Predictive Data Mining Techniques for Economic Evaluation of Unconventional Resources: The Tight Gas of Argentina
- Reynaldo José Cervantes Bravo (Consultant) | Edgar Jimenez Nieves (Consultant) | Ludmer Arcaya Arhuata (Consultant)
- Document ID
- Offshore Technology Conference
- Offshore Technology Conference Brasil, 29-31 October, Rio de Janeiro, Brazil
- Publication Date
- Document Type
- Conference Paper
- 2019. Offshore Technology Conference
- Tight Gas, Argentina, Economic Evaluation of Unconventional resources, Predictive Data Mining Techniques
- 3 in the last 30 days
- 88 since 2007
- Show more detail
- View rights & permissions
Tight gas reservoir has potential to provide a significant contribution to meet the global energy demand. Unconventional resource plays and in particular tight gas reservoir are generally characterized by lower geologic risk but higher commercial risk. For that reason, a precise understanding of the potential range can lead to the commercial success; this weighs on the economic evaluation process.
The cutting-edge method "Technical Datamining" (DM), use artificial intelligence, statists, and algorithm of learning machines to accomplish new knowledge of clustering and predictive types. Neural networks-DM are computational models that have been used in different research fields with outstanding results. Thus, models of temporal series are pursued to develop to achieve reliable estimations of the main economic indexes: NPV, IRR, Payout and investment performance in the high-risk Oil & Gas portfolios, in particular economic evaluation of unconventional/Tight Gas resources, which is our concern. Neural networks learn from experience and errors: when more wells of the investment's portfolios are added, the experience will improve.
The process of knowledge improvement begins with the extraction, transformation and loading data to the collection of the resultant model and its analysis. This involves an exhaustive work with the exploration and evaluation with the behavior of independent variables (Capex, Opex, Reserves, Gas Price and Time), the outliers, the normalization, variability and the distributions. Furthermore, it is vital to maintain a complex and extensive training of the neural network model with different parameters and iterations, using the previous experience's expert. Our study has 4 years and a monthly seasonality for processing the data in the search to optimize decision making.
The model application will be developed in the sectoral block of the Lajas Formation of the Neuquén Basin, with six wells in production, the GOIS value above 3000 MMm3 and the current recover factor estimated in 19 %. In addition to this, are expected the incorporation of new wells to the block to increase the recovery factor above 35 % and thus improve the return on investment (NPV / Investment). Finally, the construction of neural network model will provide predictive values more precisely through a time series using 80 % focusing on tasks for training and 20% for testing, with minor errors of 5 %.
Extracting hidden knowledge or information not trivial of dataset to be used in making decision.
Discovery of unknown models   in order to discover meaningful patterns and rules .
|File Size||1 MB||Number of Pages||20|