Software Development using Machine Learning to Predict PPFG and 3D Geomechanics
Abstract
Pressure prediction plays a fundamental role to design mud weight and well trajectory for wellbore stability and prevents stuck pipe. Some manual calculations are only able to calculate on certain condition such as clean-shale formation and under-compaction mechanism formation. Unfortunately, real formation can be very heterogenic. A method to produce independent formation type shall be developed solve the issue. Therefore, software using machine learning (ML) were developed to generate scrupulous pressure prediction.
Logging data (e.g., Density, Sonic, Gamma Ray) and drilling parameter (e.g., ROP, RPM, WOB) from 2 wells (MLC-01, TTA-01) were used as machine learning input. In this research, 3 methods which are Artificial Neural Network (ANN) Feedforward type, Random Forest (RF), and Support Vector Machine (SVM) were applied.
The result exhibits (1) ANN showed the least Root Square Mean Error (RSME) of 0.11401 in comparison to the other 3 methods, Determination Coefficient (R2) 0.9789. Thus, ANN will be used for the rest of the analysis. (2) 4 data (Density, Sonic, Gamma Ray, Depth) together achieve the most precise with actual condition with RSME 0.0714 and R2 0.9826. (3) After plotting the result in one graph, pore pressure prediction from ANN method is closer to actual pore pressure rather than manual calculation result.
It is to conclude that this software gives a promising result to predict Pore Pressure, Fracture Gradient, and Shear Failure Gradient. The comparative analysis results show that ANN Feedforward type has the feature estimation by its shorter time prediction and high accuracy (a coefficient of determination of 0.99 and RSME 0.08 – 0.23. The overpressure prediction, XRD and Geomechanics can be analysis in one integrated software.
