Research on a prediction model for compressive strength of CFRP-confined concrete based on Bayesian optimized CNN-LSTM network
TL;DR
The results show that the optimized model significantly outperforms traditional methods in prediction accuracy, achieving a coefficient of determination (R2) of 0.957 and a root mean square error (RMSE) of 7.42 MPa on the test set.
Abstract
This paper proposes a hybrid model (Bayes-CNN-LSTM) integrating Bayesian optimization, a convolutional neural network (CNN), and a long short-term memory (LSTM) network for the intelligent prediction of the compressive strength of carbon fiber-reinforced polymer (CFRP)-confined concrete. Based on 518 sets of experimental data, the model uses CNN to extract local correlations among features and LSTM to capture underlying dynamic evolution, while Bayesian optimization is employed for adaptive hyperparameter tuning. The results show that the optimized model significantly outperforms traditional methods in prediction accuracy, achieving a coefficient of determination (R2) of 0.957 and a root mean square error (RMSE) of 7.42 MPa on the test set. This study provides an efficient and reliable intelligent framework for predicting the mechanical performance of FRP-confined concrete.