Research on Performance Prediction of Ultra-High Performance Concrete based on Machine Learning

Ultra-High Performance Concrete (UHPC) has become a key material in the construction field due to its excellent performance, but traditional methods face numerous challenges in predicting its performance. In recent years, machine learning (ML) has provided a new path for UHPC performance prediction with its advantages, and related research has shown a multi-dimensional development trend. This paper reviews the latest progress of machine learning in the field of UHPC performance prediction, covering basic models, ensemble and meta-learning frameworks, and small data solutions, and uses methods such as SHAP and LIME to reveal the relationship between material components and performance. By analyzing the advantages and limitations of different technologies, it provides a reference for researchers in technology selection. Studies have shown that machine learning significantly improves the accuracy of UHPC performance prediction, and interpretability methods help analysis the mechanism of material action. Small data solutions and other technologies promote the implementation of related technologies. In the future, it is necessary to break data barriers, integrate physical constraints, develop multi-scale modeling, and combine explainable artificial intelligence and generative AI to promote UHPC design into an intelligent era of "on-demand customization", realize digital intelligent management and control throughout the life cycle, and provide support for green and low-carbon buildings.