Development of Machine Learning Models for Sandface Pressure Prediction in Oil Well
The Oil & Gas (O&G) industry is increasingly leveraging Machine Learning (ML) techniques to predict well performance indicators, estimate missing operational metrics, and mitigate unexpected operational failures. However, the availability of extensive and high-quality datasets remains a majo...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
AIDIC Servizi S.r.l.
2025-07-01
|
| Series: | Chemical Engineering Transactions |
| Online Access: | https://www.cetjournal.it/index.php/cet/article/view/15436 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The Oil & Gas (O&G) industry is increasingly leveraging Machine Learning (ML) techniques to predict well performance indicators, estimate missing operational metrics, and mitigate unexpected operational failures. However, the availability of extensive and high-quality datasets remains a major challenge due to the diversity of well characteristics and proprietary industry constraints. This study employs the Society of Petroleum Engineers' (SPE) Rate Transient Analysis (RTA) dataset to predict Sandface Pressure or Bottom Hole Pressure (BHP) using Decision Tree (DT) models. Initially, a comprehensive literature review on RTA was conducted, followed by an in-depth evaluation of the dataset and its variables. Feature selection was performed based on data availability, Spearman’s correlation analysis, and Principal Component Analysis (PCA), leading to the exclusion of the "Oil Volume" variable from model training to improve predictive performance. The optimal DT model configuration was determined through cross-validation, utilizing Scikit-learn’s GridSearchCV for hyperparameter optimization. The best-performing model achieved an ??2 score of 0.982 and a Mean Squared Error (MSE) of 4.878 × 10?4, demonstrating that RTA data effectively supports BHP prediction and that DT models are well-suited for this application. These findings highlight the potential of data-driven approaches in enhancing predictive analytics for well performance monitoring and optimization in the O&G sector. |
|---|---|
| ISSN: | 2283-9216 |