As a cleaner energy source, the importance of shale gas as a fuel has increased in the past few decades with the main challenge faced being its quantification. While many accurate models exist for the estimation of the free gas inside the pore space of shales, only a limited number of empirical models with insufficient accuracy exist that can quantify the adsorbed gas. Also, adsorption experiments are complicated, time consuming, and prone to many sources of error including gas leakage during coring and preparation of samples. An attempt was made to develop a computational model to predict shale gas adsorption using statistical learning with a large data set consisting of 301 entries. The work done in this study attempted to develop artificial intelligence (AI) models that outperform the existing statistical learning model using the same data set. Three AI techniques were utilized in this study namely artificial neural networks (ANN), adaptive network-based fuzzy logic (ANFIS), and functional networks (FN). Using each AI technique, a submodel was developed for the prediction of both the Langmuir pressure and volume. The Langmuir pressure submodel used temperature (T), total organic carbon (TOC), and vitrinite reflectance (Ro) as input variables while only T and TOC were used as input for the volume submodel. Results showed that a model developed using ANFIS for the prediction of Langmuir pressure had an average absolute percentage error of only 15.26% compared to 27.67% of the statistical learning model. As for the volume submodel, ANN produced the best results with an AAPE of 21.34% compared to the original 23.76%. While FN showed better results in pressure prediction compared to statistical learning, the opposite was true for volume prediction. Also, in both prediction models (pressure and volume), FN was the least accurate our of the three AI methods.

You can access this article if you purchase or spend a download.