Abstract: With the continuous promotion of electric vehicles (EV), its strong random charging load poses challenges to the operation of the distribution network. To improve the reliability and economy of the distribution network operation, an interval prediction method for the EV charging load based on scene generation with multiple correlation days was proposed. First, the Spearman rank correlation coefficient was used to analyze the correlation between the EV charging load on the day to be forecasted and those on the historical days being examined to find the historical days that have strong correlations with the day to be forecasted. The original multiple correlation day charging scene set (OMCDCSS) was constructed to describe the charging behaviors of EVs. Then, large numbers generating multiple correlation day charging scenes (GMCDCS) with similar probability distributions and different timing distributions to the original multiple correlation day charging scenes set were obtained based on the beta-variational auto-encoder (β-VAE). Finally, appropriate scenes with high relevance to the EV charging load data of the known historical days were selected from the generated set to construct a relevant scene set. Based on the data average and data interval of the last day of the relevant scene set, the deterministic and interval prediction results of the EV charging load for the day to be forecasted were obtained. The contrast experiment proves that the proposed method provides a more reliable prediction interval and narrower prediction interval width.