Analyzing the temperature evolution during the construction of mass concrete and establishing accurate prediction models are essential for ensuring structural quality and construction safety. This paper proposes a temperature prediction model for mass concrete based on Random Forest Transfer Learning. By integrating data preprocessing and feature extraction, the model leverages a transfer learning mechanism to enhance generalization performance under small–sample conditions, enabling high–precision short–term temperature forecasting. Experimental results demonstrate that the proposed model achieves significant predictive accuracy, with a coefficient of determination (R2) exceeding 95%. Compared to conventional machine learning models, the model exhibits superior robustness, with prediction residuals confined within the range of –2 to 2°C. Furthermore, the predicted values for unmonitored locations closely align with observed temperature patterns. This high–fidelity prediction enables engineers to implement proactive thermal control measures, effectively mitigating the risk of structural cracking and enhancing the long–term safety and durability of mass concrete structures.
Gao et al. (Thu,) studied this question.