Evolutionary Programming with Double Exponential Probability Distribution

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


Evolutionary programming, Evolutionary computation, Exponential mutation, Function optimization, Double exponential distribution


In this paper, a new evolutionary programming with the mutation operator based on dou ble exponential probability distribution, is investigated. In conventional evolutionary programming, the mutation operator is mainly based on normal probability distribution or Cauchy probability distribution to evolve solutions for given optimization problems. The double exponential probability distribution with one positive real valued parameter has some positive amount second moment and is symmetric with respect to origin. Although the variance of this probability distribution is neither infinite as Cauchy distribution, nor unit as standardized normal distribution, the amount of this variance is be controllable by the value of this parameter. This fact plays an important role at the evolution process in evolutionary programming. The results of computational experiment show that our proposed evolutionary programming with double exponential probability distribution performs much better than the conventional evolutional programming when ap plied to the optimization problems which are well known as the benchmark problems in this research field.

Important Links:

Go Back