Measurement-Based Load Modelling of Oil Drilling Rig Microgrid Using Machine Learning

What is it about?

In remote oil drilling applications, the efficient operation of their associated islanded microgrids is critical to ensuring uninterrupted energy supply, optimizing resource utilization, and reducing operational costs. The load profile needed by an oil well drilling rig fluctuates wildly especially during lifting or lowering the heavy drill string by the draw-works resulting in more power demand. This fluctuation in power demand can cause the diesel engine that powers the drilling rig to operate inefficiently. Hence, accurate load modelling and prediction for this pulsating load are essential for optimizing the microgrid performance of the oil drilling rig and facilitating effective energy management. This research paper presents a comprehensive study on load modeling and predictive analytics for microgrids in the context of oil drilling rigs. The measurement-based data and advanced machine learning techniques are used to develop an accurate load model. The proposed model can be utilized for studies of dynamic stability, energy saving solutions and creating a new adaptive protection approach. The simulation results have shown that the exponential GPR regression model proves to be the most accurate, one out of the 18 regression methods studied, with an error rate of 3.09%.

Featured Image

Photo by WORKSITE Ltd. on Unsplash

Photo by WORKSITE Ltd. on Unsplash

Why is it important?

The significance of load modeling and accurate predictions in the challenging conditions of oil drilling rig microgrids is rooted in their vital role in securing reliability, cost optimization, enhanced safety, and support for environmental objectives within these demanding operations. This research aims to offer valuable insights into the development of precise load models and predictions for microgrids in oil drilling rigs. The findings of this study lead to enhance performance of oil drilling rig operations. This research has employed a range of regression models, including linear models, artificial neural networks, support vector machine, ensemble methods, decision trees, and gaussian process regression. Detailed descriptions and results of each model are provided in this study. Out of the 18 regression methods tested to capture the load profile, the exponential GPR regression model proves successful in identifying the pattern of the oil drilling rig's load profile. It achieved a low NRMSE of 3.09%, indicating high accuracy. The NRMSE value of 3.09% and an R-Square value of 0.99 emphasize the precision of the exponential GPR regression model.

Perspectives