Pub Date : 2026-01-31DOI: 10.1016/j.eswa.2026.131460
Shukun Yu , Quanwang Wu , Kun Cai , Taolin Guo , Zhuo Jiang , Tianhao Sun
Although private clouds provide workflow execution environments with enhanced data security, greater administrative control, and predictable resource provisioning, they often suffer from limited elasticity and struggle to adapt to dynamic and bursty workload patterns. Serverless computing, characterized by elastic scaling, pay-per-use pricing model, and event-driven execution, presents a promising paradigm to address these limitations. Hence, this paper explores integrating private clouds with serverless computing to enable efficient workflow offloading. A systematic workflow offloading framework is established for minimizing monetary costs under workflow deadlines. A Deadline-Aware Task Offloading (DATO) heuristic algorithm is proposed, which strategically offloads workflow tasks between serverless platforms and private cloud resources. It employs latest-start-time prioritization to sort tasks from different workflows, and dynamically determines optimal offloading decisions by balancing task characteristics, resource availability, and cost considerations. Evaluation experiments have been carried out with realistic workflows and platform settings. The results demonstrate the excellent performance of DATO in reducing costs under deadline constraints compared with traditional approaches.
{"title":"Efficient workflow offloading in private clouds using serverless computing","authors":"Shukun Yu , Quanwang Wu , Kun Cai , Taolin Guo , Zhuo Jiang , Tianhao Sun","doi":"10.1016/j.eswa.2026.131460","DOIUrl":"10.1016/j.eswa.2026.131460","url":null,"abstract":"<div><div>Although private clouds provide workflow execution environments with enhanced data security, greater administrative control, and predictable resource provisioning, they often suffer from limited elasticity and struggle to adapt to dynamic and bursty workload patterns. Serverless computing, characterized by elastic scaling, pay-per-use pricing model, and event-driven execution, presents a promising paradigm to address these limitations. Hence, this paper explores integrating private clouds with serverless computing to enable efficient workflow offloading. A systematic workflow offloading framework is established for minimizing monetary costs under workflow deadlines. A Deadline-Aware Task Offloading (DATO) heuristic algorithm is proposed, which strategically offloads workflow tasks between serverless platforms and private cloud resources. It employs latest-start-time prioritization to sort tasks from different workflows, and dynamically determines optimal offloading decisions by balancing task characteristics, resource availability, and cost considerations. Evaluation experiments have been carried out with realistic workflows and platform settings. The results demonstrate the excellent performance of DATO in reducing costs under deadline constraints compared with traditional approaches.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131460"},"PeriodicalIF":7.5,"publicationDate":"2026-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122665","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-30DOI: 10.1016/j.eswa.2026.131425
Arslan Saleem, Cem Direkoglu
Traffic anomaly detection (AD) is essential for improving public safety, reducing risks, and enabling quick responses in intelligent surveillance systems. Aerial traffic monitoring, particularly using Unmanned Aerial Vehicles (UAV), has gained attention due to its potential to address challenges like dynamic urban environments, yet it remains underexplored. Detecting anomalies in drone-captured video involves unique obstacles: rare events, small and overlapping objects, multi-scale targets, and complex backgrounds. To address these challenges, we propose the Depthwise Convolutional Variational Autoencoder (DwCVAE), a novel model designed to enhance AD in drone-based traffic surveillance. DwCVAE leverages depthwise convolutions, which allow efficient and detailed feature extraction, improving model sensitivity to subtle and multi-scale anomalies. The proposed DwCVAE adopts an encoder-latent-decoder VAE architecture, in which stacked depthwise convolutional layers in the encoder emphasize spatially localized feature learning while maintaining channel-wise efficiency, and a compact variational latent space captures the distribution of normal traffic dynamics. Built on variational autoencoder (VAE) architecture, DwCVAE creates compact latent representations that capture normal traffic patterns, enabling reliable detection of deviations. Anomalies are identified through reconstruction-based scoring, where events that deviate from the learned normal representations yield higher reconstruction errors. This depthwise approach marks a key innovation, optimizing both computational efficiency and detection accuracy. We design four additional models: Convolutional Variational Autoencoder (CVAE), Dilated Convolutional VAE (DCVAE), Separable Convolutional VAE (SCVAE), and Convolutional LSTM VAE (CLSTMVAE) to systematically assess the effectiveness of DwCVAE. Additionally, we evaluate DwCVAE against state-of-the-art weakly supervised and unsupervised models on two benchmark datasets, Drone-Anomaly and UIT-Adrone. DwCVAE achieves an AUC of 74.95 with an EER of 0.30 on Drone-Anomaly, and an AUC of 79.77 with an EER of 0.27 on UIT-ADrone, demonstrating its superior performance in complex aerial surveillance tasks.
