Taxi Scheduling Based On Vehicle Sharing Network

With the development of national economy and the acceleration of urbanization,the level of motorization has been continuously improved.Urban traffic congestion is also increasingly serious.Taxi is an important part of urban traffic and its operational efficiency and capacity size directly affect the degree of urban traffic congestion.The operational efficiency is closely related to the capacity scale.When the operational efficiency increases,the capacity scale can be appropriately reduced.On the contrary,when the operational efficiency decreases,the capacity scale should be appropriately increased to meet the existing travel needs.From the perspective of vehicle sharing network,this dissertation optimizes taxi scheduling to improve taxi operation efficiency and reduce taxi capacity.This dissertation mainly studies the following issues.Firstly,the concept of vehicle sharing network is introduced,and the basic method for constructing vehicle sharing network is proposed.Secondly,a static scheduling method is constructed based on the historical taxi travel data and urban road network data.The historical taxi travel data are matched with the urban road network data,and the travel time estimation model is established by minimizing the travel time error.The differential evolution algorithm is used to solve the model.According to the schedulable conditions composed of the predicted length parameter,the empty driving time threshold parameter and the delay parameter,the taxi scheduling problem is transformed into the corresponding vehicle sharing network,and the equivalent minimum path coverage problem is solved by minimizing the number of vehicles.The Hopcroft-Karp algorithm based on bipartite graph matching is used to solve the vehicle sharing network to obtain the scheduling scheme and the required number of vehicles.The interval maximum overlap number problem is introduced to approximate the number of vehicles operating simultaneously in the road.Then,a dynamic scheduling strategy is proposed based on deep spatio-temporal prediction.The spatio-temporal visualization analysis of travel data is carried out by using spatio-temporal cube technology and dimension reduction technology.The multi-graph construction and multi-graph fusion mechanism are added to the existing spatio-temporal prediction graph network model of deep learning.Different graph structure matrices are constructed based on DTW distance,parameter learning and random generation methods,respectively.Different deep spatiotemporal prediction models are established to predict the spatio-temporal demand of future taxi travel,and guide the real-time dynamic taxi scheduling.Finally,relevant experiments are carried out on the taxi travel data set in Manhattan,New York.The results show that the scheduling method based on vehicle sharing network used in this dissertation can improve the efficiency of taxi operation without changing the travel habits of passengers.Under the condition of known future travel data,at least 4496 vehicles are needed in the time span of experimental data,and the maximum number of vehicles operating simultaneously in the road network is 4337.The results of spatio-temporal prediction of different models show that multi-graph construction and multi-graph fusion mechanism can significantly improve the prediction performance of the model.The mean square error loss of the model using DTW similarity distance and original spatial adjacency relationship is 9%lower than that of the original model using only spatial adjacency relationship.