web analytics
Advancing Seismic Research with Modular Frameworks

Advancing Seismic Research with Modular Frameworks

By: Felix Balderas, Geophysical Insights
Published with permission: Oilfield Technology
September 2011

In many disciplines, a greenfield project is one that lacks any constraints imposed by prior work. The analogy is to that of construction on ‘greenfield’ land where there is no need to remodel or demolish an existing structure. However, pure greenfield projects are rare in today’s interconnected world. More often one must interface with existing environments to squeeze more value from existing data assets or add components to a process, manage new data, etc. Adding new technologies to legacy platforms can lead to a patchwork of increasingly brittle interfaces and a burgeoning suite of features that may not be needed by all users. Today’s challenge is to define the correct ‘endpoints’, which can join producer and consumer components in a configurable environment.

This article highlights a strategy used to develop new seismic interpretation technology and the extensible platform that will host the application. The platform, which is code-named Paradise, includes an industry standard database, scientific visualization, and reporting tools on a service-based architecture. It is the result of extensive research and technology evaluations and development.

Geophysical Insights develops and applies advanced analytic technology for seismic interpretation. This new technology, based on unsupervised neural networks (UNN), offers dramatic improvements in transforming and analyzing very large data sets. Over the past several years, growth in seismic data volumes has multiplied in terms of geographical area, depth of interest, and multiple attributes. Often, a prospect is evaluated with a primary 3D survey along with five to 25 attributes serving general and unique purposes. Self-organizing maps (SOM) are a type of UNN applied to interpret seismic reflection data. The SOM, as shown in Figure 1, is a powerful cluster analysis and pattern recognition method developed by Professor Teuvo Kohonen of Finland.

SOM analysis for seismic interpretation

SOM analysis for seismic interpretation

UNN technology is unique in that it can be used to identify seismic anomalies through the use of multiple seismic attributes. Supervised neural networks operate on data that has been classified so the answer is known in specific locations, providing reference points for calibration purposes. With seismic data, a portion of a seismic survey at each logged well is known. UNN however, do not require the answer to be ‘known’ in advance and therefore are unbiased. Through the identification of these anomalies, the presence of hydrocarbons may be revealed. This new disruptive technology has the potential to lower the risk and time associated with finding hydrocarbons and increases the accuracy of estimating reserves.

The company decided to build new components separately, then, with loosely coupled interfaces, add back the legacy components as services. In its efforts to build a new application for the neural network analysis of seismic data, Geophysical Insights struggled to find a suitable platform that met the goals of modularity, adaptability, price, and performance. While in the process of building new technologies to dramatically change seismic interpretation workflow, an opportunity arose for a new approach in advancing automation, data management, interpretation and collaboration using a modular scientific research platform with an accessible programming interface.

Seismic interpretation workflow with Paradise

Figure 2. A software framework for hosting oil and gas software applications

With this new technology concept underway, an infrastructure was needed to support a core technology and make that infrastructure available for others. To deploy their own core technology, they would potentially need databases, schemas, data integration tools, data loaders, visualization tools, licensing, installers, hard copy, and much more. While not everyone would need the numerous lower level components, they would find that there is more work to be done on the supporting infrastructure than on the core technology itself. Without a platform, each vendor would have to undergo the long process of gathering requirements, developing, testing, and evaluating numerous frameworks, all for something that is not their core product. This is not only a major distraction from developing the core technology, but also an expensive endeavor that most are not ready to make, and in some cases perhaps a deal breaker for the project. The company decided to move forward, developing a platform for itself that would be useful for others. The basic concepts around the chosen architecture are depicted in Figure 2. The goal was to build an affordable, yet powerful platform that could be used by small and large organizations alike, for building and testing new software technologies and shortening the time between design and deployment of new components. Developing a platform separate from the core component meant that it was possible to overlap the development activities for the core component and platform. This minimized the impact that changes in the platform had on the science component and vice versa, thus reducing delivery time. Similar platforms already existed but due to their price, these were out of reach for many smaller vendors and potential end users. Any vendor wanting to promote a simple tool integrated on pricey platforms would find a limited audience based on who could afford the overall platform. End users would probably pay for extra but perhaps unused features. One of the company’s goals was for a modular, affordable overall platform. A vendor of a new component can choose to license portions of the Geophysical Insights’ platform as needed. A good software design practice is to include end users early in the process, making them part of the team. One thing that they made clear was their sensitivity to price, particularly maintenance costs.

The new generations are more accustomed to working with social collaboration and mobility tools. No longer can a scientist bury himself behind a pile of literature in a dark office to formulate a solution. With the changing demographics of geoscientists entering the workforce and declining research funds, the lag time between drawing a solution on the white board and when it can be visualized remotely across many workstations must be reduced. These are some of the challenges this platform tackles.

Design and architecture are all about trade-offs. One of the earliest decision points was the fundamental question of whether to go with open source or proprietary technology. This decision had to be made at various levels of the architecture, starting with the operating system, i.e. Linux versus Microsoft or both. Arguments abound regarding the pros and cons of open source technologies such as security, licensing, accountability, etc. In the end, although it was felt that Linux dominated server applications, when looking at the potential users, the majority would be using some version of Windows OS. This one early decision shaped much of the future direction, such as programming languages and development tools.

Evaluations were conducted at various architecture levels, taking time to try out the tools. The company designed data models and evaluated databases. For programming languages and development platforms, C++, C# with MS Studio IDE and Java with Eclipse IDE were evaluated to search for mixed-language interoperability, reliability and security. Java/Eclipse IDE did not meet all the set goals, instead better mixed language programming support was discovered between managed C#/.NET code and unmanaged Fortran for some scenarios. Other scenarios required multiple simultaneous processes.

