Abstract: Photovoltaic power plants are prone to degradation from module aging, environmental disturbances, and equipment failures during long-term operation. Traditional inspection methods often suffer from latency and limited accuracy in fault detection and localization. To address these challenges, this paper proposes a fast fault localization technology for photovoltaic power plant operation and maintenance based on an intelligent diagnosis system. An equivalent power deviation model and a health index model are constructed to mitigate environmental interference, while multi-source feature fusion combined with probabilistic localization criteria enables efficient fault identification and localization at the module, inverter, and system levels. Furthermore, collaborative algorithms are introduced to enhance recognition accuracy and response timeliness. Experimental validation on an actual photovoltaic power plant demonstrates that the proposed method significantly reduces detection time, improves localization accuracy, and lowers operation and maintenance costs.