基于神经网络的转炉冶炼终点锰、磷静态预测算法

Static Prediction Algorithm of Manganese and Phosphorus at the End Point of Converter Smelting on Neural Network

在国内重工业领域中,很多钢铁企业所采用的转炉大部分为最小型的转炉,由于容量有限无法对转炉冶炼结束时的锰、磷进行静态预测,进行影响了冶炼的精度。然而,传统算法用于实现锰和磷的冶炼终点。因此,充分利用最近开发的人工神经网络技术,基于Visual Basic编程语言,神经网络模型用于预测转炉冶炼结束时的锰和磷状态。针对半钢炼钢分开建立锰、磷含量、温度预测模型,确定输入层参数有37个,中间隐藏层参数有30个,输出层参数有两个3层BP神经网络。模型在30 000炉样本的基础上做数据训练,对权值、阈值进行修正,并保存100炉未训练过的学习样本作为模型网络训练依据,对转炉冶炼进行在线训练,通过训练的模型可以很好的适应转炉冶炼多变的生产条件。

In domestic heavy industry,most of the converters adopted by many iron and steel enterprises are the smallest ones. Because of the limited capacity,the static prediction of manganese and phosphorus at the end of converter smelting can't be carried out,which affects the smelting accuracy. However,the traditional algorithm is used to realize the end point of manganese and phosphorus smelting. Therefore,making full use of the recently developed artificial neural network technology and based on Visual Basic programming language,the neural network model is used to predict the manganese and phosphorus states at the end of converter smelting. The prediction models of Mn,P content and temperature are established for semi-steel steelmaking separately. 37 input layer parameters and intermediate hidden layer are determined. There are 30 parameters and two three-layer BP neural networks. The model has made the data training on the basis of 30,000 furnace samples to modify the weights and threshold values in saving 100 untrained learning samples as the basis of model network training to carry on-line training for converter smelting. The training model can adapt to the changeable production conditions of converter smelting.