Skip to Main Content
A search method using an evolutionary algorithm such as a genetic algorithm (GA) is very effective if the parameter is appropriately set. However, the optimum parameter setting was so difficult that each optimal method depending on each problem pattern must be developed one by one. Therefore, this has required special expertise and large amounts of verification experiment. In order to solve this problem, a new method called "adaptive parameter control" is proposed, which adaptively controls parameters of an evolutionary algorithm. However, since this method just increases the selection probability of a search operator that generated a well evaluated individual, this is apt to be a shortsighted optimization method. On the contrary, a method is proposed to realize longsighted optimal parameter control of GA using reinforcement learning. However, this method does neither consider the calculation cost of search operators nor multipoint search characteristics of GA. This paper proposes a method to efficiently control parameters of an evolutionary algorithm by using the reinforcement learning where the reward decision rules are elaborately incorporated under the consideration of GA's multipoint search characteristics and calculation cost of the search operator. It is expected that this method can efficiently learn parameters to optimally select search operators of GA for approximately solving Travelling Salesman Problems (TSPs).