Pub Date : 2026-01-15Epub Date: 2025-12-18DOI: 10.1016/j.jfranklin.2025.108359
Yuyang Zhao , Dawei Gong , Jiaoyuan Chen , Shijie Song , Minglei Zhu
This study presents a novel data-driven fault-tolerant control (FTC) algorithm for addressing the output containment control problem in multi-agent systems (MASs) with unknown dynamics and actuator faults. To overcome the modeling challenges associated with such systems, a dynamic linearization technique is employed to construct an online model using only real-time input-output data, thereby eliminating the need for prior knowledge of agent dynamics. Building on this framework, an adaptive fault estimator is developed to identify actuator fault information based solely on measurable input-output data. The estimated fault signals are then integrated into the control law to compensate for actuator failures and ensure robust system performance. A rigorous theoretical analysis proves that both the containment error of the MASs and the estimation error of the adaptive fault estimator remain bounded, guaranteeing system stability. Finally, three simulation studies, including two numerical examples and one real-system experiment, are conducted to validate the proposed method, demonstrating its effectiveness, robustness, and practical feasibility.
{"title":"Data-driven output containment fault-tolerant control of unknown multi-agent systems with actuator faults","authors":"Yuyang Zhao , Dawei Gong , Jiaoyuan Chen , Shijie Song , Minglei Zhu","doi":"10.1016/j.jfranklin.2025.108359","DOIUrl":"10.1016/j.jfranklin.2025.108359","url":null,"abstract":"<div><div>This study presents a novel data-driven fault-tolerant control (FTC) algorithm for addressing the output containment control problem in multi-agent systems (MASs) with unknown dynamics and actuator faults. To overcome the modeling challenges associated with such systems, a dynamic linearization technique is employed to construct an online model using only real-time input-output data, thereby eliminating the need for prior knowledge of agent dynamics. Building on this framework, an adaptive fault estimator is developed to identify actuator fault information based solely on measurable input-output data. The estimated fault signals are then integrated into the control law to compensate for actuator failures and ensure robust system performance. A rigorous theoretical analysis proves that both the containment error of the MASs and the estimation error of the adaptive fault estimator remain bounded, guaranteeing system stability. Finally, three simulation studies, including two numerical examples and one real-system experiment, are conducted to validate the proposed method, demonstrating its effectiveness, robustness, and practical feasibility.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"363 2","pages":"Article 108359"},"PeriodicalIF":4.2,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145837884","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Medical report generation is a currently popular research direction that combines image analysis and natural language processing technologies. It aims to relieve the pressure on doctors in writing reports and provide a diagnostic reference for them. Although current report generation technologies have made significant progress, there are still some problems. For example, the extraction of local lesion areas in medical images is insufficient, cross-modal alignment is difficult, and the extracted visual features are single-angled. Considering these issues, we propose the Hierarchical Multi-scale Visual Feature Flow (HMVF) model for medical report generation. Firstly, we introduce three feature extraction branches from different angles: disease label features, lung border features, and local lesion information features. This endows the model with the ability to focus on features from different perspectives of the image. To our knowledge, this is the first time that lung border features have been introduced into medical report generation for the MIMIC-CXR dataset. Then, the contents extracted from each branch are successively transmitted to the subsequent branches to achieve information interaction. At the same time, according to the characteristics of these visual features from different angles, we enhance the visual features of each branch respectively to prevent irrelevant information from diluting key information. Finally, we transmit the information at each level to different layers of the cross-modal information fusion module to generate the final text report. Extensive experiments on authoritative datasets demonstrate that our method outperforms others across multiple language evaluation metrics. On the IU-Xray dataset, our model scored 0.512, 0.365, 0.181, and 0.410 for BLEU-1/2/4, and ROUGE respectively. On the MIMIC-CXR dataset, scores were 0.391, 0.239, 0.161, and 0.115 for BLEU-1/2/3/4. These results confirm that the multi-branch feature extraction and hierarchical cross-modal fusion of HMVF effectively address the limitations of existing methods, providing a more robust solution for automated medical report generation. The model’s design also offers a reference for multi-perspective feature utilization in other cross-modal generation tasks. Source code is available here.
