Novel deep neural network automatic modeling method applied to microwave device

The invention relates to a novel deep neural network automatic modeling method applied to a microwave device, belongs to the field of microwave device modeling, and the method is used for solving theproblem that an existing shallow neural network modeling method cannot meet the new requirements of high model dimension, large working range, high reliability and the like in the field of microwave device modeling. The invention particularly provides a novel hybrid deep neural network structure comprising a Batch Normalization (BN) layer and a Sigmoid hidden layer and a training method of the hybrid deep neural network structure. The problems of gradient disappearance and training prevention caused by increase of the number of layers of the neural network can be solved; and an adaptive algorithm is adopted to automatically determine the number of BN layers and Sigmoid hidden layers in the deep neural network, so that the automation of the training process is realized. The deep neural network model obtained by the method can accurately represent complex, continuous and highly nonlinear device characteristics of a microwave device, is high in calculation speed, and can effectively replace an original device to perform simulation and design of a subsequent circuit or system.