Neuro-simulation modeling of chemical flooding

Chemical flooding has proved to enhance oil recovery of reservoirs considerably. Development strategies of this method are more efficient when they consider both aspects of operation (recovery factor, RF) and economics (net present value, NPV). In this study, a multi-layer perceptron (MLP) neural network is developed for modeling of chemical flooding using surfactant and polymer via prediction of both RF and NPV in a unique model. The modeling algorithm is divided into three processes: training, generalization, and operation. In training process, the initial structure of the network is trained, and then the architecture of the trained network is optimized for reduction of prediction errors in generalization process. Furthermore, the optimum structure is compared with other methods like Radial Basis Function (RBF) neural network, quadratic and multi-objective regressions. The optimum architecture of the network contains one hidden layer with 8 neurons and training function of Bayesian regularization. In operation process, sensitivity analysis is studied for evaluating of effective parameters (inputs) on the performance of chemical flooding. The error is always less than 5% during the implementation of all processes. The results demonstrate that neuro-simulation of chemical flooding is reliable, inexpensive, fast in computational effort, and capable in accurate prediction of both RF and NPV in one model.