ABSTRACT
Accurate prediction of traffic volume is an important function of the intelligent expressway system, which can provide a data basis for the formulation of traffic guidance strategies. Expressway network toll data comprise timely update, large amount of data, and convenient access. The purpose of this paper is to predict the traffic volume of expressway OD (origin and destination) with toll stations as the start and end points, so that the traffic management department can manage and control the expressway traffic. Effective data is obtained through the preprocessing of massive expressway network toll data. Based on the traffic data obtained through data collection and preprocessing, the distribution characteristics of expressway traffic volume are analyzed, and the traffic volume under different time scales are analyzed as well as the cycle similarity of travel demand and the distribution of mileage. Through the discussion of the realization principle of the prediction function of the Long-Short Term Memory model (LSTM), the one-dimensional convolution filter is applied in the input feature extraction process of the long-term and short-term neural network model. Then, hybrid LSTM models, which only consider several periods before the prediction point of the day and consider both the period before the prediction point of the day and the historical data of the same period, are constructed respectively. The input and output of the model are determined, and the index of model prediction effect evaluation are selected. Through the case study of the expressway in Province J, it is proved that the established traffic volume prediction model has smaller prediction errors than other models, and can provide management and control basis for traffic managers.
