Engineering Applications of Artificial Intelligence最新文献_第7页

Enriched multi-view ensemble approach for high-dimensional imbalanced data classification 高维不平衡数据分类的丰富多视图集成方法

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

Engineering Applications of Artificial Intelligence

Pub Date : 2026-01-23 DOI: 10.1016/j.engappai.2026.113940

Yuhong Xu , Dongyi Ding , Peijie Huang , Zhiwen Yu , C.L. Philip Chen

High-dimensional imbalanced data classification is a challenging issue in real-world applications, where massive invalid features and class imbalance severely impede the behavior of classifiers. Due to high-dimensional features, imbalanced approaches suffer hardship in yielding adequate results. To tackle these issues, this paper proposes an enriched multi-view ensemble approach (EMEA), aiming to construct an accurate and resilient classifier ensemble system for high-dimensional class-skewed data. First, an enriched multi-view optimization (EMO) is designed to extract effective and diverse features from high-dimensional imbalanced data, it promotes the classification ability through subview learning on multiple diverse scenarios. Then a prioritized integration of subviews (PIS) is developed to conduct selective integration for subviews, aiming to construct a high-quality view that enhances decision-making for high-dimensional imbalanced data classification. Finally, EMEA employs resampling to construct a balanced subset, mitigating the impact of class imbalance on the base classifier. The experiments on 16 high-dimensional class-skewed datasets demonstrate that EMEA is superior to other mainstream imbalanced ensemble approaches.

在现实应用中，高维不平衡数据分类是一个具有挑战性的问题，大量无效特征和类不平衡严重阻碍了分类器的行为。由于高维特征，不平衡的方法很难产生足够的结果。为了解决这些问题，本文提出了一种丰富的多视图集成方法（EMEA），旨在为高维类倾斜数据构建一个准确、有弹性的分类器集成系统。首先，设计了一种丰富的多视图优化（EMO）算法，从高维不平衡数据中提取有效且多样的特征，通过对多个不同场景的子视图学习，提高分类能力；在此基础上，提出了子视图优先级集成（PIS）方法，对子视图进行选择性集成，构建高质量的子视图，增强对高维不平衡数据分类的决策能力。最后，EMEA采用重采样来构建一个平衡子集，减轻类不平衡对基分类器的影响。在16个高维类偏斜数据集上的实验表明，EMEA方法优于其他主流的不平衡集成方法。

{"title":"Enriched multi-view ensemble approach for high-dimensional imbalanced data classification","authors":"Yuhong Xu , Dongyi Ding , Peijie Huang , Zhiwen Yu , C.L. Philip Chen","doi":"10.1016/j.engappai.2026.113940","DOIUrl":"10.1016/j.engappai.2026.113940","url":null,"abstract":"<div><div>High-dimensional imbalanced data classification is a challenging issue in real-world applications, where massive invalid features and class imbalance severely impede the behavior of classifiers. Due to high-dimensional features, imbalanced approaches suffer hardship in yielding adequate results. To tackle these issues, this paper proposes an enriched multi-view ensemble approach (EMEA), aiming to construct an accurate and resilient classifier ensemble system for high-dimensional class-skewed data. First, an enriched multi-view optimization (EMO) is designed to extract effective and diverse features from high-dimensional imbalanced data, it promotes the classification ability through subview learning on multiple diverse scenarios. Then a prioritized integration of subviews (PIS) is developed to conduct selective integration for subviews, aiming to construct a high-quality view that enhances decision-making for high-dimensional imbalanced data classification. Finally, EMEA employs resampling to construct a balanced subset, mitigating the impact of class imbalance on the base classifier. The experiments on 16 high-dimensional class-skewed datasets demonstrate that EMEA is superior to other mainstream imbalanced ensemble approaches.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"167 ","pages":"Article 113940"},"PeriodicalIF":8.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038587","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

Revolutionizing artificial intelligence enabled predictive analytics with smart consumer electronics for real-time healthcare monitoring 革命性的人工智能通过智能消费电子产品实现预测分析，实现实时医疗监控

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

Engineering Applications of Artificial Intelligence

Pub Date : 2026-01-23 DOI: 10.1016/j.engappai.2025.113712

Ala Saleh Alluhaidan , Amal M. Aqlan , Mashael Maashi , Ahmed Alsayat , Mashail N. Alkhomsan , Faten Derouez , Rakan Alanazi , Tawfiq Hasanin

The healthcare field has undergone a significant shift in the last few years with the advent of data streaming expertise. Data streaming refers to the constant transfer and study of real-time data from multiple sources. In the healthcare environment, data streaming enables healthcare providers to monitor patients' health, predict health issues, and provide personalized care. Real-time observation of patient well-being and predictive analytics for disease analysis and prevention have become gradually significant in healthcare, as they permit healthcare providers to perceive probable health problems before they arise and occur before they become severe. Consumer electronics health technology has transformed health monitoring by permitting constant tracking of crucial signs, physical activity, and other health restrictions. Incorporating artificial intelligence (AI) and deep learning (DL) into consumer electronic devices promises to improve personalized healthcare by aiding real-time data study and early recognition of health problems. In this manuscript, a Personal Health Monitoring with Predictive Analytics and Consumer Electronics using Dimensionality Reduction and Ensemble Classifiers (PHMPACE-DREC) model is presented. The intention is to propose a consumer electronics method for real-time health monitoring and predictive analytics using advanced models to enable proactive and personalized healthcare solutions. To accomplish that, the PHMPACE-DREC model involves a data pre-processing stage initially by applying min-max normalization to convert the input data into a suitable format. Next, the feature selection step is applied, which is a critical stage as it decreases the data dimensionality and enhances efficiency by using three methods, such as Fast Correlation-based Filter Feature (FCBF), Recursive Feature Elimination (RFE), and Least Absolute Shrinkage and Selection Operator (LASSO). Finally, the classification process is performed by the three ensemble classifiers, such as Elman Neural Network (ENN), Deep Q-Network (DQN), and Conditional Variational Autoencoder (CVAE). The experimental analysis of the PHMPACE-DREC approach portrayed a superior accuracy value of 99.11 % over existing methods under the Wearables dataset.

