System Identification And Control Based On Neural Network

Modeling and control of nonlinear systems are the focus and difficulty in the field of control.Neural network is widely applied in system identification,control and optimization for its learning,adaptation and powerful nonlinear mapping capabilities.Meanwhile,there are random noises and disturbances in many practical systems inevitably.Therefore,the neural network identification and control of practice system with random noise become a difficulty.Feedforward neural network has been widely used in approximation,recognition,classification,data compression and so on for its simple structure and strong plasticity.This paper focuses on the structure and learning algorithm of BP neural network and its application in adaptive control.However,feedforward neural network is defecting in identification and control of nonlinear dynamic system,so this paper studies the structure of recurrent neural network.Then this paper focus on the structure and learning algorithm of discrete state space neural network,and applys it to nonlinear system identification.On the other hand,the error correction learning algorithm is widely used to train the neural network.The training criteria is generally a function of error and traditional minimum mean square error criteria capture only to the second-order statistics in the stochastic data,could be inadequate for arbitrary non-Gaussian and nonlinear system.Information measure can describe the uncertainty of random variables well,so this paper introduces the new measures such as entropy,information potential and survival information potential in details and applies them to identification and control of neural networks.Finally,the inlet temperature and flow rate of the flue gas which will affect the temperature of evaporator's outlet,may be subject to non-Gaussian distribution in ORC system.An adaptive control method based on neural network is applied to control the degree of the superheated and ensure that the random of the tracking error is minimized in this paper.