Given the nonlinear power load, diversity of prediction conditions, subjectivity of parameter setting of the prediction model, etc., an ISSA-GRU (ISGU) hybrid model based on the combination of a date mapping method based on daily average load with strong adaptability, a Gate Recurrent Unit (GRU) with high non-linear fitting performance and an improved sparrow search algorithm (ISSA) with strong search ability are proposed for short-term load forecasting (STLF). First, the data mapping method based on daily average load is used to map the week-holiday factor to solve the problem that it is difficult to input the prediction network because of non-digitization. Then, highly relevant eigenvalues are selected from many correlated factors to deal with the diversity of prediction conditions. Finally, the GRU network is constructed for load forecasting, and the ISSA algorithm is used to configure GRU network parameters objectively. To verify the effectiveness of the ISGU hybrid model, we use the Singapore power load data experiment, and compare the experimental results with the existing algorithms. The results show that this method has good performance for STLF and effectively improves the accuracy of STLF statistical standards. This work is supported by the National Natural Science Foundation of China (No. 51977039).