An MOEA/D with Information Fusion in Offspring Generation

In decomposed based multi-objective evolutionary algorithms (MOEAlDs), the reproduction operator has a signif-icant influence on the performance of algorithms. Information fusion with both global and local information is a practical way to generate promising offspring solutions, and most existing MOEA/Ds conduct information fusion in offspring generation by selecting mating parents from the neighbors or the whole population for each subproblem. However, the balance between exploitation and exploration is a challenging task in these MOEA/Ds. To tackle this challenge, this work proposed a novel information fusion reproduction operator MOEA/D, called MOEA/D-IFR. In the algorithm, a probability model is built for each subproblem to extract the population distribution information, and then new trial vectors are sampled from the built model to combine some individuals to generate offspring by the proposed reproduction operator. To validate the performance of MOEA/D-IFR, two test suites of 19 benchmark instances with complicated characteristics are employed, and the experimental results suggest that the proposed reprodcution operator is promising.