The System Modeling And Controller Design Of Time-Delay System Based On Neural Network

In the modern industrial production process, the controlled object with time-delay characteristics is very common. Because of time-delay, the controlled process can not reflect the disturbance of system in time, which brings worse control performances, such as obvious overshoot and longer adjusting time well be longer. The research on the system modeling and control method of the time-delay system has been a research emphasis and difficulty in the filed of control. Therefore, it has great theoretical and practical significance with the research on modeling and control methods for a time-delay system.The neural network has extensive application in modeling of time-delay system with its unique ability of nonlinear mapping, stronger ability of self-learning and capacity of precision, while the dynamic matrix control has unique advantages in solving the problem of the control of time-delay system because of its prediction properties. In this thesis, a predictive control strategy is proposed which based on the self-adaptive mutation PSO-BP neural network though combining DMC with self-adaptive mutation PSO-BP neural network.Firstly, the standard error back propagation algorithm and the application of BP neural network for the model identification are introduced in this thesis. Then the principle and design method of BP neural network for the system identification is expounded. Then a model of time-delay system is build based on BP neural network. The simulation results show that the standard BP neural network can approximate time-delay system, but this method has low precision and low convergence rate.Secondly, this thesis explores the modeling of time delay system based on adaptive mutation PSO-BP neural network. Through analyzing the disadvantage of BP neural network, this thesis proposes an improved adaptive mutation particle swarm optimization applying for the optimization of BP neural network. The mutation operator is introduced to the algorithm analogous to the genetic algorithm. The mutation rate is determined according to the population diversity index. This strategy balances the global and local search ability of particle swarm preferably and improves the identification performance of BP neural network. It proves that the BP neural network based on adaptive mutation PSO-BP algorithm has high precision and high convergence rate for the identification of time-delay system.Finally, this thesis presents a predictive control method based on adaptive mutation PSO-BP neural network identification according to the characteristics of time-delay systems. Firstly, we build the multi-step prediction model of the controlled object by neural network for this method with combining the dynamic matrix predictive control with neural network prediction model. Then the optimal control law is computed by DMC algorithm to realize the control of time-delay systems. The simulation results show that the dynamic matrix predictive control based on neural network has better control effects for the time-delay system.