Reservoir Pressure Gradient Trend Prediction for the Potash Area of Delaware Basin Using Artificial Neural Network and Geophysical Log Cross Sections

Abstract

Historically, there has been controversies between the oil & gas companies and potash miners in the Secretarial Order Potash Area (SOPA) of Delaware basin. Mostly, these disputes are based on high pressure related operational failures in the area. To reduce these operational anxieties, it is vital to calculate the reservoir pressures, verify the pressures with machine learning predictions, and use the verified pressures to build pressure trend profiles using geophysical log cross-sections. To fulfil the above-mentioned objectives, the methodology used in the process starts with the calculation of reservoir pressures for the area using drilling data. The calculated pressures are then verified with Artificial Neural Network (ANN) machine learning model predictions utilizing well logs and drilling parameters. The verified reservoir pressures are then used to build pressure trend profiles using geophysical log cross-sections. Parameters used in building the ANN include deep, medium, & shallow laterolog resistivity logs, gamma ray log, neutron & density porosity logs, calculated overburden stress, cable tension log, well, caliper log, depth, lithology, mud weight, photoelectric cross-section log, calculated average porosity, calculated water saturation, corrected bulk density log, and bulk density log. Potash is mined in a limited area in the southeast portion of the state of New Mexico. This "potash area" has been afforded special status through the Department of the Interior through several Orders authored by the then Secretary of the Interior. In this work, this "potash area" will be known as the Secretarial Order Potash Area or SOPA. The reservoir pressure gradients were calculated according to the hydrostatic gradients of over 229 selected wells drilled and completed within the SOPA. The ANN model was built using 3 steps including data manipulation, analysis, and deployment. The reservoir pressures were predicted by the Artificial Neural Network (ANN) with high accuracy. The correlation coefficient, R for the training, validation, and testing are 0.978, 0.985, and 0.976, respectively. The Mean Square Error (MSE) was 2.9129 after 136 epochs optimum number of iterations. The overall correlation coefficient (R) is greater than 0.979. These results show that ANN models predicted the measured reservoir pressures accurately for the potash area. Next, the geophysical log cross-sections were created in 2-Dimensional and 3-Dimensional profiles with the verified reservoir pressures using Petra, Matlab, IHS Kingdom, and R machine language. Three west to east cross-sections were created for the three portions of the area namely Back-reef, Reef, and Basin separately. The fourth cross-section was created from the North (Back-Reef) to the South (Basin) through the Reef. The cross sections are displayed showing formation strata, depths, and pressure trends. The information gained from this study will be used to optimize the economic recovery of oil and gas and potash resources from this area which is rewarding to the American Public. It will also promote the safety of underground mining and reduce surface environmental impacts to specific Drilling Islands within the designated development areas. This will bring about safely concurrent development of both resources unachievable without the machine learning model applied in this study.