Building a General and Sustainable Machine Learning Solution in a Real-Time Drilling System
- Yuxing Ben (Occidental Petroleum Corporation) | Weilu Han (APEX) | Chris James (Occidental Petroleum Corporation) | Dingzhou Cao (Occidental Petroleum Corporation)
- Document ID
- Society of Petroleum Engineers
- IADC/SPE International Drilling Conference and Exhibition, 3-5 March, Galveston, Texas, USA
- Publication Date
- Document Type
- Conference Paper
- 2020. IADC/SPE International Drilling Conference and Exhibition
- Machine Learning, Real Time Drilling System, Rig State Classification
- 50 in the last 30 days
- 112 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 5.00|
|SPE Non-Member Price:||USD 28.00|
One of the core components of a real-time drilling analytics system is online rig state classification. We have deployed a rig state classification algorithm which leverages machine learning on a real-time drilling system. With new data continuously flowing through the machine-learning-based algorithm, monitoring performance and accuracy is a time-consuming proposition. Traditionally, misclassified results have been reviewed and corrected offline, manually. This paper will present a general and sustainable machine learning solution to address this challenge.
A convolutional neural network (CNN) model was deployed to classify rig states in a real-time drilling analytics system. After observing wells with misclassified results, the time-series data was aggregated and corrected offline. With this new, corrected data set, the original model was retrained including both the new labeled data and the original data set used to develop the model. In parallel, an offline model based on semantic segmentation and batch data aggregation was developed to continuously monitor the performance of the online model. Finally, a post-processing procedure was implemented to improve the results which were identified as inaccurate.
After studying drilling time series data from over 40 wells with over 30 million timestamps across three US onshore basins, several problems were identified that led to lower than expected CNN model inference accuracy. Those include: data integrity, data quality (unit of measure), model uncertainty, and sample distribution shift. After these issues were addressed, a new ML model was developed which had an accuracy of over 99% and f1 score of 0.99-1. The offline model, based on semantic segmentation, had better accuracy than the online model, resulting from batch data processing across entire drilling intervals. The post-processing procedure, based on domain knowledge, was applied to the data mislabeled by the online CNN model. The combination of the three techniques has led to a highly-reliable prediction which forms the foundation of drilling analytics across the entire rig fleet.
This paper presents a multi-layered approach to developing and monitoring a ML based rig state inference system for real-time drilling analytics. With rig states forming the foundation of drilling analytics, a reliable, accurate, scalable solution is required to minimize manual effort required to clean and classify data prior to analysis. The authors are not aware of any real-time drilling applications of CNN-based inference or semantic-segmentation-based post-processing to classify and monitor time-series rig state prediction.
|File Size||1 MB||Number of Pages||9|
Ben, Y., James, C., Cao, D. 2019. Development and Application of a Real-time Drilling State Classification Algorithm with Machine Learning. Presented at the SPE/AAPG/SEG Unconventional Resources Technology Conference, Denver, 22-24 July. URTEC-2019-253-MS. https://doi.org/10.105530/urtec-2019-253.
Brownlee, J. 2017. How to Get Reproducible Results with Keras. https://machinelearningmastery.com/reproducible-results-neural-networks-keras/.
Jordan, J. 2018. An Overview of Semantic Image Segmentation. https://www.jeremyjordan.me/semantic-segmentation/
Ronneberger O., Fischer, P., Brox T. 2015. U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Navab N., Hornegger J., Wells W., Frangi A. (eds) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. MICCAI 2015. Lecture Notes in Computer Science, vol 9351. Springer.