Passenger Flow Prediction Based on Composite Model

Abstract

The article aims to investigate the performance combination model and traditional single prediction model in traffic flow prediction, providing reliable and accurate traffic flow prediction data for intelligent transportation systems. In this article, a fusion of these two models urban rail transit passenger flow prediction model (TCN-BiLSTM) is proposed based on temporal convolutional network (TCN) and bi-directional long and short term memory network (BiLSTM). Firstly, using TCN to extract hidden information and temporal relationships of urban rail passenger flow, capturing the dependency relationships between long-distance features. Secondly, the BiLSTM model further obtains the before and after connections between the passenger flow characteristics of urban rail transit. Finally, the output results of BiLSTM are incorporated into an attention mechanism to allocate different permissions, increase the weight of important features, and reduce the impact of redundant information on the prediction accuracy of TCN-BiLSTM. This article evaluates the model's predictions using MAE (Mean Absolute Error), RMSE (Root Mean Square Error), R2 (Coefficient of Determination), and MAPE (Mean Absolute Percentage Error).The results of the ablation experiment showed that the TCN-BiLSTM prediction model reduced RMSE, MAE, and MAPE by 10.3, 6.6, and 1.9, respectively. Meanwhile, R2 also reached 0.904. When using the TCN-BiLSTM prediction model for prediction at different sites, the R2 values are all above 0.9. By comparison, the TCN-BiLSTM prediction model has the smallest error, highest fit, and high portability in predicting metro rail passenger traffic passenger flow.