Surrogate-based Method For Waterflooding Production Optimization

With the rapid advancement of computer computing power and theoretical technology,the deep integration of big data and artificial intelligence has promoted profound changes in the intelligent development of science and industry.While the current oil and gas field development is turning to deep-seated,deep-sea and other difficult-to-exploit resources,the intelligent transformation and upgrading of the oil and gas industry will also be an inevitable trend to face the challenge of long-term low oil prices.How to use artificial intelligence technology to maximize the potential of oil reservoirs and improve the recovery of existing oil fields is very important.To realize the smart oil and gas field development,the upgrading and improvement of production optimization methods for water-flooding reservoir development urgently need to be further explored.The large-scale production optimization problem for actual reservoir is a typical complex system optimization problem.Researchers have carried out a large number of related studies using conventional optimization theory combined with reservoir numerical simulation,but it is difficult to effectively solve the large-scale,strong nonlinear,and timeconsuming actual reservoir block production optimization problems.In this paper,a novel algorithm is proposed for the high-dimensional,strongly nonlinear,and time-consuming water-flooding reservoir production optimization problems.Machine learning based method is adopted as proxy model to approximate the input/output relationship of the numerical simulator,and then assists in the evolution in the limited number of numerical simulation runs to accelerate the optimization process.This paper proposed a surrogate-assisted optimization algorithm combined with dimensionality reduction algorithm,and used Gaussian process as proxy model to embed Sammon mapping to improve the accuracy of the proxy model on high-dimensional optimization problems.Targeting at the strongly nonlinear and expensive optimization problems,this paper proposed a global and local surrogate-assisted evolutionary algorithm.By constructing a global surrogate model,the local optimal is smoothed out and the global exploration performance of the proposed algorithm is improved.Local surrogate model is constructed in a small local promising area and optimal position of the surrogate model is sampled to accelerate the convergence process.In order to verify the performance,the algorithm was applied to test functions and reservoir production optimization problems to verify the effectiveness of the method.In addition,a hierarchical surrogate-based sampling framework is proposed.The radial basis function is used as surrogate model to select the best offspring individuals,the Euclidean distance is used to quantify the uncertainty of each offspring individual among the offspring population,and the individual with the highest uncertainty is selected.Subsequently,the solutions closest to the current optimal solution and the solutions with the highest fitness value are adopted to construct two local surrogate models,and the optimal of the two local proxy models are sampled and evaluated to accelerate the convergence process.The proposed algorithm is evaluated on 20-100 dimensional CEC2005 benchmark function set.Compared with 5 state-of-the-art algorithms,the proposed algorithm showed best performance on 14 among 25 problems benchmark functions.