Pub Date : 2026-01-27DOI: 10.1016/j.eswa.2026.131320
Erfan Amani Bani, Kourosh Eshghi
Mathematical modeling and the subsequent development of optimization algorithms for problems have been the core focus of operations research scientists. However, the challenges of solving complex models promptly have always sparked numerous innovations in this field. Quantum computing has been proposed as an alternative to binary computing for several decades. In recent years, operations researchers have paid special attention to applying and integrating this logic with optimization. Specifically, many quantum-based optimization algorithms have been developed; however, little attention has been given to modeling optimization problems using quantum variables. In this paper, a practical problem, the dynamic ride-sharing problem, is redefined and then modeled with the help of quantum variables. Based on quantum variables, the resulting model is fully compatible with quantum algorithms. Subsequently, quantum algorithms based on Benders’ decomposition have been developed. Despite the limitations of access to quantum computing hardware, from a theoretical perspective in terms of computational complexity and solving a simple example, the performance of the algorithms has been demonstrated.
{"title":"Quantum modeling of the dynamic ride-sharing problem: Development of quantum benders decomposition methods","authors":"Erfan Amani Bani, Kourosh Eshghi","doi":"10.1016/j.eswa.2026.131320","DOIUrl":"10.1016/j.eswa.2026.131320","url":null,"abstract":"<div><div>Mathematical modeling and the subsequent development of optimization algorithms for problems have been the core focus of operations research scientists. However, the challenges of solving complex models promptly have always sparked numerous innovations in this field. Quantum computing has been proposed as an alternative to binary computing for several decades. In recent years, operations researchers have paid special attention to applying and integrating this logic with optimization. Specifically, many quantum-based optimization algorithms have been developed; however, little attention has been given to modeling optimization problems using quantum variables. In this paper, a practical problem, the dynamic ride-sharing problem, is redefined and then modeled with the help of quantum variables. Based on quantum variables, the resulting model is fully compatible with quantum algorithms. Subsequently, quantum algorithms based on Benders’ decomposition have been developed. Despite the limitations of access to quantum computing hardware, from a theoretical perspective in terms of computational complexity and solving a simple example, the performance of the algorithms has been demonstrated.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"310 ","pages":"Article 131320"},"PeriodicalIF":7.5,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080874","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 : 2026-01-27DOI: 10.1016/j.eswa.2026.131287
Yu Yang , Suxia Zhu , Guanglu Sun , Zian He , Xinyu Liu , Kai Zhou , Xiaojuan Cui
Federated learning (FL) enables privacy-preserving fine-tuning of multimodal large language models (MLLMs) on edge devices; however, the limited computational resources of edge clients, coupled with inherent modality and data heterogeneity across clients, pose major challenges for federated multimodal fine-tuning and lead to performance degradation. To tackle these issues, we propose DP-HM2F, a data-driven LoRA framework with a dual-projection representation mechanism for heterogeneous multimodal federated fine-tuning. Specifically, DP-HM2F establishes a dual-projection architecture that exploits a global feature pool and client-specific local feature pools, where the global pool encodes privacy-agnostic shared representations and each edge client dynamically maintains a local pool to refine heterogeneous multimodal representations. The architecture enables projection-based retrieval between the global and local pools to improve representation alignment, while introducing additional computational overhead on resource-constrained devices. To mitigate this limitation, DP-HM2F integrates a data-driven LoRA module that adaptively scales the number of trainable parameters based on local data, thereby alleviating computational constraints across heterogeneous clients. Furthermore, to address semantic conflicts induced by high-dimensional representation spaces during federated aggregation, we introduce a positive-vector collaborative optimization strategy to alleviate conflicting client updates. Extensive experimental results demonstrate that DP-HM2F, with only 7.05% of trainable parameters (a 0.3% reduction compared with conventional LoRA-based methods), achieves a performance improvement of 4.1 points under heterogeneous multimodal settings.
