With the application of sensor technology and IoT (Internet of things) technology in power systems, the power system has developed from the traditional physical system to a highly networked system, also known as the power CPS (Cyber Physical Systems). Simultaneously, many information security problems have been introduced. The existing research on information security of power systems mainly focuses on the network layer, while sensors in the physical layer of the power system are in face of information security issues as well. In this paper, we propose a false data injection method aiming at sensors of power system based on robust principal component analysis. Without mastering any information of the power system, our method allows attackers to structure false data attack vectors which is capable of bypassing bad data detection mechanism of power system only using eavesdropped sensor data. Compared with conventional principal component analysis methods, our method is of better robustness that it can effectively filter out most of the sharp noise in eavesdropped sensor data.