Diversity Preservation in Genetic Algorithm by Lifespan Control

Abstract

Genetic algorithms (GA) application to a real world problem which possesses various complicated constraints often requires to design each problem dependent genetic operations to keep the feasibility of the individual, which could decrease the population diversity and cause the early convergence. We propose Life Control Genetic Algorithm (LCGA) to maintain the diversity under such biased operations, by setting the lifespan of each individual depending on the relative fitness. LCGA is first applied to a typical functional optimization with largely biased crossover, and compared with several types of GA including uniform lifespan. Next, it is applied to a building design optimization as an example of a real world combinatorial optimization with complicated constraints, and the effectiveness is studied by numerical experiments.