A Patch-wise Coarse-Fine Grain Fusion Prediction Networks for Short-term Household Load Forecasting

Accurate household-level load forecasting is crucial for the stability and flexibility of smart grids, facilitating efficient electricity distribution and the integration of renewable energy sources. Capturing the temporal dependency in household load data serves as the cornerstone for gaining insights into fundamental household load patterns. However, the limitations of a single methodology and the volatility and uncertainty due to seasons, weather and consumer behavior still make it difficult to capture time dependence. To address these constraints, we propose a novel Patch-wise Coarse-Fine Grain Fusion Prediction Network. This network captures temporal dependency at both coarse-grained and fine-grained levels through two branches, thereby achieving a comprehensive understanding of household load consumption patterns. Experiments conducted on real datasets demonstrate that the performance of the proposed method outperforms various existing short-term load forecasting methods.