When faults occur in microgrids (MGs), MG reconfiguration is required. To improve the efficiency of reconfiguration, a hierarchical optimization method for MG reconfiguration is proposed, where deep belief networks (DBNs), a typical model for deep learning, is introduced, which offers a way to avoid calculating power flow in the process of the MG reconfiguration. In this method, the first level optimization is obtaining the on/off status of breakers based on the supply-demand balance in MGs. In the second level optimization, the trained DBNs are used to obtain network losses and node voltages, etc., when the on/off status of breakers, are given to the DBNs, which is a good way to reduce time of reconfiguration. In addition, an evaluation method is also proposed, where the analytic hierarchy process (AHP) is employed to adjust the weights of preferences in a comprehensive evaluation function, so it offers the diversity of decisions for the MG reconfiguration. Finally, to test the proposed method, a modified IEEE 33-bus system is built in MATLAB for simulations. Compared to those methods where power flow is recalculated under different on/off status of breakers, our method can obtain the same reconfiguration scheme, but the time of reconfiguration is only one-seventieth of that of other methods. Furthermore, in terms of the proposed evaluation method, reconfiguration schemes can be selected under different preferences.