随着数据流专业知识的出现，医疗保健领域在过去几年中经历了重大转变。数据流是指对来自多个来源的实时数据进行不断的传输和研究。在医疗保健环境中，数据流使医疗保健提供者能够监控患者的健康状况、预测健康问题并提供个性化护理。对患者健康状况的实时观察和疾病分析和预防的预测分析在医疗保健中逐渐变得重要，因为它们允许医疗保健提供者在可能出现的健康问题出现之前和在它们变得严重之前发现它们。消费电子健康技术通过允许持续跟踪关键体征、身体活动和其他健康限制，改变了健康监测。将人工智能（AI）和深度学习（DL）整合到消费电子设备中，通过帮助实时数据研究和早期识别健康问题，有望改善个性化医疗保健。在这份手稿中，个人健康监测与预测分析和消费电子产品使用降维和集成分类器（PHMPACE-DREC）模型提出。其目的是提出一种用于实时健康监测和预测分析的消费电子方法，使用先进的模型来实现主动和个性化的医疗保健解决方案。为了实现这一点，PHMPACE-DREC模型涉及一个数据预处理阶段，首先应用最小-最大归一化将输入数据转换为合适的格式。接下来是特征选择步骤，这是一个关键阶段，因为它通过使用快速相关滤波特征（FCBF）、递归特征消除（RFE）和最小绝对收缩和选择算子（LASSO）三种方法来降低数据维数并提高效率。最后，采用Elman神经网络（ENN）、Deep Q-Network （DQN）和条件变分自编码器（CVAE）三种集成分类器进行分类。实验分析表明，在可穿戴设备数据集下，PHMPACE-DREC方法的准确率比现有方法高99.11%。

{"title":"Revolutionizing artificial intelligence enabled predictive analytics with smart consumer electronics for real-time healthcare monitoring","authors":"Ala Saleh Alluhaidan , Amal M. Aqlan , Mashael Maashi , Ahmed Alsayat , Mashail N. Alkhomsan , Faten Derouez , Rakan Alanazi , Tawfiq Hasanin","doi":"10.1016/j.engappai.2025.113712","DOIUrl":"10.1016/j.engappai.2025.113712","url":null,"abstract":"<div><div>The healthcare field has undergone a significant shift in the last few years with the advent of data streaming expertise. Data streaming refers to the constant transfer and study of real-time data from multiple sources. In the healthcare environment, data streaming enables healthcare providers to monitor patients' health, predict health issues, and provide personalized care. Real-time observation of patient well-being and predictive analytics for disease analysis and prevention have become gradually significant in healthcare, as they permit healthcare providers to perceive probable health problems before they arise and occur before they become severe. Consumer electronics health technology has transformed health monitoring by permitting constant tracking of crucial signs, physical activity, and other health restrictions. Incorporating artificial intelligence (AI) and deep learning (DL) into consumer electronic devices promises to improve personalized healthcare by aiding real-time data study and early recognition of health problems. In this manuscript, a Personal Health Monitoring with Predictive Analytics and Consumer Electronics using Dimensionality Reduction and Ensemble Classifiers (PHMPACE-DREC) model is presented. The intention is to propose a consumer electronics method for real-time health monitoring and predictive analytics using advanced models to enable proactive and personalized healthcare solutions. To accomplish that, the PHMPACE-DREC model involves a data pre-processing stage initially by applying min-max normalization to convert the input data into a suitable format. Next, the feature selection step is applied, which is a critical stage as it decreases the data dimensionality and enhances efficiency by using three methods, such as Fast Correlation-based Filter Feature (FCBF), Recursive Feature Elimination (RFE), and Least Absolute Shrinkage and Selection Operator (LASSO). Finally, the classification process is performed by the three ensemble classifiers, such as Elman Neural Network (ENN), Deep Q-Network (DQN), and Conditional Variational Autoencoder (CVAE). The experimental analysis of the PHMPACE-DREC approach portrayed a superior accuracy value of 99.11 % over existing methods under the Wearables dataset.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"167 ","pages":"Article 113712"},"PeriodicalIF":8.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038714","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

Fourier-enhanced sequence-to-sequence latent graph neural networks for multi-node spatiotemporal forecasting in a hydroelectric reservoir 基于傅里叶增强序列对序列潜在图神经网络的水电站多节点时空预测

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

Engineering Applications of Artificial Intelligence

Pub Date : 2026-01-23 DOI: 10.1016/j.engappai.2026.113939

Laio Oriel Seman , Stefano Frizzo Stefenon , Kin-Choong Yow , Leandro dos Santos Coelho , Viviana Cocco Mariani

This paper presents a Fourier-enhanced dynamic sequence-to-sequence latent graph neural network (Seq2SeqLatentGNN), a deep learning architecture for multi-node spatiotemporal forecasting in hydroelectric reservoir systems. The model integrates three key components: (i) a custom Fourier layer that analyzes global temporal patterns through frequency-domain transformations, (ii) a latent correlation graph convolutional network that infers relational structures between monitoring stations without requiring predefined adjacency matrices, and (iii) an attention-based sequence-to-sequence model that processes temporal dependencies while enabling multi-step forecasting. The architecture simultaneously learns graph structure and forecasting tasks, adapting to changing spatial relationships between reservoir nodes. The proposed architecture was evaluated using a comprehensive dataset derived from 19 interconnected hydroelectric reservoirs located in southern Brazil. The dataset encompasses multiple years of high-resolution (hourly) measurements, including reservoir water levels, inflow and outflow rates, precipitation records, and energy production metrics. Experimental results demonstrate that Seq2SeqLatentGNN achieves superior performance compared to conventional statistical models and contemporary machine learning methods, as measured by standard error metrics. Analysis of the learned latent correlations reveals meaningful spatial dependencies that align with hydrological principles. The model exhibits consistent performance across varying temporal patterns, adapts to regime transitions, and captures both periodic and nonstationary dynamics. The proposed architecture contributes to spatiotemporal forecasting by combining spectral processing, dynamic graph learning, and sequence modeling in a unified framework applicable to systems with evolving connectivity patterns.

