Waterflooding Optimization using ANN Proxies and Mean-Standard Deviation Objective Function with a Sharpe Ratio

Summary In recent years, artificial intelligence (AI) techniques, particularly artificial neural networks (ANNs), have been widely applied in the oil and gas industry for tasks like Enhanced Oil Recovery (EOR) and production optimization. Key studies have demonstrated the utility of ANNs for tasks such as well placement optimization, reservoir simulation replacement, and history matching, often combined with genetic algorithms for improved efficiency. This study applies a finance-inspired portfolio optimization approach to production strategy, using a ‘mean-standard deviation’ model for Net Present Value (NPV) to balance expected return against risk. An ANN is utilized to predict NPV’s average and standard deviation across geological realizations, allowing rapid optimization without extensive direct reservoir simulations. The methodology includes a parameter to adjust the risk-return trade-off, with higher values of the parameter favoring NPV maximization and lower values focusing on minimizing risk. Using the Brugge field as a case study, this approach has proven effective and may extend to more complex reservoirs. This framework enables engineers to assess production strategies with a focus on economic risk, optimizing for a balance between expected returns and the reliability of outcomes.