• Journal of KIISE:Software and Applications
• /
• v.37 no.8
• /
• pp.591-598
• /
• 2010

In this study, we propose a new method for generating candidate solutions based on the Cauchy probability distribution in order to complement the shortcoming of the solutions generated by the normal distribution. The Cauchy probability distribution has infinite mean and variance, and it has rather large probability in the tail region relative to the normal distribution. Thus, the Cauchy distribution can yield higher probabilities of generating candidate solutions of large-varied variables, which in turn has an advantage of searching wider area of variable space. In order to compare and analyze the performance of the proposed method against the conventional method, we carried out an experiment using benchmarking problems of real valued function. From the result of the experiment, we found that the proposed method based on the Cauchy distribution outperformed the conventional one for all benchmarking problems, and verified its superiority by the statistical hypothesis test.

• Journal of KIISE:Software and Applications
• /
• v.37 no.9
• /
• pp.694-705
• /
• 2010

The mutation operation is the main operation in the evolutionary programming which has been widely used for the optimization of real valued function. In general, the mutation operation utilizes both a probability distribution and its parameter to change values of variables, and the parameter itself is subject to its own mutation operation which requires other parameters. However, since the optimal values of the parameters entirely depend on a given problem, it is rather hard to find an optimal combination of values of parameters when there are many parameters in a problem. To solve this shortcoming at least partly, if not entirely, in this paper, we propose a new mutation operation in which the parameter for the variable mutation is theoretically estimated from the self-adaptive perspective. Since the proposed algorithm estimates the scale parameter of the Cauchy probability distribution for the mutation operation, it has an advantage in that it does not require another mutation operation for the scale parameter. The proposed algorithm was tested against the benchmarking problems. It turned out that, although the relative superiority of the proposed algorithm from the optimal value perspective depended on benchmarking problems, the proposed algorithm outperformed for all benchmarking problems from the perspective of the computational time.