A Traffic Flow Prediction Approach: LSTM with Detrending

Abstract

Traffic flow prediction plays a key role in many Intelligent Transportation System research and applications. It aims to forecast the forthcoming traffic conditions with the help of historical data. Urban traffic always has its morning and afternoon peak hours. We also observed that the urban traffic flow can always be divided into main trend data and its residual part. The main trend data presents a similar trend on different days. The residual data is time-variant part which reflects the short-term fluctuation of traffic condition over each day. Enlighted by detrending, Principal Component Analysis (PCA) method is applied to extract the main trend data in this paper. The residual data is obtained by subtracting the main trend data from the overall traffic flow data. Then Long Short-Term Memory (LSTM) model is proposed to predict the residual data. With main trend data and predicted residual data, the urban traffic flow can be predicted by the joint PCA and LSTM approach. Finally, the empirical study demonstrates the propose method outperforms similar traffic prediction models.