As a key technology for typhoon disaster risk mitigation introduced by the insurance industry at present, similar typhoon retrieval has obvious shortcomings in the core link of insurance business-disaster loss assessment. To address this industry pain point, this study comprehensively considers the actual data situation of the insurance industry and constructs an artificial intelligence algorithm for similar typhoon retrieval suitable for insurance practice: Specifically, an intelligent retrieval algorithm system is built based on the multi-dimensional characteristics of typhoons to achieve minute-level dynamic retrieval; furthermore, it integrates an insurance industry-specific retrieval database constructed with a 100,000-year typhoon random event set, significantly expanding the retrieval coverage. Experimental verification shows that the response time of the proposed algorithm reaches the minute level. The insurance loss assessment value obtained by target retrieval based on the optimal paths of multiple sets of representative typhoons has a deviation of ≤ 5% (Top1) from the actual insurance loss. When dealing with typhoon forecasting scenarios with uncertainties, dynamic retrieval is conducted for different forecasting stages, and the relative error of the estimated insurance loss is ≤ 25% (Top1), which can effectively envelop the range of actual insurance losses, thereby reducing the uncertainty in the insurance industry's disaster loss prediction process under the background of typhoon disasters. The research indicates that the intelligent retrieval algorithm proposed in this study for catastrophe insurance meets the industry application standards in terms of efficiency, accuracy, and practicality, and provides an innovative solution for typhoon disaster risk management.