A framework to optimize a designed geotechnical system probabilistically using MLP-ANN and ELECTERE decision making – a nailed wall study

This research presents a comprehensive framework for the optimization of a designed geotechnical systems probabilistically, combining the Non-dominated Sorting Genetic Algorithm II (NSGA-II) with the ELECTRE decision-making method. By incorporating the Single Random Variable (SRV) approach and Latin Hypercube Sampling (LHS), the framework is adaptable to various geotechnical applications. A case study on a nailed wall system builds on prior research, optimizing critical design factors such as nail length (L), diameter (D), and horizontal spacing (S h ). The optimization aims to maximize the Factor of Safety (FS) while minimizing maximum horizontal displacement (H max ), maximum moment (M max ), and maximum shear force (V max ). To improve optimization efficiency, a Multi-Layer Perceptron Artificial Neural Network (MLP-ANN) is utilized. The SRV approach generates cumulative distribution functions (CDFs) for the objectives, with Conditional Value at Risk (CVaR) applied at a 95% confidence level to address extreme event risks. The results are ranked using the ELECTRE method, providing valuable insights into selecting Pareto-optimal solutions based on the importance of different objectives. This framework enhances reliability and safety under uncertainty and can be adapted to a range of geotechnical system designs. Step-by-step framework for optimization of a designed geotechnical system under uncertainty