神经网络在线工业跟踪在加热炉控制中的应用

基于BP算法的神经控制器是目前最常用、最成熟的神经控制结构,它可分为普通学习法和专门学习法两种结构,后者又分为直接和间接两种学习结构。普通学习法和间接学习法还没有形成真正的自适应控制特性,它们一般不能实现在线控制。当事先知道对象的定性知识后,直接专门算法能实现自适应控制,但此法会使执行器饱和,引起对象输出不稳定。针对这种问题,提出了一种神经网络在线工业跟踪控制方法。可解决BP算法中误差函数对权值的偏导数计算问题以及执行器输出的脉动变化和饱和问题。将此算法用于加热炉控制,仿真结果表明了它的可行性和自适应性,提出的神经网络在线工业跟踪控制性能优于常规PID控制器。

A Neural Network Approach for on-Line Industrial Tracking Controlling and Its Application on Heating Furnace

The neural controller based on BP algorithm is the most commonly used for the neural controlling structure at present, it may be divided into two structures-basic learn method and special learn method, the latter is divided again into direct learn structure and indirect learn structure. Self-adaptive controlling using common learn method and indirect special learn method is not yet formed. They usually connot achieve on-line controlling. When the qualitative knowledge of the object has been obtained, the self-adaptive controlling can be fulfilled throug direct special algorithm. But this algorithm may cause the excutor to become saturated, thus the object output will be not stable. To solve these problems, a neural network approach for on-line industrial tracking controlling is presented. It can be used to solve two problems, that is, the evaluation of the partial derivative of error function with respect to weight in BP algorithm, and the fluctuation and saturation of the actuator output. Simulations on a heating furnace controlling have been investigated; and the results demonstrate the feasibility and adaptive property of the proposed scheme. Performance of the neural nwtwork approach for on-line industrial tracking controlling is better than that of the conventional PID controller.