{"title":"Bridging Visual Analysis and Text Generation: A Hierarchical Multi-scale Visual Feature Flow Model for Accessible Radiographic Report Automation","authors":"Hailong Zuo , Zhi Weng , Yunlu Duan , Zijing Lv , Feifan Bi , Zhiqiang Zheng","doi":"10.1016/j.jfranklin.2025.108328","DOIUrl":"10.1016/j.jfranklin.2025.108328","url":null,"abstract":"<div><div>Medical report generation is a currently popular research direction that combines image analysis and natural language processing technologies. It aims to relieve the pressure on doctors in writing reports and provide a diagnostic reference for them. Although current report generation technologies have made significant progress, there are still some problems. For example, the extraction of local lesion areas in medical images is insufficient, cross-modal alignment is difficult, and the extracted visual features are single-angled. Considering these issues, we propose the Hierarchical Multi-scale Visual Feature Flow (HMVF) model for medical report generation. Firstly, we introduce three feature extraction branches from different angles: disease label features, lung border features, and local lesion information features. This endows the model with the ability to focus on features from different perspectives of the image. To our knowledge, this is the first time that lung border features have been introduced into medical report generation for the MIMIC-CXR dataset. Then, the contents extracted from each branch are successively transmitted to the subsequent branches to achieve information interaction. At the same time, according to the characteristics of these visual features from different angles, we enhance the visual features of each branch respectively to prevent irrelevant information from diluting key information. Finally, we transmit the information at each level to different layers of the cross-modal information fusion module to generate the final text report. Extensive experiments on authoritative datasets demonstrate that our method outperforms others across multiple language evaluation metrics. On the IU-Xray dataset, our model scored 0.512, 0.365, 0.181, and 0.410 for BLEU-1/2/4, and ROUGE respectively. On the MIMIC-CXR dataset, scores were 0.391, 0.239, 0.161, and 0.115 for BLEU-1/2/3/4. These results confirm that the multi-branch feature extraction and hierarchical cross-modal fusion of HMVF effectively address the limitations of existing methods, providing a more robust solution for automated medical report generation. The model’s design also offers a reference for multi-perspective feature utilization in other cross-modal generation tasks. Source code is available <span><span>here</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"363 2","pages":"Article 108328"},"PeriodicalIF":4.2,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145837955","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2026-01-15Epub Date: 2025-12-17DOI: 10.1016/j.jfranklin.2025.108369
L.J. Alvarez-Vázquez , A. Martínez , M.E. Vázquez-Méndez
We present here a novel mathematical approach to deal with the optimal performance of a raceway pond (an open-channel pond where the circulating wastewater, during its purification process, is used to grow algae that will be used as a source for the production of bioenergy). The maximization of algal productivity is addressed here within an optimal control framework for partial differential equations. Thus, after formulating the real-world control problem in a rigorously mathematical way, we prove the existence of optimal solutions, we propose a full computational algorithm for its numerical solution and, finally, we present several results for the numerical optimization of a realistic case.
{"title":"Optimizing algaculture in open-channel raceway ponds for the production of bioenergy","authors":"L.J. Alvarez-Vázquez , A. Martínez , M.E. Vázquez-Méndez","doi":"10.1016/j.jfranklin.2025.108369","DOIUrl":"10.1016/j.jfranklin.2025.108369","url":null,"abstract":"<div><div>We present here a novel mathematical approach to deal with the optimal performance of a raceway pond (an open-channel pond where the circulating wastewater, during its purification process, is used to grow algae that will be used as a source for the production of bioenergy). The maximization of algal productivity is addressed here within an optimal control framework for partial differential equations. Thus, after formulating the real-world control problem in a rigorously mathematical way, we prove the existence of optimal solutions, we propose a full computational algorithm for its numerical solution and, finally, we present several results for the numerical optimization of a realistic case.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"363 2","pages":"Article 108369"},"PeriodicalIF":4.2,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145837948","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2026-01-15Epub Date: 2025-12-20DOI: 10.1016/j.jfranklin.2025.108357
Fanyueyang Zhang , Jun-e Feng , Yuanhua Wang
The finite multi-objective potential game (FMOPG) with different objective sets is proposed in this paper. Leveraging the semi-tensor product of matrices, two necessary and sufficient conditions for verifying FMOPGs have been presented, which provide the theoretical foundation for the research on potential-based multi-objective distributed optimization. In addition, the Pareto equilibrium in this class of games is examined by establishing two equivalent verification criteria. Significantly, the relationship between Pareto equilibrium and potential functions is further characterized, which demonstrates that this model can exhibit richer equilibrium behaviors and model more complex interaction patterns compared to FMOPGs with a common objective set for all players. Finally, taking multi-objective facility-based systems as an example, the applicability of the results is validated.
