For multi-period dynamic var control optimization, a dynamic optimization model is used, in which the objective function is the minimum summation of network loss coat and regulating cost of controlling devices. A Fisher ordered clustering algorithm is employed to segment a load curve, making the load curve close to the actual load change. A two-phase optimization method is presented to acquire a multi-period coordinated optimization based on a modified genetic algorithm and a quasi-dynamic programming. In order to take into account both solution speed and accuracy, the number of optimal paths to be recorded for all stages can be set flexibly according to the system size, which is suitable for the dynamic var control optimization of a large-scale system. The results show that the proposed method is practical and effective.