Optimal Prediction Model of Cold-Rolled Strip Performance Based on PSO-BP Optimized Neural Network

This paper presents a PSO-optimized BP neural network model for the property prediction of cold rolled strip steel. Cold rolled strip steel is of paramount importance within the arc of steel manufacturing, and so many factors affect its performance, including the physical properties of the raw material and the parameters involved in the rolling process. Such a complex nonlinear relationship is difficult to describe with the traditional physical model or empirical formula. Therefore, this paper adopts the data-driven machine learning approach to build a prediction model. The BP neural network has been widely adopted because of its strong adaptability in nonlinear relationship modeling but suffers from slow training speed and a tendency to fall into the local minima. The PSO algorithm, inspired by the foraging behaviors of bird flocks, features a strong global search capability. It can effectively optimize the initial weights and bias of the BP neural network and improve training efficiency and prediction accuracy. This paper first used Spearman’s rank correlation coefficient to analyze the relationship between variables of the dataset, then built up and trained the models of the BP neural network and PSO-BP neural network. The experimental results show that PSO-BP neural network performs better in predicting cold rolled strip steel properties with a great value of practical application and research.