Layer | Fill | Outline |
---|
Map layers
Theme | Visible | Selectable | Appearance | Zoom Range (now: 0) |
---|
Fill | Stroke |
---|---|
Collaborating Authors
Fusion Based Classification Method And Its Application
Jin, Long (Jackson School of Geosciences, The University of Texas at Austinproduction.) | Sen, Mrinal K. (Jackson School of Geosciences, The University of Texas at Austinproduction.) | Stoffa, Paul L. (Jackson School of Geosciences, The University of Texas at Austinproduction.)
Introduction Summary Classification algorithms have many applications both in exploration and production seismology. Many classification algorithms have been reported in the literature, such as, seismic facies identification, lithology/fluid prediction, etc. However, improper choice of an algorithm and parameters for a specific problem will create incorrect classification results. Here, we elaborate on some of these issues. Further, we propose combing multiple classifiers with Dempster- Shafer theory (DS) to increase the accuracy of the classification. The philosophy of our approach is that different classifiers may offer complementary information about the patterns to be classified, combining classifiers in an efficient way can achieve better classification results than any single classifier. The effectiveness of this method is demonstrated with a synthetic data test. Several methods of classification or pattern recognition exist that can automatically extract information from seismic data or well logging data. They have been used in seismic facies, lithology/fluid prediction, and time-lapse seismic anomaly classification and identification, etc. Over the years, researchers have made rigorous efforts to apply advanced classification methods in exploration and production seismology, including cluster, discriminant analysis, Bayesian classification, neural network, and support vector machine etc. There are numerous papers on this aspect. For example, Sagga (2003) presented an approach based on competitive neural network for the classification and identification of reservoir facies from seismic data. Li and Castagna (2004) presented the application of support vector machine in AVO classification of gas and wet sand. Avseth and Mukerji (2002) used three different methods to classify six different facies based on well-log measurements of p - wave velocity and gamma ray. However, there are many uncertainties in the conventional classification methods. One of them is that the accuracy of classification relies on the choice of different classification algorithms and also parameters related to a specific algorithm. The quality of data for classification affect the accuracy and stability. For supervised classification, the choice of learning data will also influence classification results. Given so many uncertainties, it is a challenging task to get accurate classification results. We propose a fusion based classification method which can address some of the uncertainties. Algorithm Dempster-Shafer Theory The Dempster-Shafer(DS) theory is a mathematical theory of evidence based on belief functions and plausible reasoning, which is used to combine separate pieces of information (evidence) to calculate the probability of an event. The theory was developed by Arthur P. Dempster and Glenn Shafer. In the following, the basic concept and theory are presented. There are three important functions in DS theory: the basic probability assignment function (bpa or m ), the belief function(Bel), and the plausibility function(Pl). Let X be the universal set, the set of all states under consideration. The power set, P(X), is the set of all possible sub-sets of X , including the empty set, F. The basic probability assignment is a primitive of evidence theory. Fusion based classification To account for the uncertainties of choosing different classification algorithms, different parameters, and different input attributes, a fusion based classification method is presented.
Sample application To illustrate the method, we consider the simple synthetic experiment described in Figure 3. The data, displayed in References Figure 4, consists of two sets of seismic traces with indicated Gardner, L. W., 1949, Seismograph determination of salt dome boundary deep on the dome flank: Geophysics, v. 14, p. 29-38. Each set contains one trace for each of 15 May, B. T., and Covey, J. D., 1981, An inverse ray method for receiver positions, with one set of traces for each of two computing geologic structures from seismic reflections-Zerooffset source positions. Each trace records geophone response as a case: Geophysics, v. 46, p. 268-287. To conserve Waters, K. H., 1981, Reflection seismology: John Wiley & Sons, New York.
