Research And Application On Neural Network Optimizing Theory

Neural Network owns strongly self-learning ability, which can adapt to complex conditions and meet multi-target controlling requests. And it is able to approximate arbitrarily any nonlinear function well. As a result, Neural Network is widely applied in controlling field.In this paper, the central purpose is optimizing multi-layer feed-forward neural network globally. An applied coding scheme of genetic algorithm is proposed and the genetic operators are improved to optimize the BPNN and RBFNN's topology structures. The simulation results indicate that these optimizing methods are effective. Shape patterns recognition is the key to the shape control. Based on the neural network optimizing theory, the shape signals recognition network with 6-input and 3-output is established. In the training this network's topology structure and mapping function are the most optimal all along. And the learning velocity and precision are improved. With this recognition method shape pattern information and magnitude can be received rapidly and exactly, which can provide reliable data for later shape controlling. Hydraulic bend roller is the basic segment of AFC system. Its dynamic and static characteristic is important to the whole AFC. For this a neural network IMC strategy is applied. And its NNC and NNI are both established with the optimal neural network. So the hydraulic bend roller system's dynamic response velocity and static tracking precision are advanced. Hydraulic bend roller force can operate better and mill's dynamic characteristic is improved.