Hybrid optimization approach using evolutionary neural network & genetic algorithm in a real-world waterflood development

Mohammed Al-Aghbari, Ashish M. Gujarathi*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

The hybrid optimization method of using evolutionary neural network (EvoNN) and NSGA-II algorithms is applied in two case studies. The first optimization study is applied in a benchmark model of the Brugge field consisting of 20 oil producers and 10 water injectors. The two objective functions are defined as maximizing short-term net present value (NPVS) and maximizing long-term NPV (NPVL). The second study is applied to a sector model of a Middle Eastern oil field developed by waterflooding to maximize cumulative oil production and minimize cumulative water production. The real field sector model consists of four producers and three injectors and it is run for ten years with 20 time steps. Bottom-hole pressure (BHP) for producers and water injection rates (qwi) for injectors are the decision variables used in the two studies. EvoNN data-driven model is based on the predator-prey genetic algorithm used in the training and optimization of the data. The optimization results obtained by the EvoNN algorithm are then used as guiding input in the NSGA-II optimization to re-initialize the population. Overall, the Pareto optimal solution obtained by the EvoNN guided NSGA-II has a more optimal solution with better convergence and diversity compared to the NSGA-II solution. The hybrid approach of using EvoNN guided NSGA-II resulted in a 70% improvement in the convergence and the computation demand for the Brugge field model. For the real field sector model, EvoNN guided NSGA-II algorithm resulted in a better convergence obtained at all generations compared to the NSGA-II algorithm solution. The maximum total oil production determined by EvoNN guided NSGA-II is 550.6 Mm3 compared to 522 Mm3 by NSGA-II. Water oil ratio (WOR) is reduced with lower water production obtained by the EvoNN guided NSGA-II algorithm compared to the NSGA-II algorithm. The best optimal solution from the EvoNN guided NSGA-II optimization for the real field sector is determined by the net flow method (NFM) at 521.25 Mm3 oil and 5208.6 Mm3water. The Pareto optimal solutions obtained by the EvoNN guided NSGA-II algorithm provide multiple optimum solutions for the decision-maker to manage the production and injection of the wells in the waterflood development based on the requirements and operational conditions.

Original languageEnglish
Article number110813
JournalJournal of Petroleum Science and Engineering
Volume216
DOIs
Publication statusPublished - Sep 2022

Keywords

  • Evolutionary neural network
  • EvoNN
  • Multi-objective optimization
  • NSGA-II
  • Reservoir simulation
  • Waterflood optimization

ASJC Scopus subject areas

  • Fuel Technology
  • Geotechnical Engineering and Engineering Geology

Cite this