Information frequently is locked away in company archives and databases. Using artificial intelligence can automatically unlock hidden value in producing fields.

No matter how well a company's data is retained, organized or catalogued, the mere presence of a data warehouse is only the first step. The critical task is to extract the information out of the warehouse and make it work for the company by turning it into value.
Data mining is one of the best ways to operate on large stores of data to derive pertinent trends and patterns. Companies in many industries are taking advantage of large amounts of data from their own records or from external vendors by using data mining tools and techniques to make key business decisions.
What is data mining?
Data mining uses artificial intelligence (AI) tools to automatically find underlying structures and relationships in large amounts of data. By using pattern recognition technologies as well as statistical and mathematical techniques to sift through warehoused information, data mining helps analysts recognize significant facts, relationships, trends, patterns, exceptions and anomalies that might otherwise go unnoticed. By studying these patterns and trends, petroleum engineers can predict with increasing precision how wells and reservoirs will react to any adjustments. Companies also can use data mining techniques to visualize data or present it in an easily digestible format, as well as to check for and fill holes in the underlying data store.
Keep in mind that accurate data and appropriate models are necessary for data mining to work. There is no hidden power that will transform poor input data into useful output information. The data mining process has its own set of limitations that must be recognized.
Data mining tools
Some applications on the market are sometimes falsely referred to as data mining tools. Reporting and online analytical processing products are examples. These are useful for extracting information, but the user still has to expect and focus on a pattern beforehand. If it is not expected, it will not be detected. True data mining techniques must automate pattern detection.
Many data mining tools are available. They overlap in the problems they can solve, but for any given problem there is usually one "best" tool. Consider three of the tools popular in the oil and gas industry: artificial neural networks, genetic algorithms and hybrid systems.
Artificial neural networks are computational models broadly inspired by the organization of the human brain. Unlike conventional problem-solving algorithms, artificial neural networks can be trained to perform a particular task. This is done by presenting the system with a representative set of examples describing the problem, namely pairs of input and output samples. The artificial neural network then will extrapolate the mapping between input and output data. After training, the artificial neural network can be used to recognize data that is similar to any of the examples shown during the training phase.
Genetic algorithms, or evolutionary computing models, employ survival-of-the-fittest methods to evolve an algorithm to address a problem or "breed" a solution.
Hybrid systems incorporate user-driven analysis tools and data-driven analysis tools to accomplish a task. These systems can incorporate the best of both worlds: the neural net's ability to recognize patterns and the expert's decision-making capabilities.
Data mining in E&P
E&P companies are at different stages in their data management implementations. Some companies are more advanced than others in using data warehouses or repositories for archival and retrieval. When it comes to trending and modeling tools, production and reservoir engineers are still using conventional tools. These tools are usually spreadsheets, production-specific graphing and mapping applications and parametric simulators. Conventional E&P data visualization and surveillance software have no data mining techniques, modeling or optimization capabilities. These tools rely on the users' knowledge and experience to detect patterns and trends in wells and reservoirs. If the user does not suspect a pattern, he or she will never find it.
Can business data mining software be used in the oil and gas industry? Sure, but many of the software programs in the market today are industry-specific or tailored to suit a certain industry. And conventional artificial neural network software and other software programs based on nonparametric techniques are good for data mining and modeling but lack data preparation and data visualization tools suitable for the oil and gas industry.
A few E&P service companies are incorporating artificial neural networks and AI in their software. Most of the available commercial software, however, is optimized for geoscience applications or restricted to solving a particular problem, for example, well test analysis.
To gain the most from data mining tools, a user must be able to explore different classes of data: depth-dependent, time-dependent and static. This requires that data mining tools not be restricted to a specific application or to certain sets of input and output parameters. One new oil and gas application that adheres to this is Decide! Oil & Gas.
Mining production data
Until a well is drilled and production is started, the reservoir structure and dynamics in any given field are based on assumptions and greatly influenced by the knowledge and experience of geoscientists. However, measured field and well data such as production rates, injection rates and wellhead pressures are real and not influenced by any geological assumptions or petrophysical interpretations. Therefore, using data mining techniques on production data could unlock new business opportunities that were not anticipated before.
An application may detect patterns, but the user must be knowledgeable enough to separate meaningful patterns and trends from meaningless ones. This requires a good grounding in petroleum engineering concepts and the algorithms underlying the data mining application in use.
By using measured well data and statistical data mining tools, engineers can answer questions such as:
• Does it pay to test all producing wells periodically, or could the operator be more selective in the testing campaign?
• Which well properties were most successful for fracture stimulation?
• What other wells in the field could yield successful results similar to the best producers?
• Most importantly, why is a certain trend happening?
One of data mining's most valuable traits is that results sometimes challenge commonly held assumptions. Engineers can gain new knowledge from their assets. Users must be careful about the way they interpret the results from data mining tools.
Field examples
A highly fractured North Sea field has 28 wells and uses water injection and gas lift to enhance production. Data mining tools were used instead of reservoir simulators and decline curve analyses to forecast field production and optimize the ratio of water injection to oil production.
A neural network model (Figure 1) was trained to learn the relationships between the input (average reservoir pressure, measured wellhead pressures for all producing wells in the field injection rates of all the injectors in the field and gas lift rates) and the output (measured production rates for oil and water).
The model was trained on 80% of the available historical data, and its accuracy to predict was tested on the remaining data. Once the prediction results from the neural network model were satisfactory, the model was used for forecasting.
This model is used for more reliable forecasting of the field's production when varying the input parameters. Note that no assumptions (geological data) or calculated data (wells back-allocated production data) were used, hence the high degree of accuracy.
Then the genetic algorithm technique was used to optimize the field's water-injection-to-production ratio. In this technique, a user defines his objective for the model's output(s) and constraints for the model's input(s), then lets the model come back with the results. In the case study, the objectives were to achieve a minimum water-injection rate and a maximum oil-production rate, or at least maintain existing production rates. The constraints were to maintain existing reservoir pressure and wellhead pressures. The model also was given a range of values for water injection and gas lift rates to minimize the number of modeling runs and guide it toward realistic results faster.
A study of the above field size could be completed in about10 to 15 work days, provided all data to be used in the model is accurate and available electronically. Compared to the conventional simulation model, data-driven modeling could be at least 70% faster.
Set goals
Data mining works best when clear and measurable goals are established, such as to decrease the number of well failures per year or reduce lifting cost per barrel. Defining the goal will determine what data types and data mining methods are necessary. Data mining is by no means magic; you cannot throw any question at it and have it come back with answers and solutions. Users are required to be knowledgeable in statistics and petroleum engineering.
Recommendation
Facing lucrative oil and gas prices, production companies are looking for new ways to revive and optimize field production, faster and more effectively. Data mining allows companies to unlock hidden opportunities in their assets.

Comments