multi-class classification 多类分类的最大边界超椭球支持向量机

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-04-01 Epub Date: 2026-02-12 DOI: 10.1016/j.engappai.2026.114055

Ting Ke, Mingzhu Meng, Feifei Yin

To address the low efficiency issues of support vector machine (SVM)-based multi-classification methods, the hyper-sphere support vector machine has been widely adopted. However, it still suffers from challenges such as feature correlation and inconsistent feature scales. To overcome these limitations, this paper proposes a maximal margin hyper-ellipsoid support vector machine (M³HE-SVM) approach. Unlike conventional methods that use Euclidean distance, this approach employs Mahalanobis distance for optimal margin measurement, aimed at not only decorrelating features, eliminating dimensional discrepancies, and achieving implicit feature selection, but also further capturing the geometric information of data and the probability distribution of the population. Extensive experiments are conducted on three categories of datasets: (1) a variety of representative synthetic datasets covering scenarios with linear separability, nonlinear distributions, class imbalance, non-spherical structures, and high-dimensional multi-class data; (2) multiple real-world datasets from the University of California, Irvine (UCI) Machine Learning Repository; and (3) large-scale real-world datasets and NDC datasets. Experimental results demonstrate that M³HE-SVM consistently outperforms the maximal margin hypersphere support vector machine (M³HS-SVM) and other traditional methods in both classification accuracy and testing efficiency, exhibiting strong robustness and generalization ability.

为了解决基于支持向量机（SVM）的多分类方法效率低的问题，超球面支持向量机被广泛采用。然而，它仍然面临着特征相关性和特征尺度不一致等挑战。为了克服这些局限性，本文提出了一种极大余量超椭球支持向量机（M3HE-SVM）方法。与传统方法使用欧几里得距离不同，该方法采用马氏距离进行最优边缘度量，不仅可以去除相关特征，消除维度差异，实现隐式特征选择，而且还可以进一步捕获数据的几何信息和总体的概率分布。在三类数据集上进行了大量的实验：(1)涵盖线性可分性、非线性分布、类不平衡、非球形结构和高维多类数据场景的各种具有代表性的合成数据集；(2)来自加州大学欧文分校（UCI）机器学习存储库的多个真实数据集；(3)大规模真实世界数据集和NDC数据集。实验结果表明，M3HE-SVM在分类精度和测试效率上均优于M3HS-SVM等传统方法，具有较强的鲁棒性和泛化能力。

{"title":"Maximal margin hyper-ellipsoid support vector machine for multi-class classification","authors":"Ting Ke, Mingzhu Meng, Feifei Yin","doi":"10.1016/j.engappai.2026.114055","DOIUrl":"10.1016/j.engappai.2026.114055","url":null,"abstract":"<div><div>To address the low efficiency issues of support vector machine (SVM)-based multi-classification methods, the hyper-sphere support vector machine has been widely adopted. However, it still suffers from challenges such as feature correlation and inconsistent feature scales. To overcome these limitations, this paper proposes a maximal margin hyper-ellipsoid support vector machine (M<sup>3</sup>HE-SVM) approach. Unlike conventional methods that use Euclidean distance, this approach employs Mahalanobis distance for optimal margin measurement, aimed at not only decorrelating features, eliminating dimensional discrepancies, and achieving implicit feature selection, but also further capturing the geometric information of data and the probability distribution of the population. Extensive experiments are conducted on three categories of datasets: (1) a variety of representative synthetic datasets covering scenarios with linear separability, nonlinear distributions, class imbalance, non-spherical structures, and high-dimensional multi-class data; (2) multiple real-world datasets from the University of California, Irvine (UCI) Machine Learning Repository; and (3) large-scale real-world datasets and NDC datasets. Experimental results demonstrate that M<sup>3</sup>HE-SVM consistently outperforms the maximal margin hypersphere support vector machine (M<sup>3</sup>HS-SVM) and other traditional methods in both classification accuracy and testing efficiency, exhibiting strong robustness and generalization ability.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114055"},"PeriodicalIF":8.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174799","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Self-discharge estimation for lithium-ion batteries based on formation data in production 基于生产地层数据的锂离子电池自放电估计