本文提出了一种基于傅里叶增强的动态序列到序列潜在图神经网络（Seq2SeqLatentGNN），这是一种用于水电水库系统多节点时空预测的深度学习架构。该模型集成了三个关键组件：(i)通过频域变换分析全球时间模式的自定义傅立叶层，（ii）推断监测站之间关系结构的潜在相关图卷积网络，而不需要预定义的邻接矩阵，以及（iii）基于注意力的序列到序列模型，该模型在实现多步骤预测的同时处理时间依赖性。该体系结构同时学习图结构和预测任务，以适应水库节点之间不断变化的空间关系。使用来自巴西南部19个相互连接的水力发电水库的综合数据集对拟议的建筑进行了评估。该数据集包含多年的高分辨率（每小时）测量数据，包括水库水位、流入和流出率、降水记录和能源生产指标。实验结果表明，与传统统计模型和当代机器学习方法相比，Seq2SeqLatentGNN在标准误差度量方面取得了卓越的性能。对习得的潜在相关性的分析揭示了与水文原理一致的有意义的空间依赖性。该模型在不同的时间模式中表现出一致的性能，适应状态转换，并捕获周期性和非平稳动态。该体系结构将光谱处理、动态图学习和序列建模结合在一个统一的框架中，适用于具有不断变化的连接模式的系统，有助于进行时空预测。

{"title":"Fourier-enhanced sequence-to-sequence latent graph neural networks for multi-node spatiotemporal forecasting in a hydroelectric reservoir","authors":"Laio Oriel Seman , Stefano Frizzo Stefenon , Kin-Choong Yow , Leandro dos Santos Coelho , Viviana Cocco Mariani","doi":"10.1016/j.engappai.2026.113939","DOIUrl":"10.1016/j.engappai.2026.113939","url":null,"abstract":"<div><div>This paper presents a Fourier-enhanced dynamic sequence-to-sequence latent graph neural network (Seq2SeqLatentGNN), a deep learning architecture for multi-node spatiotemporal forecasting in hydroelectric reservoir systems. The model integrates three key components: (i) a custom Fourier layer that analyzes global temporal patterns through frequency-domain transformations, (ii) a latent correlation graph convolutional network that infers relational structures between monitoring stations without requiring predefined adjacency matrices, and (iii) an attention-based sequence-to-sequence model that processes temporal dependencies while enabling multi-step forecasting. The architecture simultaneously learns graph structure and forecasting tasks, adapting to changing spatial relationships between reservoir nodes. The proposed architecture was evaluated using a comprehensive dataset derived from 19 interconnected hydroelectric reservoirs located in southern Brazil. The dataset encompasses multiple years of high-resolution (hourly) measurements, including reservoir water levels, inflow and outflow rates, precipitation records, and energy production metrics. Experimental results demonstrate that Seq2SeqLatentGNN achieves superior performance compared to conventional statistical models and contemporary machine learning methods, as measured by standard error metrics. Analysis of the learned latent correlations reveals meaningful spatial dependencies that align with hydrological principles. The model exhibits consistent performance across varying temporal patterns, adapts to regime transitions, and captures both periodic and nonstationary dynamics. The proposed architecture contributes to spatiotemporal forecasting by combining spectral processing, dynamic graph learning, and sequence modeling in a unified framework applicable to systems with evolving connectivity patterns.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"167 ","pages":"Article 113939"},"PeriodicalIF":8.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038585","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

Micro ribonucleic acids-drug sensitivity prediction by variational graph auto-encoder and collaborative matrix factorization 基于变分图自编码器和协同矩阵分解的微核糖核酸药物敏感性预测

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

Engineering Applications of Artificial Intelligence

Pub Date : 2026-01-23 DOI: 10.1016/j.engappai.2026.113901

Yunyin Li , Shudong Wang , Yuanyuan Zhang , Chuanru Ren , Shanchen Pang , Tiyao Liu , Yingye Liu

The mechanisms of action for numerous drugs involve micro ribonucleic acids (miRNAs), highlighting the significance of studying miRNA-mediated drug sensitivity in drug discovery and disease treatment. Despite advancements in computational approaches, challenges persist in effectively extracting drug and miRNA features and accurately predicting their associations. The existing similarity networks of drugs and miRNAs are in urgent need of supplementing comprehensive similarity information. In addition, most computational methods extract only single-level features without combining information from different levels, limiting the performance of the models. To overcome these challenges, we combine Variational graph auto-encoder and Collaborative matrix factorization to identify MiRNA-Drug Sensitivity (VCMDS). VCMDS figures out the Gaussian Interaction Profile (GIP) kernel similarities between drugs and miRNAs and adds these measurements to each their network. By aggregating multiple sources of information, the GIP kernel similarity provides useful information by considering a wider network of interactions and measuring similarity more accurately. Subsequently, it extracts features of miRNAs and drugs at various levels by applying variational graph auto-encoder and collaborative matrix factorization. Linear and nonlinear features can be combined to produce high-quality features and thus improve the prediction performance. Finally, predicted scores are obtained using a fully connected network. VCMDS achieves an average Area Under Curve (AUC) of 0.9632 in the 5-fold Cross-Validation (CV) experiment, outperforming other competitive methods. Two types of case studies further demonstrate the effectiveness of VCMDS.

许多药物的作用机制都涉及到微核糖核酸（miRNAs），这凸显了研究mirna介导的药物敏感性在药物发现和疾病治疗中的重要意义。尽管计算方法取得了进步，但在有效提取药物和miRNA特征并准确预测它们之间的关联方面仍然存在挑战。现有的药物和mirna相似网络急需补充全面的相似信息。此外，大多数计算方法只提取单一层次的特征，而没有将不同层次的信息结合起来，这限制了模型的性能。为了克服这些挑战，我们结合变分图自编码器和协同矩阵分解来识别mirna -药物敏感性（VCMDS）。VCMDS计算出药物和mirna之间的高斯相互作用谱（GIP）核相似性，并将这些测量值添加到它们的每个网络中。通过聚合多个信息源，GIP内核相似性通过考虑更广泛的交互网络和更准确地度量相似性来提供有用的信息。随后，利用变分图自编码器和协同矩阵分解技术提取mirna和药物的各个层次的特征。线性特征和非线性特征相结合可以产生高质量的特征，从而提高预测性能。最后，利用全连接网络得到预测分数。在5重交叉验证（CV）实验中，VCMDS的平均曲线下面积（AUC）为0.9632，优于其他竞争方法。两种类型的案例研究进一步证明了VCMDS的有效性。