{"title":"DP-HM2F: Data-driven LoRA with dual-projection representation for heterogeneous multimodal federated fine-tuning","authors":"Yu Yang , Suxia Zhu , Guanglu Sun , Zian He , Xinyu Liu , Kai Zhou , Xiaojuan Cui","doi":"10.1016/j.eswa.2026.131287","DOIUrl":"10.1016/j.eswa.2026.131287","url":null,"abstract":"<div><div>Federated learning (FL) enables privacy-preserving fine-tuning of multimodal large language models (MLLMs) on edge devices; however, the limited computational resources of edge clients, coupled with inherent modality and data heterogeneity across clients, pose major challenges for federated multimodal fine-tuning and lead to performance degradation. To tackle these issues, we propose DP-HM2F, a data-driven LoRA framework with a dual-projection representation mechanism for heterogeneous multimodal federated fine-tuning. Specifically, DP-HM2F establishes a dual-projection architecture that exploits a global feature pool and client-specific local feature pools, where the global pool encodes privacy-agnostic shared representations and each edge client dynamically maintains a local pool to refine heterogeneous multimodal representations. The architecture enables projection-based retrieval between the global and local pools to improve representation alignment, while introducing additional computational overhead on resource-constrained devices. To mitigate this limitation, DP-HM2F integrates a data-driven LoRA module that adaptively scales the number of trainable parameters based on local data, thereby alleviating computational constraints across heterogeneous clients. Furthermore, to address semantic conflicts induced by high-dimensional representation spaces during federated aggregation, we introduce a positive-vector collaborative optimization strategy to alleviate conflicting client updates. Extensive experimental results demonstrate that DP-HM2F, with only 7.05% of trainable parameters (a 0.3% reduction compared with conventional LoRA-based methods), achieves a performance improvement of 4.1 points under heterogeneous multimodal settings.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"311 ","pages":"Article 131287"},"PeriodicalIF":7.5,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081695","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 : 2026-01-27DOI: 10.1016/j.eswa.2026.131343
Shuangxue Liu , Hongbin Xie , Yuzhen Lei , Jiaxing Zhao , Xuan Song
Automotive chip production involves complex interdependencies across design, manufacturing, and supply-chain processes, posing significant challenges for interpretable and consistent reasoning. To address these challenges, this paper proposes a Causal-Knowledge Heterogeneous Graph (C-KHG) framework that integrates a domain knowledge graph with a text-grounded causal event graph, capturing linguistically asserted cause-and-effect relations extracted from expert-authored technical documents. Unlike statistical causal discovery or interventional causal modeling, the proposed causal event graph focuses on causally informed semantic reasoning, emphasizing directional consistency and interpretability aligned with domain expert knowledge. Built upon the unified heterogeneous graph, we design a three-stage reasoning pipeline consisting of intent classification, graph-based adaptive retrieval, and large language model (LLM) answer generation. To evaluate its effectiveness, experiments were conducted on three representative tasks: hybrid knowledge-causal reasoning, value-chain question answering, and pure causal reasoning. Specifically, we address three types of tasks: (1) hybrid knowledge-causal reasoning, which jointly involves entity-level knowledge retrieval and cause-and-effect analysis; (2) value-chain question answering, which focuses on structured domain knowledge across the automotive chip lifecycle; and (3) pure causal reasoning, which concentrates exclusively on cause-and-effect relations without requiring explicit entity attributes. Instead of relying on direct prompt-based inference, we construct the causal knowledge graph as an explicit intermediate structured layer, efficiently bootstrapped by LLMs, which serves as a persistent and updatable domain memory. This design improves reasoning stability and directional consistency while facilitating knowledge maintenance and iterative updates without model retraining. Experimental results on automotive chip value-chain question answering tasks demonstrate that the proposed framework consistently improves reasoning accuracy, causal directionality, and interpretability compared with vanilla LLMs and conventional knowledge-graph-based retrieval methods. In particular, for the causal-knowledge fusion task, the cosine similarity of GLM4-9B improved from 9.63 to 21.75. These findings highlight the effectiveness of structured graph-based reasoning scaffolds as intermediate representations for enhancing LLM-based reasoning in complex industrial domains. Code and data are made available on https://github.com/shuangxueliu/C-KHG.
