A Parameter-free model for long-term concrete creep prediction

This paper presents a novel approach to predicting long-term concrete creep performance by leveraging early creep observations and eliminating the need for many sensing devices to collect material parameters. Our hybrid encoder-decoder model integrates a Convolutional Neural Network (CNN) for feature compression and noise reduction and a transformer for learning long-distance dependencies and global feature understanding. Additionally, the adaptive data completion method is designed to handle varying data intervals, ensuring robust predictions despite irregular sampling. Our results indicate that the proposed model significantly improves the consistency and reliability of creep predictions and comprehensively understands concrete material creep behavior over time.