Using the Producer Well Economic Indicator to Improve a Metaheuristic for Well Placement Optimization in Petroleum Fields

Resumo

Optimizing the production strategies in oil fields is a challenging problem. The underlying optimization problem usually has a large search space coupled with a high solution evaluation cost, given that it involves computationally expensive simulations in oil reservoir simulation software. At the same time, efficiently addressing the problem is highly desirable as it is directly related to the infrastructure's cost and the profits over the field's life cycle. Therefore, many methods can be found in the literature.In this paper, we propose using the Producer Well Economic Indicator (PWEI) of producer wells to improve Iterative Discrete Latin Hypercube Sampling (IDLHC), a heuristic for production optimization. Although IDLHC has been successfully applied to solve problems with a large number of decision variables and compared favorably to other heuristics from the literature, the method does not consider information from the domain of the problem in its search process. Therefore, we analyze using the PWEI, an individual well metric related to productivity, to help guide the search process. More precisely, IDLHC explores the search space by increasing the likelihood of selecting wells that appear frequently in high-quality solutions. We expand this strategy by also using the PWEI to direct the search towards promising regions of the search space. Our approach is guided by a weighted sum of both the PWEI and the frequency of wells in good solutions.We performed computational experiments with UNISIM-II-D, a well-known large benchmark from the literature, to analyze different weight assignment strategies and to determine if the PWEI or the frequency of wells should be prioritized. Based on our findings, we proposed an approach that combines the weight assignments that yielded the best results, as well as an additional approach in which the weights gradually change as the search process progresses. The results showed that our methods were able to both accelerate the early stages of the search and improve the average objective function value at the end of the optimization process, as evidenced by extensive computational experiments. Given the high relevance production strategy optimization in the industry, the proposed methodology offers a significant benefit, as it increases solution quality without requiring additional computationally-expensive simulations.