{"title":"LLM-augmented causal-knowledge heterogeneous graph framework for interpretable reasoning and collaborative knowledge fusion in automotive chip production","authors":"Shuangxue Liu , Hongbin Xie , Yuzhen Lei , Jiaxing Zhao , Xuan Song","doi":"10.1016/j.eswa.2026.131343","DOIUrl":"10.1016/j.eswa.2026.131343","url":null,"abstract":"<div><div>Automotive chip production involves complex interdependencies across design, manufacturing, and supply-chain processes, posing significant challenges for interpretable and consistent reasoning. To address these challenges, this paper proposes a Causal-Knowledge Heterogeneous Graph (C-KHG) framework that integrates a domain knowledge graph with a text-grounded causal event graph, capturing linguistically asserted cause-and-effect relations extracted from expert-authored technical documents. Unlike statistical causal discovery or interventional causal modeling, the proposed causal event graph focuses on causally informed semantic reasoning, emphasizing directional consistency and interpretability aligned with domain expert knowledge. Built upon the unified heterogeneous graph, we design a three-stage reasoning pipeline consisting of intent classification, graph-based adaptive retrieval, and large language model (LLM) answer generation. To evaluate its effectiveness, experiments were conducted on three representative tasks: hybrid knowledge-causal reasoning, value-chain question answering, and pure causal reasoning. Specifically, we address three types of tasks: (1) hybrid knowledge-causal reasoning, which jointly involves entity-level knowledge retrieval and cause-and-effect analysis; (2) value-chain question answering, which focuses on structured domain knowledge across the automotive chip lifecycle; and (3) pure causal reasoning, which concentrates exclusively on cause-and-effect relations without requiring explicit entity attributes. Instead of relying on direct prompt-based inference, we construct the causal knowledge graph as an explicit intermediate structured layer, efficiently bootstrapped by LLMs, which serves as a persistent and updatable domain memory. This design improves reasoning stability and directional consistency while facilitating knowledge maintenance and iterative updates without model retraining. Experimental results on automotive chip value-chain question answering tasks demonstrate that the proposed framework consistently improves reasoning accuracy, causal directionality, and interpretability compared with vanilla LLMs and conventional knowledge-graph-based retrieval methods. In particular, for the causal-knowledge fusion task, the cosine similarity of GLM4-9B improved from 9.63 to 21.75. These findings highlight the effectiveness of structured graph-based reasoning scaffolds as intermediate representations for enhancing LLM-based reasoning in complex industrial domains. Code and data are made available on <span><span>https://github.com/shuangxueliu/C-KHG</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"311 ","pages":"Article 131343"},"PeriodicalIF":7.5,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146070881","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 : 2026-01-27DOI: 10.1016/j.eswa.2026.131372
Xianhao Zhang, Hongfei Zhan
Injection molding is an efficient method for the mass production of plastic products, but product quality is susceptible to variations in process conditions and parameters. To improve the real-time performance and accuracy of quality control, deep learning-based data-driven prediction methods have become a research focus. Nevertheless, existing injection molding quality prediction methods tend to prematurely couple variable channels and still face limitations in information fusion and model efficiency. Therefore, this paper combines multi-source data and proposes a quality prediction method based on Multi-Stage Feature Decoupling and Fusion (MSFDF). To address the issue of premature coupling of multivariate features in injection molding, a Temporal and Channel Decoupling Based Multi-Scale Feature Extraction Module (TC-DMFE) is designed to extract multi-scale features while maintaining feature independence. In addition, to address the issue of inadequate integration of multi-scale information during the injection molding process, a Channel-wise Multi-scale Feature Fusion Module (CMFF) is proposed, which fully integrates multi-scale features through a channel by channel fusion strategy and enhances the model’s comprehensive understanding of injection molding process variables under multi-scale variation patterns. On this basis, a Deep Feature Guided Channel Attention Recoupling Module (DCAR) is further constructed to learn inter-channel dependencies and apply channel weighting to achieve more effective variable recoupling. The model proposed in this paper effectively reduces training time while maintaining prediction accuracy and possesses the ability to quickly adapt to injection molding production scenarios.
