Research on interval prediction method of railway freight based on big data and TCN‐BiLSTM‐QR

Abstract With the rapid development of logistics, the categories of goods and the frequencies of train transportation in railway freight have increased significantly. The volatility and uncertainty of railway freight transportation have become even greater. Accurately predicting railway freight volume in the medium to long term has become increasingly challenging. On the basis of traditional prediction models, this paper introduces the concepts of interval and probability prediction, and proposes a temporal convolutional network (TCN)‐bi‐directional long short‐term memory (BiLSTM) interval prediction method for medium and long‐term railway freight volume. The method uses grey relational analysis for data dimensionality reduction and feature extraction, and TCN, BiLSTM, and quantile regression for modelling. Through a case study of freight transportation on the Shuohuang Railway, the results show that the TCN‐BiLSTM model achieves higher accuracy in point prediction and better performance in interval prediction compared to other general prediction models. The interval prediction can provide references for freight volume fluctuations in periods with significant volatility, which can assist railway transportation companies in better scheduling and planning based on such information.