Abstract
This study proposes a rolling multi-objective optimization framework that integrates CNN–BiLSTM–Attention-based solar radiation forecasting for operation optimization of multi-energy heating systems. The framework coordinates the control of solar collectors, air-source heat pumps, gas boilers, and thermal storage tanks to mitigate operational challenges caused by solar irradiance uncertainty and load fluctuations. By incorporating hourly irradiance forecasts, the NSGA-II rolling optimization algorithm is employed to minimize operating costs, carbon emissions, and temperature deviations, with back-testing conducted using real-world meteorological data. The results demonstrate that this forecast-driven strategy significantly outperforms an ideal sunny-day dispatch strategy, reducing total operating costs by up to 48.1% and carbon emissions by up to 84.3%.