Short-term traffic flow prediction based on phase space reconstruction and particle swarm optimization BP neural network

A back propagation neural network (BPNN) model based on phase space reconstruction (PSR) and particle swarm optimization (PSO) is introduced to fully excavate the spatiotemporal features of short traffic flow and to improve the prediction accuracy. The phase-space reconstruction is used for reconstruction of the traffic flow data. The spatiotemporal features obtained from the reconstruction are trained in a neural network model and optimized by particle swarm optimization algorithm. The trained model is used to predict the traffic flow with a short future time at the target detection point. In this paper, the data used in the model comes from the Canadian Whitemud Drive highway. According to the different characteristics of the data, four datasets are constructed: namely weekday congestion (WC), weekday non-congestion (WNC), non-work congestion (NWC) and non-work non-congestion (NWNC). Based on these datasets, this paper designs comparative experiments between various models, and the results show that: the PSR-PSO-BPNN model has better prediction performance in WC than the pre-improved BPNN, PSR-BPNN, and PSO-BPNN models, and the mean absolute percentage error (MAPE) is reduced by 7.13%, 10.08% and 2.3%, respectively. In addition, the model proposed in this paper outperforms the baseline models such as the long short-term memory model (LSTM) and the gated recurrent unit model (GRU), and the MAPE is reduced by 4.07% and 4.35%, respectively. Finally, on the abovementioned four datasets, the PSR-PSO-BPNN model shows higher performance in terms of stability and accuracy, which verifies the effectiveness of the combined algorithm of phase space reconstruction and particle swarm optimization.