{"title":"A quality prediction method for injection molding products based on multi-stage feature decoupling and fusion","authors":"Xianhao Zhang, Hongfei Zhan","doi":"10.1016/j.eswa.2026.131372","DOIUrl":"10.1016/j.eswa.2026.131372","url":null,"abstract":"<div><div>Injection molding is an efficient method for the mass production of plastic products, but product quality is susceptible to variations in process conditions and parameters. To improve the real-time performance and accuracy of quality control, deep learning-based data-driven prediction methods have become a research focus. Nevertheless, existing injection molding quality prediction methods tend to prematurely couple variable channels and still face limitations in information fusion and model efficiency. Therefore, this paper combines multi-source data and proposes a quality prediction method based on Multi-Stage Feature Decoupling and Fusion (MSFDF). To address the issue of premature coupling of multivariate features in injection molding, a Temporal and Channel Decoupling Based Multi-Scale Feature Extraction Module (TC-DMFE) is designed to extract multi-scale features while maintaining feature independence. In addition, to address the issue of inadequate integration of multi-scale information during the injection molding process, a Channel-wise Multi-scale Feature Fusion Module (CMFF) is proposed, which fully integrates multi-scale features through a channel by channel fusion strategy and enhances the model’s comprehensive understanding of injection molding process variables under multi-scale variation patterns. On this basis, a Deep Feature Guided Channel Attention Recoupling Module (DCAR) is further constructed to learn inter-channel dependencies and apply channel weighting to achieve more effective variable recoupling. The model proposed in this paper effectively reduces training time while maintaining prediction accuracy and possesses the ability to quickly adapt to injection molding production scenarios.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"310 ","pages":"Article 131372"},"PeriodicalIF":7.5,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080788","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 : 2026-01-27DOI: 10.1016/j.eswa.2025.131034
Jun Guan , Shuanghui Ye , Wenjun Yi
Particle swarm optimization (PSO) is widely applied to various practical problems due to its strong optimization capability and flexibility. However, when tackling complex optimization tasks, it suffers from shortcomings such as premature convergence and an imbalance between global exploration and local exploitation. To address these issues, this study proposes a multi-strategy cooperative particle swarm optimization algorithm (MSCPSO). MSCPSO divides the population into leaders and followers based on fitness and integrates diverse learning strategies to enhance performance. First, a nonlinear adaptive inertia weight is proposed to dynamically adjust inertia according to particle roles, effectively balancing exploration and exploitation. Second, a weighted learning strategy is introduced, which assigns weights based on leader fitness values to guide particles more efficiently toward promising solution regions. Third, a fitness-distance balance mechanism is designed to maintain population diversity in the early stage, accelerate convergence in the later stage, and reduce the probability of falling into local optima. Finally, in the later iterations of the algorithm, a terminal replacement mechanism is designed to replace the worst global particle, reducing population diversity to accelerate convergence. Comparative experiments on CEC2014, CEC2017, and CEC2022 test suites against seven heuristic algorithms, eleven PSO variants, and eight state-of-the-art algorithms show that multi-strategy cooperation significantly enhances PSO performance. MSCPSO outperforms most compared algorithms. Finally, MSCPSO is applied to 3D UAV path planning in complex environments. Across 12 scenarios of varying complexity, MSCPSO demonstrates the ability to generate more feasible and efficient paths in most cases.
{"title":"MSCPSO: A multi-strategy cooperative particle swarm optimization algorithm for UAV path planning","authors":"Jun Guan , Shuanghui Ye , Wenjun Yi","doi":"10.1016/j.eswa.2025.131034","DOIUrl":"10.1016/j.eswa.2025.131034","url":null,"abstract":"<div><div>Particle swarm optimization (PSO) is widely applied to various practical problems due to its strong optimization capability and flexibility. However, when tackling complex optimization tasks, it suffers from shortcomings such as premature convergence and an imbalance between global exploration and local exploitation. To address these issues, this study proposes a multi-strategy cooperative particle swarm optimization algorithm (MSCPSO). MSCPSO divides the population into leaders and followers based on fitness and integrates diverse learning strategies to enhance performance. First, a nonlinear adaptive inertia weight is proposed to dynamically adjust inertia according to particle roles, effectively balancing exploration and exploitation. Second, a weighted learning strategy is introduced, which assigns weights based on leader fitness values to guide particles more efficiently toward promising solution regions. Third, a fitness-distance balance mechanism is designed to maintain population diversity in the early stage, accelerate convergence in the later stage, and reduce the probability of falling into local optima. Finally, in the later iterations of the algorithm, a terminal replacement mechanism is designed to replace the worst global particle, reducing population diversity to accelerate convergence. Comparative experiments on CEC2014, CEC2017, and CEC2022 test suites against seven heuristic algorithms, eleven PSO variants, and eight state-of-the-art algorithms show that multi-strategy cooperation significantly enhances PSO performance. MSCPSO outperforms most compared algorithms. Finally, MSCPSO is applied to 3D UAV path planning in complex environments. Across 12 scenarios of varying complexity, MSCPSO demonstrates the ability to generate more feasible and efficient paths in most cases.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"310 ","pages":"Article 131034"},"PeriodicalIF":7.5,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080873","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}
