Abstract: When distributed power generation, primarily represented by wind and photovoltaic energy, is connected to the distribution network, it tends to exacerbate system voltage fluctuations and readily causes local voltage to exceed permissible limits. In order to ensure the accuracy of the solution after relaxation, the relaxation technique represented by the second-order cone optimization has extremely strict constraints, and the dual gap will become larger under extreme power injection. To solve the above problems, this paper proposes an active-reactive power optimization strategy for active distribution network based on improved convex internal approximation method. First, based on the branch power flow equation, the convex hull of nonlinear terms related to branch current, node voltage and apparent power flow is defined, which is then combined with the remaining linear elements to form convex internal approximation, and a mature mathematical model is established. Furthermore, the network tolerance criterion and the KKT condition are introduced to shrink the dual gap of the optimal solution. Finally, the IEEE 33-node test system and the PG&E 69-node test system are used for simulation calculation, and the output interval of each decision variable in the power grid is obtained by using Yalmip and other platforms, which verifies the stability of optimization, accuracy of relaxation and efficiency of calculation.