A cretaceous reservoir has been strongly modified by post depositional processes (PDP). According to previous study, PDP change the reservoir petrophysical properties and significantly impact subsequent flow simulation. Classify the lithofacies from logs is difficult due to the complex carbonate facies structures, strong PDP overprints, and challenging logging conditions in carbonate reservoirs, particular with vugs and fractures.
In this study, a workflow including core description, logcore analysis and probabilistic neural network facies analysis are introduced to accurately predict the log lithofacies. First, we analyzed five geological factors impact on reservoir qualities. Study shows dolomite has better reservoir quality than limestone. Fracture/microfracture and Vug significantly improve the porosity and permeability. However, anhydrite does not significantly reduce the reservoir quality. There is no significant correlation between reservoir qualities and the five major lithofacies: grainstone, muddy packstone, packstone, wackstone, and mudstone. PDP overprints complicated the reservoir quality-lithofacies correlation and should be removed from training data in later log faciesclassification.
Second, multivariate statistic algorithms and neural network method are tested for log facies classification. Discriminate analysis and multi-logistic regression had poor to fair prediction accuracy. Because of the capability to delineate complex nonlinear relationship between facies and log data, Probabilistic Neural Network (PNN) outperformed other tested algorithms and archived reasonably good prediction accuracy above 70%.
Finally, the predicted geological facies are cross validated with a hold off well. Furthermore, two log lithological indicators as well as the predicted facies are mapped by each zone using simple kriging method. The log indicator maps and facies proportion maps agree with the geological conceptual model. Thus the facies prediction could be used for subsequent reservoir modeling and flow simulation.
A cretaceous reservoir has complex sedimentary structure and has been strongly modified by post depositional processes (PDP).
Because the complex carbonate facies structures, strong PDP overprints, and challenging logging conditions in carbonate reservoirs, particular with vugs and fractures, it is challenging to classify facies using log data. Furthermore, the large amount of seismic energy is absorbed by overlaying formations. Therefore, the carbonate internal structures cannot be identified from 3D seismic interpretation. The PDP not only changes the reservoir quality but the petrophysical responses. Therefore, the traditional facies based modeling method should be improved to predict the spatial distribution of reservoir quality. The overprints of PDP should be removed from facies classification and be integrated into later model building. It is necessary to briefly review the reported carbonate reservoir modeling methods.
Due to complex carbonate facies model, several methods are introduced to classify carbonate facies using well log data. These methods can be generalized in two major directions:
using empirical equation such as Archie equation or rock fabric equation to classify facies (Asquith, 1985; Holtz and major 2004; Lucia, 2005)
using multivariate statistics and neural network method to characterize the complex nonlinear relationship of carbonate facies and log parameters.
Here we focus on the second direction.
Lim and others, 1997, initiated a statistic workflow using descriptive statistical analysis, principle component analysis, and
