Traffic Flow Prediction Using Spatiotemporal Analysis and Encoder-Decoder Network

Abstract

Intelligent transportation is an important part of a smart city. Due to the traffic flow sequence has characteristics of periodicity, nonlinearity and easily affected by external factors, improving the accuracy of traffic flow prediction in traffic hub network is important research content of intelligent transportation. For traffic flow prediction problem, an end-to-end framework called DeepTFP is proposed. Specifically, extracting spatiotemporal characteristics of traffic flow data as input of the model through spatiotemporal analysis. Then, a cross-entropy loss function based on error updating for the encoder-decoder network is designed to generate traffic flow predictions, encoder using Bi-direction long-short term memory(BiLSTM), decoder using long-short term memory(LSTM). We conducted extensive experiments on real datasets. The experiment results show that DeepTFP outperforms the other traffic flow prediction methods in terms of prediction error.