Bus Arrival Time Prediction Model And Empirical Study

Giving priority to the development of public transport is an effective method to change the traffic growth pattern, improve the efficiency of resource utilization, reduce traffic pollution and ease traffic congestion, and it is the principal thinking of the urban transportation development in our country. The key to the development of public transport is providing passengers with good services, and accurate bus arrival time prediction can provide better services by promoting reasonable arrangements for passengers and real-time deployments for the buses.This paper matches the GPS location data of bus operation with the bus line, obtains the bus location information per second by interpolating, gets the relationships between buses according to the sequence of the buses arrive at the stops, and analyzes bus operating characteristics and the factors affecting the bus arrival time systematically.There are two limitations of the existing researches:rarely explicitly study on bus dwell time and passing time of the intersection; seldom consider the influence of road traffic flow state on bus arrival time. Therefore, this paper first proposes a bus arrival time prediction model based on the ε-SVR (ε-Support Vector Regression) without intersections, using genetic algorithm to optimize the model parameters and input variables, and verifies the validity of the model based on the data of Line300internal respectively according to the weekday peak periods (7:00-9:00/17:00-19:00), flat periods, and the weekends. The prediction results indicate that using the traffic detector data can effectively reduce the prediction error. Average relative prediction error of the three periods can be reduced by at most1.55%,1.50%and1.25%respectively.Further, this paper puts forward a bus arrival time prediction model with intersections and verifies the validity of the model based on the VISSIM simulation. The simulation results reveal that the prediction model can obtain high prediction accuracy when the traffic flow is unblocked and slow-moving. As for the congestion state, the average relative prediction error is relatively large for the secondary queuing phenomenon may happen.