Automated Artificial Intelligent Pressure Gradient Analysis for Fluid Contact and Compartmentalization Analysis

Abstract

The reservoir compartmentalization structure and fluid contacts of a field are essential for determining the value of a reservoir asset and provide the two primary purposes of pressure gradient determination. Several new straightforward data-analytical methods have been developed to extract pressure gradient information based on physical properties of the reservoir and meta-analysis of derived pressure gradient information. These methods can be used to provide near real-time feedback about the pressure measurements quality. This paper describes two distinct methods to determine reservoir compartmentalization structure and fluid contacts. The first method implements statistical evolution to rapidly identify pressure gradients. The second method transforms identified pressure measurements into a meta-analytical visual representation of pressure gradients vs. depth with additional input from measurement consistency. Both methods rely on the accurate removal of pressure outlier data, such as that attributable to supercharging. A new technique using expert knowledge of physical constraints was implemented for reliable outlier removal. The two methods then diverge in subsequent conditioning of the data, but re-converge in adapting an efficient fitting method to extract the desired information. Both methods provide reliable removal of pressure data that are not related to formation fluid densities, regardless of reservoir number, fluid number, or fluid type. To date, the removal procedure removes more than 95% of outliers and retains more than 90% of accurate pressure data. Both methods also return the correct number and types of fluids. Although pressure gradient estimation can vary by up to 50% for fluid zones of less than 50 ft, the estimation error of the pressure gradients is reduced to less than 3% for fluid zones greater than 100 ft. Furthermore, fluid breaks can be calculated to within 8 ft for the statistical evolution method and to within 30 ft using the visual method. Finally, although the statistical evolution method is markedly faster than the visual method, both techniques provide feedback within a few minutes. The methods discussed provide feedback about the necessity to retake or take more pressure data during formation-pressure surveys within minutes. This feedback eliminates the delay in reservoir property estimation and greatly increases the reliability and quality of pressure data obtained. The methods presented also use a new application of data meta-analysis to reduce processing time and increase reliability.