The use of alternating conditional expectation to predict methane sorption capacity on coal

Abstract

Conventional approaches to determine methane sorption capacity, including manometric, volumetric and gravimetric methods, require complex procedure of preparation of coal samples and long-term sorption measurement. This article proposed the use of alternating conditional expectation (ACE) algorithm to relate methane sorption capacity (VL) to coal composition, vitrinite reflectance and temperature without conducting sorption tests, which minimizes the work volume and time required in conventional measurement method.The basic idea behind the ACE is to estimate a suit of optimal transforms of a dependent and a set of independent variables that result in a linear correlation between the transformed independent and dependent variables with minimum error. Underlying effect can be uncovered of the control of each independent variable on dependent variable through the transform. 139 sets of proximate analysis, maceral analysis and methane sorption data from previous studies were acquired. Ash, fixed carbon, moisture, vitrinite content, vitrinite reflectance and temperature were selected as independent variables to predict VL. The resulted ACE transforms have a correlation coefficient R2 of 0.91, indicating an excellent match between the predicted and measured VL values. Normality and homoscedasticity were verified by Lilliefors- and F-test, which further confirmed the capability of ACE as being a correlation tool. The effects of independent variables on VL observed from the transforms show an agreement with previous studies.Based on the ACE transform results, an explicit VL model was proposed bearing a polynomial correlation with the independent variables. The validity of the proposed model was proved by fitting it to another 43 data sets. Additionally, outlier diagnose was conducted through standardized residuals and their effect on prediction accuracy was investigated.