In the process of intelligent mine construction, it is essential to analyze massive mine monitoring data from a global perspective so as to observe the mine safety situation comprehensively. To address this issue, a multi-level safety situation awareness system for mines is proposed. The system analyzes the monitoring data of each subsystem within the mine area by deploying a local safety situation awareness model in the fog computing facility to observe the local safety situation of the mine. The local safety situation is gathered to the cloud computing facility through the high-speed communication network of the mine. The global safety situation is further achieved by the global safety situation awareness model deployed in the cloud computing facility. The local and global safety situation awareness model processes the correlation between the data by using the encoder and decoder based on the gated recurrent unit. The Attention mechanism is applied in the model so as to filter the key data and improve the model’s computing speed. Meanwhile, in order to make the model run in the best state, the particle swarm algorithm is used to find the optimal hyperparameters of the model. The simulation results show that the safety situation awareness model has high accuracy.