This paper investigates a single batch-processing machine scheduling problem with uncertain processing time. The uncertain processing time is characterized by interval grey number. A grey mixed integer linear programming model is established to formulate this uncertain scheduling problem to minimize the makespan. To solve this problem, a genetic algorithm with targeted population generation and neighbourhood search is designed. The results of experiments demonstrate that the proposed algorithm has excellent performance in both efficiency and stability. The resulting scheduling scheme can be shown through the Gantt chart with interval grey processing time, offering a novel approach for visualizing scheduling schemes with uncertain processing time.
{"title":"Single batch-processing machine scheduling problem with interval grey processing time","authors":"Naiming Xie , Yihang Qin , Nanlei Chen , Yingjie Yang","doi":"10.1016/j.asoc.2024.112661","DOIUrl":"10.1016/j.asoc.2024.112661","url":null,"abstract":"<div><div>This paper investigates a single batch-processing machine scheduling problem with uncertain processing time. The uncertain processing time is characterized by interval grey number. A grey mixed integer linear programming model is established to formulate this uncertain scheduling problem to minimize the makespan. To solve this problem, a genetic algorithm with targeted population generation and neighbourhood search is designed. The results of experiments demonstrate that the proposed algorithm has excellent performance in both efficiency and stability. The resulting scheduling scheme can be shown through the Gantt chart with interval grey processing time, offering a novel approach for visualizing scheduling schemes with uncertain processing time.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"170 ","pages":"Article 112661"},"PeriodicalIF":7.2,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143213273","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 : 2025-02-01DOI: 10.1016/j.asoc.2024.112660
Ruizhao Zheng , Yong Zhang , Xiaoyan Sun , Lei Yang , Xianfang Song
Design change planning is an inevitable part of the product development process. Evolutionary algorithms (EAs) have been widely adopted to search for optimal change paths due to their strong global search capabilities. However, many existing approaches overlook key environmental factors like carbon emissions. Furthermore, EAs often struggle with premature convergence when solving complex design problems. This paper aims to develop an effective algorithm for green product design changes by incorporating carbon emission metrics and reinforcement learning techniques. Firstly, a constrained multi-objective optimization model for the green product change planning problem is built for the first time. Besides change cost and duration, a green indicator, i.e., carbon emissions, is introduced into the model, which can make obtained change plans more suitable for actual needs. Next, a multi-strategy self-switching multi-objective evolutionary algorithm assisted by reinforcement learning (R-MSMOEA) is developed to improve the performance of EA on solving the above model. Finally, the proposed model and algorithm are applied in the design change problem of a specific type of Skyworth TV, and experimental results verify their feasibility and effectiveness.
{"title":"A reinforcement learning-assisted multi-objective evolutionary algorithm for generating green change plans of complex products","authors":"Ruizhao Zheng , Yong Zhang , Xiaoyan Sun , Lei Yang , Xianfang Song","doi":"10.1016/j.asoc.2024.112660","DOIUrl":"10.1016/j.asoc.2024.112660","url":null,"abstract":"<div><div>Design change planning is an inevitable part of the product development process. Evolutionary algorithms (EAs) have been widely adopted to search for optimal change paths due to their strong global search capabilities. However, many existing approaches overlook key environmental factors like carbon emissions. Furthermore, EAs often struggle with premature convergence when solving complex design problems. This paper aims to develop an effective algorithm for green product design changes by incorporating carbon emission metrics and reinforcement learning techniques. Firstly, a constrained multi-objective optimization model for the green product change planning problem is built for the first time. Besides change cost and duration, a green indicator, i.e., carbon emissions, is introduced into the model, which can make obtained change plans more suitable for actual needs. Next, a multi-strategy self-switching multi-objective evolutionary algorithm assisted by reinforcement learning (R-MSMOEA) is developed to improve the performance of EA on solving the above model. Finally, the proposed model and algorithm are applied in the design change problem of a specific type of Skyworth TV, and experimental results verify their feasibility and effectiveness.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"170 ","pages":"Article 112660"},"PeriodicalIF":7.2,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143213396","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 : 2025-02-01DOI: 10.1016/j.asoc.2025.112692