{"title":"Multi-objective potential games: model, equilibrium, and applications","authors":"Fanyueyang Zhang , Jun-e Feng , Yuanhua Wang","doi":"10.1016/j.jfranklin.2025.108357","DOIUrl":"10.1016/j.jfranklin.2025.108357","url":null,"abstract":"<div><div>The finite multi-objective potential game (FMOPG) with different objective sets is proposed in this paper. Leveraging the semi-tensor product of matrices, two necessary and sufficient conditions for verifying FMOPGs have been presented, which provide the theoretical foundation for the research on potential-based multi-objective distributed optimization. In addition, the Pareto equilibrium in this class of games is examined by establishing two equivalent verification criteria. Significantly, the relationship between Pareto equilibrium and potential functions is further characterized, which demonstrates that this model can exhibit richer equilibrium behaviors and model more complex interaction patterns compared to FMOPGs with a common objective set for all players. Finally, taking multi-objective facility-based systems as an example, the applicability of the results is validated.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"363 2","pages":"Article 108357"},"PeriodicalIF":4.2,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145838022","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2026-01-15Epub Date: 2026-01-02DOI: 10.1016/j.jfranklin.2025.108381
Yu Wang, Huiliao Yang, Dawei Wu, Yuquan Chen
This article presents a control strategy for quadrotor unmanned aerial vehicle (UAV) to achieve autonomous landing on moving unmanned surface vehicle (USV). First, to address the challenge of safe landing due to USV deck oscillations induced by wind and waves, the Long Short-Term Memory (LSTM) network is employed to predict motion trajectories and identify optimal landing window. Next, a modified practically predefined time stable (PPTS) system is established as theoretical basis of subsequent control design. Subsequently, by incorporating neural networks into the adaptive command-filtered backstepping control framework, the system’s unknown dynamics terms (UDT) and differential explosion phenomena are effectively addressed. Stability analysis demonstrates that the closed-loop system is practically predefined-time stable (PPTS), with tracking errors converging to a small residual set within the predefined time. Finally, simulation results verify the effectiveness and superiority of the proposed method.
{"title":"Autonomous shipboard landing of quadrotor UAV: A neuroadaptive predefined-time approach with LSTM prediction","authors":"Yu Wang, Huiliao Yang, Dawei Wu, Yuquan Chen","doi":"10.1016/j.jfranklin.2025.108381","DOIUrl":"10.1016/j.jfranklin.2025.108381","url":null,"abstract":"<div><div>This article presents a control strategy for quadrotor unmanned aerial vehicle (UAV) to achieve autonomous landing on moving unmanned surface vehicle (USV). First, to address the challenge of safe landing due to USV deck oscillations induced by wind and waves, the Long Short-Term Memory (LSTM) network is employed to predict motion trajectories and identify optimal landing window. Next, a modified practically predefined time stable (PPTS) system is established as theoretical basis of subsequent control design. Subsequently, by incorporating neural networks into the adaptive command-filtered backstepping control framework, the system’s unknown dynamics terms (UDT) and differential explosion phenomena are effectively addressed. Stability analysis demonstrates that the closed-loop system is practically predefined-time stable (PPTS), with tracking errors converging to a small residual set within the predefined time. Finally, simulation results verify the effectiveness and superiority of the proposed method.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"363 2","pages":"Article 108381"},"PeriodicalIF":4.2,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926861","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2026-01-15Epub Date: 2025-12-14DOI: 10.1016/j.jfranklin.2025.108327
Jumei Yue, Xiaoyang Xu, Yongyi Yan, Zengqiang Chen, Xiaona Song
Traditional methods of constructing a parallel network of finite state machines (PNoFSMs) fail to reveal the logical relation between the dynamics of the entire network and the dynamics of the component finite state machines (FSMs). This paper approaches the problem from an algebraic perspective and, based on the dynamic behavior of FSMs, classifies PNoFSMs into Type I and Type II parallel networks. Building on the bilinear algebraic model of FSMs established in recent years, the concept of structure matrix operator is introduced and extended into expanded structural matrix. On this basis, the ideas of "local construction" and "global construction" are introduced, and algebraic construction formulas for Type I and Type II PNoFSMs are established. The results show that the formulated method significantly simplifies the construction process. More importantly, it algebraically clarifies the logical dynamic relation between the parallel network and its component FSMs. Finally, the method is further extended to the case of non-deterministic FSMs, and an algebraic construction method for non-deterministic PNoFSMs is developed.
