ニューコンセプトによる衝突回避問題解決へのアプローチ : 状況判断ニューラルネットワーク

概要

論文の詳細を見る
This paper explores a new possibility of building a collision avoidance system using a neural network. The reason using a neural network technique is aimed at parallel processing for many vehicles and at the possibility of making an integrated circuit which stored collision avoidance judgment criteria. As the first step, authors have built a circumstantial judgment sub-system using a neural network. The neural network is composed of two backpropagation neural networks, which are degree of scope network and action level network. The supervised training data is acquired from an expert of a ship operator. The summary of the results are shown below. (1) An output from the networks is in agreement with the expert's decision making when the continuous circumstances change is given in the networks. (2) When unknown data is included in the networks input, the lower correct recognition rate is given. To solve the lower correct recognition rate problem, it is necessary to have an addition to the supervised training data which gives consideration to unknown data pattern. (3) The quality of supervised training data is discussed with randomness scale. (4) The possibility of building a collision avoidance sub-system using a neural network is found.
公益社団法人日本航海学会の論文
1991-09-25