An Improved Dynamic Process Neural Network Model Identification Method

With the gradual improvement of the industrial level, modern science and technology has a rapid development, the current industrial products gradually achieve intelligence and automation. At present, the traditional neural network model identification methods generally adopt the sum of the square errors of model output and sample output as the performance index. In the process of neural network training, if the training accuracy is too high, the phenomenon of over-fitting may occur. As a result, the generalization ability of the neural network is reduced, and the error is reduced when the training is small, and the convergence speed is also significantly reduced. To solve these problems, an improved dynamic process neural network model identification method is proposed in this paper. Compared with the traditional neural network model identification methods, the proposed method can effectively improve the data fitting ability and generalization ability of the model under the same identification accuracy.