神经网络磁链估计的感应电机反步法研究  

Study on Induction Motor Backstepping Method Based on Neural Network Flux Estimator

作　　者：沈艳霞[1] 林瑾[1] 纪志成[1] 

机构地区：[1]江南大学控制科学与工程研究中心,江苏无锡214122

出　　处：《控制与决策》2006年第7期833-836,共4页Control and Decision

基　　金：江苏省自然科学基金项目(BK2004021)

摘　　要：为实现感应电机的位置渐近跟踪,基于反步法并取转矩和磁链控制信号作为虚拟控制,设计了感应电机位置控制系统.采用多层前馈神经网络估计转子磁链,以L evenberg-M arquardt算法训练网络并调整权值.最后基于Lyapunov稳定性理论证明了系统的稳定性.仿真结果表明,所设计的神经网络磁链观测器具有良好的估计效果,位置跟踪误差迅速收敛,具有较优的伺服跟踪特性.In order to implement position asymptotic tracking, based on the backstepping method, a position control system of induction motor is designed using the flux and torque control signals as fictitious signals. A multi-layer feedforward neural network is designed to estimate the rotor flux. Levenberg-Marquardt optimum algorithm is used to train the network and adjust the weights. The stability of this system is proved by Lyapunov stable theory. The simulation results show that, the proposed neural network flux observer has good estimation effect, the position tracking error converges quickly and the system can obtain better servo performance.

关 键 词：感应电机 神经网络 反步法 LEVENBERG-MARQUARDT算法 

分 类 号：TM34[电气工程—电机]