{"title":"Micro ribonucleic acids-drug sensitivity prediction by variational graph auto-encoder and collaborative matrix factorization","authors":"Yunyin Li , Shudong Wang , Yuanyuan Zhang , Chuanru Ren , Shanchen Pang , Tiyao Liu , Yingye Liu","doi":"10.1016/j.engappai.2026.113901","DOIUrl":"10.1016/j.engappai.2026.113901","url":null,"abstract":"<div><div>The mechanisms of action for numerous drugs involve micro ribonucleic acids (miRNAs), highlighting the significance of studying miRNA-mediated drug sensitivity in drug discovery and disease treatment. Despite advancements in computational approaches, challenges persist in effectively extracting drug and miRNA features and accurately predicting their associations. The existing similarity networks of drugs and miRNAs are in urgent need of supplementing comprehensive similarity information. In addition, most computational methods extract only single-level features without combining information from different levels, limiting the performance of the models. To overcome these challenges, we combine Variational graph auto-encoder and Collaborative matrix factorization to identify MiRNA-Drug Sensitivity (VCMDS). VCMDS figures out the Gaussian Interaction Profile (GIP) kernel similarities between drugs and miRNAs and adds these measurements to each their network. By aggregating multiple sources of information, the GIP kernel similarity provides useful information by considering a wider network of interactions and measuring similarity more accurately. Subsequently, it extracts features of miRNAs and drugs at various levels by applying variational graph auto-encoder and collaborative matrix factorization. Linear and nonlinear features can be combined to produce high-quality features and thus improve the prediction performance. Finally, predicted scores are obtained using a fully connected network. VCMDS achieves an average Area Under Curve (AUC) of 0.9632 in the 5-fold Cross-Validation (CV) experiment, outperforming other competitive methods. Two types of case studies further demonstrate the effectiveness of VCMDS.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"167 ","pages":"Article 113901"},"PeriodicalIF":8.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038590","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

Research on cable terminal interface defect state detection based on electric field characteristics and multi-core improved support vector machine 基于电场特征和多核改进支持向量机的电缆终端接口缺陷状态检测研究

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

Engineering Applications of Artificial Intelligence

Pub Date : 2026-01-23 DOI: 10.1016/j.engappai.2026.113739

Yujing Tang , Yang Fu , Qin Cai , Jieping Wu , Qi Wang , Guoqiang Gao

As key equipment for high-speed rail power transmission and the connection of high-voltage systems, the cable terminals are crucial to ensuring the stable operation of the railway system. However, the existing detection methods for cable terminals are easily affected by on-site noise and have low detection accuracy. Therefore, this paper proposes a method for detecting interface defect status of high-speed cable terminals based on the electric field strength feature set and multi-kernel support vector machine (MK-SVM). Firstly, a spatial electric field detection platform was built to extract the electric field intensity of the prefabricated defective cable terminals of different lengths. Secondly, the optimization of the characteristic parameters of electric field strength of defective cable terminals was realized based on the Pearson coefficient method. In order to improve the recognition effect and model generalization ability, a MK-SVM combining linear kernel function and radial basis kernel function was proposed. Finally, a comparative study was conducted on the optimization effects of particle swarm algorithm, firefly algorithm, simulated annealing algorithm and genetic algorithm on MK-SVM. Research has shown that using genetic algorithm for parameter optimization of multi-core SVM has the best performance, with recognition accuracy, average precision, average recall, and average F1 score of 95.6 %, 96 %, 95.6 %, and 0.96, respectively. Compared with the unoptimized SVM, the four feature parameters increased by 8.9 %, 7.9 %, 8.9 %, and 9.6 %, respectively.

电缆终端作为高铁输电和高压系统连接的关键设备，对保证铁路系统的稳定运行至关重要。但现有的电缆终端检测方法容易受到现场噪声的影响，检测精度较低。为此，本文提出了一种基于电场强度特征集和多核支持向量机（MK-SVM）的高速电缆终端接口缺陷状态检测方法。首先，建立空间电场检测平台，提取预制不同长度缺陷电缆端子的电场强度；其次，基于Pearson系数法实现了缺陷电缆端子电场强度特征参数的优化。为了提高识别效果和模型泛化能力，提出了线性核函数和径向基核函数相结合的MK-SVM算法。最后，对比研究了粒子群算法、萤火虫算法、模拟退火算法和遗传算法对MK-SVM的优化效果。研究表明，采用遗传算法对多核支持向量机进行参数优化的效果最好，识别准确率为95.6%，平均精密度为96%，平均查全率为95.6%，平均F1分数为0.96。与未优化支持向量机相比，4个特征参数分别提高了8.9%、7.9%、8.9%和9.6%。

{"title":"Research on cable terminal interface defect state detection based on electric field characteristics and multi-core improved support vector machine","authors":"Yujing Tang , Yang Fu , Qin Cai , Jieping Wu , Qi Wang , Guoqiang Gao","doi":"10.1016/j.engappai.2026.113739","DOIUrl":"10.1016/j.engappai.2026.113739","url":null,"abstract":"<div><div>As key equipment for high-speed rail power transmission and the connection of high-voltage systems, the cable terminals are crucial to ensuring the stable operation of the railway system. However, the existing detection methods for cable terminals are easily affected by on-site noise and have low detection accuracy. Therefore, this paper proposes a method for detecting interface defect status of high-speed cable terminals based on the electric field strength feature set and multi-kernel support vector machine (MK-SVM). Firstly, a spatial electric field detection platform was built to extract the electric field intensity of the prefabricated defective cable terminals of different lengths. Secondly, the optimization of the characteristic parameters of electric field strength of defective cable terminals was realized based on the Pearson coefficient method. In order to improve the recognition effect and model generalization ability, a MK-SVM combining linear kernel function and radial basis kernel function was proposed. Finally, a comparative study was conducted on the optimization effects of particle swarm algorithm, firefly algorithm, simulated annealing algorithm and genetic algorithm on MK-SVM. Research has shown that using genetic algorithm for parameter optimization of multi-core SVM has the best performance, with recognition accuracy, average precision, average recall, and average F1 score of 95.6 %, 96 %, 95.6 %, and 0.96, respectively. Compared with the unoptimized SVM, the four feature parameters increased by 8.9 %, 7.9 %, 8.9 %, and 9.6 %, respectively.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"167 ","pages":"Article 113739"},"PeriodicalIF":8.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038528","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

