Optimization of Inflow Control Device Completion Design Using Metaheuristic Algorithms and Supervised Machine Learning Surrogate

Authors:
  • CTC. Tan
    C. Tan
  • ACA. Chua
    A. Chua
  • SMS. Muniandy
    S. Muniandy
Publisher website
Google Scholar
Abstract
Inflow control device (ICD) completion design optimization is crucial for maximizing hydrocarbon recovery and net present value (NPV) in horizontal wells with heterogeneous reservoirs properties. However, conventional workflows often suffer from high computational costs and limited exploration of the vast solution space. This study presents an innovative approach that integrates machine learning surrogate modeling with the African Buffalo Optimization (ABO) algorithm to address these challenges. A case study on a synthetic reservoir model based on the Gullfaks oil field demonstrates the effectiveness of this approach. Latin Hypercube Sampling was initially used to generate training data for surrogate models, but the bias towards low-NPV cases limited their predictive accuracy. Incorporating high-quality training data with more high-NPV examples significantly improved performance, with LightGBM consistently outperforming other algorithms. The enhanced ABO-LGBM optimizer achieved a 29.77% increase in NPV compared to the open hole (OH) case, showcasing the superior capability of the integrated workflow. While computationally more intensive than simpler models, the improved accuracy of LightGBM justified the trade-off. The study also revealed the complex interaction of fluid flow dynamics and reservoir heterogeneity, emphasizing the need for further research in ICD optimization. The proposed workflow significantly reduced computational time and SME man-hours for up to 75%, offering a promising pathway for enhancing the efficiency and effectiveness of ICD completion design in the oil and gas industry. Future work could explore advanced sampling techniques, alternative machine learning (ML) algorithms, hybrid optimization approaches, uncertainty quantification, real-field validation and incorporation of operational constraints to further advance this field.