Up till now from 2012, New Mexico State and the Bureau of Land Management (BLM) have permitted drilling activities in the Potash Area (PA) of the Delaware Basin up to the Second Bone Spring formation. This permission was based in part on the available technology and data at the time and in part on the high-pressure concern by potash miners. This high pressure can cause kicks, a blow out, or casing to buckle. Thus, loss of wellbore integrity can lead to loss of zone isolation or fluid migration by gas channeling or micro annulus into the potash formation which is above the oil and gas formations. But with the advent of current technological advancement leading to improved drilling up to and below the Wolfcamp more data is now available.
Based on the needs mentioned above, the objective of this research is to develop a methodology for predicting the reservoir pressure gradient trend for the Potash Area using Artificial Neural Network (ANN) and Multilinear Regression machine learning models. The decision on what model should be used was guided by the efficiency and accuracy of the models. The study utilized drilling and well logs parameters such as Deep & Shallow Laterolog Resistivities, Gamma Ray log, Neutron & Density Porosity Limestone logs, Sonic logs, caliper log, depth, lithology, mud weight, Photoelectric Cross-section, average porosity, water saturation, corrected bulk density log, and bulk density log. All the wells were drilled and completed within the potash area to at least the base of the Wolfcamp formation.
Data manipulation, analysis, and deployment were the three steps used for building the model. The Artificial Neural Network (ANN) predicted the reservoir pressures with high accuracy where the correlation coefficient, R, for the training, testing, and validation are greater than 0.978, 0.976, and 0.985 respectively. The overall correlation coefficient, (R) is greater than 0.979 with the Mean Square Error (MSE) of about 2.9129 after 136 epochs optimum number of iterations. Similarly, the Multilinear Regression model has a high accuracy with coefficient of determination (R2) greater than 0.990. These results show that both models can predict the reservoir pressures accurately for the potash area.
The reservoir pressure predicted from the drilling and well logs parameters using machine learning models would not only improve permit and execution stages quality decision making but also lead to safe and improved concurrent recovery of oil and gas and potash resources in the area.