Prediction of Electric Multiple Unit Fleet Size Based on Convolutional Neural Network

TL;DR

This paper develops a convolutional neural network model to accurately predict the future fleet size of electric multiple units in China, aiding railway planning and investment decisions.

Contribution

It introduces a CNN-based approach for EMU fleet prediction and compares its performance with a BPNN, demonstrating CNN's superior accuracy and generalization.

Findings

CNN outperforms BPNN in prediction accuracy

CNN shows better generalization ability

The model aids in efficient EMU fleet planning

Abstract

With the expansion of high-speed railway network and growth of passenger transportation demands, the fleet size of electric multiple unit (EMU) in China needs to be adjusted accordingly. Generally, an EMU train costs tens of millions of dollars which constitutes a significant portion of capital investment. Thus, the prediction of EMU fleet size has attracted increasing attention from associated railway departments. First, this paper introduces a typical architecture of convolutional neural network (CNN) and its basic theory. Then, some data of nine indices, such as passenger traffic volume and length of high-speed railways in operation, is collected and preprocessed. Next, a CNN and a backpropagation neural network (BPNN) are constructed and trained aiming to predict EMU fleet size in the following years. The differences and performances of these two networks in computation experiments are analyzed in-depth. The results indicate that the CNN is superior to the BPNN both in generalization ability and fitting accuracy, and CNN can serve as an aid in EMU fleet size prediction.