Mengjun Xu , Lei Liu , Pengfei Xia , Ziqiang Li , Bin Li
Adversarial transferability is an intriguing yet dangerous property of deep neural networks (DNNs), enabling the potential for black-box adversarial attacks. To better safeguard DNN-based AI models against such attacks, adversarial defense is explored through the lens of reducing the transferability of adversarial examples. From the perspective of decision boundaries in convolutional neural network classification models, we propose a novel method called Adversarial Path Transferability Reduced (APTR). This method is grounded in rigorous theoretical derivation and empirical validation, utilizing an untrained surrogate model as a reference. During training, the surrogate model and the target model are jointly trained using cross-entropy loss, allowing the target model to learn to maintain an orthogonal decision boundary relative to that of the surrogate model while preserving classification performance. This approach effectively cuts off adversarial transferability from unknown networks. Extensive experiments on CIFAR-10, SVHN, and DeepFake datasets demonstrate that APTR outperforms three state-of-the-art (SOTA) optimization baselines against seven black-box attacks. Notably, the results indicate that networks trained with a surrogate model of one architecture can also defend against black-box attacks from models of different architectures, where the APTR method exhibits black-box defense capabilities comparable to adversarial training without the trade-off problem of adversarial training.
{"title":"Decreasing adversarial transferability using gradient information of attack paths","authors":"Mengjun Xu , Lei Liu , Pengfei Xia , Ziqiang Li , Bin Li","doi":"10.1016/j.asoc.2025.112692","DOIUrl":"10.1016/j.asoc.2025.112692","url":null,"abstract":"<div><div>Adversarial transferability is an intriguing yet dangerous property of deep neural networks (DNNs), enabling the potential for black-box adversarial attacks. To better safeguard DNN-based AI models against such attacks, adversarial defense is explored through the lens of reducing the transferability of adversarial examples. From the perspective of decision boundaries in convolutional neural network classification models, we propose a novel method called <strong>A</strong>dversarial <strong>P</strong>ath <strong>T</strong>ransferability <strong>R</strong>educed (<strong>APTR</strong>). This method is grounded in rigorous theoretical derivation and empirical validation, utilizing an untrained surrogate model as a reference. During training, the surrogate model and the target model are jointly trained using cross-entropy loss, allowing the target model to learn to maintain an orthogonal decision boundary relative to that of the surrogate model while preserving classification performance. This approach effectively cuts off adversarial transferability from unknown networks. Extensive experiments on CIFAR-10, SVHN, and DeepFake datasets demonstrate that APTR outperforms three state-of-the-art (SOTA) optimization baselines against seven black-box attacks. Notably, the results indicate that networks trained with a surrogate model of one architecture can also defend against black-box attacks from models of different architectures, where the APTR method exhibits black-box defense capabilities comparable to adversarial training without the trade-off problem of adversarial training.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"170 ","pages":"Article 112692"},"PeriodicalIF":7.2,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143213468","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 : 2025-02-01DOI: 10.1016/j.asoc.2025.112704
Junyu Li , Changshi Liu , Kunxiang Yi , Lijun Fan , Zhang Wu
Current research on the electric vehicle routing problem (EVRP) predominantly focuses on customer characteristics or the diversity of charging mechanisms, while relatively insufficient attention is paid to the influence of energy interactions facilitated by vehicle-to-grid (V2G) technology on route planning. This study presents a novel approach to EVRP with charging/discharging based on time-of-use (TOU) electricity pricing and diverse charging stations. The proposed method enables electric vehicles to select charging stations for charging or discharging en route, depending on electricity price fluctuations, thus offering opportunities for cost reduction and profit enhancement in logistics distribution. A tailored adaptive non-dominated sorting genetic algorithm-Ⅱ (ANSGA-Ⅱ) is developed to address the problem, which integrates adaptive probability calculation, hybrid population generation, and neighborhood search operators. Testing on benchmark instances demonstrates that the proposed ANSGA-Ⅱ effectively addresses the problem, exhibiting strong convergence. The optimized routing allows vehicles to efficiently engage in vehicle-grid interactions, incentivized by TOU pricing, yielding significant profits for logistics companies, amounting to approximately 20.82 % of total logistics costs. This approach provides a new strategic avenue for optimizing logistics operations. Ultimately, sensitivity analysis elucidates the correlation among TOU electricity pricing, logistics costs, and discharging profits.
