The Prediction of Traffic Flow Based on Long Short-Term Memory Network for All Weather

Abstract

To set up a schedule of toll collector in advance and reduce the queue length and traffic congestion of toll stations, this paper proposes a short-term forecasting method for toll stations based on long short-term memory network. First, we analyze the quality of highway toll data. Then, a data preprocessing method based on the cubical smoothing algorithm with five-point approximation is designed. Moreover, we establish the traffic data set which is associated with the toll stations information and time. Then we construct a traffic flow prediction model. Taking the Airport Station of Airport Highway in Guangzhou as an example, then test the validity and real-time performance of our model. The results show that the mean absolute percentage error of the prediction is about 3.6% when the forecast horizon of prediction is 5 minutes; when the forecast horizon of prediction is 10 minutes, the mean absolute percentage error is about 6.07%; when the forecast horizon of prediction is 15 minutes, the mean absolute percentage error is about 8.68%, therefore, the model can accurately predict the traffic flow of toll stations. At the same time, compared with the KNN algorithm and GBDT algorithm, the model of this paper not only has higher prediction accuracy, but also has better adaptability to predict the peak, and when weather is adverse, the algorithm of this paper also can predict accurately through extracting the relevance of time in data set effectively.