Abstract:Deep learning method can help to learn the deep features of power load data and improve the accuracy of prediction, but it also brings problems such as difficult to adjust super parameters and poor interpretability of the model. To solve these problems, this paper introduces the deep forest model for short-term load forecasting. Based on the multi-Grained Cascade forest model, the multi-granularity window scanning method is improved, and the window size and sliding step size are adjusted, so that the model can extract the periodicity characteristics of power load data in different time scales.In addition, the calculation method of deep forest output layer is improved, and the output result is improved from discrete class vector to continuous predicted value, so as to improve the accuracy of the model. Finally, the feasibility and effectiveness of the proposed method are verified in the measured data of northeast China power grid.According to the experimental results, the improved deep forest algorithm can achieve higher accuracy with higher prediction accuracy, and has faster learning speed than the deep neural network.