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-04-01 Epub Date: 2026-02-12 DOI: 10.1016/j.engappai.2026.114180

Haoyuan Zheng , Shaobin Yang , Weihua Xue , Shouzhen Xiao , Ding Shen , Wei Dong , Xu Zhang

Global annual shipments of lithium-ion batteries reached 1545.1 GW-hours (GWh) in 2024, representing a substantial increase. Notably, the energy-storage segment alone experienced a year-on-year growth of 64.9 %. Prior to dispatch, lithium-ion batteries must undergo self-discharge testing to ensure safety and reliability. In practice, identifying the approximately 2% of batteries exhibiting excessive self-discharge requires a prolonged resting period (10-30 days) to track self-discharge voltage drop (SDV-drop), which accounts for nearly two-thirds of the overall production cycle and severely limits manufacturing efficiency. Rapid and accurate prediction of self-discharge behavior has thus become a pressing engineering challenge. This study presents an artificial intelligence enabled framework that predicts a 28-day voltage drop using formation-stage data, thereby obviating the prolonged rest period. The approach integrates latent feature extraction from charge-discharge curves, unsupervised clustering, and transfer learning. Specifically, both comprehensive temporal and static features are automatically extracted from current, voltage, and capacity trajectories, along with scalar performance indicators. A hybrid K-means-t-distributed stochastic neighbor embedding (t-SNE) algorithm partitions the dataset into internally homogeneous clusters, enhancing intra-cluster consistency and inter-cluster separability. During transfer learning, maximum mean discrepancy aligns feature distributions between source and target domains, while a feature-label consistency constraint further mitigates domain shift and improves generalization. Comparative experiments demonstrate that the proposed model markedly outperforms state-of-the-art baselines in predicting SDV-drop. This framework thus provides a theoretical foundation and practical pathway for rapid self-discharge assessment, which enables significant reductions in production cycle time and improves manufacturing efficiency.

2024年，全球锂离子电池年出货量达到1545.1吉瓦时（GWh），大幅增长。值得注意的是，仅储能部分就经历了64.9%的同比增长。锂离子电池在出厂前必须进行自放电测试，以确保安全性和可靠性。在实践中，识别大约2%的电池表现出过度自放电，需要很长时间（10-30天）来跟踪自放电电压降（SDV-drop），这占整个生产周期的近三分之二，严重限制了制造效率。因此，快速准确地预测自放电行为已成为一项紧迫的工程挑战。该研究提出了一种人工智能框架，可以使用地层阶段数据预测28天的电压降，从而避免了长时间的休息时间。该方法集成了从充放电曲线中提取潜在特征、无监督聚类和迁移学习。具体来说，从电流、电压和容量轨迹以及标量性能指标中自动提取综合的时间和静态特征。混合k均值-t分布随机邻居嵌入（t-SNE）算法将数据集划分为内部均匀的簇，增强了簇内一致性和簇间可分离性。在迁移学习过程中，最大平均差异使源域和目标域之间的特征分布保持一致，而特征标签一致性约束进一步减轻了域移动并提高了泛化。对比实验表明，所提出的模型在预测sdv下降方面明显优于最先进的基线。该框架为快速自放电评估提供了理论基础和实践途径，可显著缩短生产周期时间，提高制造效率。

