A double-model differential evolution for constrained waterflooding production optimization.

Recently, evolutionary algorithms (EAs) are becoming more and more popular in oil industry. It should be noted that the existence of various state constraints largely decreases the efficiency of EAs. In this paper, a double-model differential evolution (CSDE) is proposed for waterflooding production optimization with nonlinear inequality constraints. There are two main stages in CSDE: 1) the boundary of feasible region is explicitly constructed by a classifier, support vector machine (SVM), to prescreen the feasible solutions from current population, and 2) the objective function is approximated by radial basis function (RBF) surrogate model which can be used to select the best individual among feasible solutions. It is worth noticing that problems with a large number of constraints can be significantly simplified by distinguishing the feasible solutions and infeasible solutions with one unique SVM classifier. To the best of our knowledge, this is the first foray to integrate SVM and RBF into evolutionary algorithm for waterflooding optimization problem. The CSDE method has been tested on three artificial cases: three-channel, PUNQ-S3 and SPE9. The test results demonstrate that CSDE algorithm can handle constraints more efficient and achieve higher net present value (NPV) compared with the original EAs and other single-model algorithms. • A classification-based surrogate-assisted differential evolution (CSDE) is proposed. • This is the first foray to integrate SVM and RBF into evolutionary algorithm for constrained production optimization. • The experimental results demonstrate that the proposed method can handle constraints better and achieve higher NPV. [ABSTRACT FROM AUTHOR]