Intelligent Optimization Of Well Placement And Controls For Water Flooding Reservoirs

With the developing of the intelligent technology in oil and gas industry,the combination of reservoir numerical simulation and optimization algorithms has become an important method in the construction of ‘Smart Oilfield'.However,several problems remain in the use of the intelligence technology in well placement and parameter optimization.These problems include the complexity of the variable design(integer and non integer variables),dimensionality(number of wells and control duration),local minimum problem,and computational complexity of the reservoir simulation,which impact the efficiency and quality of the reservoir optimization.In this study,we introduce a systematic and efficient method,which combines reduced order model and high-fidelity surrogate model,to optimize the well placement parameters remained in the production-injection controlling for water-flooding reservoir.The main achievements are as follows:(1)The reduced order model is based on POD and DEIM,which build a reduced order reservoir simulation model for water flooding.The ROM method can increase the computational speed(10-20 times)with a relatively lower loss of accuracy(<5%).(2)Considering the high accuracy of the full-order model and the high efficiency of the low-order model,through the design of sampling strategy,the base point sample point and the non-base point sample point are defined,and a multi-fidelity reservoir Gaussian process surrogate model is established.In the surrogate model,NPV is the objective function,and well placement parameters and injection-production parameters are design variables.(3)Based on the constructed multi-fidelity surrogate model of water flooding reservoirs,combined with the point-adding criterion based on the upper confidence limit,the important design domain method,and the differential evolution non-gradient optimization algorithm retaining the elite population,a reservoir placement based on the dynamic surrogate model is established and applied to the optimization design of vertical and horizontal wells.Compared with traditional optimization algorithms,this method performs better in terms of convergence speed,optimization efficiency,and global optimization capabilities.(4)Because the method of reservoir optimization design through the surrogate model can not effectively solve the problem of‘dimension disaster' of variables,when considering multi-time step control parameters,the adjoint gradient optimization method,trajectory piecewise linear approximation(TPWL)and POD are combined,and a new controls parameters optimization method based on piecewise linear approximation macro model is proposed.Compared with the conventional adjoint gradient method,this method can reduce the optimization calculation time by about 90% when obtaining the approximate optimization scheme.(5)Using MRST toolbox and MATLAB,the proposed optimization method of well placement and control parameters is applied to a three-dimensional water flooding reservoir(Egg model),which verifies the effectiveness and applicability of the method.