Prediction Of Bus Arrival Time Based On Spatio-temporal Information Neural Network

With the improvement of national economic strength and scientific and technological level,people’s living standards have been significantly improved,and the number of private cars has also increased significantly,which makes the pressure on road traffic become greater and greater.The large increase in the number of private cars will inevitably bring about many problems such as traffic congestion,environmental pollution and even traffic accidents,which will have various negative impacts on people’s lives.It is very important to develop intelligent public transportation system and record the running data and monitoring information of bus in real time for efficient bus operation and scheduling.At the same time,it is conducive to the realization of economical public transportation system,avoiding the waste of resources and promoting the green and sustainable development of urban public transportation.Intuitively speaking,there are many random factors that affect the arrival time of urban public transport vehicles,such as weather,traffic conditions,intersections,morning and evening rush hours,and so on.Therefore,using the real-time running data of public transport vehicles to study the prediction model of vehicle arrival time and provide the accuracy of vehicle arrival time prediction can greatly facilitate passengers’ travel and greatly improve users’ riding experience.Firstly,this paper gives the current technical application and research status in the field of bus arrival prediction at home and abroad,and compares and analyzes the existing achievements.The theoretical basis,advantages and disadvantages of several common arrival time prediction techniques are emphatically analyzed.Secondly,based on the original data processing and feature vector acquisition,a neural network based on spatio-temporal information is proposed to predict the arrival time of bus.The GPS data of public transport vehicles in Tianjin from December 1st to December15 th,2018 are used for model training.Compared with other forecasting methods using single bus data,the input of this paper uses more GPS data and the data is more complete.At the same time,the running status of all vehicles on the same route is used to express the current road condition implicitly,which reduces the difficulty and complexity of feature value extraction.Finally,the model in this paper is compared with Simple NN and XGBoost models.According to the error distribution chart,the difference comparison chart between the predicted value and the true value,the average absolute error and the root mean square error.The results show that the forecasting model proposed in this paper has better performance and more accurate forecasting effect,and is not affected by factors such as working days and peak hours.Compared with other models that only analyze a single line,the forecasting model in this paper has wider applicability.