Abstract: At present, research on predicting the charging load of electric vehicles has not solved the problem of not being able to obtain real-time origin destination (OD) data in practical applications and considering the actual perception and decision-making psychology of car owners. In response to this issue, based on comprehensive consideration of dynamic traffic information, environmental temperature, real-time vehicle flow, queuing theory, and other methods, this paper establishes a new in-depth learning framework for urban transportation system OD flow prediction, and completes the spatiotemporal distribution prediction of electric vehicle charging load. This paper firstly analyzes the impact of various factors such as urban traffic network information, day type, and weather on the driving rules of electric vehicles. The starting and ending points of corresponding private car and taxi driving behaviors are obtained through a bidirectional short-term memory recursive neural network algorithm. Secondly, a link impedance and node impedance model considering dynamic traffic information and intersection flow is introduced, as well as an air conditioning energy consumption model considering ambient temperature and vehicle real-time speed. The real-time Dijkstra algorithm is used to plan the minimum travel cost travel path for the starting and ending points of electric vehicles, simulating the driving behavior of electric vehicle users. Finally, path planning experiments and charging demand prediction experiments for different types of electric vehicles are completed in different application scenarios. The results show that the spatiotemporal distribution characteristics of the charging demand obtained are consistent with the objective demand.

Key words: electric vehicle, dynamic traffic information, charging load, neural network