Optimization of Well Location in W Reservoir Based on Machine Learning Agent Model

As the blood of industry, petroleum is very important to the development of national economy and National Energy Security. In the process of oilfield production and development, the determination of injection-production well location is the basis of the whole subsequent development plan of the oilfield. Therefore, the decision-making of reservoir well location has a great influence on the overall production effect. In the engineering development site, numerical simulation trials are usually carried out based on the experience of engineers or a limited number of solutions to determine the deployed well position, but it is often difficult to obtain the optimal well location plan. In this paper, a straight well position optimization method based on machine learning agent model is established, which characterizes the straight well position through the well location in two coordinate parameters in the horizontal and longitudinal directions, and forms an optimal decision space according to the area range of the straight well to be optimized, and uses the Latin hypercube method to directly sample the entire feasible decision space, and after sampling, the continuous variable is converted into discrete variable by rounding operation to obtain the sample scheme of straight well deployment. Combined with the reservoir numerical simulator, the obtained sample scheme is evaluated, and a radial basis neural network surrogate model is constructed based on the real sample library. The surrogate model coupled with the differential evolution algorithm is used to prescreen the newly generated candidate schemes, and continue to iteratively optimize until the optimal well position is obtained. This method was applied to the W reservoir in an actual oilfield in China, and the optimization design of the new infill straight well scheme was carried out. Compared with the comparison well location scheme designed according to the field development experience, the optimization scheme could obtain more cumulative oil and the displacement was more balanced. It can be used to guide the optimal deployment of straight well positions in oilfield sites.