Automatic Parameter Identification of an Advanced Plasticity Model for Saturated Soils in Offshore Geomechanics
Automatic Parameter Identification of an Advanced Plasticity Model for Saturated Soils in Offshore Geomechanics
Accurate offshore geotechnical analysis relies on precise constitutive models, yet parameter identification for advanced models remains challenging, subjective, and often reliant on trial and error. Automating this process presents significant obstacles, as traditional gradient-based and variational methods can be limiting, while gradient-free techniques, such as particle swarm optimization and genetic algorithms, lack mechanisms to retain insights from previous iterations. This paper introduces a novel approach inspired by DeepMinds AlphaZero algorithm, known for its success in complex strategic games like chess. By conceptualizing parameter identification as a strategic game, we integrate deep learning with Monte Carlo Tree Search to enable iterative self-learning and refinement, thereby reducing dependency on user input and extensive experimental data. Specifically, we investigate the use of Extreme Gradient Boosting (XGBoost) to capture and retain learned insights throughout the self-learning process, analyzing how its configuration impacts the accuracy and efficiency of model calibration.
Javad Ghorbani; Majidreza Nazem; Sara Moridpour
5th International Symposium on Frontiers in Offshore Geotechnics (ISFOG2025)
5 - Data Analytics and Machine Learning
https://doi.org/10.53243/ISFOG2025-156
Contact Information
- Secretary General - Dr A M McNamara
- Email: [email protected]
- Phone: +44 20 7040 8154
- City, University of London
- Northampton Square
- London EC1V 0HB. United Kingdom