Microgrid Net Load Forecasting Method Based on Extreme Value Feature Interaction

Microgrids, by integrating distributed energy resources, significantly enhance the efficiency, reliability, and sustainability of power systems. Net load refers to the difference between the user load and the output of renewable energy in a microgrid. As the penetration of renewable energy increases, net load forecasting becomes increasingly important for the operation and management of microgrids. However, the highly volatile nature of net load presents a significant challenge for accurate forecasting. To address this issue, this paper proposes a net load forecasting model for microgrids based on adaptive feature fusion. The model uses a Temporal Convolutional Network (TCN) as the baseline, designing an Extreme Feature Interaction Network (EFIN) to capture extreme point features. It combines Convolutional Neural Networks (CNN) and an improved Attention mechanism to extract both local and global features effectively. Finally, the extracted features are input into the TCN through an adaptive fusion network to generate the forecast. Experimental results demonstrate that the proposed model outperforms traditional forecasting methods in terms of prediction accuracy and convergence speed, exhibiting superior accuracy and robustness.