Abstract: In order to be able to assess the risk of different fire hazards in cruise ship cabins, a new neural network model that can classify the fire risk level of cabins in real time is proposed. A physical model of cruise ship cabin fire was established by FDS（Fire Dynamics Simulator）, and the safety indexes such as smoke temperature, CO volume fraction and visibility during fire were numerically simulated, and the fire risk level was classified into four levels based on the degree of its impact on human body. Then, a novel Attention-BP Neural Network model is designed for analyzing the collected multi-source fire information and classifying the hazard levels of different cabins in real time, which integrates the diagnosis results of multiple neural network models through the self-Attention mechanism, adaptively distributes the weight of each BP neural network model. Experimental results show that the proposed Attention-BP Neural Network model can effectively realize the early warning of fire risk level, and the classification accuracy of this proposed model achieves 97.32%. Compared with other machine learning algorithms, it has the highest stability and accuracy, and reduces the uncertainty of cabin fire early warning.