A 4D Geomechanics Modelling Approach Using Machine Learning for Drilling Events Predictions in the Giant North Kuwait Jurassic Field

Abstract

Abstract Drilling new wells in the deep Jurassic formations of the North Kuwait fields has gained increased importance given its high gas production potential, which represents a strategic target for Kuwait's domestic gas independency. However, reaching these formations bears severe drilling risks and challenges caused by differentially depleted reservoir sections that lead to arbitrary stress distributions and high pore pressure zones. This study presents an integrated 4D geomechanical model that includes a machine learning model for the development of a drilling expert system. This allows to predict drilling events along any arbitrary well trajectory in the 3D field model by analysing the stress field based on the construction of a mechanical earth model. The model further assesses completion quality to hydraulically fracture the rock mass. The simulation covers a 34 million grid cell model and is based on high-resolution geological information, seismic inversion data, rock core laboratory data featuring uniaxial, triaxial, multiaxial geomechanical tests, scratch, creep behaviour and thick-wall cylinder tests as well as petrophysical logs from 54 deep Jurassic wells. Seismic acoustic impedance data served as a proxy for co-kriging the spatial distribution of geomechanical properties. In addition, fault and fracture models were incorporated at reservoir levels. The drilling expert system allows for an easily applicable and efficient extraction of the required drilling parameters and rock properties along any arbitrary well trajectory to conclude on the probability of drilling events occurring. The derived wellbore stability analyses reduce the costly and time-consuming risks of wellbore collapses when passing through high pressure formations. The simulated 4D stress field yields a strike-slip stress regime with a minimum horizontal stress direction of 135°. Identified zones in the reservoir sections show high hydraulic fracture potential given by the under- and overlying stress barriers in the anhydrite formations for stress containment. The evaluation of critically stressed, natural fractures as well as faults lead to the conclusion of potential reactivation due to high pressure injections in the Najmah reservoir for hydraulic fracturing. The study successfully showcased the integration of domain geomechanics expertise and data science, to derive a 4D geomechanical earth model as input for training machine learning models to evaluate drilling integrity and hydraulic fracturing potential.