At the GUI level, the company looked at Qt, WinForm and WPF. It was decided to use WPF because it allows for a richer set of UI customization including integration of third party GUI controls, which was also evaluated. Licensing tools, visualization tools and installers were also examined. (All of this is a bit too much to discuss here in detail, but Geophysical Insights advocates taking the time to evaluate the suitability of the technology to the application domain.)

The company also considered standards at different levels of the architecture. There is usually some tension between standards and innovation, so caution was needed about where to standardize. One component that appeared as a good candidate for standardization was the data model. Data assets such as seismic and well data were among the data that needed to be worked with, but the information architecture also required new business objectives that were not common to the industry. For example, the analytical data resulting from the neural net processes. A data model was required, which was simultaneously standard, yet customizable. It also needed to have the potential to be used as a master data store.

Professional Petroleum Data Model (PPDM) is a great, fully documented, and supported master data store, which shares a lot of common constructs with several other proprietary data stores, and has a growing list of companies using it. PPDM builds a platform and vendor independent, flexible place to put all the E&P data. The company actively participates in that community, helping to define best practices for the existing tables while proposing changes to the model.

Research, including attendance and participation at industry conferences and discussions with people tackling data management issues, made it clear that the amount of data, data types and storage requirements are growing exponentially. The ‘high-water’ marks for all metrics are moving targets. It will be a continuing challenge to architect for the big data used by the oil and gas business. ‘Big data’ refers to datasets that are so large they become awkward to work with using typical data management and analysis techniques. Today’s projects may include working with petabytes of data. Anyone building a boutique solution today will have to be prepared for rising high-water marks, and if they depend on a platform, they should expect the platform to be scalable for big data and extensible for new data types.

Neural networks in general, when properly applied, are adept at handling the big data issues through multidimensional analysis and parallelization. They also provide new analytical views on the data while automated processing eliminates human-induced bias, enabling the scientist to work at a higher level. Using these techniques, the scientist can arrive at an objective decision at a fraction of the time. In the face of a data deluge and a predicted shortage of highly skilled professionals, automated tools can assist in achieving the increasing productivity demands placed on people today.

The usefulness of a platform depends heavily on the architecture. Geophysical Insights has witnessed how rigid architectures in other software projects can become brittle over time, causing severe delays for new enhancements or modifications. However, business cannot wait for delayed improvements. Rigid architectures limit growth to small incremental steps and stifle the deployment of innovations. Today’s technology change rates call for a stream of new solutions, with high-level workflows including the fusion of multi-dimensional information.

A well designed architecture allows for interoperability with other software tools. It encompasses the exploration, capture, storage, search, integration, and sharing of data and analytical tools to comprehend that data, combined with modern interfaces and visualization in a seamless environment.

Good guidelines for a robust architecture include Microsoft’s Oil and Gas Upstream IT Reference Architecture. Another is IBM’s Smarter Petroleum Reference Architecture.

It was decided to implement the platform on Microsoft frameworks that support a service-oriented architecture. A framework is a body of existing source code and libraries that can be used to avoid having to write an application from scratch. There are numerous framework and design pattern choices for different levels of the architecture and too many for a review here. Object Relational Mapping is a good bridge between the data model and the application logic and the company also recommends N-Tiered frameworks.

Personnel who understand the business domain and the technology must carry out the implementation – otherwise one must plan to spend extra time discussing ontology and taxonomies. They must adhere to efficient source code development practices. The changing work environment will require tools and practices to deal with virtual teams, virtual machines and remote access. The company is using a test driven development (TDD) approach. This approach increases a developer’s speed and accuracy. It keeps requirements focused and in front of them, eliminating time spent on unnecessary features. It also enables parallel development of interdependent systems. In the long run, it yields dividends by reducing maintenance and decreasing risk. Using TDD, a developer can deliver high quality code with certainty.

Ultimately, the science has to come down to business. A good licensing strategy is one that will maximize revenues and allow users to buy products a-la-carte as opposed to a one-size-fits-all approach. Some vendors attempt to be creative with bundles for different levels of upgrades, but a configurable platform allows maximum user choice among available, even perhaps competing technologies. The market will favor vendors that innovate and manage data and licenses well.

Geophysical Insights’ neural network application presents an opportunity to examine seismic data in ways and means orthogonal to those of the legacy systems today. The research platform enabled the company to use this application as a configurable service. Making the right choices in information and application architectures and frameworks was the key to achieving the business objectives of modular services. The company can now move forward with additional science modules and tangential neural network processes, servicing a rapidly changing landscape, licensed to fit specific needs.

References

1. Kohonen, T., Self-Organising Maps, 3rd ed. (2001).

2. American Geosciences Institute, ‘Geoscience Currents’.

3. The Professional Petroleum Data Management Association (Accessed on 29 July 2011).

Seismic Attribute Analysis Can Benefit From Unsupervised Neural Network

Seismic Attribute Analysis Can Benefit From Unsupervised Neural Network

Process identifies anomalies from original data without bias

By: Tom Smith, Ph.D. Geophysical Insights and Deborah Sacrey, Auburn Energy
Published with permission: Offshore Magazine
September 2011

The primary task facing a seismic interpreter is to recognize and attribute a geologic significance to observable patterns in the seismic response. The most apparent patterns are found in seismic reflections. In recent years, the industry is using more subtle patterns and connecting them to such attributes as porosity, lithology, and fluid content, as well as underground structure.

The separation of such patterns and their use as potential identifiers of subsurface characteristics comprises attribute analysis, a customary instrument in the geoscientist’s toolkit. Over the years, seismic data volumes have increased in terms of geographic area covered, depth of interest, and the number of attributes. New and potentially disruptive technologies have developed to take advantage of all the attributes available in the seismic data.

