Optimization and Non-Linear Identification of Reservoir Water Flooding Process
In this study, dynamic optimization and identification of petroleum reservoir waterflooding using receding horizon (RH) principles was examined. Two forms of the strategy were compared on a realistic reservoir model. Sequential quadratic programming (SQP) was applied to optimize net present value (NPV) using water injection rates as the variables. MRST from SINTEF was used for the reservoir modeling. The identification of the reservoir was performed using nonlinear autoregressive with exogenous input (NARX) neural network from MATLAB. Data for the network training and validation was obtained by carrying out a numerical experiment on a high fidelity model of the reservoir. This model was developed with Eclipse Reservoir Simulator from Schlumberger. From the results obtained, moving-end RH gave a higher NPV than fixed-end RH with a margin of $0.5 billion. The identification algorithm was very much effective and near perfect for the studied reservoir.