Aiming at the state master-slave synchronization problem of discrete-time output-coupled neural networks (OCNNs), this study proposes a novel control framework. Deviating from the existing studies, this study aims at the constraint that the output state cannot obtain all neuron state information, innovatively constructs full-dimensional observers in the master/slave system respectively to realize state reconstruction, which solves the problem of missing reference trajectory caused by incomplete state information of the master system, and realizes state synchronization for the first time under the output-coupled framework. A round-robin (RR) protocol is introduced to dynamically schedule the output-coupled communication among neurons to address the transmission efficiency bottleneck under limited bandwidth resources. To allocate communication resources more intelligently, a novel event-triggered (ET) mechanism is designed. Under this mechanism, the triggering threshold is constructed based on the last triggering instant and is updated according to the most current triggering instant. It significantly reduces overall resource consumption while ensuring control performance. Then, the joint Lyapunov function is constructed based on the designed observer-controller-protocol interaction dynamic model. It derived a quantitative relationship between the triggering frequency and the error decay boundary, thereby providing a quantifiable basis for precision tuning in network control. Finally, the effectiveness of the proposed method is verified by circuit simulation.
{"title":"Dual-observer-based event-triggered state synchronization for discrete-time output-coupled neural networks under Round-Robin protocol","authors":"Zhihong Liang , Huaguang Zhang , Juan Zhang , Qiongwen Zhang","doi":"10.1016/j.eswa.2026.131155","DOIUrl":"10.1016/j.eswa.2026.131155","url":null,"abstract":"<div><div>Aiming at the state master-slave synchronization problem of discrete-time output-coupled neural networks (OCNNs), this study proposes a novel control framework. Deviating from the existing studies, this study aims at the constraint that the output state cannot obtain all neuron state information, innovatively constructs full-dimensional observers in the master/slave system respectively to realize state reconstruction, which solves the problem of missing reference trajectory caused by incomplete state information of the master system, and realizes state synchronization for the first time under the output-coupled framework. A round-robin (RR) protocol is introduced to dynamically schedule the output-coupled communication among neurons to address the transmission efficiency bottleneck under limited bandwidth resources. To allocate communication resources more intelligently, a novel event-triggered (ET) mechanism is designed. Under this mechanism, the triggering threshold is constructed based on the last triggering instant and is updated according to the most current triggering instant. It significantly reduces overall resource consumption while ensuring control performance. Then, the joint Lyapunov function is constructed based on the designed observer-controller-protocol interaction dynamic model. It derived a quantitative relationship between the triggering frequency and the error decay boundary, thereby providing a quantifiable basis for precision tuning in network control. Finally, the effectiveness of the proposed method is verified by circuit simulation.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"310 ","pages":"Article 131155"},"PeriodicalIF":7.5,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080881","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 : 2026-01-26DOI: 10.1016/j.eswa.2026.131334
Xiong Liqin , Chen Xiliang , Luo Xijian , Cao Lei
Multi-UAV cooperative pursuit is an important branch in the field of multi-robot collaboration, widely applied in critical tasks such as cooperative reconnaissance and emergency rescue. Existing researches primarily focus on the constant-speed cooperative pursuit problem in fully observable environments, while paying less attention to the issue of pursuing fast-evading targets in partially observable settings. Therefore, this paper proposes a perception-enhanced multi-agent deep reinforcement learning method to enable pursuing UAVs to learn cooperation through local observations. Firstly, a pursuit judgment model based on Apollonius circle is constructed for the bounded multi-UAV cooperative pursuit problem, and a rigorous theoretical proof is provided for the boundary condition of successful pursuit. Subsequently, a dynamic multi-agent interaction graph is established based on the real-time connectivity among the pursuing UAVs, and then information features are extracted from neighbors using a two-layer graph attention network to enhance their perceptual capability. Finally, a joint reward function incorporating multiple types of rewards is designed to reflect task requirements, and the framework of centralized training with decentralized execution is utilized to train the policies of pursuing UAVs, promoting them to learn autonomous cooperation. To verify the effectiveness of our method, extensive comparative experiments are conducted in various scenarios with different evasion strategies. Experimental results show that in almost all scenarios, our method outperforms other methods in terms of success rate, stability, and time consumption. Notably, it improves the success rate by up to several times (over ten times in some cases) and reduces the average pursuit steps by a maximum of 71.42%.