One new technology, based on unsupervised neural networks (UNN), reveals deeper insights into the seismic response and thereby reduces exploration risk. Unsupervised neural network technology can help interpreters recognize seismic anomalies that may indicate the presence of hydrocarbons, often when conventional techniques fall short. This new technology may also find application in the prediction of lithologies and fluid properties, as well as in estimating the size of reservoirs.

The self-organizing map (SOM), a form of UNN and a powerful pattern recognition method, was initially developed by Prof. Teuvo Kohonen of Finland during the 1970s-80s. Based on this approach, UNNs assist the interpreter in prospect evaluation by:

  • Enabling the rapid comparison of large sets of seismic attributes
  • Identifying combinations of attributes that reveal seismic anomalies
  • Distilling the interpretation process to identify hydrocarbons with greater speed and accuracy.

Supervised vs. unsupervised UNNs

For several years there have been a few commercial tools in the upstream industry that are based on “supervised” neural networks. A supervised neural network (SNN) operates on data that has been classified, i.e., “ground truth” is known in specific locations, providing reference points for calibration of the network. In terms of seismic data, for example, a segment of a seismic survey at each logged well is employed to calibrate the SNN. Supervised neural networks unite the seismic data at the well to the known conditions from the well, while inferring geophysical properties away from wells.

But what happens if there is no well log or other forms of ground truth available, such as in a greenfield exploration?

This is the advantage of unsupervised neural networks, which do not need an “answer” beforehand and cannot, therefore, be biased.

seismic interpretation of a salt dome

UNNs can be applied to “unclassified” data, which operates on the seismic data response alone.  Hence, the technology can be used to identify where even exploratory wells should be drilled by identifying geobodies that are different than their surroundings.

UNN analysis - offshore West Africa

Offshore examples

The edge of salt has been interpreted in this example along with several channel systems working through the data. Operating on this same data using a UNN produces results where the same edge of the salt outline can be seen, but “anomalous” areas (in white) stand out.

A few geobodies are associated with the previously interpreted channels in the original amplitude data.  The red horizontal line across marks the time slice which displays a horizontal view of the same salt dome.  In this time slice, the edge of the salt body has been interpreted along with a few of the major faults created by the salt uplift.

The white regions have been identified by the UNN as “anomalies” that have distinctly different properties than their surroundings, potentially indicating the presence of hydrocarbons.  These would be regions that deserve further analysis by the interpreter.

The UNN analysis also yields a classification volume in which the interpreter can see in greater details stratigraphic and structural elements that might not have been interpreted in the original amplitude volume.

The UNN Process

The following four tasks comprise a thorough analysis using an unsupervised neural network:

  • Carry out an assessment that reveals the right choice of seismic attributes
  • Perform an appropriate interpretation of attributes for the geologic trends of interest
  • Select the well information, where available, and calibrate the data
  • Generate new attribute volumes – The UNN classification and classification reliability.

One key to an effective analysis and interpretation is the selection of the best seismic attributes, exposed by a systematic assessment of the data.  Using Eigenvalue and Principal Component Analysis (PCA), it is practical to ascertain the relative contribution for each of the attributes to help guide the selection process.  Running multiple UNN analyses using different sets of attributes also may help understand their impact and if the results change with different sets of attributes.

A major transformation is required to take full advantage of the explosion of data in the oil field. Unsupervised neural network technology facilitates greater insights into all forms of data, but perhaps the greatest value to the oil and gas industry is derived from its application to seismic interpretation.  The technology is proving to reveal new insights into interpretation and increasing the reliability of the results.  Unsupervised neural networks can also be used to correlate well information with well log data while improving the value of reservoir simulation.  This new technology has the potential to be “disruptive” to the industry by providing a tool that comes close to the direct detection of hydrocarbons.

 
Dr. Thomas Smith

THOMAS A. SMITH

Tom Smith received B.S. and M.S. degrees in Geology from Iowa State University. In 1971, he joined Chevron Geophysical as a processing geophysicist. In 1980 he left to pursue doctoral studies in Geophysics at the University of Houston. Dr. Smith founded Seismic Micro-Technology in 1984 and there led the development of the KINGDOM software suite for seismic interpretation.  In 2007, he sold the majority position in the company but retained a position on the Board of DIrectors.  SMT is in the process of being acquired by IHS. On completion, the SMT Board will be dissolved. IN 2008, he founded Geophysical Insights (www.geoinsights.com) where he and several other geophysicists are developing advanced technologies for fundamental geophysical problems.

The SEG awarded Tom the SEG Enterprise Award in 2000, and in 2010, GSH awarded him the Honorary Membership Award.  Iowa State University awarded him Distinguished Alumnus Lecturer Aware in 1996 and Citation of Merit for National and International Recognition in 2002. Seismic Micro-Technology received a GSH Corporate Star Award in 2005.  In 2008, he founded Geophysical Insights to develop advanced technologies to address fundamental geophysical problems. Dr. Smith has been a member of the SEG since 1967 and is also a member of the HGS, EAGE, SIPES, AAPG, GSH, Sigma XI, SSA, and AGU.

Deborah Sacrey DEBORAH SACREY is the Frank and Henrietta Schultz Chair and Professor of Geophysics in the ConocoPhillips School of Geology & Geophysics at the University of Oklahoma. He has devoted his career to seismic processing, seismic interpretation and reservoir characterization, including attribute analysis, multicomponent 3-D, coherence and spectral decomposition. Marfurt began his career at Amoco in 1981. After 18 years of service in geophysical research, he became director of the University of Houston’s Center for Applied Geosciences & Energy. He joined the University of Oklahoma in 2007. Marfurt holds an M.S. and a Ph.D. in applied geophysics from Columbia University.

