The widespread application of Distributed Generation (DG) and Electric Vehicles (EVs) puts higher requirements on the optimal operation of an active distribution network, and distribution network reconfiguration is one of the key technologies to cope with this. Considering the uncertainties of DG, EV and other loads, the correlation of random variables generated by Latin Hypercube Sampling (LHS) is considered. Then, a bi-layer optimization model is established by minimizing the integrated operation cost and the system nodes’ voltage difference. Finally, the Improved Harmony Search Algorithm (IHSA) based on the strategy of breaking up the whole into parts is used in dynamic reconfiguration of distribution network in different time intervals. Through the simulation of an IEEE33 node system, this paper proves that the above method can effectively improve the economics and reliability of distribution network operation while enhancing the utilization rate of renewable energy. This work is supported by National Natural Science Foundation of China (No. 61672337).