Title: |
First Year Report of EDGE Project: an International Research Coordination Network for Geothermal Drilling Optimization Supported by Deep Machine Learning and Cloud Based Data Aggregation |
Authors: |
Rolando CARBONARI, Dang TON, Alain BONNEVILLE, Daniel BOUR, Trenton CLADOUHOS, Geoffrey GARRISON, Roland HORNE, Susan PETTY, Robert RALLO, Adam SCHULTZ, Carsten F SØRLIE, Ingolfur Orn THORBJORNSSON, Matt UDDENBERG, Leandra WEYDT |
Key Words: |
machine learning, data analysis, deep learning, well optimization |
Conference: |
Stanford Geothermal Workshop |
Year: |
2021 |
Session: |
Modeling |
Language: |
English |
Paper Number: |
Carbonari |
File Size: |
1610 KB |
View File: |
|
Drilling optimization may have several definitions, but what they all have in common is the concept of drilling time and well failure reduction, which is of fundamental importance in reducing the overall costs of a geothermal project. To this purpose, an international research coordination network aimed at developing machine learning strategies to improve geothermal drilling efficiency has been established under the support of the EDGE Program of the US Department of Energy (DOE) Geothermal Technologies Office. The EDGE research collaboration involves two US Universities, a DOE National Laboratory and four Geothermal and Oil and Gas companies from several countries (Iceland, Norway, USA).The first year of the project consisted of four major tasks: 1) Data gathering from more than 100 wells from different companies and geothermal fields; 2) Exploratory Data Analysis (EDA) to assess both the quality and the structure of the data, i.e. the presence of gaps, outliers, typos, the correlation between variables and their distribution, and also to define which variables might be more useful for drilling efficiency prediction in such a way that a data format/structure can be defined as a standard for the machine learning procedures; 3) Development of a well data repository for the data products (data, code, analysis workflows and models) developed during the project; and 4) Initial development and testing of machine learning (ML) techniques. The main findings of the exploratory data analysis and initial machine learning testing can be summarized as follows: i) information related to drill bit life cycle and bottom hole assembly are necessary to improve the data clustering as well as to improve the accuracy of machine learning algorithms; ii) lithological classifications usually used to describe well cuttings are too specific and idiosyncratic to be useful for machine learning purposes in their raw form; iii) both Random Forest and Deep Learning models were tested. At present, their accuracy in predicting drilling parameters is similar overall, with Deep Learning models slightly outperforming the Random Forest ones as the number of input parameters increases. With regard to idiosyncratic lithological information as appearing in the raw mudlog data, we have tried both dummy encoding and text embedding to encode the lithological information but none of them has resulted in an improvement in the accuracy of the machine learning algorithms in predicting drilling parameters. A new “rock-strength” descrption needs to be defined for this purpose.
Press the Back button in your browser, or search again.
Copyright 2021, Stanford Geothermal Program: Readers who download papers from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the original publisher.
Attend the nwxt Stanford Geothermal Workshop,
click here for details.