A Novel Automatic Detection and Diagnosis Module for Quantitative Lithofacies Modeling

Abstract

Abstract Quantitative lithofacies modeling is an essential part of reservoir characterization that involves recognition of lithofacies along with depth. It is a difficult task due to diverse, nonlinear, high-dimensional, noisy and imbalanced nature of sensory well logs data. To overcome the above-mentioned issues, we propose a novel automatic detection and diagnosis module (ADD), comprising of Wavelet associated Twin support vector machine (TwinSVM) for Quantitative Lithofacies Modeling. ADD module comprises of three stages i.e. pre-processing stage, model development stage, and post-processing stage. In the pre-processing stage, resampling, data normalization, wavelet package denoising, and feature extraction (Relief algorithm) have been performed on input sets for data smoothening and enhancement of lithofacies detection capability of ADD module. During the model development stage, TwinSVM has been trained and tested on processed data obtained from preprocessing stage. The parameters of TwinSVM were properly tuned using Genetic algorithm. The optimized ADD module has also been tested on an additional set of unseen well logs data to validate its effectiveness and reliability during the post-processing stage. The performance of five conventional classifiers models has also been compared with the proposed ADD module. The test outcomes clearly indicate the superiority of ADD module over other conventional classifiers for Quantitative lithofacies modeling.