Research On Commuting Private Car Ride-sharing Matching Based On Electronic Registration Identification Data Of Motor Vehicles

With social progress and economic development,private car travel has become the primary travel mode for residents.The rapid increase of private car ownership has brought severe traffic problems.These problems would become particularly prominent in the morning and evening peak hours,owing to large-scale commuting.Therefore,ride-sharing for commuting,which can help reduce traffic flow,is often considered an effective means to alleviate traffic problems during peak hours.Electronic registration identification(ERI)of motor vehicles is an emerging means of traffic information acquisition based on Radio Frequency Identification(RFID),tracking all the motor vehicles in the city,including private car.Therefore,this thesis employs ERI data to research ride-sharing matching for commuting private car based on reinforcement learning.During the research,we meet two major difficulties:1.How to identify commuting private car.2.How to carry out commuting private car ride-sharing matching with an overall perspective instead of focusing on instant satisfaction.Around these two issues,our primary work consists of the following aspects:(1)This thesis proposes a commuting private car identification model based on the random forest.The identification model takes the travel regularity as the input feature.The travel regularity is described from two aspects: the whole travel and single travel attributes.The regularity of the whole travel is obtained by the information entropy,while the regularity of single travel attributes is obtained by the coefficient of variation.Through related experiments,the proposed model is verified to have good identification performance.Its AUC(area under ROC curve)reaches 94.43%,better than other models.Finally,the trained model is employed to identify 231,190 commuting private cars.On the basis of identified commuting private car,the commuting travel mode of private car is analyzed,and the influence of commuting trips of private car is explored.(2)This thesis proposes a data-driven ride-sharing matching model based on reinforcement learning to carry out commuting private car ride-sharing matching with an overall perspective instead of focusing on instant satisfaction.The model takes the driver’s spatiotemporal information as the state,takes the four designed ride-sharing matching actions as the action,selects the opposite number of the average ride time of the ride-sharing participants as the reward function and ε-Greedy algorithm as the strategy function.The model is trained by Deep Q Network.Finally,a ride-sharing matching experiment is carried out on the ERI data set of Chongqing City.Then,we evaluate the ride-sharing strategy from four aspects: average ride time sum,total vehicle travel time,total vehicle travel distance,and the number of vehicles.The result shows that the number of commuting private car on roads in the morning and evening peaks decreased by 21.45% and 21.01%,respectively,after ride-sharing matching.