Tight Reservoir History Matching And Productivity Uncertainty Analysis Based On Surrogate Model

As the most critical step in the process of reservoir numerical simulation,history matching is a process of re-understanding the reservoir through dynamic data and numerical simulation methods.Automatic history matching(AHM)can make up for the randomness,blindness and subjectivity of artificial history fitting,but its matching accuracy and time cost cannot be taken into account.As a common model simplification method in engineering design,surrogate model can balance the contradiction between matching accuracy and time cost.Aiming at the multi-objective optimization problem in history matching,this paper establishes a surrogate model based on radial basis function,and takes Pareto optimization as the optimization criterion,and puts forward five sampling strategies for history matching.For Griewank and Sphere functions,the performance of surrogate model algorithm and traditional algorithm in single objective and multi-objective optimization problems is compared.The results show that surrogate model algorithm is better than traditional algorithm in optimization accuracy and time cost,and random perturbation strategy is the best.Aiming at the characteristic model of fractured tight reservoir,the embedded discrete fracture model is established,which mainly determines six uncertain parameters and multiple objective functions such as oil rate and water rate.The influence of super parameters on matching accuracy is studied,so as to obtain the best radial basis function type,number of candidate points,weight for response surface criterion and perturbation range multiplier.Different sampling strategies are compared,the results show that the random perturbation strategy is the best.Compared with the traditional optimization algorithm,the surrogate model algorithm has higher matching accuracy and lower time cost.Finally,the surrogate model is used to analyze the productivity uncertainty of tight reservoir,and the cumulative oil and water productions of P10,P50 and P90 are determined.According to the actual reservoir,Petrel is used to establish the geological model,matching the production data of 15 wells for 20 years,and 45 uncertain parameters and two objective functions of daily oil production and daily water production are determined.The surrogate model algorithm is applied for AHM,using random perturbation strategy.After 500 iterations,the matching error is 3.36%,which meets the actual requirements of the filed,and the surrogate model is used to analyze the productivity uncertainty of tight reservoir.The research shows that the surrogate model algorithm can be used to carry out the history matching,which has higher matching accuracy and lower time cost than the traditional algorithm.Among the five sampling strategies,the random perturbation strategy and genetic algorithm strategy are the best.This method can be used for the AHM of different types of oil reservoirs or the optimization design of other related petroleum engineering schemes.