Balance divergence for knowledge distillation 知识蒸馏的平衡发散

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

Engineering Applications of Artificial Intelligence

Pub Date : 2026-01-23 DOI: 10.1016/j.engappai.2026.113943

Yafei Qi , Chen Wang , Zhaoning Zhang , Yaping Liu , Yongmin Zhang

Knowledge distillation (KD) represents a fundamental artificial intelligence (AI) technique for model compression and optimization. In computer vision AI applications, most KD methods use Kullback–Leibler (KL) divergence to align teacher–student output probabilities, but often neglect crucial negative aspects of teacher “dark knowledge” by underweighting low-probability signals. This limitation leads to suboptimal logit mimicry and unbalanced knowledge transfer to the student network. In this paper, we investigate the impact of this imbalance and propose a novel method, named Balance Divergence Distillation (BDD). By introducing a compensatory operation using reverse KL divergence, our method can improve the modeling of the extremely small values in the negative from the teacher and preserve the learning capacity for the positive. Furthermore, we test the impact of different temperature coefficients adjustments, which can lead to further balance in knowledge transfer. The evaluation results demonstrate that our method achieves accuracy improvements of

1 % \sim 3 %

for lightweight student networks over standard KD methods on both Canadian Institute for Advanced Research 100 classes(CIFAR-100) and ImageNet datasets. Additionally, when applied to semantic segmentation, our approach enhances the student by 4.55% in mean Intersection over Union (mIoU) compared to the baseline on the Cityscapes dataset. These experiments confirm that our method provides a simple yet highly effective solution that can be seamlessly integrated with various KD frameworks across different vision tasks.

知识蒸馏（Knowledge distillation， KD）是一种用于模型压缩和优化的基本人工智能技术。在计算机视觉人工智能应用中，大多数KD方法使用kullbackleibler （KL）散度来对齐师生输出概率，但往往通过低估低概率信号而忽略了教师“暗知识”的关键负面方面。这种限制导致次优的逻辑模仿和不平衡的知识转移到学生网络。在本文中，我们研究了这种不平衡的影响，并提出了一种新的方法，称为平衡发散蒸馏（BDD）。通过引入反向KL散度的补偿操作，我们的方法可以改进对来自教师的负值极小值的建模，并保留对正值的学习能力。此外，我们还测试了不同温度系数调整对知识转移的影响，从而进一步平衡知识转移。评估结果表明，在加拿大高级研究所100类（CIFAR-100）和ImageNet数据集上，我们的方法在轻量级学生网络上的准确率比标准KD方法提高了1% ~ 3%。此外，当应用于语义分割时，与cityscape数据集的基线相比，我们的方法将学生的平均交叉口比联盟（mIoU）提高了4.55%。这些实验证实，我们的方法提供了一种简单而高效的解决方案，可以与不同视觉任务的各种KD框架无缝集成。

{"title":"Balance divergence for knowledge distillation","authors":"Yafei Qi , Chen Wang , Zhaoning Zhang , Yaping Liu , Yongmin Zhang","doi":"10.1016/j.engappai.2026.113943","DOIUrl":"10.1016/j.engappai.2026.113943","url":null,"abstract":"<div><div>Knowledge distillation (KD) represents a fundamental artificial intelligence (AI) technique for model compression and optimization. In computer vision AI applications, most KD methods use Kullback–Leibler (KL) divergence to align teacher–student output probabilities, but often neglect crucial negative aspects of teacher “dark knowledge” by underweighting low-probability signals. This limitation leads to suboptimal logit mimicry and unbalanced knowledge transfer to the student network. In this paper, we investigate the impact of this imbalance and propose a novel method, named Balance Divergence Distillation (BDD). By introducing a compensatory operation using reverse KL divergence, our method can improve the modeling of the extremely small values in the negative from the teacher and preserve the learning capacity for the positive. Furthermore, we test the impact of different temperature coefficients adjustments, which can lead to further balance in knowledge transfer. The evaluation results demonstrate that our method achieves accuracy improvements of <span><math><mrow><mn>1</mn><mtext>%</mtext><mo>∼</mo><mn>3</mn><mtext>%</mtext></mrow></math></span> for lightweight student networks over standard KD methods on both Canadian Institute for Advanced Research 100 classes(CIFAR-100) and ImageNet datasets. Additionally, when applied to semantic segmentation, our approach enhances the student by 4.55% in mean Intersection over Union (mIoU) compared to the baseline on the Cityscapes dataset. These experiments confirm that our method provides a simple yet highly effective solution that can be seamlessly integrated with various KD frameworks across different vision tasks.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"167 ","pages":"Article 113943"},"PeriodicalIF":8.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038535","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

Predicting dielectric properties of polyetherimide-based composite via combined molecular dynamics simulation and machine learning 基于分子动力学模拟和机器学习的聚醚酰亚胺基复合材料介电性能预测

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

Engineering Applications of Artificial Intelligence

Pub Date : 2026-01-23 DOI: 10.1016/j.engappai.2026.113935

Yue Zhang , Zheng Gong , Changhai Zhang , Yongquan Zhang , Chao Yin , Xubin Wang , Tiandong Zhang , Xiajie Yi , Huajie Yi , Qi Wang

The design of high-performance polymer dielectrics for capacitor energy storage is crucial but often hindered by time-consuming, resource-intensive development cycles. Polyetherimide is a promising matrix material, yet its performance is limited by a low dielectric constant and breakdown strength. To accelerate the design process, we propose and validate an integrated computational framework combining molecular dynamics simulations with interpretable machine learning. In terms of the Artificial Intelligence contribution, a weighted ensemble model was developed from a dual database of molecular dynamics parameters and molecular descriptors to predict dielectric property in Polyetherimide-based composites. The model was then deconstructed using the SHapley Additive exPlanations framework, which unveiled a multi-scale design hierarchy. This analysis revealed that filler weight fraction and intrinsic dielectric constant are the most dominant predictors, followed by interfacial compatibility and molecular polarity. Regarding the engineering application, to validate our computational approach, model-selected Benzil and Acetophenone were fabricated into composite films. Experimental results confirmed the model's high accuracy, identifying optimal contents of weight percent of 15 wt% for Benzil and 10 wt% for Acetophenone. Notably, the Polyetherimide-based composite with 10 wt% of Acetophenone achieved an excellent discharge energy density of 10.3 J/cm³, representing a 58 % enhancement over pristine Polyetherimide. Ultimately, this study not only developed a promising material but established a reliable data-driven methodology providing clear guidance for designing next-generation polymer dielectrics.

