Due to their relatively enclosed characteristics, tunnels facilitate rapid spread of fire and accumulation of harmful smoke gases, posing a significant threat to the safety of human lives and property. The heat release rate (HRR) is a critical parameter that characterizes the scale and severity of a tunnel fire incident. In order to predict the HRR and meet the demand for rapid real-time feedback in tunnel fire scenarios, this study was conducted on three tunnel fire datasets established through numerical simulations and on-site experiments. Then, a deep neural network model, Informer, was applied to train a predictive model for tunnel fire HRR for the first time based on the datasets mentioned above. The results indicate that this method which applies the Informer model demonstrates high accuracy. In short-term predictions, the R² values on multiple tunnel datasets exceeded 0.85. This study contributes to aiding rescue personnel in promptly obtaining information on the subsequent development trends of a tunnel fire incident.