Developing a support vector regression model via optimization algorithms to appraise the hardness properties of high‐performance concrete

• Published in: Structural concrete : journal of the FIB Vol. 24; no. 3; pp. 4047 - 4063
• Main Authors: Yan, Xiaoyan, Lu, Shasha
• Format: Journal Article
• Language: English
• Published: Weinheim WILEY‐VCH Verlag GmbH & Co. KGaA 01-06-2023; Wiley Subscription Services, Inc
• Subjects: Algorithms; Artificial intelligence; Compressive strength; Concrete aggregates; Concrete properties; Environmental effects; Hardness; high‐performance concrete; improved grey wolf optimization; Kernel functions; Machine learning; Mechanical properties; Modelling; Optimization; Optimization algorithms; Prediction models; Radial basis function; Regression models; Silica fume; slump flow; Support vector machines; support vector regression
• ISSN: 1464-4177; 1751-7648
• DOI: 10.1002/suco.202200779

High‐performance concrete (HPC) as a highly sophisticated aggregate in constructional projects has made modeling given mechanical properties a very complex problem. Declaring by many studies, mechanical features of HPC are not only characterized by the maximum size of coarse aggregate and water amount since influencing by the other components. Using fly‐ash and silica fume as the key constituents can simultaneously increase the hardness aspects and the environmental effects. Considering the compressive strength and slump flow of concrete should be investigated before performing any practical practices. Artificial intelligence approaches with precise and low‐cost methods can replace the costly experimental ways. Therefore, the present paper has aimed to link a prediction model with optimization algorithms to accurately appraise the hardness properties of HPC samples rarely found in literature like this way. In this regard, a machine learning approach of Support Vector Regression using two kernels of Gaussian and radial basis function is coupled with matheuristic algorithms to optimize the modeling process of compressive strength and slump flow of HPC samples. The internal settings of SVR would be tuned at optimal rate by optimizers to function efficiently. To investigate the performance of hybrid frameworks developed in this research, several indicators evaluated the results of hybrid models. Therefore, the R2 of the models was calculated averagely at 0.91 with a maximum difference rate of 11% for the testing phase. While the RMSE index assessed the models with higher values of 16.56 mm for slump and 12.86 MPa for compressive strength. Generally, using smart approaches with high‐accuracy performance has been proposed to be used instead of physical procedures increasing the productivity of concrete compressive strength in terms of time, energy, and cost criteria.