Total organic carbon characterization using neural-network analysis of XRF data

The co-occurrence of kerogen and nonkerogen minerals in shale poses a great challenge; most importantly, different scales of measurements and ranges of analytical instruments become the prerequisite for characterization. The traditional shale-characterization technique adopts mineralogical analysis for the inorganic constituent and the total organic carbon (TOC) for the organic matter (kerogen). However, despite modern laboratory analytical techniques, the direct and simultaneous determination of the organic and inorganic constituents of a shale formation may be costly and unrealistic. Hence, the use of the costeffective and more efficient X-ray fluorescence (XRF) tools for the elemental characterization of shale. The missing TOC problem, typical of well logging analysis, albeit, is a major challenge of this method. The objective of this work is to carry out quantitative analysis and interpretation of geochemical and mineralogical composition for the evaluation of organicrich shale formations using a neural network (NN) with the primary interest to optimize the performance of the XRF technique. For this purpose, a machine-learning artificial neural network (ANN) method has been devised to map easy-to-measure nondestructive XRF data of organic-rich shale to total organic carbon. Subsequently, the developed model, based on the existing dataset, was used to predict missing TOC. In the data-driven ANN model, 70% of the dataset was used for training, 15% for validation and 15% for testing. The accuracy and improvement of the NN model was established based on statistical parameters as performance metric. Furthermore, a quantitative calibration function to map the XRF data to TOC is developed based on the extraction of weights and biases of the NN. The implementation of the proposed calibration function for the calculation of TOC in comparison to the measured TOC was obtained with the coefficient of determination (R2) and mean absolute percentage error (MAPE) of 0.974 and 14.54%, respectively. These results confirm the applicability of the proposed calibration function in facilitating the simultaneous measurement of TOC and elemental concentration of shale for both laboratory and field-scale applications.