PID Neural Network Control Research Based on Fuzzy Neural Network Model

Abstract

A kind of PID neural network control on the basis of fuzzy neural network model is proposed in this paper, which combines the fuzzy neural network model with the PID neural network. PID neural network weight is tuned online by using fuzzy neural network model and gradient descent method, and the proposed method is applied to the bed temperature control of CFB boilers. The simulation results show that the proposed controller produces better effects than traditional controllers in both steady performance and dynamic performance , including shorter steady-state time , non-overshot, non- oscillator, higher dynamic tracking rate , etc. Currently, there exist three methods for the combination of neural network with PID as follows: 1. The neural network identification PID control. The three parameters of PID controller are regulated by kinds of different neural network identification, and the most prominent difference between the traditional controllers and the given controller is that the ratio parameter, the integral parameter and the differential parameter are adjustable; 2.The PID control with single neuron structure, the simple combination of neural network with PID control, whose inputs are errors, differential errors and integral errors. And weights are ratio coefficient, integral coefficient, differential coefficient respectively; 3. PID neural network control, put forward by Shu Huailin for the first time, is the complete combination of neural network with PID control. Every neuron is equipped with one of the functions of ratio, integration, differentiation. In (3), the first two methods are applied to the bed temperature control of CFB boilers .The comparison with traditional PID control shows that the first method has shorter response time, but exists over-shot and oscillator. Though the response time in the second method matches that of traditional PID controller, there is non-overshot, non- oscillator. On the other hands the two methods both have disadvantages. PID neural network control based on fuzzy neural network model is proposed in this paper, which combines the fuzzy neural network model with the PID neural network. PID neural network weight is adjusted online by using fuzzy neural network model and gradient descent method, and the proposed method is applied to the bed temperature control of CFB boilers. II. PID CONTROL SYSTEM BASED ON FUZZY NEURON MODEL PID neural network is the combination of traditional PID and neural network, where PID functions are imported into basic single neuron, and then a PID neuron is formed. New neural network is formed d by these basic neurons according to PID control rules (4). The block diagram of fuzzy neural model based PID neural network control system is shown as Fig.1, where r, u, y, are system input, controller output and system output respectively. And PIDNN is the PID neural network controller, FNNM being the fuzzy neural model with the controller output and the system output to be its inputs .In order to establish the control system, firstly, FNNM should be established according to the collected field data; Secondly, PID neural network controller should be introduced on the basis of model, and the controller parameters are tuned online in light of the model output and the error between system input and output.