Abstract: To achieve accurate description and rapid response to civil aviation cabin fire scenarios, this paper constructs a cabin fire case retrieval model that integrates knowledge meta and aggregated K nearest neighbor algorithm. By mining the attribute features of cabin fires, performing numerical conversion and preprocessing, similar case retrieval and level prediction are carried out. The results indicate that based on the theory of disaster element deconstruction, cabin fire cases can be deconstructed into a four-dimensional knowledge element framework containing causative factors, disaster bearing bodies, disaster prone environments, and emergency response. Combined with the characteristics of fire events, the entire process of cabin fire disaster formation, development, and response can be subdivided into 8 sub knowledge elements and 12 attribute features; At the same time, the Euclidean distance between the randomly predicted target case and the total number of cases was calculated. The majority voting method and aggregation sub model were used to obtain the final prediction value. A dataset of 1 500 simulation cases was constructed for training, validation, and optimization. 24 real cases were selected for testing, and the best K value of 3 was obtained through cross validation using the leave one method. The model's prediction accuracy reached 0.916 7. Based on accuracy, precision, recall, and F1 score settings, a multi model comparative evaluation shows that the proposed aggregated K nearest neighbor model is significantly superior to other models. Therefore, the proposed cabin fire case retrieval model can provide support for predicting the level of civil aviation cabin fires and emergency decision-making.

Key words: cabin fires, knowledge meta model, aggregation K nearest neighbor algorithm, case retrieval, event level