Reference Point Based Distributed Computing for Multiobjective Optimization


As the computational complexity of the problem and/or the number of objectives increases, a large population has to be evaluated at each generation of algorithm, and this process needs more computational resources, or requires more time for the same computational resource. However, distributing the tasks into different processors (or cores) is a good solution in speeding up the process overall. In this study, a novel and pragmatic distributed computing approach for multiobjective evolutionary optimization algorithm is proposed. Instead of dividing the objective space into pre-defined cone-domination principles, as proposed in an earlier study, a distribution of reference points initialized on a hyper-plane spanning the entire objective space is assigned to different processors and the R-NSGA-II procedure is invoked to find respective partial efficient fronts. Our results show that the proposed distributed computing approach reduces the overall computational effort compared to that needed with a single-processor method.