Sammendrag
Generally, optimal well controls to maximize net present value (NPV) are obtained by coupling of numerical reservoir simulation with an optimization algorithm. This approach requires a significant number of numerical reservoir simulations that are computationally expensive and time-consuming due to complex flow behavior of reservoir. As a result, it becomes a necessity to develop a fast and accurate alternative. Light mathematical models, such as proxy models, have a high capability to identify very complex and non-linear behavior in short time, such as complex dynamic flow behavior of reservoir.
This study proposes a methodology that begins by developing smart proxy models (SPMs) for a synthetic field model, based on Artificial Neural Networks (ANNs) and then integrate the established proxy models with Genetic Algorithm (GA) to solve the well control optimization problem. Three ANN models are developed using reservoir simulator data to predict field production profiles, i.e., field oil production rate, field water injection rate, and field water production rate based on sets of well control values, i.e., bottom-hole pressures (BHP). Latin hypercube sampling is used to prepare the database utilized for constructing SPMs. Hyperparameter optimization study assists in finding the best ANN architecture for each proxy. Various performance metrics are explored to comment on "goodness" of the proxy models. The proposed methodology also includes sensitivity study of GA control parameters using SPM-GA coupling and introduces the possibility for occasional retraining and multiple quality checks of ANNs as more data is gathered. From SPM-GA coupling, the optimum well control parameters, namely bottom hole pressures of injectors and producers, which maximize net present value, are investigated.
The developed proxy models produce outputs within seconds, while the reservoir simulator takes an average time of 30 minutes for the synthetic field i.e. Olympus. SPM-GA coupling works well for well control optimization study by finding BHP configuration that gives a significant increase of 35% in net present values (NPV) and requires fewer numerical simulations compared to the traditional approach. The results show that the established proxy models are found to be robust and efficient tools for mimicking the numerical simulator performance in well control optimization. Significant reduction in computational time and resources is observed.