Prediction Model of Urban Traffic Performance Index Using ARIMAX

摘要

The road traffic performance index (TPI) is a comprehensive index to evaluate the congestion level of the road network. With regard to the historical continuity of TPI, and the external variables such as month, holiday, and weather, the autoregressive moving average with external variables (ARIMAX) model is developed to forecast urban traffic congestion. Based on the data set of 620 days, the ARIMAX (2, 0, 0) and seasonal ARIMAX (1, 0, 1) * (2, 1, 3) are established using a data-driven approach. Both models are used to predict the TPI for the next 90 days. The results show that the average absolute error (MAPE) of the two models is about 10%. Compared with the seasonal ARIMAX, the prediction accuracy of ARIMAX is more ideal. In short, the TPI is mainly related with the time factor, and the holidays come in second, while the month, rainfall, and humidity also have some impact.