Abstract
Lithofacies identification in carbonate reservoirs using conventional well logs is a typically complex nonlinear problem due to influences of multiple factors, such as fluids and fractures. Kernel Fisher discriminant analysis (KFD) is a useful nonlinear single-kernelled method to classify lithofacies. However, the prediction capacity of a single kernel is limited to some extent, especially for complicated lithofacies identification problems. To alleviate this issue, a multiple kernel Fisher discriminant analysis (MKFD) method, an improved KFD, is introduced in this work. MKFD utilizes multi-scaled kernel functions to realize the optimal nonlinear mapping instead of a single kernel used by KFD, which can extract more efficient nonlinear lithologic features. It first increases dimension of input data to obtain more nonlinear features and then reduces dimension to extract effective information from these features for lithofacies identification. To examine the effectiveness of MKFD for lithofacies identification in carbonate reservoirs, a conventional log dataset labelled by rock cores from carbonate reservoirs of Asmari Formation in Paleogene Oligocene-Neogene Miocene in A Oilfield, Zagros Basin, Iraq is used. Both statistical (confusion matrix) and geological evaluations (blind well test) indicate that MKFD outperforms KFD and the prediction results of MKFD are more consistent with rock cores. It has been demonstrated that MKFD can provide an accurate and effective means for lithofacies identification in carbonate reservoirs.
NSTL
Elsevier