{"title":"An algebraic mechanism for constructing parallel network of finite state machines","authors":"Jumei Yue, Xiaoyang Xu, Yongyi Yan, Zengqiang Chen, Xiaona Song","doi":"10.1016/j.jfranklin.2025.108327","DOIUrl":"10.1016/j.jfranklin.2025.108327","url":null,"abstract":"<div><div>Traditional methods of constructing a parallel network of finite state machines (PNoFSMs) fail to reveal the logical relation between the dynamics of the entire network and the dynamics of the component finite state machines (FSMs). This paper approaches the problem from an algebraic perspective and, based on the dynamic behavior of FSMs, classifies PNoFSMs into Type I and Type II parallel networks. Building on the bilinear algebraic model of FSMs established in recent years, the concept of structure matrix operator is introduced and extended into expanded structural matrix. On this basis, the ideas of \"local construction\" and \"global construction\" are introduced, and algebraic construction formulas for Type I and Type II PNoFSMs are established. The results show that the formulated method significantly simplifies the construction process. More importantly, it algebraically clarifies the logical dynamic relation between the parallel network and its component FSMs. Finally, the method is further extended to the case of non-deterministic FSMs, and an algebraic construction method for non-deterministic PNoFSMs is developed.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"363 2","pages":"Article 108327"},"PeriodicalIF":4.2,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926860","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2026-01-15Epub Date: 2025-12-29DOI: 10.1016/j.jfranklin.2025.108377
Yonghong Lan, Shikang Zheng
Model-free adaptive control (MFAC) can provide advantages such as lower operational costs, higher scalability and easier implementation. However, how to further improve the controller performance and optimize the controller parameters is still an open problem. In this paper, a high-order pseudo-partial derivative estimation-based improved fractional-order model-free adaptive control (HOPPD-IFOMFAC) scheme and linear matrix inequality (LMI)-based parameters optimization method are proposed for a class of single-input single-output discrete-time nonlinear systems. Firstly, a compact form of fractional-order dynamic linearization model is used to equivalently describe the nonlinear system, which relates the variation of the output signal with the fractional-order variation of the input one. Considering more information of the previous time, a high-order pseudo-partial derivative parameter estimator algorithm is designed. Then, with the introduction of the observer, a control input criterion function, which includes the observer tracking error and the rate of its change is presented. On the base of this, the HOPPD-IFOMFAC scheme is derived, which not only utilizes more control knowledge of the previous time in the control law, but also uses more information of the previous time in the estimation algorithm, which can effectively improve control performance. Furthermore, some of the controller parameters are optimized by using the LMI convex optimization technique. Finally, two numerical examples demonstrate that the proposed HOPPD-IFOMFAC scheme can track the desired trajectory with improved convergence and tracking performance.
