Self-Adaptive Parameter Control in Genetic Algorithms Based on Entropy and Rules of Nature for Combinatorial Optimization Problems

The parameter control problem is crucial to the performance of genetic algorithms. In this paper, we propose a self-adaptive approach based on entropy and rules of nature to control the parameters of algorithms. This approach utilizes the entropy of both the population and each genetic locus as the feedback to evaluate the state of algorithms. Then, parameters are adjusted according to the state of algorithms and rules of nature. This strategy avoids the impact of randomness when evaluating the status of algorithms and tracks the development of each gene in time to prevent premature and nonconvergence on a certain gene. Furthermore, this method can not only maintain the solutions with good quality but also increase the probability that the solutions with poor quality change. The experimental results demonstrate that the proposed parameter-controlling strategy is valid for the algorithm to enhance the performance for solving a variety of combinatorial optimization problems.