{"title":"A depthwise convolutional variational autoencoder for anomaly detection in complex traffic scenarios from UAV views","authors":"Arslan Saleem, Cem Direkoglu","doi":"10.1016/j.eswa.2026.131425","DOIUrl":"10.1016/j.eswa.2026.131425","url":null,"abstract":"<div><div>Traffic anomaly detection (AD) is essential for improving public safety, reducing risks, and enabling quick responses in intelligent surveillance systems. Aerial traffic monitoring, particularly using Unmanned Aerial Vehicles (UAV), has gained attention due to its potential to address challenges like dynamic urban environments, yet it remains underexplored. Detecting anomalies in drone-captured video involves unique obstacles: rare events, small and overlapping objects, multi-scale targets, and complex backgrounds. To address these challenges, we propose the Depthwise Convolutional Variational Autoencoder (DwCVAE), a novel model designed to enhance AD in drone-based traffic surveillance. DwCVAE leverages depthwise convolutions, which allow efficient and detailed feature extraction, improving model sensitivity to subtle and multi-scale anomalies. The proposed DwCVAE adopts an encoder-latent-decoder VAE architecture, in which stacked depthwise convolutional layers in the encoder emphasize spatially localized feature learning while maintaining channel-wise efficiency, and a compact variational latent space captures the distribution of normal traffic dynamics. Built on variational autoencoder (VAE) architecture, DwCVAE creates compact latent representations that capture normal traffic patterns, enabling reliable detection of deviations. Anomalies are identified through reconstruction-based scoring, where events that deviate from the learned normal representations yield higher reconstruction errors. This depthwise approach marks a key innovation, optimizing both computational efficiency and detection accuracy. We design four additional models: Convolutional Variational Autoencoder (CVAE), Dilated Convolutional VAE (DCVAE), Separable Convolutional VAE (SCVAE), and Convolutional LSTM VAE (CLSTMVAE) to systematically assess the effectiveness of DwCVAE. Additionally, we evaluate DwCVAE against state-of-the-art weakly supervised and unsupervised models on two benchmark datasets, Drone-Anomaly and UIT-Adrone. DwCVAE achieves an AUC of 74.95 with an EER of 0.30 on Drone-Anomaly, and an AUC of 79.77 with an EER of 0.27 on UIT-ADrone, demonstrating its superior performance in complex aerial surveillance tasks.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131425"},"PeriodicalIF":7.5,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122667","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-30DOI: 10.1016/j.eswa.2026.131329
Longxin Zhang , Shuping Ye , Benlian Xu , Shuting Le , Xu Zhou , Mingli Lu , Jinliang Cong , Jian Shi
Recent advancements in 3D Gaussian Splatting (3DGS) have demonstrated significant potential for high-fidelity scene reconstruction and real-time rendering. However, applying 3DGS-based Simultaneous Localization and Mapping (SLAM) in dynamic environments remains challenging due to disturbance from moving objects, which corrupt the Gaussian initialization and optimization processes. To address these issues, we propose a dynamic 3D Gaussian SLAM framework that integrates a dynamic pollution suppression mechanism and an incremental Gaussian optimization strategy. Our approach leverages semantic segmentation and depth fusion to actively mask dynamic regions during initialization, ensuring the integrity of the foundational Gaussian model. Furthermore, we introduce an image inpainting network to restore static content in masked areas, enabling high-quality densification of under-rendered regions. Experimental results on the TUM RGB-D dataset and real-world dynamic scenes demonstrate that our method significantly outperforms state-of-the-art approaches in both localization accuracy and rendering quality, achieving robust performance in highly dynamic environments.
