Genetic Algorithms (GAs) are powerful and usable algorithms for the nonlinear optimization problem, and some studies of application of GA to the control problem were reported. The search process of GA depends on the fitness value which is assigned to each candidate solution by the fitness function. But as the control objective is more complicated, the design of the proper fitness function becomes more difficult.
In this paper, we propose a new search method of GA which reduces the difficulties of the design of the fitness function. In our method, the control objective is divided into some intermediate objectives according to the control strategy, and the search process of GA proceeds with the fitness function for the intermediate objective. The search process is controlled by switching the fitness function based on the average fitness value of the current candidate solutions so that the optimum solution is found. Thus, the search space is structured by using the fitness function and the structure is changed by switching the fitness function based on the quality of the current candidate solutions. In order to confirm the availability of the proposed method, the swing-up control problem of a pendulum is used as an application example and the simulation results are given.
J-STAGEがリニューアルされました！https://www.jstage.jst.go.jp/browse/-char/ja/
