A novel multi-objective optimization method for well control parameters based on PSO-LSSVR proxy model and NSGA-II algorithm

Single-objective well control problems have been studied for many years as one of the most typical optimization problems by researchers worldwide. However, single-objective optimization often could not meet the needs in practical application processes leading to multi-objective optimization problems which have better adaptability in practical applications. In this study, a new method for multi-objective well controls optimization problem using support-vector regression (SVR) proxy model and non-dominated sorting genetic algorithm-II (NSGA-II) was developed. This method was named the multi-objective optimization with proxy model (MOO-PM). In the MOO-PM method, the net present value (NPV) and cumulative oil production (COP) were considered as the objective functions while the bottom hole pressure of production wells and water injection rate of injection wells were chosen as the optimization variables. To the best of our knowledge, this is the first time that SVR proxy model and NSGA-II are applied simultaneously to the optimization of well controls problems. Compared to reservoir simulation model, the SVR proxy model was computationally more efficient. This meant the large calculation time spent on reservoir simulation operation for multi-objective optimization was sharply reduced by using the SVR proxy model. Furthermore, the accuracy of the SVR proxy model was proved by comparing it to the results of simulation of optimized cases and non-optimized cases for a synthetic field as well as a real reservoir. It was concluded that MOO-PM method could obtain better results with higher efficiency.

Automated summary

This study introduced a new method for multi-objective well controls optimization using SVR proxy model and NSGA-II algorithm, named MOO-PM. By considering NPV and COP as objective functions, and choosing bottom hole pressure of production wells and water injection rate of injection wells as optimization variables, the MOO-PM method achieved better results with higher efficiency.