In order to solve the problem of large prediction errors in existing power load forecasting methods, a short-term power load forecasting model based on improved spotted hyena algorithm optimized-extreme learning machine (ISHO-ELM) is proposed. Firstly, the quasi-inverse learning strategy and elite strategy are introduced into the spotted hyena optimization algorithm to improve the search capability of the algorithm, and the effectiveness of the algorithm is verified by the benchmark function. Secondly, using the algorithm of ISHO optimization of ELM random parameters in order to improve the prediction accuracy and stability of the model. Lastly, the advancement and practicability of the constructed short-term power load forecasting model are verified by the actual measurement data. The results show that the fitting coefficients of the proposed ISHOELM model are 1.6% and 1.7% higher than those of the existing ELM and SVM models, respectively. This study is of great significance to improve the stability of power system operation.