{"title":"Dynamic 3D Gaussian SLAM via motion suppression and incremental optimization","authors":"Longxin Zhang , Shuping Ye , Benlian Xu , Shuting Le , Xu Zhou , Mingli Lu , Jinliang Cong , Jian Shi","doi":"10.1016/j.eswa.2026.131329","DOIUrl":"10.1016/j.eswa.2026.131329","url":null,"abstract":"<div><div>Recent advancements in 3D Gaussian Splatting (3DGS) have demonstrated significant potential for high-fidelity scene reconstruction and real-time rendering. However, applying 3DGS-based Simultaneous Localization and Mapping (SLAM) in dynamic environments remains challenging due to disturbance from moving objects, which corrupt the Gaussian initialization and optimization processes. To address these issues, we propose a dynamic 3D Gaussian SLAM framework that integrates a dynamic pollution suppression mechanism and an incremental Gaussian optimization strategy. Our approach leverages semantic segmentation and depth fusion to actively mask dynamic regions during initialization, ensuring the integrity of the foundational Gaussian model. Furthermore, we introduce an image inpainting network to restore static content in masked areas, enabling high-quality densification of under-rendered regions. Experimental results on the TUM RGB-D dataset and real-world dynamic scenes demonstrate that our method significantly outperforms state-of-the-art approaches in both localization accuracy and rendering quality, achieving robust performance in highly dynamic environments.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"311 ","pages":"Article 131329"},"PeriodicalIF":7.5,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081700","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The emotion-cause pair extraction (ECPE) task aims to identify emotion clauses and their corresponding cause clauses from document-level text. It has important applications in a wide range of scenarios, including public opinion monitoring and user feedback analysis. Although research has made initial progress on this task, existing methods still face challenges in identifying implicit emotions. Firstly, the lack of explicit semantic guidance leads to insufficient discriminative power, especially when dealing with ambiguous emotional expressions. Secondly, existing methods primarily focus on modeling intra-sentence relationships, which limits their ability to jointly capture cross-sentence temporal dependencies and global semantic information. To address the challenges of emotion-cause pair extraction, we propose a question-guided dual-channel contrastive learning framework, DCCL. Firstly, the DCCL employs a question formulation based on machine reading comprehension (MRC) to guide the model in capturing the emotion-cause relationship between clauses. Furthermore, task-specific queries are explicitly injected into the input, making the model more aware of the task objective. Secondly, in DCCL, we design a dual-channel network combining query-aware clause-level Transformer and BiLSTM to enhance the model’s ability to capture temporal and global contextual dependencies, which enables DCCL to capture the temporal and global contextual relationships between clauses more fully. Thirdly, the DCCL incorporates supervised contrastive learning. We leverage positive and negative samples to incorporate contrastive learning into each channel, which optimizes the representation space and enhances the model’s ability to recognize ambiguous emotions and boundary conditions. We conducted experiments on three mainstream tasks, namely emotion cause pair extraction, emotion extraction, and cause extraction, on the ECPE benchmark dataset. The results show that DCCL improves the F1 scores of the best baseline models such as CD-MRC, SEG, ect by 1.53%, 4.41%, respectively in the emotion-cause pair extraction task, 0.81%, 4.37%, respectively in the emotion extraction task, and 0.62%, 1.27%, respectively in the cause extraction task. Moreover, compared with the large language model baseline LLM-MTLN, DCCL further improves F1 by 2.48%, 4.50%, and 0.63% on these three tasks, respectively.
