抄録
A gradient-based method for global optimization of nonlinear programming problems, named "Modal Trimming Method," has been proposed. It has been shown that the capability of global search is because the renewal of the values of variables based on an extended Newton-Raphson method creates a chaotic behavior. In some cases, however, this method also creates a zigzag behavior, which may result in the deterioration of the capability of global search. In this paper, a method of searching for feasible solutions toward the Newton direction is proposed to avoid the zigzag behavior and enhance the capability of global search. In addition, the modal trimming method with this strategy is applied to some test problems, and its validity is ascertained.
