Bus Arrival Time Prediction Model Based on Bidirectional Long Short-term Memory Network

To improve the accuracy of bus arrival time prediction and increase the bus usage in the cities, this paper proposes a bus arrival time prediction model based on a bidirectional Long Short-term Memory (BiLSTM) neural network and the hyperparameter search. The improved seagull algorithm optimization adding Attention mechanism to bidirectional LSTM (ISOA-BiLSTM-Attention) prediction model was developed by introducing nonlinear convergence factor, sine cosine operator, and adaptive parameters to improve the seagull algorithm to achieve hyperparametric optimization of the bidirectional LSTM model. The Attention mechanism was added to improve the information processing ability of bidirectional LSTM. Then, the trajectory data of bus route 220 in Nanchang, Jiangxi Province of China, were used to predict the bus arrival time for different directions and time to validate the model prediction accuracy. The results show that, the proposed model has better performance than the traditional bidirectional LSTM model. The improved seagull algorithm can achieve a better optimization effect on the bidirectional LSTM-Attention model. Compared with the existing model and seagull algorithm (SOA) optimized bidirectional LSTM-Attention model, the mean absolute percentage error was reduced by 5.96%, the root mean square error was reduced by 9.87%, and the mean absolute error was reduced by 7.99% in the ISOA-BiLSTM-Attention for bus arrival time prediction. Moreover, the ISOA-BiLSTM-Attention has the largest model decision coefficient R2 value, which indicates the good generalization ability and stability of the proposed model, and can provide good fitness of accuracy for bus arrival time. © 2023 Science Press. All rights reserved.