Application of LS-PCP model based on EWM in predicting settlement of high-speed railway roadbed

Abstract

Accurate prediction of roadbed settlement is of great significance to the maintenance of high-speed railway roadbeds and the safe operation of trains. This study proposes a long- and short-term parallel combined prediction (LS-PCP) model based on the prediction characteristics of the LSTM model, GM(1.1) model, and ESP model and applies it to the prediction of roadbed settlement of high-speed railways. First, according to the spatiotemporal characteristics, slow-varying characteristics, and short valid data characteristics of the settlement process of a high-speed railway roadbed, this study designed a combined form of long-term LSTM prediction and short-term GM(1.1) and ESP sliding prediction to overcome the problem of large prediction errors when roadbed settlement enters different stages. Next, the mutual inclusiveness of the member models’ prediction results is tested by the principle of inclusiveness test, and the combination weights are determined by considering the information entropy of the member models through the entropy weighting method. Finally, the combined prediction results of the proposed LS-PCP model are verified from the actual monitoring data of a high-speed railway in Hebei Province and the Guiguang High-speed Railway. The results prove that the proposed LS-PCP combined model has higher prediction accuracy, and the prediction data of this model have important reference significance for the maintenance of high-speed railway roadbeds and safe vehicle operation.