Research on Single-Section Traffic Flow Prediction Based on BiLSTM

Traffic flow prediction has always been an integral part of intelligent transportation systems, which can help decision-makers to formulate effective traffic management strategies. In this paper, the Bidirectional Long and Short Term Memory Networks is proposed to solve the problem that the prediction accuracy decreases when the prediction sequence is too long in the Long short-term memory algorithm. In terms of the selection of optimizer and parameter, this paper picks out the best optimizer and parameters by continuously adjusting the optimizer and its parameter settings, calculating the MAE value of the future first-day prediction, and comparing them in each experiment. Finally, the paper applies the BiLSTM-based single-section traffic flow prediction model to predict the sensor 0, sensor 60, and sensor 120 traffic flow data respectively, from the dataset PeMS08 of the California highway network in the United States. By comparing the prediction results of MAE, RMSE, and MAPE, we found that the BiLSTM neural network made effective improvements in the degradation of prediction accuracy when the LSTM neural network sequence is too long. Thus, the model has a better prediction performance.