高性能聚合物电介质电容器储能的设计是至关重要的，但往往受到耗时，资源密集的开发周期的阻碍。聚醚酰亚胺是一种很有前途的基体材料，但其性能受到介电常数和击穿强度低的限制。为了加速设计过程，我们提出并验证了一个将分子动力学模拟与可解释机器学习相结合的集成计算框架。在人工智能贡献方面，从分子动力学参数和分子描述符的双重数据库中开发了加权系综模型来预测聚醚酰亚胺基复合材料的介电性能。然后使用SHapley加性解释框架解构该模型，该框架揭示了一个多尺度设计层次。分析表明，填料质量分数和本征介电常数是最主要的预测因子，其次是界面相容性和分子极性。在工程应用方面，为了验证我们的计算方法，将模型选择的苯并和苯乙酮制备成复合薄膜。实验结果证实了该模型的高准确度，确定了苯甲醚的最佳质量分数为15 wt%，苯乙酮的最佳质量分数为10 wt%。值得注意的是，含有10 wt%苯乙酮的聚醚酰亚胺基复合材料的放电能量密度达到了10.3 J/cm3，比原始聚醚酰亚胺提高了58%。最终，这项研究不仅开发了一种有前途的材料，而且建立了一种可靠的数据驱动方法，为设计下一代聚合物电介质提供了明确的指导。

{"title":"Predicting dielectric properties of polyetherimide-based composite via combined molecular dynamics simulation and machine learning","authors":"Yue Zhang , Zheng Gong , Changhai Zhang , Yongquan Zhang , Chao Yin , Xubin Wang , Tiandong Zhang , Xiajie Yi , Huajie Yi , Qi Wang","doi":"10.1016/j.engappai.2026.113935","DOIUrl":"10.1016/j.engappai.2026.113935","url":null,"abstract":"<div><div>The design of high-performance polymer dielectrics for capacitor energy storage is crucial but often hindered by time-consuming, resource-intensive development cycles. Polyetherimide is a promising matrix material, yet its performance is limited by a low dielectric constant and breakdown strength. To accelerate the design process, we propose and validate an integrated computational framework combining molecular dynamics simulations with interpretable machine learning. In terms of the Artificial Intelligence contribution, a weighted ensemble model was developed from a dual database of molecular dynamics parameters and molecular descriptors to predict dielectric property in Polyetherimide-based composites. The model was then deconstructed using the SHapley Additive exPlanations framework, which unveiled a multi-scale design hierarchy. This analysis revealed that filler weight fraction and intrinsic dielectric constant are the most dominant predictors, followed by interfacial compatibility and molecular polarity. Regarding the engineering application, to validate our computational approach, model-selected Benzil and Acetophenone were fabricated into composite films. Experimental results confirmed the model's high accuracy, identifying optimal contents of weight percent of 15 wt% for Benzil and 10 wt% for Acetophenone. Notably, the Polyetherimide-based composite with 10 wt% of Acetophenone achieved an excellent discharge energy density of 10.3 J/cm<sup>3</sup>, representing a 58 % enhancement over pristine Polyetherimide. Ultimately, this study not only developed a promising material but established a reliable data-driven methodology providing clear guidance for designing next-generation polymer dielectrics.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"167 ","pages":"Article 113935"},"PeriodicalIF":8.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038589","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

Few-shot semantic segmentation for clearance intrusion risk detection in metro tunnel point clouds 基于语义分割的地铁隧道点云清除入侵风险检测

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

Engineering Applications of Artificial Intelligence

Pub Date : 2026-01-23 DOI: 10.1016/j.engappai.2026.113909

Wenbo Qin , Yuxiang Wang , Shangbin Gao , Cheng Zhou

A wide range of mechanical, electrical, and plumbing (MEP) components are mounted along metro tunnel linings, where subtle spatial displacements caused by loosening or deformation may intrude into the train clearance envelope. Detecting such early-stage deviations is challenging due to occlusion, low illumination, and the dense arrangement of facilities in tunnel point clouds acquired using simultaneous localization and mapping (SLAM). To address this problem, this study proposes a training-free few-shot semantic segmentation framework for clearance intrusion detection. The model integrates geometry-based descriptors (linearity, planarity, verticality), a scale factor control mechanism for multi-scale feature enhancement, and a confidence-based filtering strategy to suppress uncertain predictions. Experiments were conducted on metro tunnel point clouds acquired using a backpack-mounted light detection and ranging (LiDAR) SLAM system, with segmentation performed using 1 m under a single-class few-shot setting. The proposed method achieves a mean intersection over union (mIoU) of 78.4 %, while requiring only a small support set of 15 blocks, and the reconstructed axes of MEP facilities enable deviation detection below 1 cm relative to reference inspection epochs. These results demonstrate that the proposed framework provides a practical and robust solution for early-stage clearance intrusion risk assessment in metro tunnel environments.

沿着地铁隧道衬砌安装了各种机械、电气和管道（MEP）组件，在那里，由松动或变形引起的细微空间位移可能会侵入列车间隙包络。由于使用同步定位和测绘（SLAM）获得的隧道点云中遮挡、低照度和密集的设施安排，检测此类早期偏差具有挑战性。为了解决这一问题，本研究提出了一种用于清除入侵检测的无需训练的少镜头语义分割框架。该模型集成了基于几何的描述符（线性、平面、垂直）、用于多尺度特征增强的尺度因子控制机制以及用于抑制不确定预测的基于置信度的过滤策略。实验采用背包式光探测与测距（LiDAR） SLAM系统获取的地铁隧道点云，在单类少射设置下，以1米为单位进行分割。该方法实现了78.4%的平均交联（mIoU），而只需要15个块的小支持集，并且MEP设施的重建轴线相对于参考检测时代可以检测到小于1 cm的偏差。结果表明，该框架为地铁隧道环境下的早期间隙入侵风险评估提供了一种实用、稳健的解决方案。

{"title":"Few-shot semantic segmentation for clearance intrusion risk detection in metro tunnel point clouds","authors":"Wenbo Qin , Yuxiang Wang , Shangbin Gao , Cheng Zhou","doi":"10.1016/j.engappai.2026.113909","DOIUrl":"10.1016/j.engappai.2026.113909","url":null,"abstract":"<div><div>A wide range of mechanical, electrical, and plumbing (MEP) components are mounted along metro tunnel linings, where subtle spatial displacements caused by loosening or deformation may intrude into the train clearance envelope. Detecting such early-stage deviations is challenging due to occlusion, low illumination, and the dense arrangement of facilities in tunnel point clouds acquired using simultaneous localization and mapping (SLAM). To address this problem, this study proposes a training-free few-shot semantic segmentation framework for clearance intrusion detection. The model integrates geometry-based descriptors (linearity, planarity, verticality), a scale factor control mechanism for multi-scale feature enhancement, and a confidence-based filtering strategy to suppress uncertain predictions. Experiments were conducted on metro tunnel point clouds acquired using a backpack-mounted light detection and ranging (LiDAR) SLAM system, with segmentation performed using 1 m under a single-class few-shot setting. The proposed method achieves a mean intersection over union (mIoU) of 78.4 %, while requiring only a small support set of 15 blocks, and the reconstructed axes of MEP facilities enable deviation detection below 1 cm relative to reference inspection epochs. These results demonstrate that the proposed framework provides a practical and robust solution for early-stage clearance intrusion risk assessment in metro tunnel environments.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"167 ","pages":"Article 113909"},"PeriodicalIF":8.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038575","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

