Self-Adaptive Parent to Mean-Centric Recombination for Real-Parameter Optimization

Most real-parameter genetic algorithms (RGAs) use a blending of participating parent solutions to create offspring solutions through its recombination operator. The blending operation creates solutions either around one of the parent solutions (having a parent-centric approach) or around the centroid of the parent solutions (having a mean-centric approach). In this paper, we argue that a self-adaptive approach in which a parent-centric or a mean-centric approach is adopted based on population statistics is a better procedure than either approach alone. We propose a self-adaptive simulated binary crossover (SA-SBX) approach for this purpose. On a suite of eight unimodal and multi-modal test problems, we demonstrate that a RGA with SASBX approach performs reliably and consistently better in locating the global optimum solution than the RGA with the original parent-centric SBX operator and the well-known CMA-ES approach.