{"title":"DCCL: Question-guided dual-channel contrastive learning framework for emotion-cause pair extraction","authors":"Hongyang Wang, Yajun Du, Jia Liu, Xianyong Li, Xiaoliang Chen, Yanli Lee, Qing Qi, Wanjie Zhang","doi":"10.1016/j.eswa.2026.131357","DOIUrl":"10.1016/j.eswa.2026.131357","url":null,"abstract":"<div><div>The emotion-cause pair extraction (ECPE) task aims to identify emotion clauses and their corresponding cause clauses from document-level text. It has important applications in a wide range of scenarios, including public opinion monitoring and user feedback analysis. Although research has made initial progress on this task, existing methods still face challenges in identifying implicit emotions. Firstly, the lack of explicit semantic guidance leads to insufficient discriminative power, especially when dealing with ambiguous emotional expressions. Secondly, existing methods primarily focus on modeling intra-sentence relationships, which limits their ability to jointly capture cross-sentence temporal dependencies and global semantic information. To address the challenges of emotion-cause pair extraction, we propose a question-guided dual-channel contrastive learning framework, DCCL. Firstly, the DCCL employs a question formulation based on machine reading comprehension (MRC) to guide the model in capturing the emotion-cause relationship between clauses. Furthermore, task-specific queries are explicitly injected into the input, making the model more aware of the task objective. Secondly, in DCCL, we design a dual-channel network combining query-aware clause-level Transformer and BiLSTM to enhance the model’s ability to capture temporal and global contextual dependencies, which enables DCCL to capture the temporal and global contextual relationships between clauses more fully. Thirdly, the DCCL incorporates supervised contrastive learning. We leverage positive and negative samples to incorporate contrastive learning into each channel, which optimizes the representation space and enhances the model’s ability to recognize ambiguous emotions and boundary conditions. We conducted experiments on three mainstream tasks, namely emotion cause pair extraction, emotion extraction, and cause extraction, on the ECPE benchmark dataset. The results show that DCCL improves the F1 scores of the best baseline models such as CD-MRC, SEG, ect by 1.53%, 4.41%, respectively in the emotion-cause pair extraction task, 0.81%, 4.37%, respectively in the emotion extraction task, and 0.62%, 1.27%, respectively in the cause extraction task. Moreover, compared with the large language model baseline LLM-MTLN, DCCL further improves F1 by 2.48%, 4.50%, and 0.63% on these three tasks, respectively.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"311 ","pages":"Article 131357"},"PeriodicalIF":7.5,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146070882","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-29DOI: 10.1016/j.eswa.2026.131381
Jiajia Jiang , Min Zhan , Gaocan Gong , Lin Wang , Quanbo Zha
Preference learning involves developing a model that reflects a decision-maker's preference based on the provided information. Existing preference learning models for multi-criteria classification overlook exploring the membership degree of an alternative to a predefined class, especially in the face of heterogeneous types of criteria information. To address this issue, this paper proposes a preference learning model based on membership degree maximization with heterogeneous information. First, based on the additive utility function, we construct a preference model that integrates numerical and linguistic criteria information to express the utilities of alternatives. Within this model, four types of utility functions are considered to describe the variation characteristics of criteria. Next, triangular fuzzy numbers are employed to capture the membership degree of each alternative within the predefined classes, and a learning model is developed by maximizing the membership degrees of alternatives to their corresponding predefined classes. Finally, the proposed model is applied to landslide early warning to verify its feasibility.
