Аннотации:
© 2019, International Multidisciplinary Scientific Geoconference. All rights reserved. We consider a problem of reservoir facies classification based on well logs. The classification procedure assumes that there is a train sample, which has a corresponding lithotype for every log entry. Since the physical nature of observed rocks is very complex, there are currently no adequate mathematical models which can describe it. We proposed the probabilistic model which takes into consideration mutual stochastic dependence of adjacent log entries, which is represented by Markov networks. The characteristic feature of the proposed procedure is the clusterization method of well logs. Since single log entries can't be classified with desired statistical precision, one needs to use at least several entries to perform classification. Our clusterization procedure makes uses of characteristic behavior of well logs on the edges of reservoir facies. The classification itself is done via Bayesian test, which uses the aforementioned Markov network model to calculate posterior distribution. Other models of stochastic dependence of adjacent log entries, such as autoregression, are also used. The proposed procedure was tested on the well logs of Cretaceous carbonate sediments. We constructed clusters for this data and the results of the classification were compared to the fact data, obtained from core experiments.
