Ultra-short-term Load Forecasting Method Based on Improved iTransformer of Decomposition-Reconstruction

Accurate load forecasting is crucial for the safe and stable operation of power systems. In order to further improve the accuracy of load forecasting, an improved iTransformer ultra-short-term load forecasting method based on decomposition-reconstruction is proposed. Firstly, the load data is decomposed using time varying filtering-empirical mode decomposition (TVF-EMD) to obtain several intrinsic mode functions (IMFs). The IMF components are then reorganized based on their entropy values using sample entropy (SE). Secondly, the iTransformer neural network is improved by introducing a stand alone attention (SAA) mechanism to replace the self-attention mechanism in the iTransformer encoder, which effectively enhances the model's ability to capture the dependencies of different variables. Finally, the restructured components are input into the improved iTransformer for prediction, and the obtained results are superimposed to obtain the final prediction value. The actual active load dataset of 220kV substations in a region of China is taken as an example for validation and comparison with the existing mainstream models, and the results show that the prediction method adopted in this paper has better prediction performance.