Explicit Data-Based Model for Predicting Oil-Based Mud Viscosity at Downhole Conditions

Area of Science:

• Petroleum Engineering
• Rheology
• Artificial Intelligence in Geosciences

Background:

• Drilling muds require optimal viscosity for effective cuttings transport.
• Existing models for plastic viscosity (PV) of oil-based muds (OBMs) are often inaccurate, lack generalizability, and are limited to surface conditions.
• High-temperature, high-pressure (HTHP) drilling necessitates robust PV estimation models for downhole conditions.

Purpose of the Study:

• To develop a flexible, accurate, and generalizable model for estimating OBM plastic viscosity (PV) under downhole conditions.
• To overcome limitations of existing PV estimation methods, particularly for HTHP wells.
• To provide an explicit and computationally efficient model for field application.

Main Methods:

• Artificial Neural Network (ANN) technique was employed to predict PV.
• The model was trained using 88 laboratory PV measurements of OBMs from existing literature.
• A connection weight algorithm was used to determine the influence of downhole parameters on PV.

Main Results:

• The developed ANN model demonstrated high accuracy with MSE of 0.0185, RMSE of 0.136, and R of 0.967.
• The model showed good generalizability with an R value of 0.80 on a separate dataset.
• Downhole pressure (64.5%) had a greater influence on PV than downhole temperature (35.5%).

Conclusions:

• The ANN model provides a robust and efficient method for predicting OBM PV under downhole conditions.
• The model's explicit nature and low computational requirements (48 bytes memory, 12 FLOPS) facilitate easy integration into software.
• This approach eliminates the need for time-consuming laboratory measurements, enabling real-time PV data acquisition in the field.