{"title":"Self-discharge estimation for lithium-ion batteries based on formation data in production","authors":"Haoyuan Zheng , Shaobin Yang , Weihua Xue , Shouzhen Xiao , Ding Shen , Wei Dong , Xu Zhang","doi":"10.1016/j.engappai.2026.114180","DOIUrl":"10.1016/j.engappai.2026.114180","url":null,"abstract":"<div><div>Global annual shipments of lithium-ion batteries reached 1545.1 GW-hours (GWh) in 2024, representing a substantial increase. Notably, the energy-storage segment alone experienced a year-on-year growth of 64.9 %. Prior to dispatch, lithium-ion batteries must undergo self-discharge testing to ensure safety and reliability. In practice, identifying the approximately 2% of batteries exhibiting excessive self-discharge requires a prolonged resting period (10-30 days) to track self-discharge voltage drop (SDV-drop), which accounts for nearly two-thirds of the overall production cycle and severely limits manufacturing efficiency. Rapid and accurate prediction of self-discharge behavior has thus become a pressing engineering challenge. This study presents an artificial intelligence enabled framework that predicts a 28-day voltage drop using formation-stage data, thereby obviating the prolonged rest period. The approach integrates latent feature extraction from charge-discharge curves, unsupervised clustering, and transfer learning. Specifically, both comprehensive temporal and static features are automatically extracted from current, voltage, and capacity trajectories, along with scalar performance indicators. A hybrid K-means-t-distributed stochastic neighbor embedding (t-SNE) algorithm partitions the dataset into internally homogeneous clusters, enhancing intra-cluster consistency and inter-cluster separability. During transfer learning, maximum mean discrepancy aligns feature distributions between source and target domains, while a feature-label consistency constraint further mitigates domain shift and improves generalization. Comparative experiments demonstrate that the proposed model markedly outperforms state-of-the-art baselines in predicting SDV-drop. This framework thus provides a theoretical foundation and practical pathway for rapid self-discharge assessment, which enables significant reductions in production cycle time and improves manufacturing efficiency.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114180"},"PeriodicalIF":8.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174800","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Bidirectional encoder representations from transformer fusion quantum dual-stage attention bidirectional gated recurrent unit and diffusion method for short-term wind power prediction 基于变压器融合量子双阶段注意力双向门控循环单元的双向编码器表示与扩散法的短期风电预测

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-04-01 Epub Date: 2026-02-11 DOI: 10.1016/j.engappai.2026.114095

Linfei Yin, Yufeng Liu

With wind energy increasing proportion in renewable energy structure, wind energy is already a backbone in low carbon energy structure. Short-term wind power prediction can assist the demand for real-time dispatching of wind farms and power grids. Regard to the problems of low prediction accuracy and long training time of existing prediction models for short-term wind power prediction, this study proposes a large-model bidirectional encoder representations from Transformer fusion quantum dual-stage attention bidirectional gated recurrent unit and diffusion method for short-term wind power prediction. The proposed method utilizes improved complete ensemble empirical mode decomposition with adaptive noise to decompose the wind power, and then the decomposed data are input into the quantum dual-stage attention bidirectional gated recurrent unit and quantum diffusion model for training prediction; then, the bidirectional encoder representations from Transformer provides final wind power prediction. Compared with 52 prediction algorithms, the average absolute error of the proposed method is more than 30.57% less. Furthermore, the addition of parameterized quantum circuits shortens training prediction time by nearly 25%.

随着风能在可再生能源结构中的比重不断提高，风能已经成为低碳能源结构中的中坚力量。短期风电预测可以辅助风电场和电网的实时调度需求。针对现有短期风电预测模型预测精度低、训练时间长等问题，本研究提出了一种基于Transformer融合量子双阶段注意力双向门控循环单元和扩散方法的大模型双向编码器表示用于短期风电预测。该方法利用改进的全系综经验模态分解和自适应噪声对风电进行分解，然后将分解后的数据输入到量子双阶段注意力双向门控循环单元和量子扩散模型中进行训练预测；然后，变压器的双向编码器表示提供最终的风电预测。与52种预测算法相比，该方法的平均绝对误差小于30.57%以上。此外，参数化量子电路的加入使训练预测时间缩短了近25%。

{"title":"Bidirectional encoder representations from transformer fusion quantum dual-stage attention bidirectional gated recurrent unit and diffusion method for short-term wind power prediction","authors":"Linfei Yin, Yufeng Liu","doi":"10.1016/j.engappai.2026.114095","DOIUrl":"10.1016/j.engappai.2026.114095","url":null,"abstract":"<div><div>With wind energy increasing proportion in renewable energy structure, wind energy is already a backbone in low carbon energy structure. Short-term wind power prediction can assist the demand for real-time dispatching of wind farms and power grids. Regard to the problems of low prediction accuracy and long training time of existing prediction models for short-term wind power prediction, this study proposes a large-model bidirectional encoder representations from Transformer fusion quantum dual-stage attention bidirectional gated recurrent unit and diffusion method for short-term wind power prediction. The proposed method utilizes improved complete ensemble empirical mode decomposition with adaptive noise to decompose the wind power, and then the decomposed data are input into the quantum dual-stage attention bidirectional gated recurrent unit and quantum diffusion model for training prediction; then, the bidirectional encoder representations from Transformer provides final wind power prediction. Compared with 52 prediction algorithms, the average absolute error of the proposed method is more than 30.57% less. Furthermore, the addition of parameterized quantum circuits shortens training prediction time by nearly 25%.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114095"},"PeriodicalIF":8.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174803","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Exploring the sustainable development path of global digital service trade stability: A hybrid approach perspective 全球数字服务贸易稳定可持续发展路径探索：混合方法视角