{"title":"An adaptive NSGA-Ⅱ for electric vehicle routing problem with charging/discharging based on time-of-use electricity pricing and diverse charging stations","authors":"Junyu Li , Changshi Liu , Kunxiang Yi , Lijun Fan , Zhang Wu","doi":"10.1016/j.asoc.2025.112704","DOIUrl":"10.1016/j.asoc.2025.112704","url":null,"abstract":"<div><div>Current research on the electric vehicle routing problem (EVRP) predominantly focuses on customer characteristics or the diversity of charging mechanisms, while relatively insufficient attention is paid to the influence of energy interactions facilitated by vehicle-to-grid (V2G) technology on route planning. This study presents a novel approach to EVRP with charging/discharging based on time-of-use (TOU) electricity pricing and diverse charging stations. The proposed method enables electric vehicles to select charging stations for charging or discharging en route, depending on electricity price fluctuations, thus offering opportunities for cost reduction and profit enhancement in logistics distribution. A tailored adaptive non-dominated sorting genetic algorithm-Ⅱ (ANSGA-Ⅱ) is developed to address the problem, which integrates adaptive probability calculation, hybrid population generation, and neighborhood search operators. Testing on benchmark instances demonstrates that the proposed ANSGA-Ⅱ effectively addresses the problem, exhibiting strong convergence. The optimized routing allows vehicles to efficiently engage in vehicle-grid interactions, incentivized by TOU pricing, yielding significant profits for logistics companies, amounting to approximately 20.82 % of total logistics costs. This approach provides a new strategic avenue for optimizing logistics operations. Ultimately, sensitivity analysis elucidates the correlation among TOU electricity pricing, logistics costs, and discharging profits.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"170 ","pages":"Article 112704"},"PeriodicalIF":7.2,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143213628","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 : 2025-02-01DOI: 10.1016/j.asoc.2024.112664
Xuemei Cao, Xiangkun Wang, Haoyang Liang, Bingjun Wei, Xin Yang
Feature selection is a widely used data preprocessing technique, but it still faces two major challenges: (1) data in open and dynamic environments may continually emerge unknown classes, and (2) the ever-growing scale of data. To address these challenges, this paper proposes a novel Open Continual Sampling (OCS) method that combines the advantages of continual learning and three-way sampling, aiming to discover unknown knowledge and transfer known knowledge. OCS can detect unknown classes by constructing a hypersphere knowledge base and sampling the most uncertain instances at each class decision boundary from the unknown data, thereby effectively reducing redundant sample computations. Based on OCS, we introduce a rapid feature selection framework (OCS-FS). Guided by the prior knowledge base, this framework rapidly calculates the importance of a small number of candidate features on representative samples, thereby incrementally selecting the optimal feature subset for the new data. After completing the learning process for the new period, the knowledge base is updated to reinforce old knowledge and integrate new knowledge. Extensive experiments on public benchmark datasets demonstrate that our method significantly outperforms existing state-of-the-art feature selection methods in both effectiveness and efficiency.
