基于双次归一化算法的连铸坯表面纵裂纹BP神经网络

BP Neural Network of Longitudinal Crack on Continuous Casting Slab's Surface Based on Double Normalization Algorithm

通过对板坯表面纵裂纹的机理研究,确定板坯表面纵裂纹的主要影响因素。用这些影响因素作为输入量,对某钢厂现场样本数据进行了采集、归一化处理,确定BP神经网络算法的学习速率、隐藏层神经元个数。以归一化数据作为训练样本,进行训练和仿真,使神经网络提高了命中率。d_(sin)函数是根据相关冶金原理提出的一种非线性函数,d_(sin)+Premnmx双次归一化算法使神经网络准确率明显提高。

The main factors which influence the longitudinal crack on slab＇s surfaee were identified by the study of longitudinal crack causes on continuous casting slab＇s surface. With these factors as the network input, the samples＇ data of a steel plant were collected and normalized, the learning rate of BP neural network algorithm and the number of hidden layer neurons were determined. With normalized data as the training samples for training and simulation, the hit rate of neural network was improved. The d~, function is a nonlinear function which is proposed according to the metallurgical principle, and the accuracy of the neural network improves obviously with the dsin＋Premnmx double normalization algorithm.