Estimating Clean Reservoir Fluid Bubblepoint and Other Properties in Real Time Using PCA Asymptote of Optical Sensor Data and Equation of State

Abstract

Abstract Mud filtrate invasion occurs in the immediate vicinity of the well as a result of the overbalance pressure of the mud column in the well. Oil-based muds (OBM), unlike water-based muds (WBM), are miscible with reservoir fluid, and OBM contamination alters the properties of the original formation fluid. The bubblepoint of contaminated fluid is usually lower than clean fluid. While fluid is pumped out of the formation, it becomes cleaner and the bubblepoint increases; the upper limit of the increase is the clean formation fluid. While increasing the pumping rate can shorten cleanup time, pumping below the bubblepoint can modify the fluid phase behavior and cause asphaltene content in the formation fluid to precipitate out and sensor data to become erratic and noisy. Therefore, it is important not to pump below the bubblepoint, knowing the clean fluid bubblepoint in real time provides a guideline for the field engineer. The purpose of fluid sampling is to collect a representative formation fluid—samples with an acceptably low contamination. The clean fluid bubblepoint provides a lower limit on pumping pressure, which helps ensure pumping does not go below the bubblepoint and the sample is in single phase. This paper describes how clean fluid compositions are determined from the asymptote of the principal component analysis (PCA) reconstructed scores and then used as input for the equation of state (EOS) program to compute fluid properties such as bubblepoint and gas/oil ratio (GOR). The optical spectral data from the optical fluid analyzer is first despiked, and outliers from the despiked data are removed using the robust ordinary least squares regression (ROLSR) method and robust PCA (RPCA). After removing outliers, clean fluid spectra data are reconstructed using asymptotic PCA scores and PCA loadings. Using a neural network model, clean fluid compositions are determined from reconstructed fluid spectral data, and fluid compositions are used as input for the EOS program to determine fluid properties. Results confirm that the clean fluid bubblepoint and GOR do not change significantly after a few tens of liters of fluid pumpout. Analysis of the first principal component (PC1) confirms that most of the variations occur during the first few tens of liters of pumpout, indicating the predicted clean fluid compositions and properties are somewhat stable. This approach can help determine the clean fluid properties, even while pumping before taking the sample, helping ensure a monophasic fluid sample. When pumpout accumulated volume reaches 40 to 50 L—within 15 to 20 min of pumping out contaminated fluid—clean fluid compositions and properties can be estimated and used to determine reservoir continuity. Additionally, knowing the clean reservoir GOR and API gravity can help determine the type of reservoir fluid in real time.