| Resumo: | In this paper, we propose an improved version of descent direction-based multiobjective algorithm (DDMOA2). Significant modifications are introduced comparing with the originally proposed algorithm (DDMOA). DDMOA2 does not rely on the concept of Pareto dominance instead a scalarizing fitness assignment is used. Now, all population members have a probability of creating offspring. We define the concept of search matrix and population leaders for which local search is used to find descent directions. Moreover to improve efficiency, descent directions are found only for two randomly chosen objectives. The experimental study shows that the proposed approach outperforms the previous version of the algorithm with respect to the convergence to the Pareto optimal front and the diversity among obtained solutions, especially on three-objective test problems. At the same time, it provides highly competitive results with respect to other state-ofthe- art multiobjective optimizers.
|
|---|