Abstract: In view of the low accuracy of load forecasting caused by the uncertainty of power load time-series features and the poor model interpretability of existing forecasting method,the paper proposes a forecasting method for short-term load in distribution station area based on the combination method of discrete wavelet transform（DWT）,self-organizing feature map（SOM）and hierarchical fuzzy system（HFS）. Firstly,the raw load sequences in the time domain are decomposed into several components in the frequency domain by using the DWT,selecting the components as the features for the load clustering. Then the SOM algorithm is used to cluster the raw load data and divide it into different data groups with the features. After that,the HFS model is used to predict the components in all types of data groups,and the prediction results are overlapped to obtain the final load forecasting value. Finally,the simulation analysis is done with the actual load data from a distribution station area. The results demonstrate that the proposed method can effectively improve the load forecasting accuracy and has the reasonable model interpretability.