Household Power Load Forecasting Under New Power System

Accurate power load forecasting is a critical task for the efficient operation of modern power systems, which can significantly improve control, planning, and operational levels. Household-level power load forecasting, however, presents significant challenges due to the volatility and diversity in consumption patterns across different users. To address this issue, we propose the CNNMixer model in this paper, which integrates multi-scale time series to capture the complex dynamics of household power loads. The CNNmixer model is composed of two key modules: the Past Information Module (PIM) and the Future Prediction Module (FPM). In the PIM module, time series data are decomposed into seasonal and trend components across multiple scales, and these components are mixed separately using a CNN-based architecture. The FPM module integrates the predictive capabilities of these multi-scale sequences to generate robust forecasts. Experimental results demonstrate that the CNNmixer outperforms existing models by addressing issues related to overparameterization and overfitting, leading to higher efficiency and precision in power load forecasting.