Abstract: Large-scale photovoltaic（PV） power plants can be formed of numerous braches and laterals interconnected by cables and overhead power. When a fault occurs, it is hard to locate the fault. The conventional fault location methods for distribution networks are challenging to apply within the PV power plants, because centralized PV units have the different structure and controls from those of distributed PV. Therefore, a new fault location method was proposed. With the control adopted in PV inverters, the characteristics of the fault currents within the plant were analyzed firstly. Then, exploiting that PV units provide no negative-sequence currents, the node negative-sequence voltage equations which consist of the rows with measurements, was integrated with the compressive sensing theory. The Bayesian compressive sensing reconstruction algorithm was used to recover the sparse injection current vector, in which the nonzero element was used to locate the fault. The proposed method does not require the synchronous measuring information and need only short data. The results from a simulation system with 68 nodes indicate that this method can locate the faults efficiently and immune to fault types and fault resistances, as well as has a good anti-noise performance.