Validating the Grid Diversity Operator: An Infusion Technique for Diversity Maintenance in Population-Based Optimisation Algorithms

摘要

We describe a novel diversity method named Grid Diversity Operator (GDO) that can be incorporated into population-based optimization algorithms that support the use of infusion techniques to inject new material into a population. By replacing the random infusion mechanism used in many optimisation algorithms, the GDO guides the containing algorithm towards creating new individuals in sparsely visited areas of the search space. Experimental tests were performed on a set of 39 multimodal benchmark problems from the literature using GDO in conjunction with a popular immune-inspired algorithm (opt-ainet) and a sawtooth genetic algorithm. The results show that the GDO operator leads to better quality solutions in all of the benchmark problems as a result of maintaining higher diversity, and makes more efficient usage of the allowed number of objective function evaluations. Specifically, we show that the performance gain from using GDO increases as the dimensionality of the problem instances increases. An exploration of the parameter settings for the two main parameters of the new operator enabled the performance of the operator to be tuned empirically.