基于实测数据与机器学习的混凝土罐体预应力损失预测方法研究OA

To address the difficulty in accurately monitoring and predicting the prestress loss of long-distance prestressing strands in large-scale structures due to duct friction,anchor deformation,and stress relaxation,a comprehensive prediction method integrating fiber optic sensing measurement data and machine learning is proposed,based on a liquefied natural gas storage tank project.This method uses fiber optic sensing technology to conduct long-term monitoring of long-distance prestressing strands in actual engineering,obtaining prestress loss data.Based on the measured data,a numerical simulation dataset of long-term prestress loss is established.Multiple machine learning algorithms are used to construct prediction models,and the SHAP interpretability algorithm is further introduced to analyze the prediction basis and key influencing factors of the models.Experimental and simulation results show that the total losses of the circumferential and vertical prestressing strands of the storage tank are approximately 390 MPa and 320 MPa,respectively.The average percentage errors of the six established machine learning models in both the training set and the test set are below 3.5％,among which the XGBoost model has the smallest error,achieving 1.19％on the test set.Through SHAP analysis of multiple influencing factors,the length of the prestressing strand has the most significant effect on long-term loss,followed by the elastic modulus of concrete,the elastic modulus of the prestressing strand,and the tensioning control stress.The research shows that machine learning models can accurately predict prestress loss and analyze the important factors causing prestress loss.This study is of great significance for the accurate evaluation of the service performance of prestressed structures.