Evolution and model prediction of mechanical ventilation temperature field in high-geotemperature tunnels: Experimental analysis and machine learning

Some problems remain in the mechanical-ventilation cooling of high-geotemperature tunnels, such as the lack of a basis for ventilation parameter design and unclear cooling effect. These significantly affect construction progress and personnel safety. Therefore, based on a self-developed ventilation and cooling test platform for high-geotemperature tunnels, the evolution laws and prediction of the temperature field in tunnel under mechanical ventilation were studied. With a focus on two key factors, the surrounding rock temperature and ventilation wind speed of high-geotemperature tunnels, 20 types of ventilation cooling tests were designed for dry-hot high-geotemperature tunnels. Through a cross-sectional monitoring of key points, including the crown, shoulder, and side wall in the tunnel, the cooling effect of the longitudinal ambient temperature and working face area of the tunnel were studied. The results show that mechanical ventilation can effectively reduce the ambient temperature inside high-geotemperature tunnels, and the temperature drop is positively correlated with both rock temperature and wind speed. However, the cooling effect of the tunnel was limited at specific wind speeds, and ventilation alone does not result in a continuous decrease in temperature. Therefore, when surrounding rock temperature is 40 °C and the ventilation speed is 4.4 m/s, the temperature of the tunnel face area in the tunnel can be reduced to 28 °C or below. When the temperature of the surrounding rock exceeds 60 °C, ventilation alone cannot ensure that the temperature in the tunnel is suitable. On this basis, taking the historical monitoring data of the tunnel test as input parameters, a method for predicting the ambient temperature of high-geotemperature tunnels ventilation is proposed, which integrates convolutional neural network (CNN) and bidirectional long short-term memory network (BiLSTM). This realizes the prediction of the ambient temperature for the tunnel ventilation in the future. The results show that the regression value ( R 2 ), mean absolute error ( MAE ) and root mean square error ( RMSE ) of the ventilation environment temperature prediction model based on pearson correlation coefficient feature screening and CNN-BiLSTM model are 0.94,1.39 and 1.68, respectively. The error between the prediction results and the experimental monitoring values is small, and it has good prediction performance and generalization ability. These findings have practical significance for the design of ventilation duct layouts and cooling strategies in high-geotemperature tunnel constructions.