History Matching Using Face-Recognition Technique Based on Principal Component Analysis
S. Yadav · SPE Annual Technical Conference and Exhibition · 2006
Abstract This paper presents a novel methodology of history matching using the face recognition technique based on Principal Component Analysis, which is currently used in computer vision applications. During history matching the spatial reservoir parameters at grid blocks are adjusted in order to obtain a simulated response close to the observed response. This implies that the optimization problem can be prohibitively large and inefficient. In order to circumvent this problem, a set of multiple geologically plausible permeability realizations or the training images for a given depositional environment are obtained. These realizations could be quite different, yet they are not completely random. In spite of their differences, there are characteristic geological patterns, which occur in theses realizations. Such patterns could be extracted out by means of a mathematical tool called Principal Component Analysis. These characteristic geological patterns can then be combined in different combinations to obtain the new plausible realization. The goal is then to find the weights used to combine the characteristic geological patterns such that difference in the historical data and the simulated response is reduced. Therefore the history matching problem is now reduced to optimization problem in a much smaller parameter space. The number of parameters to be optimized is reduced to the number of dominating geological patterns present. The new realization is automatically constrained to the given geological environment since it is a combination of its characteristic features. Also this approach can depict patterns of geological continuity consisting of strongly connected, curvi-linear geological objects such as channels or fractures, unlike the variogram based two-point statistical covariance models. This approach mimics the "face recognition" or the "voice recognition" technique, which are already being successfully applied in their respective domains. The proposed technique has been successfully tested in a fluvial depositional environment.