Perceptron Multilayer Artificial Neural Network with Good Generalization

Abstract

Feedforward perceptron multilayer (PML) ANNs with error back propagation learning method is often used in engineering design. Unfortunately, its generalization is often poor. In this paper, some reformative methods are proposed to improve its generalization. Based on the improved PML ANN, E-plane T-kind terminal matched load of rectangular waveguide is designed successfully. The result given by the ANN is agreeable with FDTD very well.