CMAC神经网络在电动伺服摩擦补偿中的应用

基于CMAC神经网络的电动伺服摩擦补偿控制器用微分方程描述被控对象并建立其模型.控制器以系统动态误差作为CMAC的输入,用CMAC的输出与系统总输入之差调整权重.学习中,CMAC初始状态权重值为0,将误差期望值与系统当前误差量化后作为地址输入CMAC.计算CMAC的输出,然后与控制器输出相加得到系统总控制输入并进行控制.实验表明经在线学习补偿被控对象的非线性,使系统具有较强的自适应和鲁棒性.

Application of CMAC Neural Network in Friction Compensation of Electric Servo System

Controlled object was described and set up object model with differential equation in electric servo friction compensation controller based on CMAC neural network. System error is used as input of CMAC, and weighting is adjusted with the difference between output of CMAC and total input of system. In learning, the primary value of CMAC was zero. Expected value of error and current error of system was quantized and used it as address to send in CMAC. Output of CMAC was calculated, then, output of CMAC was added to output of controller, total input of system control was gained and controlled it. The test result shows that online learning can make nonlinearity compensation for controlled object, so as to system has better self-adoption and robustness.