Application of neural network model incorporating attention mechanism in concrete temperature prediction

Advance prediction is essential for intelligent control of concrete curing. The maturity method for estimating the strength value of concrete is a common non-destructive testing approach. The accuracy of this method relies on the precise prediction of concrete temperature. Therefore, it is necessary to conduct research on the prediction of concrete temperature. This study proposes a concrete temperature prediction model that integrates a convolutional neural network (CNN), a long short-term memory network (LSTM), and an attention mechanism to realise the advanced prediction of concrete strength by combining the maturity method. The model improves the time-series feature extraction capability by focusing on the critical time period of concrete temperature and humidity. The proposed prediction model outperformed the reference model, with a mean absolute error (MAE) below 0.02°C. The difference between predicted and test strength values remained within 2.91 MPa. The proposed model was applied in concrete tests under different curing conditions and all demonstrated good predictive performance, indicating that the model has good robustness. For the problem of intelligently predicting concrete strength in the field, it provides front-end technical support for intelligent curing control of concrete in winter construction and novel insights for non-destructive testing of concrete strength.