Reservoir Production Optimization Theory Based On Multi-Objective Cooperative Computing

In the real process of oilfield production and development,multiple optimization objectives are usually involved,so it is necessary to employ multi-objective optimization methods to provide multiple optimal trade-off solutions for decision makers to carry out risk assessment and asset management.However,these objectives can not be expressed explicitly,but need to be obtained by complex numerical simulations,which increases the difficulty and cost of multi-objective optimization to a certain extent.At the same time,in view of the advantages of the surrogate models in dealing with the computationally expensive multiobjective production optimization problems and the efficient global optimization ability,the surrogate-assisted multi-objective optimization algorithms have been widely used.However,with the increase of the scale and dimension of reservoir system,this kind of methods is facing with the challenge of accuracy and efficiency.Therefore,the efficient multi-objective production optimization methods based on surrogate model are studied in this paper.Firstly,the basic methods and key technologies of surrogate-assisted multi-objective production optimization are researched.Then,in response to the large-scale and computationally time-consuming characteristics of reservoir production optimization problems,a novel design idea is proposed.Aiming at the time-consuming problem of evaluating the objective functions,this paper proposes a surrogate-assisted cooperative multi-objective optimization algorithm,which adopts a divide-and-conquer strategy,uses a set of reference vectors to decompose the objective space into multiple sub-problems for cooperative computing,and trains surrogate models to dynamically plan the search area.Besides,the distribution of the reference vectors can be adaptively adjusted according to the scale of the objective functions,thus efficient global optimization is realized.Meanwhile,in view of the large-scale property of reservoir system,this paper,from the perspective of improving the surrogate model management strategy,forms a multi-objective optimization algorithm that incorporates Pareto-rank-learning scheme with surrogate-assisted infill criterion.By introducing the classification pre-screening strategy of the multi-objective algorithm,the accurate and efficient multi-class machine learning model is employed to directly predict the dominance relationship between the candidate solutions,and then the surrogate models are trained to calculate the hypervolume improvement of pre-screened solutions,in order to maintain the convergence and diversity.The introduction of the classifier can enhance the accuracy in high-dimensional decision space and reduce the computational complexity.In terms of practical production optimization applications,this paper establishes a robust production optimization system to reducethe economic and geological uncertainty,and applies the algorithms proposed in this paper to the multi-objective benchmark problems and robust production optimization problems.These results show that the proposed methods can achieve a more comprehensive and efficient reservoir management with a higher convergence speed and can solve the practical reservoir multi-objective production optimization problems.