Using PV string data to diagnose failure of solar panels in a solar power plant

Solar power plants have methods of diagnosing the failure of their solar panels. In this paper, we propose a remote automatic diagnose failure method that uses PV string data. Some string measurement devices are used for continuous remote monitoring of solar power panels. Solar panels generate low power due to panels breaking, shadows from structures, weeds, etc. If these failures can be classified by fault classification using remote string measurement data, a reduction in unnecessary repair actions and more efficient preparations are possible. We use machine learning to automatically classify causes of decreased solar power generation, such as broken panels, shadows, or weeds. We applied this diagnosis method to some large solar plants. The results of our experiment proved that the learning accuracy of this proposed method is 100%. When we tested this classification method in another solar plant, the classification accuracy was 99%. This result suggests that this diagnosis method is useful and can decrease maintenance work.