Data-Driven Bus Arrival Prediction Model Under Dual-Mode Traffic Flow

The operating status of urban public transport vehicles is affected by the social traffic flow,public transport passengers,pedestrians at intersections,traffic lights and other traffic components under the dynamic urban traffic environment,and exhibits strong randomness in both time and space dimensions.Therefore,in the actual bus operation scheduling management,the use of more scientific and effective forecasting methods to improve the ability to control the uncertainty of future operation status is to improve the efficiency and stability of bus operation and achieve more precise and controllable high-quality bus services Key means.From the perspective of actual engineering application requirements,the deployment of predictive models needs to be comprehensively measured from two aspects: predictive capability and computational efficiency,and the predictive results should provide more information for reference to assist decision-making.In response to this problem,this paper establishes a mathematical modeling method for bus arrival prediction based on basic graph theory and Copula theory,revealing the influencing factors and prediction mechanism of bus arrival behavior from the perspective of dynamic,multi-model,and randomness.Developed a rattan structure prediction system with probability distribution prediction as the core,and sacrificed smaller prediction accuracy in exchange for higher prediction efficiency and real-time prediction capabilities.The specific work is as follows:First,the offline batch road network matching algorithm and floating car penetration ratio conversion are carried out on the multi-mode traffic data based on public transportation GPS to estimate the state parameters of social traffic and public traffic flow,and propose a three-dimensional random basic graph theory to reconstruct the random speed-Density model to analyze the state characteristics of mixed traffic flow on the road section;Secondly,use the non-parametric kernel density estimation-ML method to construct a binary Copula model,and input the speed parameters of traffic flow as random variables subject to certain density conditions into the four Copula models to solve the joint probability distribution,and analyze the bus speed and social car speed The dynamic correlation structure quantifies the impact of social car speed on bus speed under different mixed traffic density.Finally,extending from the binary Copula to solving the joint probability distribution of high-dimensional variables,three kinds of rattan Copula structures are used to construct the distribution curve prediction model,and the inverse function is used to solve the quantile prediction interval.The case analysis of the mixed density-speed model and the multi-section mixed speed model are carried out on the bus operation data of Huangpu Avenue in Guangzhou,and the prediction accuracy of the model is verified by using evaluation methods such as PICP and PINAW.The experimental results show that the multi-section mixed speed model based on Rten Copula has good performance in calculation efficiency and prediction accuracy.