Provided is a photovoltaic power predication method based on a ground-based nephogram, comprising the following steps: step 1, estimating the movement of a cloud cluster; step 2, predicating ground irradiance; step 3, establishing a comprehensive photoelectric conversion model; step 4, predicating photovoltaic power. The present invention comprehensively utilizes a variety of digital image processing technologies, highlights the demands of problems, has strong relevance and practicality, and, in combination with a conventional statistical approach, provides effective technical support and methodological reference for refined predication of photovoltaic power. In addition, the present invention establishes a model in multiple steps by considering a plurality of factors, and the description of the method covers all key aspects of the field of photovoltaic predication, thus having good universality and robustness. The present invention reserves adjustment parameters for some aspects for specific application scenarios and application objects in use, thus facilitating operation and ensuring predication effect.