Valve State Estimation for a Multi-Interval Oil Production Well Using Machine Learning Techniques

B. C. Pontes (), J. S. Grosman (), E. J. R. Coutinho (), A. B. Barreto () and M. S. Carvalho ()
Additional contact information
B. C. Pontes: Pontifícia Universidade Católica do Rio de Janeiro
J. S. Grosman: Pontifícia Universidade Católica do Rio de Janeiro
E. J. R. Coutinho: Petróleo Brasileiro S.A.
A. B. Barreto: Pontifícia Universidade Católica do Rio de Janeiro
M. S. Carvalho: Pontifícia Universidade Católica do Rio de Janeiro

Chapter Chapter 22 in Integral Methods in Science and Engineering, 2026, pp 339-350 from Springer

Abstract: Abstract With the exploration of pre-salt carbonate reservoirs in ultra-deep water in Brazil, there was a significant increase in the reservoir thickness and in the levels of oil productivity per well. To optimize the production, the subdivision of the reservoir in production zones and the use of intelligent completion have been used. A typical well configuration consists of a vertical well with three zones. Each zone has a pressure and temperature sensor in the annular and an inflow control valve (ICV). Inside the tubing, there is another pressure and temperature sensor measuring the overall flow. The objective of this work is to provide an estimate of valve status using the available pressure and temperature data to correct the historical database and to try to detect spurious valve movement. The dataset consists of real pressure, temperature, valve status, and oil production data for a test that was performed with all combinations of zones producing. Initially, an exploratory analysis of the dataset was performed through a statistical analysis. Afterward, a principal component analysis (PCA) was done to determine if we could reduce the dimensionality of the problem. Also, a two-dimensional representation of the dataset was done using the t-distributed stochastic neighbor embedding (t-SNE) approach. A workflow to explore and optimize various classification algorithms was developed in Python using the sklearn library. The RandomSearchCV technique, combined with TimeSeriesSplit, was employed to obtain optimized hyperparameters for each algorithm. By combining various features, a balanced accuracy of approximately 95% was achieved on the test data of the classification model for the status of the upper zone.

Date: 2026
References: Add references at CitEc
Citations:

There are no downloads for this item, see the EconPapers FAQ for hints about obtaining it.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:spr:sprchp:978-3-032-04458-7_22

Ordering information: This item can be ordered from
http://www.springer.com/9783032044587

DOI: 10.1007/978-3-032-04458-7_22

Access Statistics for this chapter

More chapters in Springer Books from Springer
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().