A novel neural-evolutionary framework for predicting weight on the bit in drilling operations

Area of Science:

• Petroleum Engineering
• Machine Learning Applications
• Optimization Algorithms

Background:

• Accurate estimation of Weight on Bit (WOB) is crucial for optimizing drilling operations and preventing equipment damage.
• Traditional methods may lack the precision required for complex drilling dynamics.
• Artificial Neural Networks (ANNs) show promise but require effective training optimization.

Purpose of the Study:

• To compare the performance of ANNs trained with Grey Wolf Optimization (GWO), Biogeography-Based Optimization (BBO), and Levenberg-Marquardt (LM) for WOB estimation.
• To evaluate the impact of input variable selection on model accuracy and training time.
• To benchmark the optimized ANN models against other established regression techniques.

Main Methods:

• Developed and validated ANN models (LM-ANN, GWO-ANN, BBO-ANN) using drilling data (depth, speed, ROP, flow rate).
• Applied relevance tests to identify key input variables influencing WOB.
• Evaluated model performance using Mean Square Error (MSE) and Mean Absolute Error (MAE) criteria.

Main Results:

• GWO and BBO algorithms enhanced ANN accuracy, reducing training MSE by 14.62% and 24.90% respectively.
• BBO-ANN demonstrated superior performance over GWO-ANN and LM-ANN, with lower prediction errors in the testing phase.
• Using all four input variables yielded the most accurate WOB predictions with BBO-ANN, despite slightly increased training time.

Conclusions:

• Biogeography-Based Optimization is a highly effective method for training ANNs to accurately predict Weight on Bit.
• The BBO-ANN model significantly outperforms multiple linear regression, support vector regression, ANFIS, and GMDH.
• Optimized ANN models, particularly BBO-ANN, offer a robust solution for real-time WOB estimation in drilling operations.