Facies interpretation is an essential part of reservoir modeling. New software technology and enhanced data analysis promises to improve traditional interpretation methods.
Reliable lithology predictions are a central part of reservoir characterization – and a challenge. In a nutshell, traditional facies interpretation typically begins by applying certain background knowledge about rock properties and basic geological rules which provide a preliminary interpretation.
The traditional approach to facies interpretation is common, but not without its pitfalls. It often lacks the quantitative approach necessary for more in-depth insight and often relies on the interpretation of certain well logs instead of all the available data. As a result, the interpretation runs the risk of being both uncertain and biased.
Today, however, new software technology for geoscience allows users to apply statistical techniques to make the facies interpretation more robust. One example is Cegal’s Blueback Investigator, which brings robust cross-domain data analysis to Petrel* and provides unique visualization ideas to improve the understanding of your data.
Let us highlight some of the techniques that you can use with the Blueback Investigator to improve your facies or geological interpretation for well data.
The Blueback Investigator allows you to combine traditional well log or facies interpretation with specific supporting statistical methods and data analysis techniques. Additionally, it can set up unique visualizations to look at different variables together (figure 1).
One example is to visualize data in a variety of plots simultaneously. For instance, you could create a number of cross-plots and histograms that help you interrogate the data more comprehensively than a typical well section display.
Figure 1: Geological interpretation.
The Blueback Investigator is also a valuable tool for multivariate analysis, a means to identify patterns and relationships between several variables at the same time (figure 2). As most geological datasets are multivariate, this is a suitable method to identify any subtle features otherwise hidden deep in the data.
Using this method, the analysis can go beyond the standard set of well logs and include entire suites of logs which can enable subtle differences in rock properties to be identified.
Figure 2: Multivariate analysis
Extending Traditional Facies Interpretation with K-Means Clustering
Using the Blueback Python tool, simple techniques such as K-Means Clustering can be integrated. This method and other unsupervised techniques can give you a good first pass of the data or help you understand the trends in the data (figure 3).
Cegal’s Python tool allows users to bring the functionality of Python and Python Libraries into Petrel. With the PythonTool users can access libraries such as Numpy, Pandas, and Scikit-Learn to extend their workflows. Scikit-Learn contains several machine learning algorithms and unsupervised clustering methods such as K-means.
Figure 3: K-Means Clustering
Integrating Machine Learning
Using Machine learning and associated techniques, geoscientists can extend the traditional workflows of facies interpretation. These techniques enable the algorithm to find hidden patterns in and generate new insights from a wide variety of data.
Within the context of facies interpretation, machine learning can use existing interpretations of previously drilled wells to train algorithms that identify facies for you.
Cegal’s Python tool allows users to run advanced Python machine learning algorithms inside Petrel. Data from previously interpreted wells can be put into a dataset consisting of a set of features (properties) and targets (interpretation). Your targets and features are further split into a test and a training set that are used to quality control the algorithm. New properties can then be fed into the model that will predict the facies or target.
At the core, none of the techniques mentioned above are “new,” and facies interpretation will always “just” be an interpretation. But applying statistical techniques and new machine learning methods will provide you with an additional set of tools that can guide the process, improve your background knowledge, and enhance your interpretation.
Also, considerable innovation is going on in the Python world, where new approaches are can be adopted to improve the traditional techniques. In the future, these techniques could be integrated into the workflow, helping the interpreter making even smarter choices than he or she does today. Cegal’s tools help you access the latest innovations easily from Petrel.
*Petrel is a mark of Schlumberger.