In traditional mine fire simulation, the FDS simulation software has been verified by large-scale and full-size fire experiments. The resulting calculations closely align with real-world scenarios, making it a valuable tool for simulating mine fires. However, when a fire occurs in a mine, utilizing FDS software to predict the fire situation in the mine entails a sequence of steps, including modeling, environmental parameter setting, arithmetic, and data processing, which takes time in terms of days, thus making it difficult to meet the demand for emergency decision-making timelines. To address the need for rapid predictions of mine tunnel fire development, a method for swiftly estimating environmental parameters and the concentration of causative factors at various times and locations post-fire has been devised. FDS software was employed to simulate numerous roadway fires under diverse conditions. Parameters such as fire source intensity, roadway cross-sectional area, roadway wind speed, roadway inclination angle, time, and others were utilized as the input layer for a neural network. In contrast, wind flow temperature, carbon monicide (CO) concentration, fire wind pressure, visibility, and others were designated as the output layer for training the neural network model. This approach established a fire prediction model to resolve issues related to time-consuming numerical simulations and the inability to provide a rapid response to disaster emergencies. The trained neural network model can instantaneously predict the environmental parameters and concentrations of the causative factors at different times and locations. The model exhibits an average relative error of 12.12% in temperature prediction, a mean absolute error of 0.87 m for visibility, a mean absolute error of 3.49 ppm for CO concentration, and a mean absolute error of 16.78 Pa for fire wind pressure. Additionally, the mean relative error in density is 2.9%. These predictions serve as crucial references for mine fire emergency decision-making.