Physics-informed machine learning for hybrid modelling of water leakages induced by rapid valve manoeuvres in water distribution networks 基于物理的机器学习，用于水分配网络中由快速阀门操作引起的水泄漏的混合建模

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

Engineering Applications of Artificial Intelligence

Pub Date : 2026-01-23 DOI: 10.1016/j.engappai.2026.113887

Alex J. Garzón-Orduña , Oscar E. Coronado-Hernández , Alfonso Arrieta-Pastrana , Helena M. Ramos , Modesto Pérez-Sánchez

Efficient leakage management in Water Distribution Networks is essential to achieving the objectives of Smart Cities and Sustainable Development Goal 6. Conventional hydraulic models based on Extended Period Simulation cannot reproduce the short transients generated by rapid valve manoeuvres, particularly in pressure-reducing and operational valves. These events create inertial effects and pressure fluctuations that strongly influence leakage behaviour, especially when valve resistance changes with real operational timings, whether manual or actuator-driven. This study develops a hybrid Physics-Informed Machine Learning framework that couples an extended Rigid Water Column Model—derived from mass and momentum conservation and incorporating time-dependent valve resistance—with supervised Machine Learning algorithms. The approach enhances leakage prediction under transient conditions by combining physical interpretability with data-driven adaptability. Simulations from the Rigid Water Column Model and Extended Period Simulation were compared across sixteen synthetic scenarios. Four optimised Machine Learning models—Fine Tree, Bagged Trees, Exponential Gaussian Process Regression, and Wide Neural Network—were trained using physically consistent datasets. They achieved Root Mean Square Error values as low as 0.0019 L per second. Tree-based algorithms proved most robust, while Exponential Gaussian Process Regression and Wide Neural Network models showed reduced extrapolation capacity. Statistical stability verified through 95 percent confidence intervals confirmed the physical coherence of predictions. The proposed framework provides a scalable and transferable tool for real-time leakage prediction and valve operation planning within Digital Twin and Supervisory Control and Data Acquisition environments, supporting predictive control, operational decision-making, and sustainable water-infrastructure management.

高效的供水管网泄漏管理对于实现智慧城市和可持续发展目标6至关重要。基于延长周期仿真的传统水力模型不能再现快速阀门操作所产生的短暂瞬态，特别是在减压阀和操作阀中。这些事件会产生惯性效应和压力波动，严重影响泄漏行为，特别是当阀门阻力随实际操作时间变化时，无论是手动还是执行器驱动。本研究开发了一种混合物理信息的机器学习框架，该框架将扩展的刚性水柱模型与监督机器学习算法结合起来，该模型来源于质量和动量守恒，并结合了随时间变化的阀门阻力。该方法将物理可解释性与数据驱动的适应性相结合，增强了瞬态条件下的泄漏预测能力。比较了刚性水柱模型和延长周期模拟在16种综合情景下的模拟结果。四种优化的机器学习模型-细树，袋装树，指数高斯过程回归和广泛的神经网络使用物理一致的数据集进行训练。他们获得的均方根误差值低至每秒0.0019升。基于树的算法被证明是最稳健的，而指数高斯过程回归和宽神经网络模型显示出较低的外推能力。通过95%置信区间验证的统计稳定性证实了预测的物理一致性。所提出的框架提供了一个可扩展和可转移的工具，用于在数字孪生、监控和数据采集环境中进行实时泄漏预测和阀门操作规划，支持预测控制、操作决策和可持续的水基础设施管理。

{"title":"Physics-informed machine learning for hybrid modelling of water leakages induced by rapid valve manoeuvres in water distribution networks","authors":"Alex J. Garzón-Orduña , Oscar E. Coronado-Hernández , Alfonso Arrieta-Pastrana , Helena M. Ramos , Modesto Pérez-Sánchez","doi":"10.1016/j.engappai.2026.113887","DOIUrl":"10.1016/j.engappai.2026.113887","url":null,"abstract":"<div><div>Efficient leakage management in Water Distribution Networks is essential to achieving the objectives of Smart Cities and Sustainable Development Goal 6. Conventional hydraulic models based on Extended Period Simulation cannot reproduce the short transients generated by rapid valve manoeuvres, particularly in pressure-reducing and operational valves. These events create inertial effects and pressure fluctuations that strongly influence leakage behaviour, especially when valve resistance changes with real operational timings, whether manual or actuator-driven. This study develops a hybrid Physics-Informed Machine Learning framework that couples an extended Rigid Water Column Model—derived from mass and momentum conservation and incorporating time-dependent valve resistance—with supervised Machine Learning algorithms. The approach enhances leakage prediction under transient conditions by combining physical interpretability with data-driven adaptability. Simulations from the Rigid Water Column Model and Extended Period Simulation were compared across sixteen synthetic scenarios. Four optimised Machine Learning models—Fine Tree, Bagged Trees, Exponential Gaussian Process Regression, and Wide Neural Network—were trained using physically consistent datasets. They achieved Root Mean Square Error values as low as 0.0019 L per second. Tree-based algorithms proved most robust, while Exponential Gaussian Process Regression and Wide Neural Network models showed reduced extrapolation capacity. Statistical stability verified through 95 percent confidence intervals confirmed the physical coherence of predictions. The proposed framework provides a scalable and transferable tool for real-time leakage prediction and valve operation planning within Digital Twin and Supervisory Control and Data Acquisition environments, supporting predictive control, operational decision-making, and sustainable water-infrastructure management.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"167 ","pages":"Article 113887"},"PeriodicalIF":8.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038530","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

High-precision multimodal vehicle trajectory prediction model based on cross-layer interleaved spatiotemporal attention mechanism 基于跨层交错时空注意机制的高精度多模态车辆轨迹预测模型

