階層型ニューラルネットワークを用いた動的ジョブショップスケジューリング : 第2報,相互情報量に基づく入力情報の評価と不要な情報の削除(機械要素,潤滑,工作,生産管理など)

概要

論文の詳細を見る
This study deals with dynamic job shop scheduling problems using a multi-layer feed-forward neural network. In the first report, we showed that the trained neural network outperformed the best dispatching rules under various scheduling conditions. In this second report, we evaluate the significance of input information of the neural network by using mutual information, and then try to eliminate insignificant input units. Numerical experiments show that the information used in the best dispatching rules has high relevance to the output of the trained neural network. However, it is shown that the significance of the input information varies acoording to the scheduling conditions and other information such as machine load level that is not used in the dispatching rules has also high significance. It is also shown that the pruning algorithm of insignificant input units is effective.
一般社団法人日本機械学会の論文
2002-09-25