An efficient method for predicting temperature field of PC beams with CSWs using thermocouple numerical analysis and random forest algorithm

The significant difference in specific heat between concrete and steel results in non-uniform temperature fields in existing PC beam bridges with corrugated steel webs (CSWs) under combined environmental temperature, solar radiation, wind, and thermal radiation. This study proposed a numerical model using thermocouples for PC beam bridges with CSWs. The accuracy of this model was validated by comparing numerical analysis results with field measurements from the Tou-dao River Bridge in Sichuan Province, China. A training database for the temperature field was then constructed based on these numerical results. Three machine learning models, random forest, backpropagation neural network, and convolutional neural network, were employed to improve computational efficiency and address the limitations of numerical analysis in handling large datasets for temperature field simulation. Comparison of the coefficient of determination ( R 2 ) and mean squared error (MSE) for the three models indicated that the random forest model outperformed the other two in both computational efficiency and prediction accuracy. Finally, the SHapley Additive exPlanations (SHAP)function was used to interpret the trained random forest model, and the effects of girder height, bridge orientation, solar radiation, air temperature, and ground reflectivity on the vertical temperature gradient were investigated.