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

Engineering Applications of Artificial Intelligence

Pub Date : 2026-01-23 DOI: 10.1016/j.engappai.2026.113937

Fei Teng , Liqiang Jin , Junnian Wang , Feng Xiao , Mengdi Guo , Yanbo Zhou , Jin Zhang

In increasingly complex traffic environments, spatiotemporal attention mechanisms have made remarkable advancements in scene-level interaction modelling. However, the deep and multi-scale spatiotemporal representations required for safe and efficient decision-making in intelligent vehicles remain underexplored. Aiming to address this limitation, this study proposes a multimodal trajectory prediction model based on a cross-layer interleaved spatiotemporal attention (CLISTA) mechanism. Compared with conventional spatiotemporal attention frameworks, CLISTA more effectively captures multi-scale spatiotemporal interactions in complex traffic scenes through the alternating fusion of spatial and temporal features across network layers via a cross-layer interleaving structure. Firstly, spatial, dynamic and heading conflict risks are derived from the relative motion between the target vehicle and its neighbours and aggregated into a social grid weight matrix, through which the neighbours' collective influence on the target vehicle is quantified. Secondly, spatial and temporal multi-head attention modules are designed within each layer. By integrating an interleaved ‘spatial–temporal’ stacking strategy with cross-layer skip connections, the model facilitates progressive alignment and deep fusion, ranging from local interactions to long-range dependencies. Subsequently, an intention recognition module is developed. A second-order gated bilinear fusion mechanism is introduced to adaptively model higher-order couplings between local neighbour dynamics and global interaction semantics, thereby yielding a multimodal probability distribution over the target vehicle's driving intentions. Lastly, multimodal trajectory predictions are generated by decoding the fused spatiotemporal features together with the inferred intention information. Experimental results on three benchmark datasets—NGSIM (Next Generation Simulation), AD4CHE (Aerial Dataset for China Congested Highway and Expressway), and highD—demonstrate that CLISTA consistently outperforms the baseline methods. Relative to the next-best model, it reduces average/final displacement errors by 16.67 %/21.23 %, 12.99 %/21.14 % and 10.53 %/21.59 % on NGSIM, AD4CHE and HighD, respectively. Overall, CLISTA offers reliable multi-hypothesis trajectory priors for safe and efficient decision-making in complex traffic scenarios.

在日益复杂的交通环境中，时空注意机制在场景级交互建模方面取得了显著进展。然而，智能车辆安全高效决策所需的深度和多尺度时空表征仍未得到充分探索。针对这一局限性，本研究提出了一种基于跨层交错时空注意（CLISTA）机制的多模态轨迹预测模型。与传统的时空注意力框架相比，CLISTA通过跨网络层的交错结构，将时空特征交替融合，更有效地捕捉复杂交通场景中的多尺度时空相互作用。首先，从目标车辆与相邻车辆的相对运动中导出空间冲突风险、动态冲突风险和航向冲突风险，并将其聚合成社会网格权重矩阵，量化相邻车辆对目标车辆的集体影响；其次，在每一层内设计时空多头注意模块；通过将交错的“时空”堆叠策略与跨层跳跃连接相结合，该模型促进了从局部相互作用到远程依赖的渐进对齐和深度融合。随后，开发了意图识别模块。引入二阶门控双线性融合机制，自适应建模局部邻域动力学和全局交互语义之间的高阶耦合，从而得到目标车辆驾驶意图的多模态概率分布。最后，将融合的时空特征与推断出的意图信息进行解码，生成多模态轨迹预测。在ngsim（下一代仿真）、AD4CHE（中国拥堵公路和高速公路航空数据集）和high - 3个基准数据集上的实验结果表明，CLISTA方法的性能始终优于基线方法。与次优模型相比，该模型在NGSIM、AD4CHE和HighD上的平均/最终位移误差分别降低了16.67% / 21.23%、12.99% / 21.14%和10.53% / 21.59%。总体而言，CLISTA为复杂交通场景下的安全高效决策提供了可靠的多假设轨迹先验。

{"title":"High-precision multimodal vehicle trajectory prediction model based on cross-layer interleaved spatiotemporal attention mechanism","authors":"Fei Teng , Liqiang Jin , Junnian Wang , Feng Xiao , Mengdi Guo , Yanbo Zhou , Jin Zhang","doi":"10.1016/j.engappai.2026.113937","DOIUrl":"10.1016/j.engappai.2026.113937","url":null,"abstract":"<div><div>In increasingly complex traffic environments, spatiotemporal attention mechanisms have made remarkable advancements in scene-level interaction modelling. However, the deep and multi-scale spatiotemporal representations required for safe and efficient decision-making in intelligent vehicles remain underexplored. Aiming to address this limitation, this study proposes a multimodal trajectory prediction model based on a cross-layer interleaved spatiotemporal attention (CLISTA) mechanism. Compared with conventional spatiotemporal attention frameworks, CLISTA more effectively captures multi-scale spatiotemporal interactions in complex traffic scenes through the alternating fusion of spatial and temporal features across network layers via a cross-layer interleaving structure. Firstly, spatial, dynamic and heading conflict risks are derived from the relative motion between the target vehicle and its neighbours and aggregated into a social grid weight matrix, through which the neighbours' collective influence on the target vehicle is quantified. Secondly, spatial and temporal multi-head attention modules are designed within each layer. By integrating an interleaved ‘spatial–temporal’ stacking strategy with cross-layer skip connections, the model facilitates progressive alignment and deep fusion, ranging from local interactions to long-range dependencies. Subsequently, an intention recognition module is developed. A second-order gated bilinear fusion mechanism is introduced to adaptively model higher-order couplings between local neighbour dynamics and global interaction semantics, thereby yielding a multimodal probability distribution over the target vehicle's driving intentions. Lastly, multimodal trajectory predictions are generated by decoding the fused spatiotemporal features together with the inferred intention information. Experimental results on three benchmark datasets—NGSIM (Next Generation Simulation), AD4CHE (Aerial Dataset for China Congested Highway and Expressway), and highD—demonstrate that CLISTA consistently outperforms the baseline methods. Relative to the next-best model, it reduces average/final displacement errors by 16.67 %/21.23 %, 12.99 %/21.14 % and 10.53 %/21.59 % on NGSIM, AD4CHE and HighD, respectively. Overall, CLISTA offers reliable multi-hypothesis trajectory priors for safe and efficient decision-making in complex traffic scenarios.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"167 ","pages":"Article 113937"},"PeriodicalIF":8.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038537","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