{"title":"Preference learning based on maximizing membership degree with heterogeneous information for landslide early warning","authors":"Jiajia Jiang , Min Zhan , Gaocan Gong , Lin Wang , Quanbo Zha","doi":"10.1016/j.eswa.2026.131381","DOIUrl":"10.1016/j.eswa.2026.131381","url":null,"abstract":"<div><div>Preference learning involves developing a model that reflects a decision-maker's preference based on the provided information. Existing preference learning models for multi-criteria classification overlook exploring the membership degree of an alternative to a predefined class, especially in the face of heterogeneous types of criteria information. To address this issue, this paper proposes a preference learning model based on membership degree maximization with heterogeneous information. First, based on the additive utility function, we construct a preference model that integrates numerical and linguistic criteria information to express the utilities of alternatives. Within this model, four types of utility functions are considered to describe the variation characteristics of criteria. Next, triangular fuzzy numbers are employed to capture the membership degree of each alternative within the predefined classes, and a learning model is developed by maximizing the membership degrees of alternatives to their corresponding predefined classes. Finally, the proposed model is applied to landslide early warning to verify its feasibility.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131381"},"PeriodicalIF":7.5,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122666","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-29DOI: 10.1016/j.eswa.2026.131388
Haixi Cheng , Chaoqun Hong , Bo Zhang , Huihui Fang , Yanwu Xu , Si Yong Yeo
Glaucoma is one of the leading causes of irreversible blindness worldwide. Color fundus photography (CFP) and optical coherence tomography (OCT) are two primary imaging modalities for glaucoma diagnosis. Recently, multi-modal approaches that combine CFP and OCT have demonstrated higher diagnostic accuracy compared to single-modal methods. However, the high similarity among medical image poses presents a challenge for extracting effective features. Additionally, low-quality features can degrade fusion performance, potentially leading to inaccurate grading results. To address these challenges, we propose a Multi-scale Attention and Gated Fusion (MAGF) framework, which incorporates a dual-branch feature extraction architecture with targeted attention, a Multi-scale Attention Fusion Module (MAFM) for enhancing OCT features, and a Gated Fusion Module (GFM) for adaptive integration of CFP and OCT modalities. Extensive experiments demonstrate that our method achieves state-of-the-art (SOTA) performance in glaucoma grading.
{"title":"MAGF: Multi-scale attention and gated fusion for multi-modal glaucoma grading","authors":"Haixi Cheng , Chaoqun Hong , Bo Zhang , Huihui Fang , Yanwu Xu , Si Yong Yeo","doi":"10.1016/j.eswa.2026.131388","DOIUrl":"10.1016/j.eswa.2026.131388","url":null,"abstract":"<div><div>Glaucoma is one of the leading causes of irreversible blindness worldwide. Color fundus photography (CFP) and optical coherence tomography (OCT) are two primary imaging modalities for glaucoma diagnosis. Recently, multi-modal approaches that combine CFP and OCT have demonstrated higher diagnostic accuracy compared to single-modal methods. However, the high similarity among medical image poses presents a challenge for extracting effective features. Additionally, low-quality features can degrade fusion performance, potentially leading to inaccurate grading results. To address these challenges, we propose a Multi-scale Attention and Gated Fusion (MAGF) framework, which incorporates a dual-branch feature extraction architecture with targeted attention, a Multi-scale Attention Fusion Module (MAFM) for enhancing OCT features, and a Gated Fusion Module (GFM) for adaptive integration of CFP and OCT modalities. Extensive experiments demonstrate that our method achieves state-of-the-art (SOTA) performance in glaucoma grading.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"312 ","pages":"Article 131388"},"PeriodicalIF":7.5,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146122714","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-29DOI: 10.1016/j.eswa.2026.131261
Shuhua Zhang , Ming Liu , Dong Li
The rapid spread of invasive plants such as Spartina alterniflora has emerged as a major ecological and economic threats to coastal wetlands, while existing management strategies often fail to adapt to dynamic invasion processes and limited financial resources. To address this challenge, this study develops a novel data-driven-simulation–optimization (DDSO) framework that enables dynamic and spatially explicit management of biological invasions. The core innovation lies in coupling data-driven ecological parameterization based on multi-source observations with a simulation model that captures life-cycle transitions and spatial dispersal, and a mixed-integer optimization module that allocates control budgets and intervention intensities across space and time. By integrating heterogeneous environmental, biological, and management data, the framework constructs time-varying ecological parameters that reflect evolving invasion conditions and underlying ecological processes. The optimization component then generates cost-effective intervention schedules under fixed budget constraints. Comparative evaluation against system dynamics (SD) and simulation–optimization (SO) models shows that DDSO outperforms conventional approaches not only in budget efficiency, but also by revealing counterintuitive management logics: management effectiveness hinges more on the presence of a coordinated optimization framework than on investment scale, and economically efficient strategies inherently favor highly uneven spatial resource allocation. These mechanism-level insights underscore the importance of early intervention and cross-regional coordination, establishing DDSO as a policy-relevant framework for adaptive invasive species management.
