Determination of Total Organic Carbon (TOC) in tight reservoir using Empirical Mode Decomposition-Support Vector Regression (EMD-SVR): A case study from XX-1 Basin, Western China

Abstract

In the process of formation evaluation for tight reservoirs, extracting quantitative information of kerogen is a potentially important factor. Moreover, Total Organic Carbon (TOC) is not strongly correlated with geophysical well logging data. In this paper, a combinatory algorithm for nonlinear regression based on Empirical Mode Decomposition (EMD) and Support Vector Regression (SVR) is proposed. On the basis of depth matching, sensitive well logging parameters are preferred by core calibration. That, which means should be used to denoise, is a key issue for acquiring precise and high quality data. Then, intrinsic mode functions (IMF) decomposed by EMD algorithm is established and applied for denosing. Further, denoised data is classified into two categories, one for training and the other for validating. Aiming for TOC predicting model, SVR is implemented both for training and predicting, and simultaneously some conventional methods such as ΔlogR, back propagation artificial neural networks(BP-ANN), and multiple linear regressions are also exerted for comparisons. The result shows that EMD-SVR is the best solution for TOC predicting, with the highest correlation coefficient and the smallest mean squared errors. Likewise, this algorithm is applicable for other reservoirs like shale gas.