In the novel power distribution system, photovoltaic (PV) charging stations have attracted much attention as a typical distributed resource aggregation form. Since both distributed PV generation and charging loads are characterized by randomness and volatility, the load forecasting task of PV charging stations is particularly complex. In this paper, considering the dynamic impact of PV charging station access on the regional load profile, a PV charging station load forecasting method based on multi-layer limit learning machine and quantile regression theory is proposed. Firstly, the factors affecting the load of PV charging station are feature extracted and key feature quantities are extracted. Secondly, combined with the quantile regression algorithm, a multi-layer kernel limit learning machine deep neural network model is constructed to realize the load interval prediction of PV charging station under different confidence levels, and the improved sparrow optimization algorithm is used for the parameter optimization and the optimal model is selected for the load prediction. Finally, the load data of a photovoltaic charging station in a northern region of China is selected for example analysis. The results demonstrate that the proposed PV charging station load prediction method oriented to the novel distribution system achieves satisfactory forecasting performance and enables more accurate acquisition of load forecasting information.