{"title":"Open continual sampling with hypersphere knowledge transfer for rapid feature selection","authors":"Xuemei Cao, Xiangkun Wang, Haoyang Liang, Bingjun Wei, Xin Yang","doi":"10.1016/j.asoc.2024.112664","DOIUrl":"10.1016/j.asoc.2024.112664","url":null,"abstract":"<div><div>Feature selection is a widely used data preprocessing technique, but it still faces two major challenges: (1) data in open and dynamic environments may continually emerge unknown classes, and (2) the ever-growing scale of data. To address these challenges, this paper proposes a novel Open Continual Sampling (OCS) method that combines the advantages of continual learning and three-way sampling, aiming to discover unknown knowledge and transfer known knowledge. OCS can detect unknown classes by constructing a hypersphere knowledge base and sampling the most uncertain instances at each class decision boundary from the unknown data, thereby effectively reducing redundant sample computations. Based on OCS, we introduce a rapid feature selection framework (OCS-FS). Guided by the prior knowledge base, this framework rapidly calculates the importance of a small number of candidate features on representative samples, thereby incrementally selecting the optimal feature subset for the new data. After completing the learning process for the new period, the knowledge base is updated to reinforce old knowledge and integrate new knowledge. Extensive experiments on public benchmark datasets demonstrate that our method significantly outperforms existing state-of-the-art feature selection methods in both effectiveness and efficiency.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"170 ","pages":"Article 112664"},"PeriodicalIF":7.2,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143212927","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 : 2025-02-01DOI: 10.1016/j.asoc.2024.112667
Paul Arévalo , Antonio Cano , Olena Fedoseienko , Francisco Jurado
The integration of microgrids into the bulk power system introduces inherent uncertainties that challenge conventional protection systems, encompassing factors such as low fault currents, operational modes, penetration levels of renewable sources, load variations, and network topology. These uncertainties significantly impact the overall reliability of the electrical system. In the event of a fault occurrence within or external to the microgrid, swift disconnection from the primary grid is imperative. This disconnection is facilitated through the immediate operation of a static switch positioned proximate to the common coupling point. Such rapid action is essential to mitigate potential damages and expedite the restoration of electrical services. To ensure the delivery of reliable and high-quality energy to end consumers while alleviating stress on the utility grid, this paper introduces a novel methodology for the efficient detection, classification, and localization of faults in a microgrid cluster connected to the external grid. The proposed system addresses diverse irregular conditions, including conventional faults, high-impedance faults, islanding scenarios, and adverse events, covering several zones within the microgrid cluster and the external electrical grid. The proposed approach is based on a fusion of the Taguchi methodology and the discrete Wavelet transform. This combination enables the optimization of convolutional neural network training using scalograms generated from the fault signals. The results demonstrate the model’s high performance, achieving 99.25 % accuracy in fault localization and 99.13 % in fault detection and classification, all within less than 10 ms. In comparison, traditional methods like support vector machine and decision trees require over 16 ms with lower accuracy, underscoring the superior speed and precision of the proposed approach.
{"title":"A data-driven approach to microgrid fault detection and classification using Taguchi-optimized CNNs and wavelet transform","authors":"Paul Arévalo , Antonio Cano , Olena Fedoseienko , Francisco Jurado","doi":"10.1016/j.asoc.2024.112667","DOIUrl":"10.1016/j.asoc.2024.112667","url":null,"abstract":"<div><div>The integration of microgrids into the bulk power system introduces inherent uncertainties that challenge conventional protection systems, encompassing factors such as low fault currents, operational modes, penetration levels of renewable sources, load variations, and network topology. These uncertainties significantly impact the overall reliability of the electrical system. In the event of a fault occurrence within or external to the microgrid, swift disconnection from the primary grid is imperative. This disconnection is facilitated through the immediate operation of a static switch positioned proximate to the common coupling point. Such rapid action is essential to mitigate potential damages and expedite the restoration of electrical services. To ensure the delivery of reliable and high-quality energy to end consumers while alleviating stress on the utility grid, this paper introduces a novel methodology for the efficient detection, classification, and localization of faults in a microgrid cluster connected to the external grid. The proposed system addresses diverse irregular conditions, including conventional faults, high-impedance faults, islanding scenarios, and adverse events, covering several zones within the microgrid cluster and the external electrical grid. The proposed approach is based on a fusion of the Taguchi methodology and the discrete Wavelet transform. This combination enables the optimization of convolutional neural network training using scalograms generated from the fault signals. The results demonstrate the model’s high performance, achieving 99.25 % accuracy in fault localization and 99.13 % in fault detection and classification, all within less than 10 ms. In comparison, traditional methods like support vector machine and decision trees require over 16 ms with lower accuracy, underscoring the superior speed and precision of the proposed approach.