{"title":"Improved fractional model-free adaptive control and parameters optimization for a class of nonlinear discrete systems","authors":"Yonghong Lan, Shikang Zheng","doi":"10.1016/j.jfranklin.2025.108377","DOIUrl":"10.1016/j.jfranklin.2025.108377","url":null,"abstract":"<div><div>Model-free adaptive control (MFAC) can provide advantages such as lower operational costs, higher scalability and easier implementation. However, how to further improve the controller performance and optimize the controller parameters is still an open problem. In this paper, a high-order pseudo-partial derivative estimation-based improved fractional-order model-free adaptive control (HOPPD-IFOMFAC) scheme and linear matrix inequality (LMI)-based parameters optimization method are proposed for a class of single-input single-output discrete-time nonlinear systems. Firstly, a compact form of fractional-order dynamic linearization model is used to equivalently describe the nonlinear system, which relates the variation of the output signal with the fractional-order variation of the input one. Considering more information of the previous time, a high-order pseudo-partial derivative parameter estimator algorithm is designed. Then, with the introduction of the observer, a control input criterion function, which includes the observer tracking error and the rate of its change is presented. On the base of this, the HOPPD-IFOMFAC scheme is derived, which not only utilizes more control knowledge of the previous time in the control law, but also uses more information of the previous time in the estimation algorithm, which can effectively improve control performance. Furthermore, some of the controller parameters are optimized by using the LMI convex optimization technique. Finally, two numerical examples demonstrate that the proposed HOPPD-IFOMFAC scheme can track the desired trajectory with improved convergence and tracking performance.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"363 2","pages":"Article 108377"},"PeriodicalIF":4.2,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145881158","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2026-01-15Epub Date: 2025-12-25DOI: 10.1016/j.jfranklin.2025.108378
Linqing Du, Jun Fu, Huan Li
A hybrid intelligent optimization method by improving particle swarm optimization and genetic algorithm (PSO-GA) is proposed for solving multi-objective optimal control problem (MOOCP) governed by ordinary differential equations (ODEs) in nonlinear systems. This method combines the global search capability of heuristic algorithm and the rapid convergence of deterministic algorithm to generate approximately uniform Pareto front. Firstly, the MOOCP is converted into a single-objective optimal control problem (SOOCP) subject to inequality point constraints using an adaptive ε-constraint method [1]. Secondly, the enhanced PSO is combined with a two-criteria GA to locate the global optimal region of SOOCP. Thirdly, sequential quadratic programming (SQP) algorithm is applied in the global optimal region to achieve fast convergence to high local accuracy solutions satisfying optimality conditions. Finally, the effectiveness and superiority of the proposed method are ultimately validated through numerical simulations and comparative analyses against state-of-the-art algorithms. Specifically, the hybrid intelligent optimization algorithm shows a 7.65% improvement in hypervolume and a 46.60% reduction in spacing metric compared to the suboptimal algorithm.
{"title":"Hybrid intelligent multi-objective optimal control of ODE-constrained nonlinear systems","authors":"Linqing Du, Jun Fu, Huan Li","doi":"10.1016/j.jfranklin.2025.108378","DOIUrl":"10.1016/j.jfranklin.2025.108378","url":null,"abstract":"<div><div>A hybrid intelligent optimization method by improving particle swarm optimization and genetic algorithm (PSO-GA) is proposed for solving multi-objective optimal control problem (MOOCP) governed by ordinary differential equations (ODEs) in nonlinear systems. This method combines the global search capability of heuristic algorithm and the rapid convergence of deterministic algorithm to generate approximately uniform Pareto front. Firstly, the MOOCP is converted into a single-objective optimal control problem (SOOCP) subject to inequality point constraints using an adaptive ε-constraint method [1]. Secondly, the enhanced PSO is combined with a two-criteria GA to locate the global optimal region of SOOCP. Thirdly, sequential quadratic programming (SQP) algorithm is applied in the global optimal region to achieve fast convergence to high local accuracy solutions satisfying optimality conditions. Finally, the effectiveness and superiority of the proposed method are ultimately validated through numerical simulations and comparative analyses against state-of-the-art algorithms. Specifically, the hybrid intelligent optimization algorithm shows a 7.65% improvement in hypervolume and a 46.60% reduction in spacing metric compared to the suboptimal algorithm.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"363 2","pages":"Article 108378"},"PeriodicalIF":4.2,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145881157","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2026-01-15Epub Date: 2025-12-11DOI: 10.1016/j.jfranklin.2025.108284
Rui Xu , Hua Chen , Jinhai Zheng , Jisheng Zhang , Mian Xue , Xinyuan Long , Yao Tang
A model-free reinforcement learning-based iterative learning consensus control (RL-ILC) is presented for multiple unmanned surface vessels (MUSVs) under input time-varying delay and disturbances in this paper. Firstly, the MUSVs system and the nonlinear ideal learning controller are transformed into compact-form dynamic linearization (CFDL) data models and the partial form dynamic linearization (PFDL) data models in the iteration domain. Further, generalized regression neural networks (GRNNs) are applied to evaluate the unknown pseudo partial derivative (PPD) parameters of the controlled plant and disturbances, and actor-critic neural networks (ACNNs) in reinforcement learning (RL) are utilized to approximate the unknown pseudo gradient (PG) of the ideal learning controller system. The iterative learning controller, which is based on reinforcement learning and integrates neural networks, is designed to use only the information of unmanned surface vessels (USV) and its neighbors. As a result, it is proven that the consensus error and tracking error of the MUSVs with input time-varying delay are bounded. Finally, simulation results verify the efficiency of the presented control design.
