A Short-term Traffic Supply-Demand Gap Prediction Model with Integrated GCN-LSTM Method for Online Car-hailing Services

Abstract

The purpose of this paper is to explore the performance of graph neural network (GNN) based short-term traffic supply-demand gap prediction for online car-hailing. In recent years, with the rapid development of smart cities, the technology as well as the scale of online car-hailing has grown rapidly, and it is necessary to analyze the travel characteristics of online car-hailing, and secondly, there are quite few studies on the supply-demand gap prediction of online car-hailing. Based on the dataset of online car-hailing operation, we analyzed the travel characteristics of online car-hailing from several dimensions, such as traffic congestion, weather, air quality, and temperature. In response to the current road traffic congestion and the reasonable allocation of online car-hailing, this paper proposes an online car-hailing supply-demand gap prediction model based on graph convolutional neural network and long and short-term memory neural network (GCN-LSTM), with mean absolute error (MAE) and root mean squared error (RMSE) as the evaluation index, and analyzes the performance of the model through simulation. The results show that the MAE and RMSE of the proposed method is only 12.3 and 26.4, respectively, which performs better than LightGBM, LSTM and other models on this dataset. Therefore, the constructed model for predicting the supply-demand gap of online car-hailing has a high-quality prediction performance.