Multi-Task Photovoltaic Power Forecasting Based on Autoformer

This paper presents a multi-task photovoltaic power forecasting algorithm based on Autoformer, designed to jointly forecast short-term (hourly) and long-term (daily) power generation. By leveraging a multi-task learning framework, the method captures dynamic patterns across different time scales, improving forecasting accuracy for both immediate fluctuations and long-term trends. This approach enhances energy scheduling and grid stability, addressing key challenges in photovoltaic power forecasting.