Improved Particle Swarm Optimization-based GRU Networks for Short-time Traffic Flow Prediction

Abstract

Short time traffic flow forecasting is the heart of matter in intelligent transportation system (ITS). Accurate traffic flow prediction can help people to choose trip mode and trip time. Although gated recurrent unit (GRU) has outstanding performance in traffic flow forecasting, but determines the hyperparameters of the GRU rely by experience reduces the predictive effect of the model. This study uses the adaptive learning strategy improved particle swarm optimization (IPSO) algorithm to optimize the hyperparameters of GRU model. The characteristics of traffic data with network topology are matched by this algorithm, so the accuracy of traffic flow prediction can be improved. To verify the reliability of this algorithm, this study construct IPSO-GRU model by the traffic flow data from California department of transportation and compare IPSO-GRU model with other traffic flow forecasting models. The experimental results shows that, the IPSO-GRU model achieves the lowest mean square error (MSE), Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE) compared to conventional GRU model.