An modified gradient training algorithm of process neural network

摘要

Process neural network (PNN) is a new neural network. This paper intends to improve the training speed of the discrete PNN with a Levenberg-Marquardt modified gradient training algorithm. The training steps and the algorithm are illustrated. Further, an experiment for the prediction of the humidity of sealed boxes is taken as a case study. This modified algorithm is employed in the case study where its fast convergence is convinced.