Prediction of compressive strength of high-performance concrete using multi-layer perceptron

Abstract

The correlations between the mechanical properties of HPCs and their mixture compositions are complex, non-linear, and complex to characterize employing standard statistical methods. This paper aimed to estimate HPC’s compressive strength using a machine learning algorithm including Multi-layer Perceptron (MLP) with an HPC mixed collection of 168 samples via eight input variables. In addition, three meta-heuristic optimizers have been used for improving the efficiency and accuracy of MLP, which are included Dandelion Optimization (DO), Aquila Optimizer (AO), and Sooty Tern Optimization Algorithm (STOA). After fitting the presented models, the developed models’ predictive generalization and efficiency ability is evaluated against a set of performance parameters. All models used were found to perform as suitable in predicting outcomes, which can be employed for saving time and energy. As a result, Aquila’s optimization had the most accurate by MLP compared to other hybrid models. MLAO3 obtained R 2 = 0.994 and RMSE = 1.27(MPa), which are the most suitable result compared to other models.

Keywords

• high-performance concrete
• compressive strength
• multi-layer perceptron
• dandelion optimization
• aquila optimizer
• sooty tern optimization algorithm