{"title":"An integrated data-driven-simulation-optimization model: insights into controlling invasive plants in China","authors":"Shuhua Zhang , Ming Liu , Dong Li","doi":"10.1016/j.eswa.2026.131261","DOIUrl":"10.1016/j.eswa.2026.131261","url":null,"abstract":"<div><div>The rapid spread of invasive plants such as Spartina alterniflora has emerged as a major ecological and economic threats to coastal wetlands, while existing management strategies often fail to adapt to dynamic invasion processes and limited financial resources. To address this challenge, this study develops a novel data-driven-simulation–optimization (DDSO) framework that enables dynamic and spatially explicit management of biological invasions. The core innovation lies in coupling data-driven ecological parameterization based on multi-source observations with a simulation model that captures life-cycle transitions and spatial dispersal, and a mixed-integer optimization module that allocates control budgets and intervention intensities across space and time. By integrating heterogeneous environmental, biological, and management data, the framework constructs time-varying ecological parameters that reflect evolving invasion conditions and underlying ecological processes. The optimization component then generates cost-effective intervention schedules under fixed budget constraints. Comparative evaluation against system dynamics (SD) and simulation–optimization (SO) models shows that DDSO outperforms conventional approaches not only in budget efficiency, but also by revealing counterintuitive management logics: management effectiveness hinges more on the presence of a coordinated optimization framework than on investment scale, and economically efficient strategies inherently favor highly uneven spatial resource allocation. These mechanism-level insights underscore the importance of early intervention and cross-regional coordination, establishing DDSO as a policy-relevant framework for adaptive invasive species management.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"310 ","pages":"Article 131261"},"PeriodicalIF":7.5,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080880","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-28DOI: 10.1016/j.eswa.2026.131236
Mao Xi , Longyan Xu , Zaihan He , Jiali Chen , Chenyang Bai , Ren Gao
Aiming at the problems of premature convergence, insufficient population diversity and decreased convergence rate in high-dimensional optimization of metaheuristic algorithms, this paper proposes an adaptive archiving hybrid CMA-ES (A2-CMA-ES) for complex engineering design. This method improves population diversity and inhibits premature convergence by introducing external archive fusion mean update with cost information. Secondly, a step size adaptive adjustment strategy is designed, which combines nonlinear suppression and reward amplification mechanism to alleviate the abnormal fluctuation of step size and enhance the global search ability. Furthermore, a multiplicative growth strategy of population size based on statistical stagnation criterion is established, and the search distribution is dynamically expanded through soft restart mechanism to reduce estimation noise. In order to ensure the high-dimensional numerical stability, a monitoring mechanism of covariance matrix condition number and convex combination regularization correction mechanism are constructed to ensure the robustness of anisotropic sampling basis. In this stable framework, a directional mining module based on probabilistic local model is embedded, and anisotropic fine sampling is carried out in the neighborhood of historical optimal solution based on the learned local geometry. Finally, combined with the dynamic learning rate regulation mechanism, the covariance matrix update parameters are adaptively adjusted to balance the global exploration and local development. The results of ablation experiments show that the coordination mechanism of the above components makes the algorithm effectively deal with complex high-dimensional optimization problems. In the CEC2017 benchmark test, A2-CMA-ES was compared with seven advanced algorithms in multi-dimensional scenarios, and further applied to engineering problems such as pressure vessel design, cantilever beam design, three-bar truss optimization, and UAV path planning. The results show that the algorithm is competitive in terms of convergence speed, accuracy and robustness.
