Statistical method for long-term station flow prediction under foreseeable events

Long-term prediction is a challenging task affected by multiple influence factors to be considered. This paper proposed a long-term station flow forecasting model using the basic statistics to formulate the effect from multiple scenarios including holiday, weather and activities. The proposed model could learn the regularity in station flow from few historical samplings, which brings better results than classical neural networks that requires large training sets. Experiments with the data from Shenzhen Metro system have been conducted. During a rainy holiday, the hourly station flow accuracy of the whole system is above 85%, while more than 60% of stations reach the accuracy over 70%. The model also capture the peak hour of rush flow before and after a foreseeable activity, which proves the value of the proposed model for station flow management.