Predicting bentonite swelling pressure: optimized XGBoost versus neural networks.
1Department of Civil Engineering, Maulana Azad National Institute of Technology, Bhopal, 462003, India. amanutkarsh619@gmail.com.
Scientific Reports
|July 30, 2024
Summary
A new machine learning model accurately predicts bentonite swelling pressure for radioactive waste repositories. This advancement enhances the stability and sealing of crucial geological barrier systems.
Area of Science:
- Geotechnical Engineering
- Materials Science
- Machine Learning
Background:
- Bentonite's swelling pressure is vital for radioactive waste repository barrier design.
- Accurate prediction ensures long-term stability and sealing of engineered barriers.
Purpose of the Study:
- To develop a machine learning model for predicting bentonite and bentonite mixture maximum swelling pressure.
- To optimize the model using grey wolf optimization (GWO) and extreme gradient boosting (XGBoost).
Main Methods:
- Compiled a dataset of 305 experimental points with soil properties.
- Utilized a GWO-XGBoost model with a penalty term in the loss function.
- Compared performance against feed-forward and cascade-forward neural networks.
Main Results:
- The GWO-XGBoost model achieved R² of 0.9832 and RMSE of 0.5248 MPa.
- Outperformed traditional neural network models in predictive accuracy.
- Identified dry density and montmorillonite content as key predictive factors.
Conclusions:
- The GWO-XGBoost model offers a highly accurate and reliable method for predicting bentonite swelling pressure.
- This tool supports the design of effective barrier systems in geotechnical applications.
- Potential limitations exist in predicting extreme swelling values due to inherent material complexity.
