Application of Recursive Predict Error Neural Networks in Mechanical Propertise Forecasting

Abstract

Parameters control problem was crucial in rolling industrial, but the mechanical properties forecasting of strip steel was an information space incompletely and non-linear complex system which was hard for traditional method. Artificial neural networks was a non-linear system with strong non-linear modeling ability, but the traditional BP neural networks has many shortcomings like easily step into local minimum, with weak generalization ability and the middle layer neuron are hard to determine, so the artificial neural networks with recursive predict error (RPE) algorithm was proposed in this paper with the networks’ structure, algorithm, sample data selection also presented, the simulation shows its effective and can successfully applied into parameters control of rolling industrial.