In order to eliminate the impact of emergency and spatial variability of urban waterlogging events on causal analysis, a framework for constructing the event logic graph and analyzing the causes of waterlogging based on the graph was proposed in this study. The rule template library was used to extract sentences containing causal events in Chinese urban waterlogging corpus, the events in the causal sentences were extracted based on the deep neural network fusion method with voting mechanism, and after that manual rules were combined to construct the event logic graph for urban waterlogging. Then, the event logic graph was used to generate scenes centered on the waterlogging point, which was further exploited to automatically generate and train the discrete dynamic Bayesian network. Finally, this study performed causal analysis based on this network. The result shows that the event logic graph for urban waterlogging well represents the mechanism of urban waterlogging evolution. In addition, the comparison of inferred results and real results shows that this method can accurately find the causes and eliminate the influences of pseudo-positive causes.