Abstract: Dynamic reactive power optimization plays an important role in improving the voltage quality of power grid, decreasing the network loss and reducing the daily action times of discrete voltage regulators. Mathematically, it is a multi-period large-scale nonlinear mixed integer programming problem with absolute value constraints, and its efficient solution is a difficult problem. Therefore, this paper proposes a two-stage dynamic reactive power optimization algorithm based on decoupling interior point method and mixed integer programming. Firstly, the sigmoid function is used to deal with the absolute value constraint to realize the continuity of the original model, and the idea of decoupling interior point method is used to construct the diagonal band edge structure of the KKT modified equation, so as to realize the time block decoupling and efficient solution of the model. Secondly, the original model is linearized near the current continuous solution, and a mixed integer linear programming model involving all constraints of the original model is constructed, so as to determine the optimal solution of the discrete reactive power control equipment. The effectiveness of the proposed algorithm is verified through simulation of a 26 bus example.