Abstract: This paper proposes a method for the simultaneous fault diagnosis of PV array operation in a small training sample size scenario. Firstly, the PV array fault feature vetor is extracted by using the steady-state PV array electrical signal output time series to demonstrate that this feature vector represents different PV array states such as the normal state, the open-circuit fault, the line-to-line fault, and the partial shading. Secondly, due to the changing operating environments of PV arrays, a data processing method is suggested to convert the actual output into a uniform operating condition. Further, the linear discriminant analysis method is combined with the biased covariance estimation and the common singular value decomposition in a small sample size scenario. This method addresses the singularity of the sample covariance matrix and the difficulty of directly solving the discriminant function caused by the high dimension and the small sample size. Finally, an experimental platform is built on the roofs of a university in Shanghai to collect the experimental data in different PV array states. The experimental results verified the necessity of the proposed data processing method for the steady-state electrical signals application, and the applicability of the fault diagnosis algorithm to the small sample size scenario.