Hybrid and ensemble models by coupled with automated meta-heuristic algorithms for compressive and flexural strength and slump of high-performance concrete
Affiliations: [a] School of Civil Engineering, Xijing University, Xi’an, Shaanxi, PR China
| [b] School of Civil engineering & Architecture, Wenzhou Polytechnic, Wenzhou, Zhejiang, China
| [c] State Key Laboratory of Green Building in Western China, Xi’an University of Architecture & Technology, Xi’an, Shaanxi, China
Abstract: High-performance concrete (HPC) is a specialized type of concrete designed to meet stringent performance and uniformity standards that are difficult to achieve with conventional materials and standard mixing, placing, and curing methods. The testing process to determine the mechanical properties of HPC specimens is complex and time-consuming, and making improvements can be difficult after the test result does not meet the required properties. Anticipating concrete characteristics is a pivotal facet in the realm of High-Performance Concrete (HPC) manufacturing. Machine learning (ML)-driven models emerge as a promising avenue to tackle this formidable task within this context. This research endeavors to employ a synergy of ML hybrid and ensemble frameworks for the prognostication of the mechanical attributes within HPC, encompassing compressive strength (CS), slump (SL), and flexural strength (FS). The formulation of these hybrid and ensemble constructs was executed through the integration of Support Vector Regression (SVR) with three distinct meta-heuristic algorithms: Prairie Dog Optimization (PDO), Pelican Optimization Algorithm (POA), and Mountain Gazelle Optimizer (MGO). Some criteria evaluators were used in the training, validation, and testing phases to assess the robustness of the established models, and the best model was proposed for practical applications through comparative analysis of the results. As a result, the hybrid and ensemble models were the potential methods to predict concrete properties accurately and efficiently, thereby reducing the need for expensive and time-consuming testing procedures. In general, the ensemble model, i.e., SVPPM, had a more suitable performance with high values of R2 equal to 0.989 (MPa), 0.984 (mm), and 0.992 (MPa) and RMSE = 3.82 (MPa), 9.5 (mm), and 0.30 (MPa) for CS, SL, FS compared to other models, respectively.
Keywords: High-Performance dynamic properties, support vector regression, prairie dog optimization, pelican optimization algorithm, mountain gazelle optimizer