{"title":"A2-CMA-ES: a hybrid CMA-ES with adaptive archive for complex engineering design","authors":"Mao Xi , Longyan Xu , Zaihan He , Jiali Chen , Chenyang Bai , Ren Gao","doi":"10.1016/j.eswa.2026.131236","DOIUrl":"10.1016/j.eswa.2026.131236","url":null,"abstract":"<div><div>Aiming at the problems of premature convergence, insufficient population diversity and decreased convergence rate in high-dimensional optimization of metaheuristic algorithms, this paper proposes an adaptive archiving hybrid CMA-ES (A<sup>2</sup>-CMA-ES) for complex engineering design. This method improves population diversity and inhibits premature convergence by introducing external archive fusion mean update with cost information. Secondly, a step size adaptive adjustment strategy is designed, which combines nonlinear suppression and reward amplification mechanism to alleviate the abnormal fluctuation of step size and enhance the global search ability. Furthermore, a multiplicative growth strategy of population size based on statistical stagnation criterion is established, and the search distribution is dynamically expanded through soft restart mechanism to reduce estimation noise. In order to ensure the high-dimensional numerical stability, a monitoring mechanism of covariance matrix condition number and convex combination regularization correction mechanism are constructed to ensure the robustness of anisotropic sampling basis. In this stable framework, a directional mining module based on probabilistic local model is embedded, and anisotropic fine sampling is carried out in the neighborhood of historical optimal solution based on the learned local geometry. Finally, combined with the dynamic learning rate regulation mechanism, the covariance matrix update parameters are adaptively adjusted to balance the global exploration and local development. The results of ablation experiments show that the coordination mechanism of the above components makes the algorithm effectively deal with complex high-dimensional optimization problems. In the CEC2017 benchmark test, A<sup>2</sup>-CMA-ES was compared with seven advanced algorithms in multi-dimensional scenarios, and further applied to engineering problems such as pressure vessel design, cantilever beam design, three-bar truss optimization, and UAV path planning. The results show that the algorithm is competitive in terms of convergence speed, accuracy and robustness.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"310 ","pages":"Article 131236"},"PeriodicalIF":7.5,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081045","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-28DOI: 10.1016/j.eswa.2026.131291
Kai Wang , Yun Huang
To enhance the operational performance of supply chains under the trends of globalization and customization, integrated multi-factory production and distribution has recently attracted increasing attention. This paper presents a novel integrated multi-factory production scheduling and vehicle routing problem. In this problem, a set of customer orders is first assigned to several distributed factories for production, each of which is arranged as a hybrid flow shop (HFS). Owing to the technical or physical aspects, factory eligibility is considered in the production stage, where some orders can only be processed in a subset of factories. The finished products are then delivered by capacitated vehicles, subject to customer time windows. As a combination of the distributed HFS scheduling problem and the vehicle routing problem, three types of decisions have to be made, namely factory allocation, job scheduling, and vehicle assignment and routing. Considering the NP-hardness of the studied problem, a hybrid algorithm that integrates a distribution estimation algorithm (EDA) with an adaptive large neighborhood search (ALNS) is developed to generate solutions. To improve the local search capability of this algorithm, Q-Learning is employed to dynamically determine the destroy-and-repair operators of ALNS. Computational results on both small-sized and large-sized test problems indicate the superiority of the proposed algorithm.