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-04-01 Epub Date: 2026-02-13 DOI: 10.1016/j.engappai.2026.114172

Shikang Kang, Yu Shang

As the digitization process accelerates, unbalanced sustainable development conditions exacerbate the instability and risks of global digital service trade. Based on the theoretical framework of sustainable development, this study takes panel data of 161 economies from 2014 to 2023 as a sample and employs a hybrid approach of empirical analysis-dynamic fuzzy set qualitative comparisons (dynamic QCA)-artificial neural network (ANN) to identify the sustainability capabilities that drive the digital service trade stability (Dts), and to explore the sustainability portfolio paths that generate high Dts and the cases. The results show that 11 drivers are significantly and positively correlated with Dts and that a single sustainability capability does not constitute high Dts. Three combination paths exist to achieve high Dts, with industrialized innovation capabilities distributed across each path. Economic coherence, ecological sustainability, social peace and inclusion, and sustainable health and well-being as alternatives to the combination paths. The most influential antecedent condition is industrialization innovation capacity, followed by ecological sustainability. The findings demonstrate that tailored combinations of sustainable capabilities, rather than any single factor, underpin trade resilience. This study proposes and validates a hybrid research framework for artificial intelligence (AI) empowerment. This framework not only reveals the multiple driving paths of the stability of digital service trade, enriches the research of sustainable development, but also provides a new AI methodology paradigm for the interpretable causal discovery of complex socio-economic systems.

随着数字化进程的加快，不平衡的可持续发展条件加剧了全球数字服务贸易的不稳定性和风险。基于可持续发展理论框架，本研究以2014 - 2023年161个经济体的面板数据为样本，采用实证分析-动态模糊集定性比较(dynamic QCA)-人工神经网络（ANN）的混合方法，识别驱动数字服务贸易稳定性（Dts）的可持续性能力，并探索产生高Dts的可持续性投资组合路径和案例。结果表明，11个驱动因素与Dts呈显著正相关，单一的可持续能力不构成高Dts。实现高Dts存在三种组合路径，工业化创新能力分布在每条路径上。经济一致性、生态可持续性、社会和平与包容以及可持续的健康和福祉作为两者结合路径的替代方案。影响最大的前条件是工业化创新能力，其次是生态可持续性。研究结果表明，有针对性的可持续能力组合，而不是任何单一因素，是贸易韧性的基础。本研究提出并验证了人工智能（AI）授权的混合研究框架。这一框架不仅揭示了数字服务贸易稳定性的多重驱动路径，丰富了可持续发展研究，而且为复杂社会经济系统的可解释因果关系发现提供了新的人工智能方法论范式。

{"title":"Exploring the sustainable development path of global digital service trade stability: A hybrid approach perspective","authors":"Shikang Kang, Yu Shang","doi":"10.1016/j.engappai.2026.114172","DOIUrl":"10.1016/j.engappai.2026.114172","url":null,"abstract":"<div><div>As the digitization process accelerates, unbalanced sustainable development conditions exacerbate the instability and risks of global digital service trade. Based on the theoretical framework of sustainable development, this study takes panel data of 161 economies from 2014 to 2023 as a sample and employs a hybrid approach of empirical analysis-dynamic fuzzy set qualitative comparisons (dynamic QCA)-artificial neural network (ANN) to identify the sustainability capabilities that drive the digital service trade stability (Dts), and to explore the sustainability portfolio paths that generate high Dts and the cases. The results show that 11 drivers are significantly and positively correlated with Dts and that a single sustainability capability does not constitute high Dts. Three combination paths exist to achieve high Dts, with industrialized innovation capabilities distributed across each path. Economic coherence, ecological sustainability, social peace and inclusion, and sustainable health and well-being as alternatives to the combination paths. The most influential antecedent condition is industrialization innovation capacity, followed by ecological sustainability. The findings demonstrate that tailored combinations of sustainable capabilities, rather than any single factor, underpin trade resilience. This study proposes and validates a hybrid research framework for artificial intelligence (AI) empowerment. This framework not only reveals the multiple driving paths of the stability of digital service trade, enriches the research of sustainable development, but also provides a new AI methodology paradigm for the interpretable causal discovery of complex socio-economic systems.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114172"},"PeriodicalIF":8.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174864","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Transformers meet hyperspectral imaging: A comprehensive study of models, challenges and open problems 变压器满足高光谱成像：模型，挑战和开放问题的综合研究

