Hybrid machine learning models for nanocomposite drilling fluid viscosity optimization

Nanocomposite drilling fluids play a crucial role in modern oil and gas operations by offering enhanced control over rheological properties, including plastic viscosity, which is critical for efficient drilling performance. This research focuses on developing advanced machine learning models to predict the plastic viscosity of nanocomposite drilling fluids using the Gradient Boosting Machine (GBM) enhanced by cutting-edge optimization techniques such as Gaussian Process Optimization (GPO), Evolution Strategies (ES), Bayesian Probability Improvement (BPI), and Batch Bayesian Optimization (BBO). A robust experimental dataset with features including nanoparticle properties, polymer concentration, solid content, and salinity was analyzed to train and validate the models. GPO exhibited the best accuracy and generalization capabilities among the optimization methods, achieving an R² of 0.8372 on the test set and the lowest mean squared error (MSE). Specifically, the GPO-optimized GBM model achieved an R² of 0.8372 and a mean squared error (MSE) of 2.3790 on the test set, with an overall average absolute relative error (AARE%) of 4.95%. These results confirm the reliability and accuracy of the proposed predictive framework. Sensitivity and SHAP analyses highlighted polymer molecular weight and salinity as the most important parameters affecting plastic viscosity, while nanoparticle concentration showed a distinct non-linear effect: increasing concentration initially enhances viscosity, but beyond a certain limit, further increases lead to a decrease in viscosity, likely due to particle aggregation. This study significantly contributes to the field by providing a reliable predictive framework for designing and optimizing nanocomposite drilling fluids.