Abstract: In order to reduce the influence of load fluctuation characteristics on the overall operation trend of load in short-term load forecasting, a short-term load combination forecasting method oriented to fluctuation type fine division and clustering is proposed. Firstly, k-means++ is introduced to cluster the annual load according to its daily characteristics, and the clustered daily load is divided into typical load periods; Secondly, based on the idea of rain-flow counting method, the fluctuations in the typical load period are divided and combined with the fuzzy c-means clustering algorithm (FCM) to cluster the load fluctuations based on the characteristics of load fluctuations. Further, considering the relationship between the key variables and the process of load fluctuations, a fast correlation-based filter (FCBF) is applied to filter the characteristics of the corresponding correlation factors under each load fluctuation. Finally, a short-term load combination forecasting model with the daily load fluctuation and load reconstruction optimal feature set as the input and the load power as the output is established. Practical examples show that the proposed short-term load combination forecasting method can significantly improve the accuracy of short-term load forecasting.