Deep Online Recommendations For Electric Taxis By Coupling Spatiotemporal Graph Neural Network And Reinforcement Learning

As an important part of urban public transportation,the taxi industry has adopted environmentally friendly and efficient electric taxis as an effective measure to achieve emission peak and carbon neutrality.Compared with traditional internal combustion engine vehicles,electric taxi has longer charging time and charging queuing time,which reduces the operation efficiency of electric taxi.Moreover,The spatiotemporal dynamic changes of taxi travel demand and charging facilities service capacity in the city,and the long-term influence of different decision-making actions in the operation process,makes the decision of electric taxi operation more complicated.To this end,this study constructs an electric taxi operation optimization model based on spatiotemporal graph network and reinforcement learning to improve the operational efficiency of electric taxis in the following three aspects.1.This study extracts residents’ taxi travel demand based on taxi trip data,analyzes the spatiotemporal distribution pattern of travel demand,and uses multi-source data sets to analyze the correlation between travel demand and multiple external factors.Based on the analysis of taxi travel demand,the travel demand prediction is performed by modeling the spatial correlation of travel demand with graph convolutional networks,modeling the global temporal correlation of travel demand with attention mechanism,and taking into account the influence of multiple external factors.By considering the spatiotemporal variation of future travel demand,the electric taxi operation is intelligently optimized.2.Based on the prediction and spatiotemporal variation of travel demand,this study establishes a spatiotemporal strategy for electric taxi operation,including cruising,charging and reward strategies.For the complex operation process of electric taxis,a Markov framework is constructed to adapt to the complex operation characteristics of electric taxis,and the operation behavior of electric taxis-travel demand-charging facilities is portrayed.Finally,an improved reinforcement learning model adapted to the complex operational characteristics of electric taxis is established in conjunction with the spatiotemporal strategy to realize intelligent optimization of electric taxi operations.Through experimental verification,the operation optimization model proposed in this study is significantly better than the comparison methods.3.Based on the operation optimization of electric taxis,this study analyzes the business and demand characteristics of the taxi operation platform,and implements the prototype system of intelligent operation of electric taxis based on the Web framework by designing different functional modules,which has certain practical application value.In general,this study follows the steps of "prediction-optimization-application" for the intelligent optimization of electric taxi operation.This study establishes the travel demand prediction model and electric taxi operation optimization model,and designs and implements the intelligent operation system of electric taxi.The research results can promote the promotion and application of electric taxis and help build an energy-saving,environmentally friendly and intelligent transportation.