基于小波神经网络的板形板厚综合系统逆控制

Strip flatness and gauge inverse control based on wavelet neural networks

针对板形板厚综合系统具有强耦合、非线性、含纯滞后环节的特点,提出一种基于小波神经网络的逆控制方案.利用两个结构相同的小波神经网络构造Smith预估器,预估器的输入参数与时延阶次无关,能较好地解决小波神经网络对维数较为敏感的问题.采用神经网络逆控制的思想设计小波神经网络控制器,引入多步预测性能指标函数对控制器权值进行在线训练.仿真研究表明,该控制方案具有较快的响应速度和良好的动态性能.

The process of automatic flatness control and automatic gauge control （AFC-AGC） is a nonlinear system with strong coupling and pure time delay. An inverse control method of AFC-AGC is developed, in which Smith predictor is designed by using two wavelet neural networks （WNN） whose structures and parameters are identical. Because inputs of network are independent of time delay order, the sensitivity of dimension for WNN is reduced. A WNN controller with the idea of neural networks inverse control is applied. Using multi-step predictive index function to train the weights of controller. Simulation results show that the system has better response performance.