Neural network notices anomalies in seismic data

Q&A with Dr. Tom Smith

By Gene Kliewer

Tom Smith and Geophysical Insights are bringing an advanced mathematical analysis method to bear on seismic interpretation. The concept of unsupervised neural networks is being extended to seismic interpretation to speed interpretation and direct the interpreter’s attention to the key areas in the data that represent the anomalies being sought as indicators of possible hydrocarbons-bearing zones.

To get a better understanding of the process, Offshore magazine visited with Smith to peek behind the curtain to see how a  self-organizing map can benefit exploration.

Offshore: When you say neural network, what do you mean? How does it work?

Smith: The inspiration for the fellow who had this breakthrough was inspired by the brain and the visual cortex in particular. The big breakthrough was to think about a learning process which would be incorporated into a two-dimensional map of neurons – computer “neurons” in this case.

Here we’re thinking about a neuron as a little depository of information. That repository is based upon what gets to it. The neuron network is presented a set of information and the neurons adapt themselves to the information presented to them.

The neurons adjust to the characteristics of the data following a set of straightforward rules.  This results in what are called self-organizing maps (SOMs).

The input data can be completely unclassified (without reference “ground truth”) and the learning progression is unattended.

Large volumes of data can be evaluated by an unsupervised neural network quickly,  making their use reasonably practical. They can also be programmed to operate unattended and report by exception when anomalies are detect.

The unsupervised neural network works without any supervision at all.  Wells are not required for this process.

Offshore: What data do you start with?

Smith: Normally these days we might start with four or five cubes of data, each a seismic attribute.  The old standard is the seismic amplitudes.  Today, interpreters work commonly with at least five and as many as 30 different attributes, all derived from a single amplitude.  Those are then all input into this unsupervised neural network process.

There are two parts to this process. The first part is the learning.  That literally is taking the attribute cubes and presenting those to the neurons. You end up with a SOM. After this self-training, the second step is then to apply the results of the two dimensional map.

How we apply it is to go back to the data cube and take every single data point from those original data cubes and compare them to the neural network result.  We find the neuron that is the closest… that learned it was the best match to that particular sample.  We keep this process up. We go through all the samples in the original data, not just the amplitude but all the attributes and we do a comparison.  This is the classification process.  Once we’re done with that we will have taken every piece of our input data and found it nearest neighbor … the neuron that is the closest match.

Offshore: What about the math involved in the program?

Smith: If you think about a particular point in a 3D survey, if you’ve got five attributes, you’ve got five numbers associated with that point in the image.  Mathematically, that just a little column vector.  So, if it’s five attributes, the neuron is also a column vector that has place for five numbers.

We have done a good deal of design and development of the technology. Today, we are delivering the results of our analysis to clients on a turn-key basis.  We also have been making substantial investments in research and development.  Going forward, Geophysical Insights will announce a dramatic, commercial-grade software framework and applications that use unsupervised neural network technology.

So, our work in geobodies has demonstrated the ability to automatically identify areas that don’t fit, where the material is substantially different than its surroundings, and differentiate between noise and valuable information. We find those areas, the dimensions, and rank them as to size and shape, and deliver those. Once we have a geobody, we go back and post it in the original volume.  If this is a geobody with some shape, that’s what makes it an anomaly.

Offshore: Why is your technology being called “disruptive?”

Smith: We have good evidence that it will substantially reduce the risk and time associated with the interpretive process. This new technology has great potential to “disrupt” the industry by providing a tool that comes close to the direct detection of hydrocarbons.

neural network process

Introduction to Self-Organizing Maps in Multi-Attribute Seismic Data

Introduction to Self-Organizing Maps in Multi-Attribute Seismic Data

By Tom Smith and Sven Treitel
Published with permission: Geophysical Society of Houston
January 2011

Unsupervised neural network searches multi-dimensional data for natural clusters. Neurons are attracted to areas of higher information density. The SOM analysis relates to subsurface geometry and rock properties while noting multi-attribute seismic properties at the wells, correlating to rock lithologies, with those away from the wells.

Computers that think like a human are well beyond our current capabilities but computers that learn are not. They are around us every day. Pocket cameras identify faces in a live digital image and automatically adjust the focus when the shutter is pressed. Post offices scan the mail and route the documents appropriately. Offices scan documents as bitmaps and convert them to text documents for editing. Web documents are indexed for content, while search engines deliver these documents through key word searches in unprecedented detail and with extraordinary speed.

We have seen a tremendous growth in the size of 3D survey seismic data volumes, and it is common today for both 2D and 3D seismic surveys to be integrated into the interpretation. Moreover, the primary survey of reflection amplitude is interpreted along with derived surveys of perhaps 5 to 25 attributes. The attributes of both 2D and 3D surveys represent multidimensional data. The problem is to keep all this data in one’s head while trying to find oil and gas. Much interpretation effort is devoted to building a geologic framework from the seismic data, identifying key reflecting intervals where oil and gas might be found and finding an interesting anomaly. At this point attributes are the framework in which we evaluate the anomaly. But this is the point where we can easily mislead ourselves. It is quite easy to build a plausible model for a prospect using only those attributes which fit our model and ignore the rest. This is bad enough, but there is even a greater crime. Lurking in the data may be combinations of attributes which are legitimate anomalies but which are never found at all.

Learning machines are artificial neural networks which can construct an experience data base from multidimensional data such as multi-attribute seismic surveys. There are two main classes of neural networks – supervised and unsupervised. With supervised neural networks, a network classifies data into groups sharing given characteristics that have already been classified by an expert. After careful processing, synthetic seismograms that are prepared at well sites serve as the expert’s data. Then the neural network is trained to classify these data at the wells. After training, the neural network literally roams the seismic data to classify areas which might be similar in some given sense to models developed at the well locations.

