Efficient Variable-Dimension Surrogate-Assisted Evolutionary Algorithm for Petroleum-Reservoir Waterflooding Production Optimization

Summary

Optimizing petroleum-reservoir waterflooding strategies is a critical technical challenge in oil- and gasfield development, directly determining production efficiency. Traditional evolutionary algorithms and conventional surrogate-assisted algorithms often suffer from an imbalance between exploration and exploitation, high evaluation costs, and low convergence efficiency. To address these issues in this paper, we propose a variable-dimension surrogate-assisted evolutionary algorithm (SAEA) integrated with a dimension evolution strategy. The proposed algorithm breaks through the constraints of a fixed solution space by adaptively adjusting the number of decision variables during the optimization process via a dimension evolution mechanism, thereby providing an automated means of searching across different dimensional spaces without requiring prior specification of the control-step number. A dual-phase collaborative optimization framework is constructed to balance exploration and exploitation: In the global exploration phase, a multisurrogate modeling strategy based on regional extreme learning machines (ELMs) is used to rapidly identify promising regions; in the local exploitation phase, a high-precision global surrogate model combined with a dynamic neighborhood shrinkage mechanism is utilized to conduct refined searches in the optimal regions. The results demonstrate that the proposed algorithm achieves competitive performance against traditional evolutionary algorithms and other surrogate-assisted algorithms across six benchmark functions and the petroleum reservoir models, although a carefully selected fixed-dimension counterpart yields a marginally higher mean net present value (NPV) in the reservoir test. Consequently, the primary value of the proposed method is interpreted as reducing reliance on prior dimensional knowledge through automated dimension search, rather than guaranteeing statistically superior convergence. This study provides an efficient and economical method for petroleum-reservoir production optimization.