Aiming at the problem of system state estimation under false data injection attacks, the mathematical model of power system was established according to the third-order model of generator and the model of automatic voltage regulator, taking the cyber-physical power system as the research object. The exponential smoothing method was used to predict the measured value, and by comparing the predicted value with the actual measured value, it detected whether there were false data injection attacks in the system. If the detection results determine that the system being subjected to false data injection attacks, the predicted value is used instead of the bad data input state estimation algorithm to restore corrupted data cansed by these attacks. Combining the exponential smoothing method with the cubature Kalman filter algorithm, an improved cubature Kalman filter algorithm was proposed to estimate the state of the system. Taking a typical five-machine power system as an example, the simulation results show that the proposed method can effectively prevent the adverse effects of false data on system state estimation.