Short-Term Subway Passenger Flow Prediction Based on GCN-BiLSTM

Abstract

Aiming at the problem that the accuracy of passenger flow prediction is not high, this paper presents a short-term passenger flow forecasting model based on Graph Convolutional Neural Network (GCN) and Bidirectional Long-term Memory Network (BiLSTM). Firstly, the historical traffic time series is divided into three time modes: recent period, daily period and weekly period; Secondly, we construct three models based on GCN and BiLSTM to capture the spatial and temporal dependence of the three patterns; Finally, the parameter matrix is used to fuse the output of the three time modes to obtain the final prediction result. By testing the data set of subway passenger flow in a city in January 2019, the experimental results show that the root mean square error of the model is reduced by 8.515% and the average absolute error is reduced by 4.239% compared with the single BiLSTM model, it has a high fitting degree with the real passenger flow value and has certain application value for the reasonable allocation of subway capacity.