{"title":"Reinforcement learning-based iterative learning consensus control for multiple unmanned surface vessels with time-varying delay","authors":"Rui Xu , Hua Chen , Jinhai Zheng , Jisheng Zhang , Mian Xue , Xinyuan Long , Yao Tang","doi":"10.1016/j.jfranklin.2025.108284","DOIUrl":"10.1016/j.jfranklin.2025.108284","url":null,"abstract":"<div><div>A model-free reinforcement learning-based iterative learning consensus control (RL-ILC) is presented for multiple unmanned surface vessels (MUSVs) under input time-varying delay and disturbances in this paper. Firstly, the MUSVs system and the nonlinear ideal learning controller are transformed into compact-form dynamic linearization (CFDL) data models and the partial form dynamic linearization (PFDL) data models in the iteration domain. Further, generalized regression neural networks (GRNNs) are applied to evaluate the unknown pseudo partial derivative (PPD) parameters of the controlled plant and disturbances, and actor-critic neural networks (ACNNs) in reinforcement learning (RL) are utilized to approximate the unknown pseudo gradient (PG) of the ideal learning controller system. The iterative learning controller, which is based on reinforcement learning and integrates neural networks, is designed to use only the information of unmanned surface vessels (USV) and its neighbors. As a result, it is proven that the consensus error and tracking error of the MUSVs with input time-varying delay are bounded. Finally, simulation results verify the efficiency of the presented control design.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"363 2","pages":"Article 108284"},"PeriodicalIF":4.2,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145798751","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2026-01-15Epub Date: 2025-12-18DOI: 10.1016/j.jfranklin.2025.108320
Huiqin Pei, Hongli Xiao
This paper investigates fault-tolerant group consensus control for hybrid multi-agent systems (HMASs) by event-triggered strategy subject to actuator faults. Firstly, to estimate the states of the agents and the actuator faults, an observer and a fault estimator are constructed using relative output estimation errors, thereby establishing the asymptotic stability of the global estimation error system. Secondly, under the event-triggered strategy, a class of fault-tolerant group consensus protocol for HMASs is proposed to compensate for actuator faults, the grouping coefficient in the protocol is used to distinguish different subgroups of agents. Based on the Lyapunov stability theorem, the stability of the closed-loop system and the convergence of consensus errors are rigorously proved. Furthermore, Zeno behavior is excluded under the proposed triggering strategy. Finally, numerical simulations are conducted to demonstrate the effectiveness of the proposed event-triggered fault-tolerant protocol.
{"title":"Fault-tolerant group consensus control for hybrid multi-agent systems by observers and event-triggered strategy","authors":"Huiqin Pei, Hongli Xiao","doi":"10.1016/j.jfranklin.2025.108320","DOIUrl":"10.1016/j.jfranklin.2025.108320","url":null,"abstract":"<div><div>This paper investigates fault-tolerant group consensus control for hybrid multi-agent systems (HMASs) by event-triggered strategy subject to actuator faults. Firstly, to estimate the states of the agents and the actuator faults, an observer and a fault estimator are constructed using relative output estimation errors, thereby establishing the asymptotic stability of the global estimation error system. Secondly, under the event-triggered strategy, a class of fault-tolerant group consensus protocol for HMASs is proposed to compensate for actuator faults, the grouping coefficient in the protocol is used to distinguish different subgroups of agents. Based on the Lyapunov stability theorem, the stability of the closed-loop system and the convergence of consensus errors are rigorously proved. Furthermore, Zeno behavior is excluded under the proposed triggering strategy. Finally, numerical simulations are conducted to demonstrate the effectiveness of the proposed event-triggered fault-tolerant protocol.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"363 2","pages":"Article 108320"},"PeriodicalIF":4.2,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145837881","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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