Abstract: With the frequent occurrence of extreme high temperatures during the summer, the loads have been growing dramatically and rapidly in recent years. As an important link to support the supply of electricity and the stable operation of power grids, the requirements for the accuracy of summer load forecasting have gradually risen. The current prediction algorithms are often not timely in tracking the rapid increase of load, and the prediction results are much lower than the actual value. Therefore, this paper proposes a two-layer optimized summer load forecasting model, which adopts a differential evolution algorithm to optimize the hyperparameters of LightGBM model in the inner layer, and optimizes the annual growth coefficients in the outer layer to reduce the impact of historical low load. The load data of the State Grid during the summer period of each year from 2021 to 2023 are selected for example verification, and the average absolute percentage error of prediction decreases by 1.82% compared with that of the LightGBM single model, which proves the validity and accuracy of the prediction model.