With the rapid development of distribution networks, power users are demanding more and more from power supply reliability.Aiming at the constraint of switch operation times in existing dynamic reconfiguration of distribution network and the distribution network power flow distribution impact of each node of the system and considering the shortage of load identification in a single clustering algorithm, a two-layer clustering algorithm is proposed based on morphology and amplitude. On outer layer, the sample data is clustered with Pearson correlation coefficient as the performance evaluation index. On inner layer, each cluster obtained from the outer clustering is clustered with Euclidean distance function as evaluation index. Multi-objective optimization mathematical model is established, which can reduce power loss of network, increase voltage stability, balance load of feeder, and minimize operation times of all switches. Multi-objective dynamic reconfiguration of distribution network is finished by improved particle swarm optimization. Simulation results show that the times of switching operation is decreased by 54.76% compared to static reconstruction, the power reduction is 15575.4 kW.h, decreased by 39.28%, the voltage deviation index is decreased by 49.1% before reconstruction, and the load imbalance is improved by 41.9%. The accuracy of the improved double-layer load clustering is improved by 11% compared with FCM clustering, and the clustering effect is closer to the original data. It is verified that the dynamic reconfiguration can effectively improve the reliability of distribution network running. This work is supported by National Natural Science Foundation of China (No. 51667013).