Abstract: Load forecasting plays an important role in the planning and economic and secure operation of power systems. However, the abnormal data in load series will result in forecasting model misspecification or incorrect model parameters estimation. So a hybrid monthly load forecasting model based on seasonal adjustment and improved holt-winters is built to solve such problems. Firstly, after seasonal adjustment, the final seasonally adjusted series where outliers or holidays effects have been removed and seasonal component series can be obtained simultaneously; secondly, the improved Holt-Winters method is used to forecast final seasonally adjusted component, and virtual regression equation is used to forecast seasonal component. Finally, the final forecasting result can be obtained by using forecasting result of seasonal component and seasonal adjusted component jointly. The case calculation results show that the proposed method can significantly improve the prediction accuracy and the forecasting performance is better than seasonal Holt-Winters, SARIMA, neural network, and support vector machine. In summary, the proposed model can be practically applied as a monthly load forecasting tool.

Key words: monthly load, Holt-Winters, seasonal adjustment, load forecasting