Prediction of the durability of high-performance concrete using an integrated RF-LSSVM model

• Published in: Construction & building materials Vol. 356; p. 129232
• Main Authors: Liu, Yang, Cao, Yuan, Wang, Lei, Chen, Zhen-Song, Qin, Yawei
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 21.11.2022
• Subjects: Chloride ion permeability coefficient; High-performance concrete; Least squares support vector machine; Prediction of permeability; Random forest; Random forest; Chloride ion permeability coefficient; High-performance concrete; Least squares support vector machine; Prediction of permeability
• ISSN: 0950-0618; 1879-0526
• DOI: 10.1016/j.conbuildmat.2022.129232

•The predction framework of RF-LSSVM is obtained.•The model can be applied to a wider range of concrete research fields.•The prediction results can provide reference for the optimization design of concrete mix.•The RF-LSSVM can be used for diversifying scenarios. In extremely cold and complex marine environments, strong concrete chloride ion penetration resistance is the key to ensuring the durability of buildings (structures). Quickly and accurately predicting the resistance of concrete to chloride penetration is critical to optimizing the concrete mix proportions. A hybrid intelligent prediction model that integrates random forest (RF) and least squares support vector machine (LSSVM) algorithms is proposed to rapidly and accurately predict the resistance of high-performance concrete (HPC) to chloride penetration. The initial index system of HPC chloride penetration resistance was established from substantial research and practical projects. An RF is employed to screen the initial index parameters and to provide an optimal index set for predicting concrete resistance to chloride penetration. Based on a sample data set, the LSSVM is implemented to establish high-precision predictions regarding the resistance of concrete to chloride penetration. The results indicate that (1) the RF method effectively screens important indicators and provides the optimal data set for the prediction of concrete resistance to chloride penetration and (2) the developed RF-LSSVM hybrid intelligent approach can effectively and accurately predict the resistance of concrete to chloride penetration. For the test set, the RMSE and R2 of the prediction model reach 0.0491 and 0.941, respectively, representing accuracy better than that typical for machine learning algorithms. The proposed RF-LSSVM hybrid intelligence model can provide a basis for optimizing the concrete mix proportion and can be applied in practical projects to help solve similar problems.