任务间共享和特有结构分解的多任务TSK模糊系统建模  

Multi-task TSK fuzzy system modeling based on inter-task common and special structure decomposition

作　　者：赵壮壮 王骏 潘祥 邓赵红 施俊[2] 王士同 ZHAO Zhuangzhuang;WANG Jun;PAN Xiang;DENG Zhaohong;SHI Jun;WANG Shitong(School of Artificial Intelligence and Computer Science,Jiangnan University,Wuxi 214122,China;School of Communication&Information Engineering,Shanghai University,Shanghai 200444,China)

机构地区：[1]江南大学人工智能与计算机学院,江苏无锡214122 [2]上海大学通信与信息工程学院,上海200444

出　　处：《智能系统学报》2021年第4期622-629,共8页CAAI Transactions on Intelligent Systems

基　　金：江苏省自然科学基金项目(BK20181339);国家自然科学基金项目(61602007);中央高校基础研究经费资助项目(JUSRP11851).

摘　　要：现有的多任务Takagi-Sugeno-Kang(TSK)模糊建模方法更注重利用任务间的相关性信息,而忽略了单个任务的特殊性。针对此问题,本文提出了一种考虑所有任务之间的共享结构和特有结构的TSK模糊系统多任务建模新方法。该方法将后件参数分解为共享参数和特有参数两个分量,既充分利用了任务间共享信息,又有效地保留了单个任务的特性。最后,本文利用增广拉格朗日乘子法(ALM)求解该最优化问题。实验结果表明,该方法比现有的模型获得了更好的表现。Existing Takagi–Sugeno–Kang(TSK)fuzzy system modeling methods pay more attention to the inter-task correlation but ignore the particularity of every single task.To address this issue,this paper proposes a novel multi-task modeling method for TSK fuzzy systems taking common and specific structures across all tasks(MTTSKFS-CS)into consideration.This method decomposes consequent parameters into shared and special ones,which not only takes advantage of the shared information among tasks but also effectively preserves the characteristics of individual tasks.Finally,the study uses the augmented Lagrange multiplier for optimization.The experimental results demonstrate the better performance of the proposed model compared with other existing methods.

关 键 词：TSK模糊系统 非线性 多任务 低秩 稀疏 参数分解 泛化性能 可解释性 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]