Short-Term Traffic Flow Prediction for Hybrid Time Series Decomposition Analysis and LSTM Neural Network

Deep learning approaches have recently demonstrated the ability to predict traffic flow with high accuracy, based on research of short-term traffic flow prediction methods, especially the long-short term memory (LSTM) neural network. However, the potential of LSTM in traffic prediction has not yet been fully exploited in terms of the depth of model structures, lack of multidimensional data, and domain knowledge fusion. In this work, a short-term traffic flow prediction model with the time series decomposition analysis and LSTM neural network is presented. The combination of in-depth information mining of traffic data and LSTM neural network is more reliable than the separate application of LSTM neural network. The effectiveness of the proposed method is validated and compared with other classical statistical and state-of-the-art neural network models, which indicates the proposed method achieves superior prediction accuracy.