120 ダイナミカルシステムによる最適化問題解決手法

抄録

A new method for solving optimization problems is proposed in terms of a dynamical system. This method aims for compatibility, which has often been problem in solutions, between two requirements : searching with a high probability for the most likely candidates of the optimal points, and searching quickly in the region. The proposed dynamical system realizes setting vlues of the visiting-weight and speed of the orbits to the system, with an assumption of ergodicity, by values of long-time limit. High values can be set for the visiting-weight to areas where the objective function takes low (high) values. Further, independently, a high-value can be set for the speed of the orbits. Consequently, the two requirements can be satisfied. Accordingly, a trade-off between a searching-weight and a searching-speed does not exist. In a numerical simulation assuming an objective function, validity within a finite time for the system formulated with the long-time limit was confirmed.