Fuzzy Neural Network And Its Application To Shape Control

The shape is an important quality criterion of cold rolled strip. With the development of instrument,electricity,automobile and light industry,more and more high-quality strip is required. But in our country,the automatic control of shape is quite an unsubstantial part. The economic losses brought by bad shape are great every year. It will bring immense economic benefits to solve the problem of shape quality. The hydraulic bending roll system is a basic part,its dynamic characteristic and steady characteristic affect the capability of entire shape control system. In view of the time-variable and nonlinear characteristics of mathematical model of hydraulic bending roll system,this thesis design a new nonlinear adaptive controler based on fuzzy neural network. The muti-layer feedforward neural network is adopted to simulate the dynamic characteristics so that the controller can obtain the Jacobian information of the plant. To overcome the shortcoming of plunging into Local Minimum easily,Gradient Descent algorithm with momentum factor is used to train the network.. The nonlinear adaptive control method based on fuzzy neural network can learn and adjust parameter of rule and parameters of member functions on line. And the control rule list is calculated according to data on the spot,which is also applied to the control on the spot. The controller is applied to hydraulic bending roller system,simulation results indicate that the adaptability,the robustness and the stability of system are satisfying,and it is appropriate to the control of the nonlinear and time-variable system. The mathematical model of the control loop of bending the intermediate rolls has time-variability and uncertainty. The routine PID algorithm can't be responsive to the transformation of system parameters rapidly and accurately,and the precision of the shape can't be assured easily. The fuzzy neural network is proposed according to the advantages and the disadvantages of fuzzy and neural network. A network is established to simulate fuzzy system and to achieve fuzzification,fuzzy inference and defuzzification. The experimental result indicates that the control effect is excellent,namely small overshoot and good robustness,and the goal of intelligent control of shape on-line is reached.