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"170 ","pages":"Article 112667"},"PeriodicalIF":7.2,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143212928","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 most important challenges in the optimization of real-time charging scheduling (CS) problems are (i) the need to model CS problems with a large number of decision variables for precise control, (ii) the increase in computational complexity with the high penetration of electric vehicles, and (iii) the lack of research on the stability and computation time of optimization algorithms on CS problems. In this paper, we design a real-time model and introduce the CS Benchmark Problems (CSBP) suite of twelve problems of four different types. Furthermore, a driver satisfaction model is introduced for the first time to analyse the impact of the results on user satisfaction. Best known solutions for all problems in CSBP are presented for the first time in this study. According to the statistical analysis results, the three competitive algorithms among 66 competitors in the optimization of CSs are LSHADE-CnEpSin, LSHADE-SPACMA and LRFDB-COA. Stability and computational complexity analyses revealed that LSHADE-SPACMA is the most successful algorithm for problems where consumers outnumber prosumer and LRFDB-COA is the most successful algorithm for problems where consumers equal or exceed prosumer. When the performance of the algorithms is evaluated regardless of the problem type, LSHADE-Spacma is the most stable algorithm with an overall success rate of 100 % on CSs. In addition, the average peak load shaving for the best known solutions of the algorithms with the highest success rate for each problem is calculated to be 94.84 %, and the average satisfaction score for all drivers is calculated to be 0.81.
{"title":"Metaheuristic search algorithms in real-time charge scheduling optimisation: A suite of benchmark problems and research on stability-analysis","authors":"Furkan Üstünsoy , H.Hüseyin Sayan , Hamdi Tolga Kahraman","doi":"10.1016/j.asoc.2025.112691","DOIUrl":"10.1016/j.asoc.2025.112691","url":null,"abstract":"<div><div>The most important challenges in the optimization of real-time charging scheduling (CS) problems are (i) the need to model CS problems with a large number of decision variables for precise control, (ii) the increase in computational complexity with the high penetration of electric vehicles, and (iii) the lack of research on the stability and computation time of optimization algorithms on CS problems. In this paper, we design a real-time model and introduce the CS Benchmark Problems (CSBP) suite of twelve problems of four different types. Furthermore, a driver satisfaction model is introduced for the first time to analyse the impact of the results on user satisfaction. Best known solutions for all problems in CSBP are presented for the first time in this study. According to the statistical analysis results, the three competitive algorithms among 66 competitors in the optimization of CSs are LSHADE-CnEpSin, LSHADE-SPACMA and LRFDB-COA. Stability and computational complexity analyses revealed that LSHADE-SPACMA is the most successful algorithm for problems where consumers outnumber prosumer and LRFDB-COA is the most successful algorithm for problems where consumers equal or exceed prosumer. When the performance of the algorithms is evaluated regardless of the problem type, LSHADE-Spacma is the most stable algorithm with an overall success rate of 100 % on CSs. In addition, the average peak load shaving for the best known solutions of the algorithms with the highest success rate for each problem is calculated to be 94.84 %, and the average satisfaction score for all drivers is calculated to be 0.81.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"170 ","pages":"Article 112691"},"PeriodicalIF":7.2,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143213079","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 : 2025-02-01DOI: 10.1016/j.asoc.2025.112698
Weishan Kong , Yanni Ju , Shiyuan Zhang , Jun Wang , Liwei Huang , Hong Qu
Traffic flow forecasting, which aims to predict future traffic patterns based on current conditions, is a crucial yet challenging task in intelligent transportation systems due to the complex spatial–temporal relationships involved. Existing methods often struggle to effectively capture these intricate spatial dependencies and temporal patterns. To address these limitations, we propose a graph enhanced spatial–temporal Transformer (GE-STT), which integrates a graph enhanced module and a spatial–temporal Transformer module for improved prediction accuracy. Specifically, the graph enhanced module combines a Graph Convolutional Network (GCN) with a Gated Recurrent Unit (GRU) to obtain enriched spatial–temporal features, and introduce the original traffic data as a correction term to deal with the errors in the enhancement process. The spatial–temporal Transformer then leverages these enhanced features for final prediction. Experimental results on four traffic datasets show that GE-STT achieves superior performance under various metrics. Compared with the best baseline in different datasets, the performance of GE-STT is improved by up to 8% under the MAE metric, highlighting its robustness and effectiveness in traffic flow forecasting tasks.
