Load Forecasting Method for Electric Vehicles Based on CNN-GRU-Attention Model

With the gradual increase in the number of electric vehicles (EV), how to accurately predict the charging load to ensure the security of the power grid and the reasonableness of the scheduling has become an urgent problem to be solved. Aiming at the problems of low accuracy of electric vehicle charging load prediction under the influence of multiple factors and the loss of important information due to the long time sequence of load sequence, this paper proposes a load prediction model based on convolutional neural network, gated recurrent unit and attention mechanism (CNN-GRU-Attention). Firstly, an input matrix including date, holiday, temperature and historical charging load data of charging stations, is constructed and CNN is used for load feature extraction. Then, the extracted features can be fully learned and the data dimensionality is reduced with the GRU. Finally, to reduce the loss of important information, the attention mechanism is considered to perform weighted averaging of feature information, with a result of realizing the load prediction of electric vehicles. Taking the public dataset provided by a U.S. public utility as a practical example, the proposed method in this paper has higher prediction accuracy compared with other models.