Alternatively, an unsupervised neural network searches multidimensional data for natural clusters. Neurons are attracted to areas of higher information density. The most popular unsupervised neural network, self-organizing maps (SOM), were introduced by Teuvo Kohonen in 1981 [1]. SOM was successfully applied to seismic facies analysis by Poupon, Azbel and Ingram in 1999 (Stratimagic) [2]. We preface recent efforts to bring SOM to bear on multiattribute seismic interpretation with a simple SOM example used by Kohonen to illustrate some of its basic features.

Quality of Life

An early problem considered by Kohonen and his research team was to identify natural clusters as they relate to quality of life factors based on World Bank data. A study that included 126 countries, considered a total of 39 measurements describing the level of poverty found in each country. While the data matrix was somewhat limited by incomplete reporting, the SOM results are still quite interesting. Shown in Figure 1 is the SOM which resulted from the learning process. Canada (CAN) and the United States of America (USA) clustered at the same neuron location shown at the 6th row of the 1st column in the figure. Ethiopia (ETH) is found on the right edge at column 13, row 5. Other country abbreviations and further details are in [3].

 

SOM

Figure 1: Self-organizing map (SOM) of World Bank quality of life data.

The reason that countries of similar quality of life cluster in similar neuron areas has to do with learning principles that are built into SOM. In this study, every country is a sample and that sample is a column vector of 39 elements. In other words, there are 39 attributes in this problem. Countries of similar characteristics (a natural cluster) plot in about the same place in attribute space. At the beginning of the learning process, neurons of 39 dimensions are assigned random numbers. During the learning process, the neurons move toward natural clusters. The data points never move. The mathematics of SOM learning define both competitive and cooperative learning. For a given data sample, the Euclidean distance is computed between the sample and each neuron. The neuron which is “nearest” to the data sample is declared the “winning” neuron and allowed to advance a fraction of the distance toward the data sample. The neuron movement is the essence of machine learning. Competitive learning is embodied in the strategy that the winning neuron moves toward the data sample.

This aspect of cooperative learning is related to the layout of the neural network. In SOM learning, the neural network is commonly a 2D hexagonal grid. This constitutes the neuron topology; the choice of a hexagonal grid rather than a rectangular grid will be apparent shortly. When a winning neuron has been found, cooperative learning takes place because the neurons in the vicinity of the winning neuron (the neighborhood) are also allowed to move toward the data sample, but by an amount less than the winning neuron. In fact, the further a neighborhood is away from the winning neuron, the less it is allowed to move. Hexagonal grids move more neurons than rectangular grids because they have 6 points of contact with their immediate neighbors instead of 4. Learning continues as winning and neighborhood neurons move toward each sample in turn until the entire set of samples has been processed. At this point, one epoch of learning has been completed. The event is marked as one time step in the learning process. For each subsequent epoch the distance a winning neuron may move toward a data sample is reduced slightly and the size of the neighborhood is also reduced. The learning process terminates when there is no further appreciable movement of the neurons. Often the number of such epochs can be in the hundreds or thousands.

As demonstrated in Figure 1, natural clustering of like-quality of life countries arises from both competitive and cooperative learning. But one may ask how is SOM learning unsupervised when the SOM map displays country labels? The answer is that in the steps just described, there is no need to order the sequence of samples in the SOM learning process. The Ethiopia sample may be processed between samples for Canada and USA with no effect on the outcome. The sample order of countries may be scrambled randomly.

ClassificationMap

Figure 2: Classification of quality of life data

salt_dome_preSOM_analysis

Figure 3: Gulf of Mexico Salt Dome

In Kohonen’s analysis of the World Bank data, the names of countries are known, however. When the SOM learning process is completed, the neuron which is closest to each country sample is labeled by the country label as shown in Figure 1. The neuron colors are arbitrary. Figure 2 is a world map in which each country is colored with the color scale used in Figure 1. Countries with similar quality of life are therefore colored similarly. Several countries which did not contribute data for the report are colored gray (Russia, Iceland, Cuba and several others). Figure 2 illustrates how the results of neural network analysis are used to classify the data. We shall see in the next section how SOM analysis and classification is an important addition to seismic interpretation.

Gulf of Mexico Salt Dome Survey

A SOM analysis was conducted on a 3D survey in the Gulf of Mexico provided by FairfieldNodal. See [4] for a description of SOM theory and a discussion of the processing steps. In particular, the introduction of a so-called curvature measure and the harvesting process are particularly relevant. Figure 3 is a vertical amplitude section across the center of the salt. Figure 4 shows the SOM analysis of 13 attributes across the same location. The SOM map is a 2D colorbar based on an 8 x 8 hexagonal grid. There are 100 epochs in the present analysis. It is readily apparent that the SOM classification is tracking seismic reflections.

SOM Classification

Figure 4: SOM classification and map. Red horizontal line marks the time of Figure 5.

Time slice

Figure 5: Time slice. Red line marks the location of Figure 4.

Shown in Figure 4 are white portions in which data have been “declassified”, a concept which we now explain. After the SOM analysis is completed, every sample in the survey is associated with a winning neuron. This implies that every neuron is associated with a given set of samples.

For any particular neuron, some samples are nearby in attribute space and others are far away. This means that there is a statistical population of distances on which to declassify what we shall call “outliers”. When a neuron is near a data sample, the probability that the sample is correctly classified is high. If a neuron and sample coincide, the probability is 100%. In Figure 4, those samples for which the probability is less than 10% are not assigned any classification. We identify such outliers as SOM anomalies. SOM anomalies are scattered about the section, with several which are larger and more compact. The horizontal red line marks the time of the time slice shown in Figure 5.

