Reservoir Engineer and Artificial Intelligence Techniques for Data Analysis
- O. Tapias (SPS Ltda) | C.P. Soto (ECOPETROL-ICP) | J. Sandoval (ECOPETROL-ICP) | H.H. Perez (ECOPETROL-ICP) | A. Bejarano (I.D.A. Ltda)
- Document ID
- Society of Petroleum Engineers
- SPE Asia Pacific Oil and Gas Conference and Exhibition, 17-19 April, Jakarta, Indonesia
- Publication Date
- Document Type
- Conference Paper
- 2001. Society of Petroleum Engineers
- 5.4.6 Thermal Methods, 6.1.5 Human Resources, Competence and Training, 5.2 Reservoir Fluid Dynamics, 5.1.5 Geologic Modeling, 7.6.6 Artificial Intelligence, 5.6.1 Open hole/cased hole log analysis, 5.6.4 Drillstem/Well Testing, 5.4.7 Chemical Flooding Methods (e.g., Polymer, Solvent, Nitrogen, Immiscible CO2, Surfactant, Vapex), 7.6.1 Knowledge Management, 7.6.4 Data Mining
- 5 in the last 30 days
- 732 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 8.50|
|SPE Non-Member Price:||USD 25.00|
Different knowledge areas on the petroleum industry require to solve problems related with data transformation processes needed to generate information and knowledge. This paper describes data analysis steps using Artificial Intelligent techniques which include the problem exploration, space analysis, surveying data source, data preparation and building the appropriate data mining model. Predictive and inferential models are illustrated by applications implemented using Unsupervised Artificial Neural Networks and a Fuzzy Rule Diagnosis System. (1) The first application is able to identify well zones potentially producing hydrocarbons in Colombian PETROLEA field. This reservoir is mainly a fractured and calcareous formation. The knowledge predictive model uses Fuzzy Learning Vector Quantization (FLVQ) built from historical production tests data and spontaneous potential, short and long resistivity well logs used for training, testing and validating the model. The final Bravais & Pearson correlation factor obtained is 0.95. (2) Finally a Fuzzy Rule Diagnosis System is applied to enhanced oil recovery screening comparing technical information from reservoir, well and oil properties. The model uses heuristic and lab results for knowledge base implementation adapted to Colombian oil fields.
Conventional methods of data interpretation in reservoir engineering use algorithmic process, stochastic and empirical models that are adjusted at the phenomena conditions. Its goal is to verify the hypothesis previously adopted with validated patterns. The source of data can be experimental or lab measurements and recording or direct observations. The data analysis must be extensive to the processes and interactions among them. These data correlate with attribute values to define an object in a time domain. Furthermore, it must be considered the complexity and uncertainty generated when the data are translated from a natural language to a machine language.
New techniques in Artificial Intelligence (IA), like neural networks, fuzzy logics, Knowledge based systems, experts systems and genetic algorithms, and others are very useful in data analysis1,2,3,4. Those techniques are applied to prediction process, diagnostic and no lineal complex transforms with a high degree of uncertainty.
Claude E. Shannon5 pioneer of the Information Theory defined it as a tool for improved uncertainty handling in the real world. These problems of patterns predictions, correlations, transformations and decisions taking are supported by the availability, quality and transformation of the data source of the information and knowledge. The data analysis methods search direct interaction among the expert, his knowledge and the available information to generate knowledge models. It initiate with the problem exploration and definition, the identity of the outside variables and the space, the characterization of de solutions methods and the applications of data mining that include the data preparation, inspection and modeling. The final solution depends directly of the expert knowledge, using two generic strategies, the first is called "down-top strategy" and uses data recorded through the time in static or dynamic way. This strategy requires data mining techniques to generate knowledge as diagnosis rules, clustering models, and non-linear predictive models. A Fuzzy Learning Vector Quantization (FLVQ) model is designed to identify well zones potentially producing hydrocarbons in Colombian PETROLEA field. The second strategy is called "top-down strategy" and starts the solution using heuristic expert knowledge represented as rules, frames, objects in a knowledge base. As an example an inferential model for Enhanced oil recovery screening is designed from bibliographic information and adapted to Colombian oil fields.
|File Size||219 KB||Number of Pages||7|