Accurate Estimation of Residual Hydrocarbon Potential by Removing the Adverse Effects of Lithological Variations on the Training Process of Adaptive Neuro-Fuzzy Inference System

With the booming exploration and development of unconventional hydrocarbon resources, the accurate estimation of source rock factors such as residual hydrocarbon potential (S2) from well logs has become increasingly important. Along with organic material properties, changes in lithology within an interval of possible source also induce well log responses. Artificial intelligence techniques may interpret these lithology-induced log responses as a signal for changes in the organic matter content and/or properties, resulting in decreasing their efficiency. In the present research, a new methodology called the litho-based method was proposed based on modeling the relationship between log data and S2 parameter for each type of lithology using Adaptive Neuro Fuzzy Inference System (ANFIS). The performance of the newly developed methodology was compared with those of the traditional ANFIS and hybrid methods for which the training process was carried out using a dataset including different lithologies. Results showed that the litho-based method successfully removed the adverse effects of lithological variations on the course of ANFIS training, resulting in estimating much more reliable S2 values. Among the traditional methods, utilizing particle swarm optimization (PSO) algorithm in conjunction with ANFIS showed higher performance. Nevertheless, the aforementioned hybrid approach is not as efficient as the litho-based method. The applicability of the proposed methodology was approved by applying it over Pabdeh source rocks for a well in SW Iran. Finally, it is recommended to use the litho-based method for estimating other geochemical factors as well as petrophysical parameters through log data.