Title:
METHOD FOR EVALUATING UNCERTAINTY OF RESERVOIR BY USING DEEP LEARNING
Document Type and Number:
WIPO Patent Application WO/2019/088543
Kind Code:
A1
Abstract:
The present invention relates to a method for evaluating uncertainty of a reservoir by using deep learning, the method being capable of: extracting, through a deep learning technique, model features for numerous reservoir models generated from static data, clustering similar reservoir models on the basis of the extracted model features, and selecting representative models so as to maintain an uncertainty range of the entire reservoir models with a small number of models; and, when dynamic data is ensured, comparing predicted dynamic data of the representative models for each cluster with observed dynamic data so as to select optimum representative models, and selecting final models having high similarity to the optimum representative models so as to improve the uncertainty range, and the method comprising: a step of preparing static data; a step of generating, by a geostatistics technique, a plurality of reservoir models utilizing the static data; a learning step of learning all the models with an auto-encoder; an encoding step of extracting, by means of the learned auto-encoder, feature vectors of the reservoir models; a step of evaluating similarity (distance) of the reservoir models according to the extracted feature vectors; a grouping step of grouping, by a clustering technique, similar models according to the similarity; a representative model selecting step of selecting representative models by grouped cluster; a first simulation step of simulating a reservoir for the representative models; a determination step of determining whether there is dynamic data observed from the reservoir; a step of evaluating uncertainty by using the simulation result of the first simulation step, when there is no observed dynamic data according to the determination result in the determination step; a step of selecting optimum representative models and final models by using the dynamic data and evaluating the uncertainty, when there is observed dynamic data according to the determination result in the determination step; and an inverse operation step of performing an inverse operation algorithm by using the representative models or the final models, and thus the uncertainty of a reservoir can be reliably evaluated with a small operation by using deep learning-based clustering.
Inventors:
LEE KYUNG BOOK (KR)
KIM JAE JUN (KR)
LIM JUNG TEK (KR)
KIM JAE JUN (KR)
LIM JUNG TEK (KR)
Application Number:
PCT/KR2018/012533
Publication Date:
May 09, 2019
Filing Date:
October 23, 2018
Export Citation:
Assignee:
KOREA INST GEOSCIENCE & MINERAL RESOURCES (KR)
International Classes:
G06F17/50; G01V1/28; G06N3/08
Foreign References:
| KR101625660B1 | 2016-05-31 | |||
| KR20130001706A | 2013-01-04 | |||
| KR20120118439A | 2012-10-26 |
Other References:
AHN SEONG IN: "Characterization of channelized reservoir using the neutral network incorporated with deep autoencoder", February 2017 (2017-02-01)
Attorney, Agent or Firm:
KIM, Jung Su (KR)
Download PDF: