The invention provides a cloud cluster movement estimation method facing to photovoltaic power prediction. The cloud cluster movement estimation method for the photovoltaic power prediction comprises the following steps: preprocessing an image of a ground-based cloud picture; thresholding the ground-based cloud picture; detecting and marking a cloud cluster and extracting center of mass of the cloud cluster; and estimating moving speed and direction of the cloud cluster. By the cloud cluster movement estimation method for the photovoltaic power prediction, each cluster in a time sequence image is tracked and studied to determine the moving speed and direction of the cloud cluster in the future, so that the cloud cluster movement in the future is estimated, and an important data source is provided for minute-grade power prediction within a super-short period of 0-4 hours for a photovoltaic power station.