Velocity model building for seismic imaging is commonly performed with tomography and full wave inversion (FWI). Both techniques are time consuming and need significant human intervention. Machine learning has been introduced into seismic imaging with the goal of reproduce the success earned in other fields. Due to the complexity of the earth, and the geological uniqueness of any one location, determining the appropriate training data can be challenging. Directly building a 3D velocity model by machine learning still has some way to go. Instead of letting machine learning do all the work, it may be more practical to only perform machine learning on the portion of model building that requires heavy human intervention. In this paper, we present a method that builds the velocity model automatically by combining novel machine learning with the mature velocity model building techniques. Presentation Date: Monday, October 12, 2020 Session Start Time: 1:50 PM Presentation Time: 1:50 PM Location: Poster Station 1 Presentation Type: Poster
Abstract Lithology identification is an important aspect in reservoir characterization with one of its main purpose of well planning and drilling activities. A faster and more effective lithology identification could be obtained from an ensemble of optimized models using voting classifiers. In this study, a voting classifier machine learning model was developed to predict the lithology of different lithologies using an assembly of different classification algorithms: Support Vector Machine (SVM), Logistic Regression, Random Forest Classifier, K-Nearest Neighbor, and Multilayer Perceptron (MLP) models. The result of the comparative analysis shows that the implementation of the voting classifier model helped to increase the prediction performance by 1.50% compared to the individual models. Despite a small significance at deployment in real scenario it improves the chances of classifying the lithology.
- Africa > Nigeria (0.69)
- North America > United States > Utah (0.29)
- North America > United States > West Virginia > Appalachian Basin > Marcellus Shale Formation (0.99)
- North America > United States > Virginia > Appalachian Basin > Marcellus Shale Formation (0.99)
- North America > United States > Pennsylvania > Appalachian Basin > Marcellus Shale Formation (0.99)
- (3 more...)
- Information Technology > Artificial Intelligence > Machine Learning > Performance Analysis > Accuracy (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Neural Networks > Perceptrons (0.72)
- Information Technology > Artificial Intelligence > Machine Learning > Statistical Learning > Support Vector Machines (0.72)
- Information Technology > Artificial Intelligence > Machine Learning > Statistical Learning > Nearest Neighbor Methods (0.69)
ABSTRACT Squeezing of rock is known to be a time-dependent phenomenon causing large deformation of rock around due essentially to excess of shear stress and occurring within a variety of failure mechanisms. It can result in economic loss and major safety concerns, as controlling the large deformations represent a rock engineering challenge. Therefore, a reliable estimate of tunnel deformation is essential for an adequate excavation, project design, and planning since it can help to build a strategy to manage the ground instability well in advance. The aim of this paper is to develop an artificial intelligence-based tool capable of recognizing the squeezing ground class as an alternative to the existing empirical charts or correlations. Historical squeezing data were compiled. On the basis of these available data, the squeezing potential of the ground was categorized into three classes namely; non-squeezing, minor, and major squeezing. A feed-forward neural network (FFN) classifier was implemented to recognize each type of squeezing class. In general, high accuracies were achieved indicating improvement over some existing squeezing models. It is suggested that the FFN-based classifier could be a useful tool in managing squeezing ground. 1. INTRODUCTION Controlling the stability of underground excavations in squeezing grounds can be challenging since the resulting large rock mass deformations cause significant operating problems. These include: difficulties for completing underground works; tunnel collapse; installation of significant amount of support; and major delays in the construction schedules, thus, resulting in an increase of the excavation costs. Squeezing grounds have severe implications for ground support because when supports are installed to contain the excessive large deformations, there is an accumulation of the support pressure and eventually the support will yield, resulting in safety concerns especially in underground mines (Dwivedi et al., 2014). Hence, a reliable estimate of tunnel deformation is important for an adequate excavation project design and planning, since it can help to build a strategy to manage the ground instability well in advance.
- Asia (0.28)
- North America > United States (0.28)
- Materials > Metals & Mining (0.67)
- Energy > Oil & Gas > Upstream (0.67)
- Information Technology > Artificial Intelligence > Machine Learning > Performance Analysis > Accuracy (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Neural Networks (1.00)
- Information Technology > Artificial Intelligence > Machine Learning > Learning Graphical Models > Directed Networks > Bayesian Learning (0.51)