{"title":"A perception-enhanced multi-agent deep reinforcement learning method for multi-UAV cooperative pursuit","authors":"Xiong Liqin , Chen Xiliang , Luo Xijian , Cao Lei","doi":"10.1016/j.eswa.2026.131334","DOIUrl":"10.1016/j.eswa.2026.131334","url":null,"abstract":"<div><div>Multi-UAV cooperative pursuit is an important branch in the field of multi-robot collaboration, widely applied in critical tasks such as cooperative reconnaissance and emergency rescue. Existing researches primarily focus on the constant-speed cooperative pursuit problem in fully observable environments, while paying less attention to the issue of pursuing fast-evading targets in partially observable settings. Therefore, this paper proposes a perception-enhanced multi-agent deep reinforcement learning method to enable pursuing UAVs to learn cooperation through local observations. Firstly, a pursuit judgment model based on Apollonius circle is constructed for the bounded multi-UAV cooperative pursuit problem, and a rigorous theoretical proof is provided for the boundary condition of successful pursuit. Subsequently, a dynamic multi-agent interaction graph is established based on the real-time connectivity among the pursuing UAVs, and then information features are extracted from neighbors using a two-layer graph attention network to enhance their perceptual capability. Finally, a joint reward function incorporating multiple types of rewards is designed to reflect task requirements, and the framework of centralized training with decentralized execution is utilized to train the policies of pursuing UAVs, promoting them to learn autonomous cooperation. To verify the effectiveness of our method, extensive comparative experiments are conducted in various scenarios with different evasion strategies. Experimental results show that in almost all scenarios, our method outperforms other methods in terms of success rate, stability, and time consumption. Notably, it improves the success rate by up to several times (over ten times in some cases) and reduces the average pursuit steps by a maximum of 71.42%.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"311 ","pages":"Article 131334"},"PeriodicalIF":7.5,"publicationDate":"2026-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146070732","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}
Soft sensors are indispensable tools for inferring hard-to-measure quality variables in modern complex industrial processes. However, the training objectives of most deep learning-based soft sensor models typically focus on minimizing point-wise errors, a strategy that often fails to capture crucial process dynamics. This limitation directly compromises the reliability of soft sensor models in practical process control. This paper introduces the Process-Dynamic Guided Fusion Loss (PD-GFL), a model-agnostic, composite loss function designed to address this gap. PD-GFL guides model optimization from three dimensions: (i) aligning slow-varying trends to capture long-term system states, (ii) matching local statistical features to preserve distributional properties, and (iii) synchronizing differential dynamic patterns to ensure the process’s intrinsic inertia and smoothness. These objectives are integrated via an adaptive uncertainty-based weighting scheme, which enables the seamless integration of PD-GFL with diverse deep learning architectures. Extensive experiments on public benchmarks and a real-world industrial dataset demonstrate the superiority of PD-GFL, yielding improvements of up to 30% in MSE and 49% in MAPE over standard training. Notably, PD-GFL empowers a simple MLP backbone to improve its MSE by approximately 6%, achieving accuracy competitive with advanced Transformer models. Our findings show that improving the training objective is an effective alternative to simply making models more complex, offering a practical way to build more reliable and efficient soft sensors for industrial applications.
