Multinodes interval electric vehicle day-ahead charging load forecasting based on joint adversarial generation

The spatial-temporal distribution of electric vehicle (EV) charging load has strong randomness and is affected by battery capacity and user behavior. In addition, the multinode charging load in the distribution network has differential correlations. A multinode charging load joint adversarial generation interval-forecasting method considering the spatial correlation of the charging load between nodes is proposed to effectively forecast the spatial-temporal distribution of EV charging load. First, the multinode joint charging scenario is constructed. Under the spatial charging load matrix, the spatial-temporal correlation between multinode charging loads in the joint charging scenario of the forecast day and the historical day is analyzed. According to the strongcorrelation historical-day multinode joint charging scenario of the forecasting day, the original multinode multiple-correlation-day joint charging scenario set, describing the charging behavior of multinode EVs, is determined. Second, a Wasserstein generative adversarial network with a gradient penalty is used to characterize the strong randomness of the spatial-temporal distribution of the charging load. A large number of joint charging scenarios with similar probability distributions but different timing distributions from the original scenario set are generated to obtain the potential spatial-temporal distribution of the multinode joint charging load. Then, based on the weighted two-dimensional correlation coefficient, the strong-correlation joint scenario set on the day to be forecast is selected from the generated multinode multiple-correlation-day joint charging scenario set. Finally, according to the strong-correlation joint scenario set on the day to be forecast, the interval-forecasting conclusion of the multinode EV charging load is obtained. To verify the effectiveness of the new multinode charging load interval-forecasting method, the simulation experiment uses the measured charging load data of 32 charging stations in a region of Zhejiang Province. The comparative experiment demonstrated that the proposed method has more-refined intervals and higher coverage than state-of-the-art interval forecasting models. Among the evaluation indexes of the charging load forecasting results of the new method, the PICP value is higher than 90.4%, and the PINAW and MAPE values are lower than 32.1% and 17.7%, respectively. The new method overcomes the limitation that the total charging load obtained by each node's charging load forecasting method is much higher than the actual demand.

Automated summary

A method for forecasting the spatial-temporal distribution of electric vehicle charging load in a multi-node charging scenario is proposed. The method considers the spatial correlation between nodes and utilizes a generative adversarial network to characterize the randomness of the charging load. Simulation results demonstrate that the method outperforms state-of-the-art models in terms of interval prediction accuracy and coverage.