With the development of the artificial intelligent technology, intelligent distribution load forecasting technology has been paid much attention. Based on the Generalized Linear Model, this article uses the historical load data of different distribution transformers as original data to establish the short-term load prediction model for the future maximum daily load of the distribution network. To choose the optimal prediction model, through model simplification and variable selection process, the k-fold Cross-Validation method is introduced. Model test results show the accuracy of the proposed method, which supports the optimization of the distribution network operation and maintenance plan.