Short-term Forecasting of Electricity Load UnderCold Wave Weather Conditions with Small SampleSets

Introduction: During cold wave weather conditions, significant changes in load patterns pose considerable challenges to the safe operation of modern power systems. Providing accurate load forecasting under these extreme weather conditions is thus crucial. Methods: In this paper, we propose a short-term electricity load forecasting method that considers small sample properties of cold wave weather conditions. In the proposed method, the definition and analysis of cold wave events are initially presented to elucidate load characteristics. Then, considering the scarcity of sample data during cold wave weather conditions, a sample-based transfer learning approach is proposed to enrich the dataset. Based on the enriched dataset, short-term forecasting models for baseline load and incremental load during cold wave events are established separately using the BPNN and XGBoost algorithms, which are integrated to form the final forecasting model. Results: Empirical validation using real-world data from cold wave events demonstrates the superior prediction performance of the proposed method compared to conventional forecasting approaches under such extreme weather conditions. Discussion: The proposed method shows strong adaptability under cold waves, and helps address the limitations of traditional models that struggle with data scarcity. Conclusion: This method enhances the stability of the power system and provides support for its safe operation.