ニューラルネットワークによる高炉炉頂部の装入物プロフィールの認識

抄録

Recently, neural network models have been applied to solve various problems and multi layer neural network with back propagation algorithm is shown to be useful for pattern recognition. In this paper, a recognition system for burden profile at the top of blast furnace using the neural network is described. Here, two problems of the neural network are studied. One is how to escape from local minima in learning process. For this purpose, a new method which updates the teaching data from simple to complicated one is proposed, and the effectiveness of the method is shown. Another problem is the selection of teaching data in recognition of scaler value, such as length. For this problem, two teaching data, which we call digital and analog type teaching data, are compared, and the digital type teaching data is shown to be better for the recognition of blast furnace data.