基于BP神经网络预报钢锭成分的软件开发

Software of Prediction Model Based on Artificial Neural Network for Steel Ingredients

采用3层BP神经网络来预测炼钢成品的C、Si、Mn成分,根据炼钢的实际生产数据,选取铁水、废钢、供氧、吹氩、硅锰合金、增碳剂等28个因素作为输入变量,对输入参数进行归一化处理,采取附加动量项和自适应学习步长的措施,解决了BP神经网络局部收敛和学习时间过长的问题,提高了神经网络预报的准确率,并用VC++语言编写程序。软件经生产现场运用后,模型预测结果表明:在规定的误差内(C±0.02%、Si±0.05%、Mn±0.06%),预报命中率达到85%以上,证明了模型的有效性。

The prediction of C,Si,Mn components of finished steel was discussed in this paper which adopted 3 layers BP neural networks. According to the actual steel production data, the molten iron, scrap, oxygen, argon blowing, Silicon-manganese alloy, carbon etc 28 factors were selected as input variables. The variables were also normalized. Additional momentum and adaptive learning step were used to solve the BP neural network local convergence and the problem of long time to learn and improve the forecasting accuracy. The software was programed by C language and was applied in produce flied. The results showed that within the specified error （C ± 0.02%, Si ± 0.05%, Mn± 0.06%）, the model reach the forecast rate over 85%.