The 0–4-h very-short-term solar irradiance forecasting is crucial for the intra-day scheduling of photovoltaic power plants. This study employs four cloud advection methods—including three optical flow algorithms and a block matching algorithm—to achieve intra-day solar irradiance forecasting based on the China’s Fengyun-4A satellite. To reduce the systematic biases introduced by different algorithms, a weighting method that combines forecasts from different cloud advection methods is used. For each advection method, a two-step approach is used: (1) the REST2 clear-sky model and the Heliosat-2 semi-empirical method are utilized to retrieve the global horizontal irradiance from the Fengyun-4A, and (2) various advection methods are employed to compute the cloud motion vector fields, allowing extrapolation for the forecast fields. All forecasts are verified against high-quality ground-based observations. The results indicate that the proposed ensemble forecasting scheme exhibits optimal performance—the 0–4-h GHI forecasts have normalized root mean square error (nRMSE) ranging from 14.08% to 22.32% and correlation coefficient ranging from 0.96 to 0.87 for—which reduces nRMSE by 2.71% in the 4-h forecasts compared to the well-tested Lucas–Kanade optical flow, and by 6.02% compared to the least-effective block matching algorithm.