Monthly Load Electricity Prediction Method Considering Extreme Weather Factors Using LSTM

Forecasting annual and monthly electricity consumption is essential for ensuring the safe and stable operation of power systems. However, this domain faces two significant challenges: the extensive time spans involved in medium- to long-term load forecasting lead to relatively scarce available data, and extreme weather factors have a substantial impact on load prediction accuracy, which existing prediction algorithms often fail to adequately address. To address these challenges, this paper introduces an innovative data augmentation method that utilizes the similarity patterns of inter-annual electricity loads to effectively expand the dataset needed for training prediction models. Furthermore, considering the fluctuation characteristics of extreme weather loads, this study proposes an impact factor correction prediction method based on normal values. By establishing a quantitative relationship model between extreme weather loads and normal loads, the proposed method significantly improves the accuracy of load predictions during extreme weather events. To verify the effectiveness and superiority of the proposed method, this study performed case simulations and result analyses using historical data from an actual power grid. The results demonstrate that, compared to traditional prediction methods, the proposed method effectively enhances the accuracy of annual and mediumterm load forecasting, providing more reliable data support and a scientific basis for power system planning and operational decision-making.