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-04-01 Epub Date: 2026-02-12 DOI: 10.1016/j.engappai.2026.113947

Guyang Zhang, Waleed Abdulla

Transformers have become the architecture of choice for learning long-range dependencies, yet their adoption in hyperspectral imaging (HSI) is still emerging. We reviewed more than 300 papers published up to 2025 and present the first end-to-end survey dedicated to Transformer-based HSI classification. The study categorizes every stage of a typical pipeline—pre-processing, patch or pixel embedding, positional encoding, spatial–spectral feature extraction, multi-head self-attention variants, skip connections, and loss design—and contrasts alternative design choices with the unique spatial–spectral properties of HSI. We map the field’s progress against persistent obstacles: scarce labeled data, extreme spectral dimensionality, computational overhead, and limited model explainability. Finally, we outline a research agenda prioritizing valuable public data sets, lightweight on-edge models, illumination and sensor shifts robustness, and intrinsically interpretable attention mechanisms. Our goal is to guide researchers in selecting, combining, or extending Transformer components that are truly fit for purpose for next-generation HSI applications.

变形器已经成为学习远程依赖关系的首选架构，但它们在高光谱成像（HSI）中的应用仍在兴起。我们回顾了截至2025年发表的300多篇论文，并提出了第一个致力于基于变压器的HSI分类的端到端调查。该研究对典型管道的每个阶段进行了分类——预处理、补丁或像素嵌入、位置编码、空间光谱特征提取、多头自关注变体、跳过连接和损耗设计——并将不同的设计选择与HSI独特的空间光谱特性进行了对比。我们将该领域的进展映射到持续存在的障碍：稀缺的标记数据、极端的光谱维度、计算开销和有限的模型可解释性。最后，我们概述了一个研究议程，优先考虑有价值的公共数据集、轻量级边缘模型、照明和传感器转换的鲁棒性，以及内在可解释的注意力机制。我们的目标是指导研究人员选择、组合或扩展真正适合下一代HSI应用的Transformer组件。

{"title":"Transformers meet hyperspectral imaging: A comprehensive study of models, challenges and open problems","authors":"Guyang Zhang, Waleed Abdulla","doi":"10.1016/j.engappai.2026.113947","DOIUrl":"10.1016/j.engappai.2026.113947","url":null,"abstract":"<div><div>Transformers have become the architecture of choice for learning long-range dependencies, yet their adoption in hyperspectral imaging (HSI) is still emerging. We reviewed more than 300 papers published up to 2025 and present the first end-to-end survey dedicated to Transformer-based HSI classification. The study categorizes every stage of a typical pipeline—pre-processing, patch or pixel embedding, positional encoding, spatial–spectral feature extraction, multi-head self-attention variants, skip connections, and loss design—and contrasts alternative design choices with the unique spatial–spectral properties of HSI. We map the field’s progress against persistent obstacles: scarce labeled data, extreme spectral dimensionality, computational overhead, and limited model explainability. Finally, we outline a research agenda prioritizing valuable public data sets, lightweight on-edge models, illumination and sensor shifts robustness, and intrinsically interpretable attention mechanisms. Our goal is to guide researchers in selecting, combining, or extending Transformer components that are truly fit for purpose for next-generation HSI applications.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 113947"},"PeriodicalIF":8.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174847","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0