Predicting Equivalent Circulating Density While Primary Cementing Job by Employing Machine Learning Techniques.

Saved in:

Bibliographic Details
Title:	Predicting Equivalent Circulating Density While Primary Cementing Job by Employing Machine Learning Techniques.
Authors:	Saad, Elfakeur Abidi¹ elfakeur@yahoo.fr, Saad, Aissa Abidi¹, Elkatatny, Salaheldin², Brahim, Khalil Moulay¹, Abid, Mohamed Nadhir³, Horra, Hichem⁴
Source:	Petroleum & Coal. 2025, Vol. 67 Issue 2, p559-573. 15p.
Subject Terms:	Machine learning, Cementation (Petrology), Model validation, Artificial neural networks, Regression analysis, Oil well drilling rigs, Decision trees, Cement slurry
Abstract:	Equivalent Circulating Density (ECD) is a key parameter in drilling and cementing operations. Previous research on ECD prediction has primarily focused on drilling rather than cementing. In practice, the miscalculation of ECD is more significant in terms of cost, leading to large losses and time consumption compared to drilling. This study aims to address this gap by applying machine learning techniques used in drilling to the primary cementing process. Due to the lack of available pressure logging tools during casing running and cementing, along with high downhole measurement costs, alternative approaches are necessary. Several regression models, including Decision Tree Regressor (DTR), Linear Regression (LR), Random Forest Regressor (RFR), Support Vector Regressor (SVR), Gradient Boosting Regressor (GBR), K-Neighbor Regressor (KNN), Ridge Regressor (RR), Artificial Neural Network (ANN), and an ensemble ML method, were applied to predict ECD before cementing. A dataset of 1,036 simulation data points was used, with 70% for training, 15% for validation, and 15% for testing. Model evaluation through statistical and graphical analyses showed that the DT model achieved the highest accuracy (R²= 0.994) for testing, followed by the RF and Ensemble models. [ABSTRACT FROM AUTHOR]
Database:	Energy & Power Source

Description
Abstract:	Equivalent Circulating Density (ECD) is a key parameter in drilling and cementing operations. Previous research on ECD prediction has primarily focused on drilling rather than cementing. In practice, the miscalculation of ECD is more significant in terms of cost, leading to large losses and time consumption compared to drilling. This study aims to address this gap by applying machine learning techniques used in drilling to the primary cementing process. Due to the lack of available pressure logging tools during casing running and cementing, along with high downhole measurement costs, alternative approaches are necessary. Several regression models, including Decision Tree Regressor (DTR), Linear Regression (LR), Random Forest Regressor (RFR), Support Vector Regressor (SVR), Gradient Boosting Regressor (GBR), K-Neighbor Regressor (KNN), Ridge Regressor (RR), Artificial Neural Network (ANN), and an ensemble ML method, were applied to predict ECD before cementing. A dataset of 1,036 simulation data points was used, with 70% for training, 15% for validation, and 15% for testing. Model evaluation through statistical and graphical analyses showed that the DT model achieved the highest accuracy (R²= 0.994) for testing, followed by the RF and Ensemble models. [ABSTRACT FROM AUTHOR]
ISSN:	13353055