Abstract: Since hypergraphs are capable of portraying higher-order interaction phenomena prevalent in complex systems compared to ordinary graphs, and because when a component in a system fails, it often affects other components related to it, thus causing cascading failure phenomena with large component damage, this paper first introduces a cascading failure model on hypergraphs, and then further proposes a new disintegration algorithm based on deep reinforcement learning. Intelligent agent is used to continuously try different node selection strategies in a small-scale synthetic hypergraph, and adjust its own behavior according to the rewards received, and finally learns the optimal hypergraph disintegration strategy. And the algorithm is validated in several datasets.