Accelerating Interactive Evolutionary Computation Convergence Pace by Using Over-sampling Strategy

摘要

Traditional evolutionary computations use random sampling strategy to generate their first generation resulting in very few (if any) good solutions found in the first generation. Over-sampling is a strategy of the deliberate selection of individuals of a rare type in order to obtain reasonably precise estimates of the properties of this type. We believe that the use of over-sampling in generating the first generation of IEC would result in better performance. We proposed two types of over-sampling process in IEC (OIEC_1 and OIEC_2), and used mineral water bottle design as a research case to verify the proposed models' performance. The initial results shown that both proposed models performed better than traditional IEC.