Abstract: A photovoltaic string online anomaly fault detection method based on current data of string combiner boxes is proposed in this paper. This method calculates the probability statistical characteristics of offline string current data and computes the discrete anomaly coefficient of online data in real time to identify abnormal faulty strings. The method first collects historical current data sets of each photovoltaic string during normal operation from the combiner boxes, performs statistical analysis on the historical current data sets to obtain adjacent data sets for each data point, calculates the similarity coefficient for each data point, and derives the empirical probability density function of the similarity coefficient. It then identifies the neighborhood data set of the current string current data and calculates its discrete anomaly coefficient. If the coefficient exceeds the threshold, the photovoltaic string is deemed abnormal. When the coefficient continuously accumulates beyond a certain value, the string is diagnosed as faulty, and the location of the faulty string is determined based on the discrete distance. This invention can reduce maintenance and inspection time, accurately identify faulty strings in real time, and minimize power generation losses caused by string faults.