AUTHORS: Raffaele Di Cuia (G.E.Plan Consulting, Italy), Denis Ferraretti (G.E.Plan Consulting, Italy), Giacomo Gamberoni (G.E.Plan Consulting, Italy), Eric Portier (Gaz de France SUEZ E&P, France), Laurent Escaré (Gaz de France SUEZ E&P Division, France)
YEAR: 2009
Presented at the 4th North African/Mediterranean Petroleum and Geosciences Conference & Exhibition Tunis (EAGE), Tunisia, 2 – 4 March 2009
Extended Abstract
The complete characterization of depositional facies and structural features is an important step in the process of understanding the reservoir potentiality. Facies distribution, depositional geometries, porosity types and fracture/stylolites identifications are key parameters to correctly describe reservoirs.
The direct subsurface information (cores) is fundamental in this process of reservoir definition but the use of other indirect tools to define and predict depositional facies and structural geometries is also important to have a more complete appreciation of the entire reservoir. In this case, it is important to properly calibrate the indirect tools with the core observations and analysis.
The FMI (Full-bore Formation Micro Imager) logs represent one of the more advanced and important indirect tools to describe the rocks characteristics; when correctly calibrated with cores and used in associations the other conventional electric logs, it can represent a key element to predict facies and characteristics in un-cored sections of the reservoir.
Image logs interpretation is a very complex task, because of the large number of variables and the huge amount of data to be analyzed. The human factor is important in the interpretation of these data not only in terms of experience of the interpreter but also in terms of ability to consider all the information available during interpretation. By analyzing the interpreter work it is possible to identify three critical factors that can change the interpretation results: the interpretation subjectivity, the large amount of time taken by the interpretation and the errors related to the wide range of values to visualize.
We have developed a new software, I2AM (Intelligent Image Analysis and Mapping) that helps the geoscientist in his interpretation task. This system aims at extrapolating the maximum amount of information from the image logs by considering not only the surfaces that cut the borehole but also the textural features of the images. In this way, we can extract information about the rock properties, avoid the subjectivity of the interpretation and reduce the interpretation time by largely automating the log interpretation, although some level of human interaction and correction is still necessary. Our approach exploits image processing algorithms to analyze borehole images and artificial intelligence techniques to classify them. The resulting implemented system produces a semi-automatic interpretation of the image logs.
This software was used over the FMI logs of four wells from the north African region in order to test the validity of the results.
METHODOLOGY AND WORKFLOW
The FMI from the two wells were interpreted using a standard geological approach. The geological interpretation focused on the identification of the main lithofacies, depositional geometries (bedding, erosional surfaces ….) and structural features (fractures, faults …). The lithofacies indentified were then calibrated with core information to better define the geological characteristics of the units.
Once the geological interpretation performed by the interpreted was finalized we applied an unsupervised approach using the I2am software to interpreted in a automatic way the FMI logs.
The I2AM approach can be summarized in four steps
1. automatic feature extraction from FMI image;
2. feature refinement and validation;
3. data analysis and clustering;
4. classes interpretation and validation.
The technique used to analyze the rock/image features applies different image processing algorithms to the image data in order to extract their properties. The image properties extracted by using these algorithms are represented by values that can be managed by an elaborator/software. The output of every algorithm has a meaning only in relationship with the results of the other ones.
This software is able to extract form FMI data the following four main features of the image:
• Contrast of the image
• Texture of the image
• Surfaces crossing the borehole (bedding, fractures …)
• Vugs or large clasts
Once the entire image log is analyzed and the system/algorithms have extracted the values that represent each image feature, the entire section of the logged borehole is divided into different classes using data mining techniques. The classes includes all the features and key elements identified in the studied image log, and form the basis of the analysis on which the interpreter carries out its considerations. The final result is a set of “image facies” that is identified along the image log and that can be calibrated to sedimentary facies, using cores, to assign them a geological meaning.
The geoscientist interaction is fundamental in two particular tasks: in step 2, after the feature extraction process, I2AM allows the interpreter to adjust some aspects of algorithm results (i.e. add/modify/remove surfaces and vugs/large clasts and remove “poor sections”). Finally, during the clustering process, it is necessary to choose the classes structure. The interpreter can select the better suggested clustering solution and modify the number of clusters/classes.
This approach represent one of the data mining methods which are able to extract useful information from large datasets or database. In particular for this analysis we used the hierarchical agglomerative clustering, an unsupervised technique. Clustering is the classification of objects (in this case all of the analyzing windows) into different groups, or more precisely, the partitioning of a data set into subsets (clusters or classes), so that the data in each subset ideally share some common characteristics. Hierarchical agglomerative clustering builds the hierarchy starting from the individual elements considered as single clusters and progressively merging clusters according to the chosen similarity measure. This measure is computed using all the extracted features, so the definition of two similar items is not based on just one characteristic but it considers a combination of all the extracted features.
The output of hierarchical clustering is a tree represented by a dendrogram (figure 1): a tree-like plot where each step of hierarchical clustering is represented as a node merging two branches into a single one.
In general, a dendrogram is displayed with a Color Mosaic (Figure 1). The Color Mosaic provides to the interpreter an aid to represent all of the features of the whole well “at a glance”. After cutting the tree at the desired level, a further information is added to the color mosaic: a “classes indicator” line, representing with a color code the “class column” assigned by the hierarchical clustering.
In order to test and validate our tool, this technique was applied, using I2AM prototype, to two FMI logs from two nearby wells, drilled over rock section of similar characteristics. For this comparison the image logs were first manually interpreted by a geologist (interpreter) that identified 8 different image facies over the logging section of both wells. This classification was based on:
• the texture of the images (fine-grained to coarse-grained);
• the organization of the texture of the images (highly organized to disorganized);
• layering (numbers of layer surfaces in a fixed interpretation window;
• resistivity contrast between the image background and the elements present in the image (low to high);
• layer thickness (thin to thick).
Some of this facies show intermediate characteristics between two end-member facies therefore for the interpreter it is sometimes difficult to correctly classify some intervals because of the complexity of the image characteristics. In addition to these 8 image facies other two types of images were identified by interpreter:
• poor quality images (the quality of the image log was poor due to acquisition or processing artifacts);
• highly fractured intervals (the rocks are intensely fracture and the characteristics of the matrix are difficult to detect on image logs).
The images from both wells were also processed using the semi-automatic methodology presented to identify different rock facies to be compared with those identified by the log interpreter.
The presence of cores over the logged intervals allowed to calibrate the image log facies and to assign to each facies a geological meaning. In this way the entire logged section is characterized in terms of geological characteristic, depositional environments and main petrophysical properties (Figure 2).
CONCLUSION
The semi-automatic unsupervised approach proposed with this study allow a better interpretation of the main geological characteristics recorded by FMI log, moreover it allows to reduce the interpretation time and subjectivity of the interpretation.
When the image facies produced by the software are correctly calibrated with core data it is possible to propagate this interpretation over un-cored sections of the well and therefore have a full interpretation of the main geological features of the well (figure 3).