{"title":"Integrated multi-factory production scheduling and vehicle routing with factory eligibility","authors":"Kai Wang , Yun Huang","doi":"10.1016/j.eswa.2026.131291","DOIUrl":"10.1016/j.eswa.2026.131291","url":null,"abstract":"<div><div>To enhance the operational performance of supply chains under the trends of globalization and customization, integrated multi-factory production and distribution has recently attracted increasing attention. This paper presents a novel integrated multi-factory production scheduling and vehicle routing problem. In this problem, a set of customer orders is first assigned to several distributed factories for production, each of which is arranged as a hybrid flow shop (HFS). Owing to the technical or physical aspects, factory eligibility is considered in the production stage, where some orders can only be processed in a subset of factories. The finished products are then delivered by capacitated vehicles, subject to customer time windows. As a combination of the distributed HFS scheduling problem and the vehicle routing problem, three types of decisions have to be made, namely factory allocation, job scheduling, and vehicle assignment and routing. Considering the NP-hardness of the studied problem, a hybrid algorithm that integrates a distribution estimation algorithm (EDA) with an adaptive large neighborhood search (ALNS) is developed to generate solutions. To improve the local search capability of this algorithm, Q-Learning is employed to dynamically determine the destroy-and-repair operators of ALNS. Computational results on both small-sized and large-sized test problems indicate the superiority of the proposed algorithm.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"310 ","pages":"Article 131291"},"PeriodicalIF":7.5,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080791","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-28DOI: 10.1016/j.eswa.2026.131361
Yazhou Du , Bokang Sun , Boyang Ma , Chao Shuai , Jianhao Yang , Yuanchao Lv , Dongchen Wang , Xuewei Wang , Mingyang Liu
Addressing the challenge where high-value fault samples of wind turbines under real operating conditions are extremely scarce-leading to ineffective feature extraction and poor generalization in existing deep learning methods due to insufficient training data-this paper proposes a Dual-Branch Dynamic Convolutional Temporal Attention Network (DBANet) for few-shot fault diagnosis. First, a time-frequency dual-branch architecture is constructed incorporating a dynamic convolution module. By utilizing an adaptive weighting mechanism of multi-expert kernels, the feature extraction strategy is dynamically adjusted to capture subtle fault signatures under few-shot conditions. Second, a Bidirectional Long Short-Term Memory (BiLSTM) network is integrated with a Multi-Head Attention (MHA) mechanism. This combination deeply mines temporal dependencies while precisely focusing on critical fault frequency bands, thereby enhancing the saliency of feature representation. Extensive validation on multiple public bearing datasets and real-world operational data from the Liaoning Datang Hongshan Wind Farm demonstrates that DBANet performs exceptionally well across diverse datasets and sample-limited environments. Specifically, in the validation using real wind farm data, the proposed method achieved an accuracy of 88.33% even in extremely data-scarce scenarios with only 5 training samples per class. The average accuracy across different training sample sizes reached 93.54%, representing a performance improvement of over 16% compared to state-of-the-art methods. These results fully demonstrate the superiority of the proposed method and its significant value for engineering applications.
{"title":"DBANet: A dual-branch dynamic convolutional temporal attention network for few-shot wind turbine bearing fault diagnosis","authors":"Yazhou Du , Bokang Sun , Boyang Ma , Chao Shuai , Jianhao Yang , Yuanchao Lv , Dongchen Wang , Xuewei Wang , Mingyang Liu","doi":"10.1016/j.eswa.2026.131361","DOIUrl":"10.1016/j.eswa.2026.131361","url":null,"abstract":"<div><div>Addressing the challenge where high-value fault samples of wind turbines under real operating conditions are extremely scarce-leading to ineffective feature extraction and poor generalization in existing deep learning methods due to insufficient training data-this paper proposes a Dual-Branch Dynamic Convolutional Temporal Attention Network (DBANet) for few-shot fault diagnosis. First, a time-frequency dual-branch architecture is constructed incorporating a dynamic convolution module. By utilizing an adaptive weighting mechanism of multi-expert kernels, the feature extraction strategy is dynamically adjusted to capture subtle fault signatures under few-shot conditions. Second, a Bidirectional Long Short-Term Memory (BiLSTM) network is integrated with a Multi-Head Attention (MHA) mechanism. This combination deeply mines temporal dependencies while precisely focusing on critical fault frequency bands, thereby enhancing the saliency of feature representation. Extensive validation on multiple public bearing datasets and real-world operational data from the Liaoning Datang Hongshan Wind Farm demonstrates that DBANet performs exceptionally well across diverse datasets and sample-limited environments. Specifically, in the validation using real wind farm data, the proposed method achieved an accuracy of 88.33% even in extremely data-scarce scenarios with only 5 training samples per class. The average accuracy across different training sample sizes reached 93.54%, representing a performance improvement of over 16% compared to state-of-the-art methods. These results fully demonstrate the superiority of the proposed method and its significant value for engineering applications.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"311 ","pages":"Article 131361"},"PeriodicalIF":7.5,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081697","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号