Current power system state estimation and control methods are susceptible to false data injection attacks, which decrease the system stability. This paper considers state estimation of power systems under attack based on dynamic fusion. First, the initial state estimation of the power system is obtained and an attack model is constructed to simulate the impact of false data injection attack. Then, the twin delayed deep deterministic policy gradient algorithm is employed to correct the state estimation under attack. Meanwhile, the alternating direction method of multipliers framework is employed to impose the optimization constraints considering the sparseness of the state space vector. Moreover, a dynamic fusion method based on the inverse of the estimation error is proposed to improve the accuracy of the state estimation. Finally, by comparisons of the simulation, it’s demonstrated that the proposed method has better state estimation performance.