{"title":"A novel process dynamic guided fusion loss for soft sensor modeling in complex industrial processes","authors":"Yulong Wang, Jiayi Zhou, Fanlei Lu, Xu Tang, Xiaoli Wang, Chunhua Yang","doi":"10.1016/j.eswa.2026.131340","DOIUrl":"10.1016/j.eswa.2026.131340","url":null,"abstract":"<div><div>Soft sensors are indispensable tools for inferring hard-to-measure quality variables in modern complex industrial processes. However, the training objectives of most deep learning-based soft sensor models typically focus on minimizing point-wise errors, a strategy that often fails to capture crucial process dynamics. This limitation directly compromises the reliability of soft sensor models in practical process control. This paper introduces the Process-Dynamic Guided Fusion Loss (PD-GFL), a model-agnostic, composite loss function designed to address this gap. PD-GFL guides model optimization from three dimensions: (i) aligning slow-varying trends to capture long-term system states, (ii) matching local statistical features to preserve distributional properties, and (iii) synchronizing differential dynamic patterns to ensure the process’s intrinsic inertia and smoothness. These objectives are integrated via an adaptive uncertainty-based weighting scheme, which enables the seamless integration of PD-GFL with diverse deep learning architectures. Extensive experiments on public benchmarks and a real-world industrial dataset demonstrate the superiority of PD-GFL, yielding improvements of up to 30% in MSE and 49% in MAPE over standard training. Notably, PD-GFL empowers a simple MLP backbone to improve its MSE by approximately 6%, achieving accuracy competitive with advanced Transformer models. Our findings show that improving the training objective is an effective alternative to simply making models more complex, offering a practical way to build more reliable and efficient soft sensors for industrial applications.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"310 ","pages":"Article 131340"},"PeriodicalIF":7.5,"publicationDate":"2026-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081049","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 : 2026-01-26DOI: 10.1016/j.eswa.2026.131351
Biao Qu , Huajun She , Qingxia Wu , Pingping Jie , Qi Yao , Liulu Zhang , Yuting Fu , Yamei Luo , Taishan Kang , Gaofeng Zheng
Radial Magnetic Resonance Imaging (MRI) enables high acceleration imaging. However, reconstructing high-quality images from highly undersampled k-space data remains a challenge due to the difficulty in accurately estimating coil sensitivity maps from limited autocalibration signal. The autocalibration signal is extracted from the central k-space in existing compressed sensing and deep learning methods. This extraction makes the low-resolution sensitivity maps lack high-frequency details, leading to suboptimal reconstructions. To address this problem, we propose a Joint Sensitivity-learning driven Deep Unfolding Network (JSDU-Net) for accelerated radial MRI. Sensitivity maps are firstly estimated from all available k-space data, including the low and high-frequency parts and then updated in the reconstruction. JSDU-Net unfolds the iterative reconstruction process into a deep neural network and introduces a novel sensitivity learning strategy that alternately updates sensitivity maps and image estimates in each iteration. This joint optimization facilitates accurate sensitivity estimation by capturing high-frequency information. Extensive experiments demonstrate that JSDU-Net achieves superior performance in detail preservation, artifact suppression, and reconstruction efficiency. Blinded evaluations by clinical radiologists show that the reconstructed images exhibit excellent diagnostic value, suggesting the potential of JSDU-Net for clinical applications of radial MRI.
{"title":"JSDU-Net: Joint sensitivity-learning driven deep unfolding network for accelerated radial MRI reconstruction","authors":"Biao Qu , Huajun She , Qingxia Wu , Pingping Jie , Qi Yao , Liulu Zhang , Yuting Fu , Yamei Luo , Taishan Kang , Gaofeng Zheng","doi":"10.1016/j.eswa.2026.131351","DOIUrl":"10.1016/j.eswa.2026.131351","url":null,"abstract":"<div><div>Radial Magnetic Resonance Imaging (MRI) enables high acceleration imaging. However, reconstructing high-quality images from highly undersampled k-space data remains a challenge due to the difficulty in accurately estimating coil sensitivity maps from limited autocalibration signal. The autocalibration signal is extracted from the central k-space in existing compressed sensing and deep learning methods. This extraction makes the low-resolution sensitivity maps lack high-frequency details, leading to suboptimal reconstructions. To address this problem, we propose a Joint Sensitivity-learning driven Deep Unfolding Network (JSDU-Net) for accelerated radial MRI. Sensitivity maps are firstly estimated from all available k-space data, including the low and high-frequency parts and then updated in the reconstruction. JSDU-Net unfolds the iterative reconstruction process into a deep neural network and introduces a novel sensitivity learning strategy that alternately updates sensitivity maps and image estimates in each iteration. This joint optimization facilitates accurate sensitivity estimation by capturing high-frequency information. Extensive experiments demonstrate that JSDU-Net achieves superior performance in detail preservation, artifact suppression, and reconstruction efficiency. Blinded evaluations by clinical radiologists show that the reconstructed images exhibit excellent diagnostic value, suggesting the potential of JSDU-Net for clinical applications of radial MRI.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"310 ","pages":"Article 131351"},"PeriodicalIF":7.5,"publicationDate":"2026-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080973","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}
