Construction of rail wear and crack initiation prediction model combined with BP neural network

Abstract

Aiming at the prediction of rail wear and crack initiation, based on the data measurement and analysis of the actual on-site wear and crack initiation of rails. A BP neural network combined with non-equidistant gray prediction model of rail wear and crack initiation is proposed. The prediction model realizes the prediction of rail wear and crack initiation by using the original test data of the rail at unequal time intervals, and takes the detected rail wear and crack initiation data as the model training sample, and then obtains the rail prediction result. After model training, it can be seen that after the BP network optimizes the residual sequence, the rail wear prediction result of the GM-BP model is basically consistent with the actual prediction curve, and the relative error result is 0.091%. Compared with the prediction result of the GM model, the accuracy is improved by 31.6%. In the prediction of rail crack initiation, the GM-BP model also has better prediction accuracy, which satisfies the construction and maintenance requirements of railway rail traffic, and has important reference significance for the development of modern railway traffic.