AbstractWith the rapid development of photovoltaic (PV) power generation, state recognition of PV arrays is of great significance to its intelligent operation and maintenance. But the random fluctuation of PV power output leads to the obvious uncertainty for performance calculation of PV arrays; this paper proposes an evaluation method of PV arrays. Firstly, the performance calculation method and its distribution model were proposed by the kernel density estimation (KDE) method. Secondly, the probability density distribution (PDD) of different indicators was analyzed to determine the suitable observation indicator so as to introduce the stability indicator, from which the benchmark array was selected. Finally, the state recognition of PV arrays was achieved. The main conclusions and innovations of this paper are the stability evaluation method of PV arrays based on the non‐parameter estimation method and the effective operating state recognition method for PV arrays. The proposed method was proven to be more reliable by comparing it to the power generation ranking method (PGRM). The effectiveness was verified by comparing the statistical and real‐time distribution of the performance ratio, and the excess rate of different confidence levels and Euclidean distance were introduced to demonstrate the accuracy of the proposed state recognition method.