Emergency logistics facility location is a critical component in the overall construction of emergency logistics system. Based on the graph neural networks, an emergency logistics facility location model was proposed, which comprehensively considered the network structure and geographic information of candidate sites. After the nodes of the emergency logistics network were encoded with their features, the model could address various practical factors affecting site selection and the structural information of complex networks. Feature maximization and feature balance were adopted as loss functions to guide the model's site selection, and the candidate addresses were ranked accordingly. The emergency logistics facility location problem in Shanghai was taken as a case study, and the city’s rail transit network was modeled and analyzed. Through rationality analysis of output results and comparison with classic models, it is verified that the proposed model can fully utilize the human-geographic information and network structural information of candidate sites to obtain better location results.