Artificial Intelligence Approaches to Modeling Equivalent Circulating Density for Improved Drilling Mud Management.

Area of Science:

• Petroleum Engineering
• Machine Learning Applications
• Drilling Optimization

Background:

• Accurate management of equivalent circulating density (ECD) is vital for preventing well control issues like lost circulation and formation fracturing.
• Traditional ECD calculation methods using downhole tools or complex models can be inefficient.
• This research focuses on a simplified approach using fewer input variables for enhanced efficiency.

Purpose of the Study:

• To develop and evaluate advanced machine learning models for predicting ECD with improved simplicity and efficiency.
• To compare the performance of various machine learning algorithms against existing empirical models.
• To identify the key input variables influencing ECD prediction and assess model operational scope.

Main Methods:

• Utilized a dataset of 2367 field measurements from two wells in an Iranian oilfield using water-based fluids.
• Applied seven advanced machine learning algorithms: CFNN, GRNN, WNN, PSO-SVR, FFA-SVR, GOA-SVR, and GMDH for correlation development.
• Employed a data split of 70% for training and 30% for testing, analyzing key variables: SPP, ROP, and MW.

Main Results:

• All applied models demonstrated high accuracy in ECD prediction.
• The GOA-SVR algorithm yielded the most reliable results with minimal average absolute percent relative errors (AAPRE).
• The GMDH model outperformed existing empirical models, especially with three key input variables; surface mud weight was the most influential factor.

Conclusions:

• Advanced machine learning models, particularly GOA-SVR, offer a robust and accurate framework for ECD prediction.
• The developed GMDH model provides a superior empirical alternative for ECD estimation.
• Leverage analysis confirmed the high operational scope of the proposed models, with a small percentage of suspicious or outlier data points.