Researches On Time-delay Control Systems Based On Neural Networks

Delay component exist universally in the process of industrial production. Bigtime-delay has a severe impact on the system stability, resulting in a bigger overshoot andlonger regulating time. Even worse, it may cause the instability of closed loop system orincapability of implementing effective control. Since the end of 50's, many control methodsconcerning the big time-lag process have been researched successfully, but few is found a realapplication in the industrial. Taking into consideration of the self-learning and self-adaptingability of neural network, this paper applies it to the control of delay component. In particular,the weight is modified by the means of combination of genetic algorithm and BP algorithm.This paper put forward the following several controlling approaches aiming atcontrolling the big time delay process. Because of the merits of the PID neural network, PIDneural network control method for time-delay systems is put forward, and In order toovercome the disadvantage of BP algorithm and genetic algorithm, the synaptic weightparameters of PID NN controller is modified by the means of combination of geneticalgorithm and BP algorithm;Combining neural network and Smith predictive control, Smithpredictive control based on NN for time-delay systems is put forward, ANN identifies themathematical model of the control object, BP neural network is used to modify the parameterof PID controller on-line, overcoming the over reliance of Smith predictive control on themodel and avoiding the weakness of traditional PID controller;Combining neural networkand internal model control, internal model control based on RBF neural network method fortime-delay systems is put forward, the simulation experiments show that the controlperformance is very good;Combining neural network and predictive control, predictivecontrol based on neural network method for time-delay systems is put forward, and thedelay-time is acquired using a 3-BP neural network online-learning, the simulationexperiments show that the control performance is very good.