ニューラルネットワークによる最短時間操船に関する研究

抄録

Recently, the minimum time optimal solutions to various maneuvering problem has been given by the authors. However, it is impossible that these solutions cover all maneuvering patterns. In order to overtake this defect, it is effective that the maneuvering system provide some interpolation ability to deduce the minimum time maneuvering solutions from the close and similar ones computed in advance. This paper proposes to apply the recent neural network theory to such minimum time maneuvering interpolation problems. The problem treated with here is a minimum time parallel shifting one. As teaching signals to the neural network, the minimum time maneuvering solutions to these problems, which are calculated by the sequential conjugate gradient restoration algorithm (SCGRA), are provided. By simulation technique, the interpolation abilities of the neural network are assessed at first and then the full-scale ship's experiment are implemented using the neural network.