Research On Data Predicting Methods For Bus Dynamic Scheduling System

Intelligent Transportation System (ITS) is attracting more and more people’s attention due to its advantages such as efficient passenger traffic organization pattern, fast and flexible response capability, perfect passenger information service and so on. As a core part of ITS, dynamic scheduling has a significant impact on operating cost, efficiency and service level in the public transport operation system.The current dynamic scheduling mostly depends on the experience of dispatchers, which is unstable, less reliable and unscientific because dispatchers are lack of data basis and blind to future passenger flow, route information and bus arrival time.Based on the current situation that dynamic scheduling is lack of data basis, this thesis studies the changes rule of passenger flow data, route data and bus arrival time data, which have an influence on the bus scheduling. Besides, this thesis establishes forecasting models which can predict the changes of passenger flow data, route data and bus arrival time in the future, and provide the data support for dynamic scheduling. The main research work has been carried out as follows:Firstly, factors that affect bus arrival time are analyzed, and finally passenger flow and real-time route information are determined as two research factors. Then, by analyzing the spatial and temporal distribution of passenger flow data and real-time route information data, data predicting ideas are put forward in this thesis.Secondly, via utilizing the periodicity and regularity of the passenger flow data and the traffic flow speed data, the traditional time series ARMA model is improved and a predicting model based on seasonal ARIMA is established. Then, the actual collected data is used to validate the passenger flow and traffic flow speed predicting models. By comparing with traditional predicting methods, seasonal ARIMA predicting model in this thesis has better prediction effect. The MAPE for passenger flow and traffic flow speed are 15.9% and 6.84% respectively, which are both within the acceptable range.Thirdly, as the passengers and drivers have their own behavior habits, there may exist certain ’self-repetition’ in the model of traffic information. This thesis takes advantage of the characteristic of ’self-repetition’ to establish the predicting model based on K-NN nonparametric regression, and studies the effect of the key parameters in the model such as the back coefficient K and the number of neighbor m. Under the optimal parameters of the model, the predicting MAPE for passenger flow and traffic flow speed are 26.1% and 22.5% respectively.Lastly, this thesis adds a dynamic adjustment section in the model while taking full account of the periodicity of bus arrival time data. The influence of passenger flow on bus stop time and the influence of traffic flow speed on route running time are analyzed, and an arrival time predicting model considering the front bus data and dynamic adjustment section is established. Then, the data generated by the simulation model is employed to validate the predicting model and the predicting results are:the predicting MAPE is 11.5% when the shift departures after 30 minutes; the predicting MAPE is 10.6% when the shift departures after 60minutes; the predicting MAPE is 5.6% when the shift departures after 90 minutes. Compared with the traditional predicting model based on GPS data, the model in this thesis has better prediction effect. Finally, three factors including fluctuation of data, change of weight and influence of the front bus data are considered in the parameters experiments, and the optimal parameters are found in a particular state.