基于神经网络的增强型自适应滑模控制策略研究  

An Enhanced Adaptive Sliding Mode Control Strategy Based on Neural Network

作　　者：万宇宾[1,2] 胡婵娟[1,2] 赵金[1] 王永骥[1,2] 万淑芸[1] 

机构地区：[1]华中科技大学控制科学与工程系,武汉430074 [2]南车株洲电力机车研究所有限公司,株洲412001

出　　处：《微电机》2011年第3期51-57,共7页Micromotors

摘　　要：针对存在显著未知惯量动态的感应电机伺服系统鲁棒跟踪控制问题,提出一种基于神经网络的增强型自适应滑模控制(EASMC)策略,根据实时控制的需要设计了可灵活配置的通用型三层前馈神经网络,并采用结构化补偿方式以充分利用其描述能力;以权值伪边界估计为基础,将不连续投影修正引入权值自适应律以实现权值估计误差有界;构造了基于改进型边界估计方法的自适应开关控制用于补偿包含重建误差、泰勒序列高阶尾项、外部扰动等在内的综合等价扰动项。仿真结果表明,该文提出的控制策略能较好地实现对未知惯量动态的拟合和补偿,有效改善了伺服系统的跟踪性能。This paper studied the precision robust tracking control of induction motor servo systems which were subject to significant unknown inertia dynamic,then proposed a novel enhanced adaptive sliding mode control（EASMC） based on neural networks.A universal 3-layer feedforward neural network topology was designed to achieve better approximation in realtime applications,and the control scheme using structural compensation was established.The boundness of estimation error of all weights was guaranteed by using the modified adaption laws based on discontinuous projection,moreover,designed an improved adaptive switching control to confront the lumped equivalent disturbance,which was composed of reconstruction error,higher-order-tails of Taylor expansion and external disturbances.Simulation results show that the proposed method can approximate and compensate for inertia dynamic,which improves the performance of servo systems effectively.

关 键 词：自适应滑模控制 神经网络 惯量动态 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]