Research On The Control Method Of Bus Antitandem In Network Environment

In urban road bus operation,the bus crosstalk phenomenon reduces the stability of bus operation and affects the development of urban public transportation.The use of reasonable anti-crosstalk control methods can effectively reduce the occurrence of crosstalk,significantly improve the efficiency of bus operation and increase the satisfaction of the public with the mode of bus travel.With the development of netconnected technology,the study of anti-crosstalk control methods for public transportation in a net-connected public transportation environment has become one of the important means to improve the efficiency of public transportation operation.This thesis explores and researches the anti-crosstalk control method for public transportation in the net-connected bus environment using reinforcement learning methods,and achieves certain results.The main work is as follows:(1)Research on the combined control method of bus stopping and flow restriction based on a single intelligent body.First of all,in order to solve the traditional deep learning algorithm estimates too high and slow convergence,DQN algorithm based on improved priority empirical playback,based on this establishment,the establishment of a single-intelligent body system of single-line network link bus anti-crosstalk control model,the bus arrival station,arrival time,the number of people on board,the number of waiting passengers and other bus recently left the station as input,considering the entire bus line The headway time distance of each vehicle,the difference between the bus headway time distance and the average value of headway time distance of all bus vehicles on the line and the passenger waiting time are used as reward functions,and the time to make extra stops or limit the number of people boarding the bus is selected as the action of the intelligent body,and the utilization rate of training samples is improved and the convergence speed of the algorithm is accelerated by improving the priority experience replay strategy,and it is compared with the bus based on the headway time distance deviation as the threshold The effectiveness of the proposed method is verified by comparing with the stopping strategy,flow restriction strategy and combined stopping and flow restriction strategy.(2)A multi-intelligence-based inter-station speed control method for networklinked buses.Firstly,based on the multi-intelligent body proximal strategy optimization algorithm(MAPPO),each vehicle in the bus line is treated as an intelligent body,the basic elements of multi-intelligent body reinforcement learning state space and action space are defined,and a reward function consisting of three components,namely forward headway deviation,bus running speed and passenger waiting time,is designed to improve headway stability and reduce passenger waiting time on the basis of avoiding long travel time between bus stops.On this basis,the introduction of a multiheaded attention mechanism allows the intelligences to pay more attention to the most useful information for themselves in the learning process and avoid the influence of other intelligences’ decisions on themselves,thus improving the training efficiency.Based on the simulation environment built to simulate the above method,the results show that the method in this paper can effectively reduce the bus headway deviation and passenger waiting time compared with the stop-to-stop speed control method based on the minimum headway deviation,the maximum speed allowed to travel between stops method and the basic MAPPO algorithm.