Abstract: In response to the complex environment in which train fires occur and the inability to ensure the effectiveness of fire recognition, this paper designed and implemented an intelligent monitoring and early warning system for train fires. The paper introduced the system architecture and functions such as real-time fire monitoring and warning, 3D visualization display, reliable connection between vehicle terminals and dispatch centers, multi-user access, and permission management. Based on the edge computing node deployment scheme, by collecting four types of data, such as image, temperature, wind speed, and smoke concentration, and using the multi-mode fusion network model based on MLP (Multilayer Perceptron), this system implemented the fusion of multi-source heterogeneous data. The simulation experiment results show that the system can effectively achieve real-time monitoring and early warning of train fires.