A New Method for Short-term Traffic Flow Prediction Based on Multi-segments Features

Abstract

Short-term traffic flow prediction is a very critical part of intelligent transportation. The data of Traffic flow is highly nonlinear and complex, so it has a great challenge to accurately predict it. This paper proposes a novel multi-segments network MCGA. This model describes the features of traffic flow data by constructing two segments including recent segments and daily segments, moreover, combines Gated Recurrent Unit (GRU) and one-dimensional Convolutional Neural Networks (CNN) to simultaneously capture the time domain features and spatial domain features of short-term traffic flow data. It also uses the Attention mechanism to fuse the features and assigns weights to the features to weaken the influence of irrelevant factors, which greatly enhances the robustness of the model. Experiments show that compared with GRU, LSTM and other time series prediction models, the MGCA model has better performance in real data.