Accurate instance segmentation of deep-hole inner surface defects is critical for ensuring the structural integrity and functional reliability of high-precision industrial components. However, this task remains highly challenging due to the extreme scarcity of annotated data, along with the irregular morphology, weak texture, and dense, random spatial distribution of defects. Existing instance segmentation methods typically rely on large-scale supervision, which is prohibitively expensive and often infeasible in real-world manufacturing. While few-shot learning offers a promising alternative, current models primarily focus on semantic segmentation and fail to delineate individual defect instances with accurate boundaries and counts. Moreover, they lack adaptive mechanisms to model fine-grained morphological variations of defect regions and are susceptible to foreground–background ambiguity induced by incomplete annotations, resulting in classification bias during inspection. To address these limitations, we propose SCID-Net, a novel few-shot defect instance segmentation framework based on multi-granularity feature coupling and instance-aware inference decoupling. Specifically, we introduce a Multi-Grained Coupling Module (GCM) to facilitate hierarchical bi-directional interaction between support and query features, enriching both class-level prototypes and instance-specific representations. Built upon this, the Instance-Aware Inference Decoupling Module (IAM) decouples dense inference into specialized pathways, and further integrates adaptive spatial modulation and prototype-driven semantic alignment to suppress noise from incomplete annotations. Extensive experiments on a proprietary industrial deep-hole defect dataset demonstrate that SCID-Net achieves state-of-the-art performance under few-shot settings. Moreover, evaluations on NEU-Seg and MS COCO further validate the exceptional generalization capability of SCID-Net, highlighting its versatility in both challenging industrial environments and diverse real-world scenarios.
{"title":"SCID-Net: Few-shot deep-hole defect instance segmentation via multi-grained feature coupling and instance-aware inference decoupling","authors":"Zongyang Zhao , Jiehu Kang , Yichen Xu , Jian Liang , Luyuan Feng , Yuqi Ren , Ting Xue , Bin Wu","doi":"10.1016/j.eswa.2026.131341","DOIUrl":"10.1016/j.eswa.2026.131341","url":null,"abstract":"<div><div>Accurate instance segmentation of deep-hole inner surface defects is critical for ensuring the structural integrity and functional reliability of high-precision industrial components. However, this task remains highly challenging due to the extreme scarcity of annotated data, along with the irregular morphology, weak texture, and dense, random spatial distribution of defects. Existing instance segmentation methods typically rely on large-scale supervision, which is prohibitively expensive and often infeasible in real-world manufacturing. While few-shot learning offers a promising alternative, current models primarily focus on semantic segmentation and fail to delineate individual defect instances with accurate boundaries and counts. Moreover, they lack adaptive mechanisms to model fine-grained morphological variations of defect regions and are susceptible to foreground–background ambiguity induced by incomplete annotations, resulting in classification bias during inspection. To address these limitations, we propose SCID-Net, a novel few-shot defect instance segmentation framework based on multi-granularity feature coupling and instance-aware inference decoupling. Specifically, we introduce a Multi-Grained Coupling Module (GCM) to facilitate hierarchical bi-directional interaction between support and query features, enriching both class-level prototypes and instance-specific representations. Built upon this, the Instance-Aware Inference Decoupling Module (IAM) decouples dense inference into specialized pathways, and further integrates adaptive spatial modulation and prototype-driven semantic alignment to suppress noise from incomplete annotations. Extensive experiments on a proprietary industrial deep-hole defect dataset demonstrate that SCID-Net achieves state-of-the-art performance under few-shot settings. Moreover, evaluations on NEU-Seg and MS COCO further validate the exceptional generalization capability of SCID-Net, highlighting its versatility in both challenging industrial environments and diverse real-world scenarios.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"310 ","pages":"Article 131341"},"PeriodicalIF":7.5,"publicationDate":"2026-01-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080964","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