{"title":"Graph enhanced spatial–temporal transformer for traffic flow forecasting","authors":"Weishan Kong , Yanni Ju , Shiyuan Zhang , Jun Wang , Liwei Huang , Hong Qu","doi":"10.1016/j.asoc.2025.112698","DOIUrl":"10.1016/j.asoc.2025.112698","url":null,"abstract":"<div><div>Traffic flow forecasting, which aims to predict future traffic patterns based on current conditions, is a crucial yet challenging task in intelligent transportation systems due to the complex spatial–temporal relationships involved. Existing methods often struggle to effectively capture these intricate spatial dependencies and temporal patterns. To address these limitations, we propose a graph enhanced spatial–temporal Transformer (GE-STT), which integrates a graph enhanced module and a spatial–temporal Transformer module for improved prediction accuracy. Specifically, the graph enhanced module combines a Graph Convolutional Network (GCN) with a Gated Recurrent Unit (GRU) to obtain enriched spatial–temporal features, and introduce the original traffic data as a correction term to deal with the errors in the enhancement process. The spatial–temporal Transformer then leverages these enhanced features for final prediction. Experimental results on four traffic datasets show that GE-STT achieves superior performance under various metrics. Compared with the best baseline in different datasets, the performance of GE-STT is improved by up to 8% under the MAE metric, highlighting its robustness and effectiveness in traffic flow forecasting tasks.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"170 ","pages":"Article 112698"},"PeriodicalIF":7.2,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143213204","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 : 2025-02-01DOI: 10.1016/j.asoc.2024.112630
Jun Cai, Shitao Zhong, Wenjing Zhang, Chenfu Yi
Harmonic noise frequently arouses by the disturbances in industrial applications, which would be a great threat to the security, stability and service life of equipment in some large and critical facilities, especially in power systems. Therefore, finding a way to resist harmonic noise is highly important. The zeroing neural networks (ZNN) have lately gained exceptional success in solving time-varying problems (TVP) as a result of its efficiency. Inspired by the effectiveness of ZNN and the dynamic system model design principles in control theory, we initially develop a coupled anti-mixed noise ZNN (AMNZNN) model that can resist the combination of single harmonic and non-harmonic noise (e.g., random noise). Then, an extended AMZNN model is further designed to remove the combination of multi-harmonic noise and non-harmonic noise. Additionally, comparisons among original ZNN (OZNN), integration-enhanced ZNN (IEZNN), harmonic-noise-tolerant ZNN (HNTZNN) and the proposed AMNZNN for time-varying matrix inversion (TVMI) under the mixture of harmonic noise and random noise are experimented to demonstrate the proposed AMNZNN model’s superior ability in resisting mixed noise. Finally, by applying the proposed extended formalism to power systems and microphone arrays in denoising, the effectiveness of the proposed method to resist multi-harmonic and random noises is further verified in scientific applications.
