Characterization of kerogen content and activation energy of decomposition using machine learning technologies in combination with numerical simulations of formation heating

Abstract

Reliable estimation of organic matter characteristics is essential in source and reservoir rocks evaluation. Their measurement is generally based on well logging and experiments after core sampling. In this study, we present a different approach from these methods to evaluate the characteristics of organic matter, which is based on the numerical simulations in combination with machine learning technologies.Formation heating processes of organic-rich shales and in-situ pyrolysis of kerogen are numerically simulated, where the time-series data of heater temperature are monitored. Here, we consider the whole organic matter composed with kerogen. The monitored heater temperature data are used as an input of inverse modeling by Artificial Neural Network (ANN) and Support Vector Machines (SVM) to figure out the characteristics of kerogen. Heater temperature acts as an indicator of type and maturity of kerogen, since it is affected by the bulk thermal conductivity of formation, which is a function of dynamically changing rock-and-pore composition by kerogen decomposition.We simulate the processes of electrical heating of 300 different formations, containing either one of Type 1, 2, or 3 kerogen with various amounts and activation energies. ANN method is employed to generate a data-driven model to estimate the unknown kerogen content and activation energy using heater temperature as an input data. SVM method, which categorizes dataset into the multiple classes by using hyperplanes, is applied to classify the dataset of heater temperature into different types of kerogens. Developed ANN and SVM models show great performances in the inversion and classification.The suggested characterization method provides a simple tool of estimating reactivity and type of kerogens. It will be used as a proxy for the inverse modeling of geochemical characterization of organic matter coupled with forward numerical simulations, by reducing the number of simulation runs required in the traditional inverse modeling.