The horizontal line in Figure 5 marks the location of the section in Figure 4. Notice the white area to the right of the salt dome crossed by the red line in Figure 4 is identified as the same white area right of the salt dome and crossed by the red line in Figure 5. We note that the SOM anomaly is a discrete geobody which appears to be related to the upturned beds flanking the salt. By geobody, we mean a contiguous region of samples in the survey which share some characteristic.

Arbitrary LineFigure 6: Arbitrary line through 3D survey passing through gas-show well (left) and producing gas (right)

SOM Classifcation of 3D Survey

Figure 7: SOM classification of 3D survey. Red horizontal line marks the time for the time slice of Figure 8

Wharton County Survey

A SOM analysis was also conducted on a 3D survey in Wharton County, Texas provided by Auburn Energy. Details of this study are found in [5]. An arbitrary line through the survey between two wells is shown in Figure 6.
The well at the left presented a gas show while the well at the right developed a single-well gas field. Note the association of gas with faults F1, F2 and F3.

Figure 7 shows a portion of the results of a SOM classification run designed in the same way as in the previous example, namely by use of the same 13 attributes, an 8 x 8 hexagonal topology of neurons and a probability cut-off of 10%. Notice that this selection of attributes did not delineate the faults very well, yet SOM anomalies are found near both wells. The time slice of Figure 8 confirms that the SOM anomaly to the left of the gas-show well (left) is a geobody. A smaller second SOM anomaly is shown right of the F2 fault. Figure 9 is a time slice through the lower SOM anomaly near the gas well (right) of Figure 7. Notice that it too is a geobody. Prior to the present SOM analysis, an earlier thorough interpretation had been conducted with all available geophysical and geological data. A large set of attributes was used, including AVO gathers, offset stacks, advanced processing as well as some proprietary attributes. As a result, four wells were drilled. Two wells had no gas shows and are not marked here. No SOM anomaly was found at or near either of the two dry wells.

Upper SOM Anomaly

Figure 8: Time slice through the upper SOM anomaly of Figure 7. The red line marks the location of the arbitrary line location.

Time slice 2

Figure 9: Time slice through the lower SOM
anomaly of Figure 7.

Further Work

The next step in this work is to gain a better understanding how the patterns obtained with SOM analysis relate to subsurface geometry and its rock properties. Research is currently underway in an attempt to answer questions of this kind. It is also important to further address the relationship between multi-attribute seismic properties at the wells, which correlate to rock lithologies, with those away from the wells.

Conclusion

Computerized information management has become an indispensable tool for organizing and presenting geophysical and geological data for seismic interpretation. Databases provide the underlying environment to achieve this goal. Machine learning is another area in which computers may one day offer an indispensable tool as well. The point is particularly germane in light of successes achieved by machine learning in other fields. The engines to help us reach this objective could well be neural networks that adapt to the data and present its various structures in a way that is meaningful to the interpreter. We believe that neural networks offer many advantages which our industry is just now recognizing.

References

1. Kohonen, T., 2001, Self-Organizing Maps, 3rd edition: Springer

2. Poupon, M., Azbel K. and Ingram, J., 1999, Integrating seismic facies and petro-acoustic modeling: World Oil Magazine, June, 1999

3. http://www.cis.hut.fi/research/som-research/worldmap.html accessed 10 November, 2010

4. Smith, T. and Treitel, S., 2010, Self-organizing artificial neural nets for automatic anomaly identification: SEG International Convention (Denver) Extended Abstracts

5. Smith, T., 2010, Unsupervised neural networks – disruptive technology for seismic interpretation: Oil & Gas Journal, Oct. 4, 2010

Dr. Thomas Smith THOMAS A. SMITH

Tom Smith received BS and MS degrees in Geology from Iowa State University. In 1971, he joined Chevron Geophysical as a processing geophysicist. In 1980 he left to pursue doctoral studies in Geophysics at the University of Houston. Dr. Smith founded Seismic Micro-Technology in 1984 and there led the development of the KINGDOM software suite for seismic interpretation.  In 2007, he sold the majority position in the company but retained a position on the Board of Directors.  SMT is in the process of being acquired by IHS. On completion, the SMT Board will be dissolved. IN 2008, he founded Geophysical Insights where he and several other geophysicists are developing advanced technologies for fundamental geophysical problems.

The SEG awarded Tom the SEG Enterprise Award in 2000, and in 2010, GSH awarded him the Honorary Membership Award.  Iowa State University awarded him Distinguished Alumnus Lecturer Aware in 1996 and Citation of Merit for National and International Recognition in 2002. Seismic Micro-Technology received a GSH Corporate Star Award in 2005.  In 2008, he founded Geophysical Insights to develop advanced technologies to address fundamental geophysical problems. Dr. Smith has been a member of the SEG since 1967 and is also a member of the HGS, EAGE, SIPES, AAPG, GSH, Sigma XI, SSA, and AGU.

Unsupervised Neural Networks – Disruptive Technology for Seismic Interpretation

Unsupervised Neural Networks – Disruptive Technology for Seismic Interpretation

By: Tom Smith, Ph.D., Geophysical Insights
Published with permission: Oil & Gas Journal
October 2010

The energy industry is faced with an exploding growth of information from a variety of sources –  seismic surveys, well logs, and field production.  A step-change in technology is being developed that has the promise of geoscientists finding hydrocarbons more rapidly and with greater certainty by utilizing this large volume of data more effectively.  Further, apart from automated tools to make better use of the data being collected, the industry risks wasting this valuable resource.  Supported by advanced software, a branch of neural networks is being found to be at least one practical solution for reducing the risk and time in finding oil and gas.  Neural network technology is used today in financial services software, pattern recognition systems, and many other settings. The general class of problems addressed by neural network technology in business is varied and diverse.  While there are several commercial tools in the upstream oil & gas industry which are based on “supervised” neural networks, this paper describes how unsupervised neural network technologies can be used with “unclassified” data, a much more difficult problem having higher value results.

