The Study Of Intelligent Control Method Based On Neural Network

With the development into large-scale and integration of modern industry, the product process tends to become complex. Many systems are short of accurate mathematic description, because the process is nonlinear, time-varying, uncertain and the strong coupling of variable. Then it is difficult to analyze and control with traditional method, so we need to research new intelligent control strategy. At present, neural network, fuzzy control and genetic algorithm have successful application in diverse fields. But when they simulate human's intelligent activities alone, there is respective localization. Hence it is essential to integrate these methods to overcome the localization. It is also active to the development of control theory and the improvement of automation level.After analyzing the control system of neural network, this paper mainly does research in two aspects. Firstly, on the basis of combination between neural network and genetic algorithm, a hybrid learning algorithm for identification of neural network is proposed. Self-adaptive hierarchical genetic algorithm which is mixed coding optimizes the structure and weights of neural network. Secondly, on the basis of combination between neural network and fuzzy logic, compensatory fuzzy operation can adjust the parameters of membership function self-adaptively and optimize fuzzy reasoning dynamically. With the recurrent nodes introduced in the second layer of the network, the recurrent compensatory fuzzy neural network has the ability of dynamic mapping. Aiming at the network, a modified relational grade clustering method is proposed to construct the initial fuzzy model of the RCFNN. Then the dynamic gradient descent method is used to tune the parameters of the network. The paper tests the performance of these two networks by the simulation of the typical nonlinear systems. Finally, the research result is applied to omethoate synthesis temperature process control. The simulation result shows that it meets the proposed standard better.The main contents are as follow:(1) The development actuality of neural network, fuzzy control and genetic algorithm is analyzed. (2) The rationale of neural network control is researched and the identification structure of neural network is analyzed, then BP neural network and its basic algorithm are given.(3) The rationale of hierarchical genetic algorithm is researched, then the operation flow of self-adaptive hierarchical genetic algorithm which optimizes the structure and weights of neural network is given. After analyzing the limitation of traditional BP algorithm, a hybrid learning algorithm for identification of neural network is proposed.(4) Fuzzy logic and neural network is integrated. On the basis of analyzing the difficulties of fuzzy neural network design, compensatory fuzzy neural network is researched. After introducing recurrent nodes, a new recurrent compensatory fuzzy neural network is designed. And a modified relational grade clustering method is used to construct the initial fuzzy model of the RCFNN. Then the dynamic gradient descent method is used to tune the parameters of the network.(5) The performance of two designed network is tested by simulation with the examples of dynamic nonlinear system.(6) The character of omethoate synthesis system is analyzed, then the neural network based on hierarchical genetic algorithm and recurrent compensatory fuzzy neural network compose the temperature control system of omethoate synthesis. Finally, it is simulated in Matlab.