Выбор и оценка ARIMA-модели для прогнозирования пассажиропотока

Abstract

Short-term forecasting of transport traffic flow is one of the main areas of research in the field of intelligent transport systems. In recent years, a large number of studies on traffic forecasting have been conducted. However, it is difficult to pick the most appropriate traffic forecasting method for a particular application. Naive methods are widely used due to their low computational costs and simplicity of implementation. But the accuracy of these methods is usually quite low. Clustering methods average out the traffic variables within a certain group of days based on similar traffic models. Macroscopic simulation models take into account only such global variables of the road network as traffic density, average speeds and flows. The advantage of nonparametric models lies in the fact that they enable us to model complex, dynamic and nonlinear processes occurring in road traffic. One of the forecasting options for passenger traffic is the forecasting on the basis of time series processing, which allows obtaining rather reliable results. Currently, considerable attention is given to the spatiotemporal forecasting. ARIMA processes are flexible linear models which can accurately describe a wide range of various time series behavior, including medium-term peaks and troughs. Despite the fact that these basic models are quite simple to describe, their evaluation requires extensive computer calculations. This paper investigates the problem of seasonality modeling when studying the time distribution of passenger traffic flow in road transport. The basic information for predicting passenger traffic flow is the data obtained during the daily parsing of the online service for searching traveling car companions (edem.rf (едем.рф)) for 12 months (February 2022 - January 2023).