A reference vector based multiobjective evolutionary algorithm with Q-learning for operator adaptation

Maintaining a balance between convergence and diversity is a challenge for multiobjective evolutionary optimization. As crossover operators can affect the offspring distribution, an adaptive operator selection and reference vector based evolutionary algorithm (OVEA) for multiobjective optimization is proposed, where adaptive operator selection (AOS) adopts Q-learning to choose crossover operators, and the reference vector assists individual selection. To make the objective vectors as close to the true Pareto Front (PF) as possible and distributed uniformly along with PF, the different crossover operators and the association between reference vectors and individuals are used to drive the population evolution. The selection range of each reference vector is controlled by the associated population subset, from which an elite individual is selected. Observing the performance of offspring, the appropriate crossover operator is picked up by Q-learning. Finally, the proposed algorithm is evaluated on the benchmark problems with different objective number ranging from 2 to 10 and compared against the state-of-the-art algorithms. The experimental results show that OVEA has remarked advantages over the compared algorithms.