Highway Short-Term Traffic Flow Prediction with Traffic Flows from Multi Entry Stations

You are here: Homepage > Search

Highway Short-Term Traffic Flow Prediction with Traffic Flows from Multi Entry Stations

Wu, BangYu / Li, Bin / Zhang, Han / Ruan, Housong / Chen, Zhu

As an important component of the Intelligent Transportation System (ITS), short-term traffic flow prediction is a key step to assess the traffic situation. It provides suggestions for travellers and helps the administrators manage the traffic effectively. Due to the availability of massive traffic data with various features, the data-driven methods have been applied widely to improve the accuracy of traffic flow prediction. However, few previous studies try to capture the information of traffic flows from multi entry stations to forecast the overall tendency of traffic flow. In this paper, we collect data at a highway exit station in Shanghai, split the data according to originating entry stations and predict the corresponding exit station traffic flow from that of the multi entry stations. Firstly, the original records are collected, preprocessed, aggregated and normalized. Secondly, the Long Short-Term Memory (LSTM) model is applied to learn from the information of the overall flow and flows from multi entry stations to predict the overall flow. The baselines are the LSTM models that only used the overall flow traffic data as the input to train them. Compared with the baselines, in other models, the flows from multi entry stations are also considered as the input to predict the traffic flow. Finally, the comparison experiments are carried out among the LSTM model, the K-NearestNeighbor (KNN) model and the Support Vector regression (SVR) model. According to the prediction results, when the information of overall flow and 13 largest flows from various entry stations is used and step is set to 10, the model prediction achieves the best performance. Compared with the best result in the baseline model, the improvement of prediction accuracy is up to 4.58 percent (The improvement is computed by Mean Absolute Error (MAE)).