Convergence Characteristics of Exponential Evolutionary Programming

H. Narihisa, K. Kohmoto, M. Tsuda, and K. Katayama (Japan)


Evolutionary programming, Evolutionary computation, Evolutionary engineering, Exponential mutation


In this paper, we present the convergence characteristics of exponential evolutionary programming (EEP) which uses a mutation based on double exponential probability dis tribution with positive parameter value. Concerning the distribution of random number which is used in mutation of evolutionary programming (EP), it is desirable that the variance of the distribution should be large in first stage of evolution and should be small in mid to final stage of evolu tion applying on optimization problems. In order to realize such distribution of random number, we propose EEP with multi switching of parameter value of the distribution from initial value to next value at some arbitrary generation so as to decrease the variance of the distribution. Experimental results show that these switching EEP can improve con vergence performance of EEP and switching effect can be recognized.

Important Links:

Go Back