Bus Arrival Time Prediction Based On ISOA Optimized LSTM Model

The increasing number of cars has brought people convenience and a series of traffic problems.Public transportation has the advantages of large passenger capacity,environmental protection,and saving road space,etc.Development of public transportation can alleviate urban traffic congestion and environmental pollution.However,inaccurate bus arrival times leading to long waiting times for passengers at stations is one of the major reasons for the decreasing share of public transportation in daily transportation trips.Therefore,it is necessary to improve the accuracy of bus arrival time prediction to attract more bus trips.Firstly,this paper analyzes the characteristics and influencing factors of the bus operation arrival process into two parts:inter-station operation and station stopping.The raw bus operation location data were then subjected to a series of analyses such as route data processing,station data processing,time transformation processing,displacement calculation processing,and quantification of influence factor feature sets to obtain a data set that can be used as input to the prediction model.The processed dataset was also characterized,and it was concluded that the different operating hours of the same bus route operation direction would cause a large difference in bus arrival time.Secondly,in order to improve the model prediction efficiency and prediction accuracy,the ISOA-LSTM prediction model was constructed by introducing a nonlinear convergence factor to improve the parameters,introducing a sine cosine operator to coordinate the global search ability of the algorithm,and introducing an adaptive distribution variation strategy to generate new solutions by improving seagull algorithm(ISOA)for the hyperparameters such as learning rate,number of iterations and number of hidden layer nodes in the long short-term memory(LSTM)model.The proposed model is applied to the prediction analysis of actual bus operation data by direction and time,and compared with the prediction effect of other models.The prediction results show that the ISOA-optimized LSTM model has the highest accuracy for bus arrival time prediction,the seagull algorithm(SOA)-optimized LSTM model is the second best,the LSTM model is better,and the BP neural network model has the worst prediction.Finally,considering that there were few studies on learning historical and future information about bus operation status,an exploratory proposal was proposed to use bidirectional LSTM(Bi LSTM)for bus arrival time prediction.Meanwhile,the Attention mechanism was added to the model to improve the information processing capability of the bidirectional LSTM,and the hyperparameters of the learning rate,the number of iterations,the number of nodes in the two hidden layers and the number of fully connected layers were optimized by applying the improved seagull algorithm to construct the ISOA-Bi LSTM-Attention bus arrival time prediction model.The proposed model and other models were used to predict the bus arrival time at different times in different directions and were compared and analyzed.The results show that the ISOA-Bi LSTM-Attention model has the highest accuracy and best data fit for bus arrival time prediction,and the overall mean absolute percentage error,root mean square error and mean absolute error values of the evaluation metrics are reduced by at least 5.96%,9.87%and 7.99%,respectively,and the overall R~2 value of the coefficient of determination increases by at least 5.78%.