Abstract: Frequent occurrences of extreme weather impose higher demands on the resilience of distribution networks, and traditional fault recovery strategies are inadequate to fully unleash the potential of active distribution networks. Therefore, a multi-stage fault recovery strategy for active distribution networks based on dynamic islanding is proposed. Firstly, an analysis is made on the feasibility region of distribution network reconstruction based on dynamic islanding, and with consideration of the temporal characteristics of distributed generations such as non-black start units (NBSU), an upper-level load restoration model is established to determine the multi-stage network topology plan. Subsequently, the lower-level model is established to provide the generation plans, which is used to examine the feasibility of the multi-stage network topology plan at the islanded flow level. And then, the established models are solved using the improved discrete particle swarm combined algorithm with the CPLEX solver. Finally, a 43-node urban distribution network is used for simulation to verify the effectiveness and superiority of the proposed method in improving resilience.