In binary multi-objective well placement optimization, multiple conflicting objective functions must be optimized simultaneously in reservoir simulation models containing discrete decision variables. Although multi-objective algorithms have been developed or adapted to tackle this scenario, such as the derivative-free evolutionary algorithms, these methods are known to generate a high number of duplicated strategies in discrete problems. Duplicated strategies negatively impact the optimization process since they: (i) degrade the efficiency of recombination operators in evolutionary algorithms; (ii) slow the convergence speed as they require more iterations to find a well-distributed set of strategies; and (iii) perform unnecessary re-evaluations of previously seen strategies through reservoir simulation. To perform multi-objective well placement optimization while avoiding duplicated strategies, this paper investigates the application of a newly proposed algorithm named MOEA/D-NFTS, with a modified diversity preservation mechanism that incorporates prior knowledge of the problem, on a multi-objective well placement optimization problem. The proposed methodology is evaluated on the UNISIM-II-D benchmark case, a synthetic carbonate black-oil simulation model in a well placement optimization problem using a binary strategy representation, indicating the presence or absence of a given candidate well position in the final strategy. The objective functions are the maximization of the Net Present Value, the maximization of the Cumulative Oil Production, and the minimization of Cumulative Water Production. The modified MOEA/D-NFTS performance is compared with a baseline algorithm without diversity preservation, and the evidence shows that the MOEA/D-NFTS produces statistically significant superior results, and is suitable for binary multi-objective well placement optimization.
Tag: Evolutionary computation
A Random Forest-Assisted Decomposition-Based Evolutionary Algorithm for Multi-Objective Combinatorial Optimization Problems
Many real-world optimization problems involve time-consuming fitness evaluation. To reduce the computational cost of expensive evaluations, researchers have been developing surrogate models to approximate the objective function values of unevaluated candidate solutions. However, most of the research has been developed for continuous optimization problems, while only a few of them address surrogate modeling for expensive multi-objective Combinatorial Optimization Problems (COPs). COPs have inherently different challenges than continuous opti- mization. For example, (i) many COPs have categorical and nom- inal decision variables; (ii) they often require the combination of both global and local search mechanisms; and (iii) some of them have constraints that make them NP-hard problems, which makes them even more difficult to solve with a reasonable number of fitness evaluations. To address these issues, this paper proposes a surrogate-assisted evolutionary algorithm that combines the decomposition-based algorithm MOEA/D, Tabu Local Search, and Random Forest as a surrogate model to approximate the objective function of unevaluated individuals on multi-objective COPs. Experiments were conducted on constrained and uncon- strained well-known multi-objective combinatorial optimization benchmark problems. The experimental results demonstrate that the proposed design outperforms state-of the-art algorithms without violating the restrictions in the number of objective function evaluations, which indicates that it may be suitable for real-world expensive multi-objective COPs.