Short-term Traffic Flow Prediction Based on ConvLSTM Model

Abstract

This paper proposes a estimation model based on Convolutional Long Short Term Memory (ConvLSTM) model to estimate short-term traffic flow. ConvLSTM is an improved algorithm based on Long Short Term Memory (LSTM) Network. It not only establishes timing characteristics like traditional LSTM models, but also depicts local spatial features like Convolutional Neural Network (CNN). The input data is extracted from the paperual operation of CNN model at the bottom, and LSTM is used instead of the pooled layer in CNN. The space-time feature of the data is further excavated, and the complete prediction model is output through the full connection layer. Experiments show that the prediction accuracy of The ConvLSTM model is higher than that of LSTM, Two-layer LSTM and the bidirectional LSTM model.