Download

Download Full Text (718 KB)

Description

The precise prediction of the rate of penetration (ROP) is of utmost importance for optimizing drilling operations and minimizing costs while increasing efficiency. However, the complex and nonlinear nature of the drilling process can pose significant challenges in achieving accurate ROP predictions. To address this challenge, multiple hybrid prediction models have been developed, and their accuracy in ROP prediction has been compared.

To accomplish this objective, we created three different hybrid models, including Artificial Neural Network – Genetic Algorithm (ANN-GA), Artificial Neural Network-Particle Swarm Optimization (ANN-PSO), and Support Vector Regression (SVR) to estimate ROP. These models were trained and tested using drilling data collected from surface sensors, including drilling parameters such as weight on bit (WOB), revolutions per minute (RPM), flow rate, ROP, and drilling torque.

The hybrid models were able to accurately estimate the ROP for the given drilling conditions and lithologies by utilizing these parameters. Furthermore, the models' accuracy and effectiveness were assessed by training and testing them using the collected drilling data.

Upon evaluating the performance of the three algorithms, our study shows that SVR (Support Vector Regression) outperformed ANN (Artificial Neural Network) in accuracy and precision when predicting the target variable. SVR consistently provided more accurate and precise predictions, capturing the underlying patterns in the data effectively. While ANN-GA (Artificial Neural Network with Genetic Algorithm) performed better than ANN-PSO (Artificial Neural Network with Particle Swarm Optimization) in the training dataset, it exhibited lower accuracy during testing. This highlights the importance of evaluating algorithm performance in both training and testing scenarios. The results also emphasize that complexity doesn't always lead to better predictions. SVR offers a promising choice for accurate and reliable predictions, but further research is needed to explore the contrasting performances and optimize these algorithms.

Publication Date

7-27-2023

City

Bismarck, ND

Disciplines

Petroleum Engineering

Comments

Adapted from extended abstract based on oral presentation given at 2023 AAPG Rocky Mountain Section Meeting, Bismarck, North Dakota, June 4-6, 2023.

Evaluation of Hybrid Prediction Models for Accurate Rate of Penetration (ROP) Prediction in Drilling Operations

Share

COinS