A supervised neural network operates on data that has been classified, i.e., the answer is known in specific locations, providing reference points for calibration purposes.  In the case of seismic data, for instance, a portion of a seismic survey at each logged well is known.  The well log provides the ground truth.  Supervised neural networks link the seismic data at the well to the known results from the well.  However, supervised neural networks have limited application since the earth is so heterogeneous thus rendering classification away from boreholes difficult.  In contrast, unsupervised neural networks do not require that the “answer” be known in advance and therefore unbiased.

The other challenge working with supervised neural networks is that statistics grow more powerful as more wells provide more classified data.  But that flies in the face of the more typical situation where our most important decisions need to be made when there are no or few wells.seismic interpretation with machine learning 01

As the number of wells increases, the value of a supervised neural network diminishes.  In contrast, unsupervised neural networks do not require drilled wells and can be run against seismic reflection data alone.

The balance of this paper describes how unsupervised neural network technology can be used to identify seismic anomalies through the use of multiple seismic attributes and how these anomalies may reveal the presence of hydrocarbons, often when conventional methods fall short.  The new technology may also find application in prediction of lithologies and fluid properties; perform a comparative analysis of wells, and select the best seismic attributes for interpretation.  In seismic interpretation, unsupervised neural networks can be used to reveal subtle geologic features that may have been missed by conventional analytic methods. Through the balance of the paper, the term “neural network” will refer to only the unsupervised form of the technology.

Case Study:  Auburn Energy

Four wells have been drilled since 2006 in the “study area” of northern Wharton County, Texas.  Using a popular industry suite of seismic interpretation software, the company interpreted several locations to be lower risk gas prospects.  The first well drilled encountered a formation that flowed.  A large quantity of gas was found; however, much of that gas was not economically recoverable.  A second well was drilled two years later that found an economic gas reservoir that has produced for over three years.  Two subsequent wells have been drilled that did not find economic reserves.  In all but one case of these wells, the original seismic interpretation indicated the presence of gas reservoirs.

“From the neural network interpretation it was clear that the two dry holes were drilled in locations that were not in economic gas concentrations”, says Deborah Sacrey, Owner of Auburn Energy.  “Applying the Geophysical Insights neural network technology to some 13 attributes, we can now see that two of the four wells would not have been drilled, saving investors about $8MM.  I had included all of the AVO attribute analysis for the area of study available at the time.  The neural network attribute analysis went well beyond conventional analysis by assimilating many more attributes than conventional software tools.  We are now expanding the study area using the neural network technology to confirm additional exploratory prospects.”

As an indication of the effectiveness of the neural network technology, Figure 1 is seismic data from the existing gas field referenced above in Wharton County.  Figure 1 is comprised of seismic reflection data and fault interpretation using conventional, commercially available, seismic interpretation software.  Since two wells were dry holes, only two of the four wells drilled in the field are shown in Figure 1.   The conventional analysis in Figure 1 indicates both reservoirs, one of which could not be produced because of very fine formation particles flowing along with the gas.  The well on the left resulted in a gas ‘show’ only, while the well on the right has been producing for three years.   In Figure 2, the original data is replaced with a neural network analysis based on a combination of 13 seismic attributes, revealing two seismic anomalies located near the two wells shown in Figure 1Figure 1 vs. 2 effectively compare conventional (“before”) and neural network (“after”) on the field of study.  Since the neural network analysis resulted in only two anomalies being indicated, it is likely that only two wells would have been drilled out of the four.  Also of interest is the position of the two wells near the edge of the two seismic anomalies, suggesting a potential “near miss” in the location of the wells.

Applying Self Organizing Maps to Seismic Interpretation

 

seismic attributes

Nature is full of examples of how animals, following a few simple rules, organize themselves into assemblages such as moving flocks, schools and herds.  Moreover, they re-organize themselves after a disruption to their normal pattern of movement.  Consider a flock of migratory geese and a school of fish.  After taking flight, the flock of geese quickly organizes into the familiar ‘V’ flying pattern.  A school of fish forms and moves about as a protection against predators.  In either the case of flying geese or school of fish, the assemblage quickly disperses at the threat of a predator and quickly re-assembles once the threat is past.  In both instances, the assemblage is robust yet each individual in the group is behaving according to a few simple instructions, i.e., ‘if not the leader, follow the individual ahead and remain to the left or right’.  The neurons in a neural network are presented with data and adapt to the data following a set of simple rules. The neural network becomes, in essence, a “learning machine” whereby the network adapts to the characteristics of the data resulting in what is called Self Organizing Maps (SOM’s).  The input data is unclassified and the learning process is unattended.  The SOM is a powerful cluster analysis and pattern recognition method developed by Professor Teuvo Kohonen of Finland during the 1970’s and 80’s.  In the case study shown above, we present results based on SOM on a 3D seismic survey consisting of a large number of seismic attributes.  These results constitute an ongoing portion of our research in this area.  Neural networks offer an automated process to assist seismic interpretation, for instance, accelerating prospect evaluation by…

  • Enabling the rapid comparison of large sets of seismic attributes,
  • Identifying combinations of attributes that reveal seismic anomalies, and
  • Distilling the interpretation process to identify hydrocarbons with greater speed and certainty.

