A NEW EVOLUTIONARY OPTIMIZATION TECHNIQUE
Rodica Ioana Lung, Dan Andrei Dumitrescu · 2003
Roaming optimization is a new evolutionary technique that is used for multimodal optimization. Roaming uses multiple subpopulations evolving in isolation in order to detect local and global optima. Within roaming a stability measure for a subpopulation is defined and used to asses whether a subpopulation has detected an optimum. An external population (archive) it is used to store the optima already found. A new method to add solutions to the archive is presented here. Experimental results prove the efficiency of the algorithm. Keywords: genetic algorithms, multimdal optimization, roaming optimization. Genetic algorithms have proven effective in solving a variety of search and optimization problems. Determining the global optima within a fitness landscape has been the subject of much research. The intrinsic parallelism in a GA suggests, however, that this method should be able to locate several optima of a multimodal function. The problem of locating multiple solutions raises in many real world applications where the knowledge of several potential solutions provides the decision maker with a better insight into the nature of the design space and perhaps suggest alternative solutions. Fitness sharing (Goldberg, Richardson, 1987) is the most popular technique for detecting multiple optima. Other approaches consider parallel subpopulations evolving in isolation or in comunication with each other in order to locate the optima. Some of the most popular parallel models are the island models (Gordon, Whitley and Bohn, 1992). 1. ROAMING OPTIMIZATION A new evolutionary technique called roaming is proposed. Roaming is inspired by movement of nomad populations trying to find best inhabitation areas by searching different regions. In a similar manner, roaming technique uses multiple subpoputations that are searching the space in order to detect the optima. Suppose one nomad population has found a potential resourcefull area, then the nomads would either populate that area or mark it as a possible dwelling area. The subpopulations used by roaming technique are detecting the potential optima in the same manner and then store them into an external population called the Archive. Just like the nomad populations that will continue to travel to different areas, after saving an optimum the subpopulations will roam to different regions of the search space.