False data injection attack is one of the important factors that threaten the safe and stable operation of power grid. The research on attack strategy and the in-depth analysis of attack mechanism can provide a basis for improving system defense strategy. Aiming at the state estimation process in the smart grid, the attacker can calculate the attack vector that satisfies the power flow constraint when he has certain resources and the attack conditions, so as to avoid bad data detection. Compared with the DC state estimation model, the attack conditions are more severe and the success rate is relatively low under the AC state estimation. This paper considers the situation that the attacker does not know the exact state variate under the AC state estimation. Firstly, the state variate is approximately calculated according to the known measurements and parameters, and then the attacker finds out the leverage measurements by the projection statistics and the hat matrix diagonal element method. Finally, an optimization model is established to attack the leverage measurements, and simulations on IEEE 14-node system verify the effectiveness of the proposed attack method.