Consider for a moment the quantity of data available from a single seismic survey and how neural networks may be applied to reveal insights in the seismic reflection data.   The main task facing a geoscientist is to identify and ascribe the geologic meaning to observable patterns in the data.  The most obvious patterns are found in seismic reflections, but in recent years the industry is using more subtle patterns and relating them to such features as porosity, lithology, and fluid content, as well as underground structure.  The isolation of such patterns and their use as possible identifiers of subsurface characteristics constitutes attribute analysis, which is a standard tool in the geoscientist’s toolkit.  Over the past several years, growth in seismic data volumes has multiplied many times in terms of geographic area covered, depth of interest, and the number of attributes.  Often, a prospect is evaluated with a primary 3D survey along with 5 to 25 attributes serving general and unique purposes.  A group of just five typical seismic reflection attributes is shown in Figure 3.

For illustration purposes, Figure 4 (left) depicts three attributes from a single 3D survey.  The three points near the center highlight one data sample for three associated attributes, aligned as parallel rectangular blocks.  Converting the three attributes into a SOM attribute perspective, as shown in Figure 4 (right), each point sample is plotted in attribute space along three attribute axes, resulting in a natural cluster of similar characteristics.  The natural clusters constitute regions of higher information density and may indicate seismic events or anomalies in the data.  Two additional natural clusters are illustrated in Figure 4 as well.  Initially, neurons are randomly placed by the algorithm in attribute space.  In the “learning” stage, neurons are attracted to the data samples in the clusters in a recursive process.  Ultimately after neuron movement has finished, the neurons reveal subtle combinations of attributes that may highlight the presence and type of hydrocarbons.  While the details of the algorithm are available in the technical literature, suffice to say that Figures 1 and 2 compare a conventional seismic data display, offering limited resolution, to a neural network classification of the same data. The neural network depiction dramatically increases the resolution and insight into the data.  

Other Applications of Neural Networks

 

three seismic attributes sorted with neural networks

Large volumes of seismic data are typically good candidates for using neural networks to identify anomalies in the data.  Beyond the most immediate opportunity of using neural networks to aid seismic interpretation, other valid applications of the technology include…

  • Identifying errors and gaps in data for quality assurance
  • Analyzing seismic attributes with well log data for better predictions away from the wells
  • Integrating seismic data in reservoir characterization and simulation
  • Incorporating micro-seismic events with other seismic data for better fracture prediction

Fortunately, large sets of data can be evaluated by a neural network rapidly, typically in a matter of minutes to a few hours, making their iterative use quite practical.  They can also be programmed to run unattended and report by exception when anomalies are encountered.

Getting Started with Neural Network Technology

Continuing with the example of seismic interpretation, the following basic steps are recommended when planning a neural network application.   Since neural networks are highly specialized technology, having a thorough understanding of the methodology of neural networks and the appropriate choice of parameters for neural network classification is strongly encouraged.  The following four general tasks outline the key steps in conducting a neural network analysis.

  1. Perform an assessment that reveals the right choice of seismic attributes
  2. Conduct an appropriate interpretation of attributes for the geologic trends of interest
  3. Select the well information, where available, for calibration purposes to bring ground truth to the seismic response
  4. Generate new attribute volumes – the neural network classification and a classification reliability

One of the keys to a successful project is selecting the best choice of seismic attributes, revealed by a thorough assessment of the data.  This step will require a deep knowledge of geophysics, of course, and is optimally conducted by domain experts.  As the neural network operates on the data, visual output from various attributes will require an interpretation of the attributes for the geologic trends of interest.  Where available, well information is then used for calibration purposes to bring the all-important ground truth to the seismic response.  The complete analysis will result in two new attribute volumes – a neural network classification and a classification reliability, which identifies uncertainty in the classification.

Conclusion

A major change is needed to take full advantage of the explosion of data in the oilfield.  Neural network technology enables greater insights into all types of data but has its greatest value when applied to seismic interpretation.  Neural networks are proving their value to reduce the time and costs for the interpretation process while increasing the dependability of the results. The technology can also be used to correlate well information with well log data and enhance the quality of reservoir simulation.  Neural networks have the promise of being a disruptive technology that will accelerate and improve the industry’s use of data from the field.

About Geophysical Insights

Serving exploration and production companies, Geophysical Insights provides specialized consulting and training in the methodology of neural networks and the appropriate choice of parameters for neural network classifications.  The company’s current work is in applying neural network technology to real problems and innovative applications.  Services are delivered through client collaboration and training.  An objective of each client engagement is to enable the client to obtain a practical understanding and use of methodologies and tools that can transform the interpretation process.

 

Dr. Thomas Smith

THOMAS A. SMITH received BS and MS degrees in Geology from Iowa State University. In 1971, he joined Chevron Geophysical as a processing geophysicist. In 1980 he left to pursue doctoral studies in Geophysics at the University of Houston. Dr. Smith founded Seismic Micro-Technology in 1984 and there led the development of the KINGDOM software suite for seismic interpretation.  In 2007, he sold the majority position in the company but retained a position on the Board of Directors.  SMT is in the process of being acquired by IHS. On completion, the SMT Board will be dissolved. IN 2008, he founded Geophysical Insights where he and several other geophysicists are developing advanced technologies for fundamental geophysical problems.

The SEG awarded Tom the SEG Enterprise Award in 2000, and in 2010, GSH awarded him the Honorary Membership Award.  Iowa State University awarded him Distinguished Alumnus Lecturer Aware in 1996 and Citation of Merit for National and International Recognition in 2002. Seismic Micro-Technology received a GSH Corporate Star Award in 2005.  In 2008, he founded Geophysical Insights to develop advanced technologies to address fundamental geophysical problems. Dr. Smith has been a member of the SEG since 1967 and is also a member of the HGS, EAGE, SIPES, AAPG, GSH, Sigma XI, SSA, and AGU.