{"title":"A coupled zeroing neural network for removing mixed noises in solving time-varying problems","authors":"Jun Cai, Shitao Zhong, Wenjing Zhang, Chenfu Yi","doi":"10.1016/j.asoc.2024.112630","DOIUrl":"10.1016/j.asoc.2024.112630","url":null,"abstract":"<div><div>Harmonic noise frequently arouses by the disturbances in industrial applications, which would be a great threat to the security, stability and service life of equipment in some large and critical facilities, especially in power systems. Therefore, finding a way to resist harmonic noise is highly important. The zeroing neural networks (ZNN) have lately gained exceptional success in solving time-varying problems (TVP) as a result of its efficiency. Inspired by the effectiveness of ZNN and the dynamic system model design principles in control theory, we initially develop a coupled anti-mixed noise ZNN (AMNZNN) model that can resist the combination of single harmonic and non-harmonic noise (e.g., random noise). Then, an extended AMZNN model is further designed to remove the combination of multi-harmonic noise and non-harmonic noise. Additionally, comparisons among original ZNN (OZNN), integration-enhanced ZNN (IEZNN), harmonic-noise-tolerant ZNN (HNTZNN) and the proposed AMNZNN for time-varying matrix inversion (TVMI) under the mixture of harmonic noise and random noise are experimented to demonstrate the proposed AMNZNN model’s superior ability in resisting mixed noise. Finally, by applying the proposed extended formalism to power systems and microphone arrays in denoising, the effectiveness of the proposed method to resist multi-harmonic and random noises is further verified in scientific applications.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"170 ","pages":"Article 112630"},"PeriodicalIF":7.2,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143213268","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 : 2025-02-01DOI: 10.1016/j.asoc.2024.112640
Di Wang , Yuanming Lu , Xiangli Yang , Die Liu , Xianyi Yang , Jianxi Yang
Structural health monitoring (SHM) technology has been widely used in civil engineering, and vibration-based damage detection (VBDD) technology is an important component of SHM research. With the advancement of deep learning, a plethora of deep learning-based algorithms have been applied to VBDD. The accuracy of VBDD is constantly improving with the assistance of various deep learning techniques. However, studies on the efficiency of VBDD tasks based on neural network are still relatively few, and lightweight network technology has been proven to be an effective way to improve efficiency of neural network. In this paper, a novel neural network based on reparameterization is presented, which can decouple the model training and deployment, and maintain high accuracy under the consideration of model inference speed. Specifically, a convolutional neural network with multiple 1 × 1 convolution is used in the training, and all layers of convolution are fused during testing and inference of the model to obtain a VGG-style network with a lighter structure and higher accuracy for deployment. Experiments on benchmark datasets from IASC-ASCE and the Z24 dataset show that the proposed method can make VBDD work better.
{"title":"Enhancing inference speed in reparameterized convolutional neural network for vibration-based damage detection","authors":"Di Wang , Yuanming Lu , Xiangli Yang , Die Liu , Xianyi Yang , Jianxi Yang","doi":"10.1016/j.asoc.2024.112640","DOIUrl":"10.1016/j.asoc.2024.112640","url":null,"abstract":"<div><div>Structural health monitoring (SHM) technology has been widely used in civil engineering, and vibration-based damage detection (VBDD) technology is an important component of SHM research. With the advancement of deep learning, a plethora of deep learning-based algorithms have been applied to VBDD. The accuracy of VBDD is constantly improving with the assistance of various deep learning techniques. However, studies on the efficiency of VBDD tasks based on neural network are still relatively few, and lightweight network technology has been proven to be an effective way to improve efficiency of neural network. In this paper, a novel neural network based on reparameterization is presented, which can decouple the model training and deployment, and maintain high accuracy under the consideration of model inference speed. Specifically, a convolutional neural network with multiple 1 × 1 convolution is used in the training, and all layers of convolution are fused during testing and inference of the model to obtain a VGG-style network with a lighter structure and higher accuracy for deployment. Experiments on benchmark datasets from IASC-ASCE and the Z24 dataset show that the proposed method can make VBDD work better.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"170 ","pages":"Article 112640"},"PeriodicalIF":7.2,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143213395","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