Research On The Prediction Method Of Bus Arrival Time Interval Based On Bayesian Estimation Theory

In order to alleviate the pressure of urban transportation,it is necessary to vigorously develop the public transportation system.However,the bus will be delayed due to various factors during the operation of the bus,which has greatly dampened the enthusiasm of travelers to choose the bus as the travel mode.In order to provide accurate and reliable arrival time information for travelers,this paper has conducted an in-depth research on the bus arrival time prediction method,with a view to encouraging more travelers to choose bus as the travel mode,thus effectively alleviating the traffic congestion problem.In this paper,the model of bus arrival time interval prediction based on Gaussian process regression(BATIP-GPR)and the model of bus arrival time interval prediction based on Bayesian neural network(BATIP-BNN)were proposed on the basis of considering the uncertainty of bus arrival time point prediction,The main contents were as follows:(1)The uncertainty of the prediction results of bus arrival time was researched,and the uncertainty of the prediction results was quantitatively analyzed by using interval prediction method.The Bayesian estimation method was selected to construct the prediction interval by comparing several commonly used interval prediction methods.The running process of the bus was analyzed,and the driving time between adjacent stations was taken as the interval prediction object.(2)This paper preprocessed the problems such as disorder,anomaly and absence in the original GPS data,and extracted the travel time between adjacent bus stations by using the processed data.Based on the analysis of the factors affecting the prediction object,the input data of the bus arrival time prediction model was determined,which was the basis for the later construction of the prediction model.(3)A BATIP-GPR model was proposed,and the actual trajectory data was used to predict the bus arrival time in working days.At the same confidence level,the interval prediction results of this model had a narrower interval range than that of an interval prediction model based on SVR combined with lower and upper bound estimation(SVR-LUBE).Compared this model with SVR-LUBE model and a traditional SVM-based point prediction model,the MAPE of this model was reduced by 0.475% and 0.566% respectively.The experimental results showed that the proposed model has the advantage of quantifiable forecasting uncertainty compared to the traditional point prediction model,and had better prediction performance compared to the interval prediction model.(4)A BATIP-BNN model was proposed,and then compared it with the BATIP-GPR model.At the same confidence level,the range of the prediction interval was narrower,and the MAPE of the point prediction results was reduced by 1.264%.Experiments showed that the BATIP-BNN model had more accurate point prediction results and more reliable interval prediction results,that is,the prediction performance had better prediction performance.