Subway Flow Prediction Based on Improved Support Vector Machine

In view of the heterogeneity of spatial and temporal elements affecting the prediction of urban population flow and the nonlinear hassle of traffic flow variation, it is very difficult to accurately predict the subway traffic flow. This paper proposes a subway flow prediction model based on grid search and an improved support vector machine, named GS-SVM. Firstly, the spatiotemporal statistics affecting the traffic flow are analyzed by discussing the nonlinearity and randomness of the traffic flow. Second, support vector machine algorithm is used to model the mapping relationship between traffic and spatiotemporal information, and considering the influence of time on traffic prediction, the traffic prediction of undetermined subway stations is realized. Third, we use grid search algorithm to dynamically optimize the kernel parameters and penalty factors of SVM to determine the relevant parameters. Compared with HA, ARIMA, LSTM, and SVM traffic prediction models, the improved GS-SVM is better than other models with smaller errors and better accuracy.