基于加工工序的高维多输入层神经网络及应用

Neural Network with High-dimension Multi-input Layers Based on Process of Working Procedure and Its Application

提出一种新的神经网络——高维多输入层神经网络,给出了这种神经网络的结构图及其学习算法.这种神经网络可看作是BP神经网络将一部分输入结点移至其某些隐层上的结果,因此具有较少的连接权;又因为这种神经网络可以依照生产流程安排除第一层以外各输入层的位置,只要位置安排合适,这种神经网络可达到比BP神经网络好的效果.文章将这神经网络按加工工序用于热轧产品质量的建模中,并将结果与BP神经网络的结果进行比较,事实表明,由于连接权值的减少,这种神经网络具有较快的学习速度,在同样的时间内可以达到比BP神经网络更好的学习效果.

A new neural network - neural network with Hight-dimension-input multi-input layers based on processing of working procedure is proposed; its constructure figure and algorithm are given as well in this paper. Because the new neural network can be seen as the result of moving some input dot of BP neural network to its hide layers, it is of few weigh value. In addition as the position of input dots in new neural network may be chosen according to production flow, as long as the chose is suitable, it can obtains a better effect than BP neural network. This paper uses the new neural network to the modeling of hot steel rolling production quality and compares the result with that of BP neural network. The fact shows that as number of weight values is decressed, the new neural network is of fast leaning rate and simultaneously can get a better result than BP neural network.