Abstract: Existing research on charging load prediction of electric vehicle （EV） lacks accurate descriptions of user travel behaviors and traffic conditions. Therefore, a spatio-temporal graph attention network is constructed to learn and predict the spatio-temporal distribution of travel demand considering urban points of interest and road traffic flow, taking into account the effect of date type, weather temperature and traffic events. The Dijkstra algorithm based on the travel time index （TTI） is used to obtain the shortest travel time. An EV energy consumption model that takes into account the impact of traffic conditions and air temperature, and a charging station selection decision model that considers distance and comprehensive charging cost, are both established. According to the actual travel demand and traffic data of the second ring area in Xian, the charging demand of electric vehicles for private cars, taxis and internet-hailed vehicles is predicted, and the changes in travel demand are analyzed for charging stations in various grid spaces in the city. The EV charging load prediction provides a reference and basis for the planning of charging facilities.

Key words: electric vehicle, charging load, spatio-temporal map attention network, urban points of interest, travel demand forecast