Adventures in pyrolysis II, modeling pyrolysis peaks of petroleum source rocks

Abstract

Programmed pyrolysis is an integral part petroleum source rock evaluation. The analysis is used to determine the potential of a source rock to generate further petroleum, and with other geochemical measurements, how much it may have generated in the past. An important part of the overall evaluation is an estimate of the source rock thermal maturity. Unfortunately, the addition of any extraneous, non-native hydrocarbons can negatively impact the yield of the samples, and more important, the thermal maturity determination. These extraneous hydrocarbons can be petroleum previously generated and retained within the source rock, fluids that migrated in the subsurface into the rocks, or organic material added during the drilling process. The objective of this study was to determine if mathematical methods could be used to remove these extraneous contributions from the true source rock yields, and arrive at a better determination of thermal maturity.Eighteen samples were studies using the standard pyrolysis method applied in the geochemical services industry. Two aliquots of each sample were analyzed: one as-received and a second after solvent extraction solvents to remove soluble organic matter but not kerogen. The pyrograms typically associate with the conversion of native organic matter in the rock were considered as the sum of up to three underlying components which could be recovered by non-linear least squares methods. Four distribution functions were used to model the pyrograms: classical Gaussian distribution, the Cauchy function, the Logistic function and the Weibull distribution. In addition to the ability of each model to match the measured data, each was evaluated for ease of use. All models reproduced the total pyrolysis yields of the test samples, and generated Tmax values with agreed within 2 °C.•The Gaussian model was marginally better at fitting the measurements than the Logistic or Weibull models, and distinctly better than the Cauchy model. Based upon experience, the Gaussian model was the most forgiving in the choice of initial parameters.•The Cauchy model performed the poorest at fitting the measurements and had the largest differences between extracted and unextracted Tmax values. The deviations from the measured values were most pronounced at high and low temperatures due the longer tail of the Cauchy function.•The Logistic model has marginally lower ability to fit the measured data than the Gaussian model, and similar to the Weibull model. However, the Logistic model was more sensitive to the choice of initial values than the Gaussian model, but less sensitive than the Weibull model.•The Weibull model was marginally poorer at fitting the measurements than the Gaussian model and similar to the Logistic model. It did less well at fitting the four component model for the more complex samples. It was most sensitive to the accurate choices for the initial conditions to insure convergence. For some samples sequential manual choices for these parameters was needed before the Gauss-Newton method could be employed.