Short-Term Abnormal Passenger Flow Prediction Based on the Fusion of SVR and LSTM

Passenger flow prediction is important for the operation of urban rail transit. The prediction of abnormal passenger flow is difficult due to rare similar history data. A model based on the fusion of support vector regression (SVR) and long short-term memory (LSTM) neural network is proposed. The inputs of the model are the abnormal features, which consist of the recent real volume series and the predicted volume series based on the periodic features. A two-stage training method is designed to train the LSTM model, which can reflect the large fluctuations of abnormal flow more timely and approximately. A combination method based on the real-time prediction errors is proposed, on which the outputs of SVR and LSTM are combined into the final outputs of the prediction model. The results of the experiments show that the SVR-LSTM model more accurately reflects the abnormal fluctuations of passenger flow, which performs well and yields greater forecast accuracy than the individual models.