Machine learning prediction of mass concrete temperature using random forest

Many methods have been proposed for predicting mass concrete temperature using various machine learning and deep learning approaches that are highly dependent on the available construction and temperature monitoring data. However, achieving accurate and stable temperature forecasting remains a major challenge due to dynamic thermal behavior, environmental variations, nonlinear heat evolution, and limited monitoring samples during mass concrete construction. In this work, we present a data-driven temperature forecasting methodology that utilizes Temporal Fusion Transformer (TFT), N-BEATS, and Quantile Regression to effectively analyze and learn temperature evolution patterns from historical concrete temperature data, improving prediction accuracy and forecasting consistency. The proposed system performs high-precision short-term temperature prediction for mass concrete structures. By utilizing significant temperature-related features, the model provides reliable and efficient forecasting for real-time thermal analysis and supports better understanding of temperature behavior during the curing process. Additionally, the system focuses on enhancing model performance, reducing unnecessary data complexity, minimizing overfitting, and improving generalization capability on unseen temperature data. The proposed framework can efficiently handle construction-related temperature datasets, ensuring scalability and stable prediction performance, ultimately supporting intelligent thermal control and improving the long-term safety and durability of mass concrete structures.

Keywords: Mass Concrete Temperature Prediction, Temporal Fusion Transformer (TFT), N-BEATS, Quantile Regression, Temperature Forecasting, Machine Learning, Deep Learning, Thermal Analysis, Concrete Temperature Evolution, Structural Safety, Construction Monitoring, Data Analysis.