During formation tester operation, the use of downhole optical spectrometry has been proven to be essential for reservoir fluid characterization. Apart from the intrinsic value of fluid profiling, obtaining fluid properties downhole in real time is of particular interest since the results may impact the decision-making process during sampling and ultimately the success of the sampling operation.

A new methodology predicts petroleum fluid composition from optical spectra acquired with wireline or whiledrilling formation testers. The method comprises fluid typing, computation of fluid composition and estimation of data specific uncertainty. The fluid typing algorithm is capable of categorizing a sample into three fluid types: gas, retrograde gas and oil. Based on the fluid type identified,the appropriate mapping matrix, which transforms optical spectra into compositions, is selected. The mapping matrix is derived from a database consisting of optical spectra, compositions and pressure/volume/temperature (PVT) properties of a wide variety of petroleum fluids. The outputs of the composition algorithm are the weight fractions of the hydrocarbon pseudo components, C₁, C₂, C₃, C₄, C₅ and C₆₊, and CO₂. The composition is used to estimate the gas-oil ratio (GOR) by means of an artificial neural network algorithm. As a measure of uncertainty, confidence intervals are computed for the predicted components of the composition and GOR. All results are available during acquisition of the data.

The accuracy of the algorithm in estimating composition, GOR and their associated confidence intervals was assessed by comparing the results of the predictions against laboratory-derived results. Several field data sets were analyzed and the results were comparedto the results obtained by PVT laboratories on the same samples. The estimated composition and GOR showed very good agreement with PVT results. Furthermore, the algorithm provides more accurate estimates of composition and GOR than are available with current downhole optical spectrometers.