Performance Prediction and Optimization Design of Ultra-High Performance Concrete Based on Multi-Scale Residual Attention Convolution Network

The design of ultra-high-performance concrete involves complex nonlinear interactions among material proportions, curing conditions, and mechanical properties, posing significant challenges for rapid and reliable decision-making in engineering practice. Traditional trial-and-error methods are time-consuming and costly, whereas existing data-driven models often lack robustness under limited data conditions and cannot provide actionable decision support for engineers. To address these limitations, this study proposes a multi-scale residual attention convolutional neural network (MSRA-CNN) as an intelligent decision-support system for ultra-high-performance concrete performance prediction. The proposed framework provides a scalable solution for intelligent construction, particularly in scenarios characterized by tight project schedules, fluctuating material properties, or limited resources, thereby improving engineering efficiency and quality control in civil engineering practice.