Drilling optimization by use of data from past drilling records is a challenging task. A sufficient amount of data from multiple sources of information is required, and the data must be standardized and properly recorded. Unfortunately, such requirements can be difficult to meet in the case of geothermal wells. This study has been an attempt to use modern data science techniques to create machine learning models that predict rate-of-penetration from past drilling records. It was necessary to overcome the limitations commonly encountered in geothermal drilling records. In addition, rate-of-penetration is not the only criterion used in drilling optimization, so other optimization objectives were also considered. In this study, machine learning models that can help to reduce nonproductive time by predicting potential tripping/problems were developed. These models were found to have satisfactory prediction accuracy to be useful in real life situations. This study also investigated the problem of incorporating “plain English” textual-type entries in drilling records into machine learning models. These textual-type entries often carry useful information about the condition of the well that may not be available elsewhere in the data records. However, the plain English remarks are recorded in nonstandardized ways, and vary greatly depending on who is making the notes. This makes the task of incorporating textual-type data into any machine learning model an expensive and time-consuming operation. We found that Bidirectional Encoder Representations from Transformers (BERT) can provide a solution for incorporating textual data into machine learning models effectively. Using textual information, together with the standardized drilling records, was found to improve the quality of predictions in most cases.

Press the Back button in your browser, or search again.

Copyright 2022, 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.