An effective method for the dynamic reactive power optimization of distribution network is proposed based on the accurate characteristics of distributed generation (DG). According to the operation control modes of different types of DG, PQ, PV, PI and PQ(V) are properly processed by means of second-order cone relaxation, and the non-convex AC power flow equation is approximately transformed into a convex second-order cone constraint. In addition, the optimization objective function is configured to drive the slack error close to zero, thus transforming the dynamic reactive power optimization problem of distribution network into a convex mixed-integer second-order cone programming problem. In this way, the solution complexity is reduced and the optimization accuracy is improved, which ensures global optimization and branch voltage and current safety constraints while considering energy storage devices. In particular, discrete reactive power compensation capacitors act on the distribution network constraints. With the minimum active power loss in multiple periods as the optimization objective, a dynamic reactive power optimization model for distributed network is built, which takes into account the accurate characteristics of DGs. Based on the DG output and load curves predicted during the scheduling period, an IEEE 69-bus system is simulated, and results verify the effectiveness of the proposed method.