Abstract: In recent years, typhoons and rainstorms have increasingly become the typical natural disasters that affect the safe and stable operation of distribution networks in coastal cities. To effectively cope with the influence of extreme disasters, a multi-stage resilience enhancement strategy is proposed for distribution networks, considering the spatiotemporal uncertainty of typhoon rainstorm disasters. Firstly, the disaster scenarios are simulated according to the historical data of typhoons, and the failure probability of distribution lines is calculated to construct an uncertainty set for the typhoon rainstorm disaster attack. On this basis, combined with the non-anticipativity characteristics of the power system scheduling process, a multi-stage robust optimization model based on the uncertainty set above is established. Secondly, a pre-extended robust dual dynamic programming algorithm is developed for solving. Then, the pre-disaster defense strategies and well-trained worst-case cost-to-go functions during disasters are obtained, which can solve the emergency response strategies for each period of extreme disasters. As a result, the optimal resilience enhancement strategies for distribution networks have been made in an organic combination of offline training and online application. Finally, the test systems for distribution networks with different scales are used to verify the effectiveness of the proposed model and method.