Research On Bus Arrival Time Prediction And Control Strategy Under The Connected Vehicles Environment

With the development of society and economy,there is an increasing degree of motor vehicles privatization and traffic congestion in urban areas.Since the 1980s,transit priority has been an economical,feasible and effective measure to alleviate traffic congestion and air pollution in many national policy documents.A large number of scholars have focused on the study of public transport priority as well.However,due to the complexity of urban road environment,the bus operation is subject to a lot of interference.The traditional measure of public transport priority is the construction of traffic infrastructure,such as bus lanes.The function of operations management is ignored.Therefore,the real transit priority is challenged.The evolution of information perception technology and vehicle-infrastructure cooperation systems can help us obtain various traffic information of bus vehicles and relevant infrastructures.The research about arrival time prediction and dynamic control of the bus system,which has important practical significance,can provide accurate travel information for passengers and decision support for bus system operators.As a result,the thesis is devoted to research on bus arrival time prediction and dynamic control under the environment of connected vehicles.Firstly,a platform is designed for bus dynamic control under the connected vehicles environment.It contains four sub-systems for data collection,data storage,data processing and data releasing.Protocols of data and communication are well explained for the platform,as well as the functions of sub-systems.The platform is designed to implement the methods of arrival time prediction and dynamic control of the bus system,which also clearly illustrates the definition of the connected vehicles environment in this research.Secondly,influence factors of bus arrival time are combed,and 20 original features are chosen for feature selection using the conventional method(analysis of variance and Spearman rank correlation analysis)and the new method Boruta.After the analyses of the data of Bus Route 1 in October 2014 in Zigong City,Sichuan Province,9 features are selected as input variables for the prediction models.It can be proved that feature selection can significantly improve the prediction accuracy of machine learning models,such as Support Vector Machine and Artificial Neural Network.Thirdly,some advantages and disadvantages are compared for serveral machine learning models,especially for Support Vector Machine and Artificial Neural Network.The part of the thesis proposes a new ensemble learning model called Bagging-ANN which is based on Artificial Neural Network.The sampling method is bootstrap,and the weights of base learners are decided by out-of-bag data.The prediction accuracy and stability of Bagging-ANN are much better than traditional machine learning models.The proposed model can still have satisfactory prediction accuracy even using the small sample data.Finally,a hybrid real-time station control strategy is proposed for the bus system,which combines both holding and skipping methods at the bus stations to get steady headways between bus vehicles.The key parameters for holding and skipping are set for chosing appropriate times to implement the strategy.Simulation is conducted for searching optimal key parameters of a high-frequency bus route in middle-sized and small cities.Comparing with groups without control strategies,the proposed strategy has a promotion by 11.43%for the evaluation index when considering passengers’benefits,as well as 8.46%when considering bus operators’ benefits.The proposed strategy proves to be effective.