TY - JOUR

T1 - Hybrid Multi-Objective Optimization Approach in Water Flooding

AU - Al Aghbari, Mohammed

AU - Gujrathi, Ashish

AU - Al Wadhahi, Majid

AU - Chakraborty, Nirupam

PY - 2021/10/25

Y1 - 2021/10/25

N2 - Non-dominated sorting genetic algorithm, second version (NSGA-II) is used as a stochastic optimization technique successfully in different engineering applications. In this study, a data-driven optimization strategy based upon evolutionary neural network (EvoNN) algorithm is developed for providing input into NSGA-II optimization. Evolutionary neural network data-driven model is built and trained using initial solutions generated by NSGA-II optimization coupled with the reservoir simulation model. Evolutionary optimization incorporated in the EvoNN strategy is applied in the trained data-driven model to generate the Pareto optimal solution, which is then used as a guiding input into NSGA-II optimization. The described method is applied in two case studies (i.e., Brugge field model and water injection pattern model). The Pareto optimal solutions obtained with data-driven model guided NSGA-II in both models show improvement in convergence and diversity of the solution. The convergence to the Pareto optimal solution has improved by 9% for case-1 (i.e., Brugge field) and by 43% for case-2 (i.e., water injection pattern model). In addition, the Pareto optimal solution obtained by the proposed hybridization has shown improvement in the water–oil ratio (WOR) up to 6% in the Brugge field and up to 97% in the water injection pattern model. This improvement can lead to wide applications in using evolutionary optimizations in real-field simulation models at acceptable computation time.

AB - Non-dominated sorting genetic algorithm, second version (NSGA-II) is used as a stochastic optimization technique successfully in different engineering applications. In this study, a data-driven optimization strategy based upon evolutionary neural network (EvoNN) algorithm is developed for providing input into NSGA-II optimization. Evolutionary neural network data-driven model is built and trained using initial solutions generated by NSGA-II optimization coupled with the reservoir simulation model. Evolutionary optimization incorporated in the EvoNN strategy is applied in the trained data-driven model to generate the Pareto optimal solution, which is then used as a guiding input into NSGA-II optimization. The described method is applied in two case studies (i.e., Brugge field model and water injection pattern model). The Pareto optimal solutions obtained with data-driven model guided NSGA-II in both models show improvement in convergence and diversity of the solution. The convergence to the Pareto optimal solution has improved by 9% for case-1 (i.e., Brugge field) and by 43% for case-2 (i.e., water injection pattern model). In addition, the Pareto optimal solution obtained by the proposed hybridization has shown improvement in the water–oil ratio (WOR) up to 6% in the Brugge field and up to 97% in the water injection pattern model. This improvement can lead to wide applications in using evolutionary optimizations in real-field simulation models at acceptable computation time.

M3 - Article

SN - 0195-0738

VL - 144

SP - 63008

EP - 63001

JO - Journal of Energy Resources Technology, Transactions of the ASME

JF - Journal of Energy Resources Technology, Transactions of the ASME

ER -