In order to accurately predict the short-term load change of a power system and provide guidance for safe, economic and efficient operation, a load forecasting method based on fuzzy clustering and random forest regression is proposed. A rough set is used to construct the compensation rules, and the prediction results are modified and compensated for. First, this paper analyzes the periodicity and weather correlation of power system load. Historical samples are clustered using C-mean fuzzy clustering. In the random forest regression prediction, similar data after clustering is used as a training set sample to build a decision tree. Taking into account the conservatism of partial random forest regression prediction and large fluctuations of power system load at the peak, the rough set theory is used to generate compensation rules after the prediction results are obtained, and load forecasting is modified. The 24-hour load forecasting of the Northern Ireland region using the above method shows that the Mean Absolute Percentage Error (MAPE) is 2.09% compared with the actual load, which verifies the effectiveness of the forecasting method. This work is supported by National Natural Science Foundation of China (No. 51877136) “Key Technologies of New-type Electrical Energy Router Adapt to Distribution Network Flexible Interconnection”.