Neural Network Inverse Model Identify And Adaptive Inverse Control Study

As a new novel controller and adjustor design method, adaptive inverse control has caught overseas and domestic scholars'more and more study interest. The development of present neural network creates condition for the nonlinear adaptive inverse study and actualization. Exploring and designing rational dynamic neural network structure and arithmetic, building more effectual system structure have become the emphasis of nonlinear adaptive inverse study. This paper has study the structure and arithmetic of neural network, and the adaptive inverse control system based on neural network. The main study content states as follow:Firstly, we study the main clustering methods of RBF network, K-mean clustering method, gradient descent method, orthogonal least squares method and dynamic clustering method. Focus on dynamic clustering method, the most defect is its threshold space is fixed. We raise an improved dynamic cluster method, which adjust the threshold according sample density. The emulation results show the method's validity and quickness.Secondly, apply RBFNN and BPNN to adaptive inverse control system. The first-order system emulation results show that the generalization of RBFNN is lower, and reduces the system control precision. Then we apply the adaptive inverse noise elimination method based on BPNN to the roller eccentricity gauge control system. The emulation results show that the criterions of this method all overmatch PID control scheme. It offers a new resolve scheme for roller eccentricity.Finally, we apply adaptive inverse noise elimination method, based on BPNN, to nonlinear system. The control error and MSE all less than other methods, which shows the adaptive inverse control method's valid to nonlinear system.