Employing multi-layer perceptron model via meta-heuristic algorithms for predicting California bearing capacity of stabilized soil

• Published in: Multiscale and Multidisciplinary Modeling, Experiments and Design Vol. 7; no. 2; pp. 1375 - 1391
• Main Author: Zhang, Lulu
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.06.2024
• Subjects: Characterization and Evaluation of Materials; Engineering; Mathematical Applications in the Physical Sciences; Mechanical Engineering; Numerical and Computational Physics; Original Paper; Simulation; Solid Mechanics; California bearing ratio; Bonobo optimizer; Dynamic control cuckoo search; Smell agent optimization; Multi-layer perceptron
• ISSN: 2520-8160; 2520-8179
• DOI: 10.1007/s41939-023-00277-3

The California bearing ratio (CBR) value is a pivotal soil characteristic for designing flexible pavements and airport runways. Additionally, it can be harnessed to ascertain the subgrade's soil reaction through correlation. This parameter is paramount in soil engineering, particularly in formulating the subgrade design for rural road networks. The CBR value of soil is subject to a multitude of influencing factors, including but not limited to maximum dry density (MDD), optimum moisture content (OMC), liquid limit (LL), plastic limit (PL), plasticity index (PI), soil type, and soil permeability. Furthermore, whether the soil is soaked or unsoaked also impacts this value. The process of CBR determination is notably protracted and demands a considerable amount of time. Recognizing the significance of this determination, the study introduces an innovative machine-learning approach. This novel method employs a multi-layer perceptron as its foundational model, harnessing the formidable capabilities of this algorithm in addressing regression challenges. To elevate the performance of the MLP and attain optimal outcomes, a hybridization approach has been employed, integrating the Bonobo Optimizer (BO), Smell Agent Optimization (SAO), and Dynamic Control Cuckoo Search (DCCS). The hybrid models proposed in this study showcase encouraging outcomes in CBR value prediction. Notably, the MLAO3 hybrid model emerges as the most precise predictor among the various models, achieving an impressive R 2 value of 0.994 and an RMSE value of 2.80.