基于BP网络误差预测的中小型转炉智能炼钢

Intelligent Steelmaking for Small and Medium-sized Converters Based on BP Network Error Prediction

目前我国中小型转炉面临无法安装副枪系统、终点控制命中率较低等问题。实现中小型转炉终点命中的关键在于钢水中的碳含量和钢水温度值的准确预测,而烟气分析系统过度依赖于现存的机理模型,预测精确度较低且鲁棒性差。为提高碳含量和温度值的预测精确性,提高其终点命中的准确性,在研究烟气数据的基础上建立BP神经网络,挖掘烟气数据同含碳量、温度之间的隐含规律,并进行数值预测。同时针对误差产生的原因进行分析和改进,运用神经网络建立误差反向预测模型改进原有模型。经研究分析表明,基于误差预测改进的含碳量和温度预测模型,具有较高的预测精度和鲁棒性。

At present, small and medium converters in China are facing problems such as unable to install secondary gun system and low hitting rate of terminal control. Accurate prediction of carbon content in molten steel and molten steel temperature is the key to achieve the final hit of small and medium-sized converters. The flue gas analysis system relies too much on the existing mechanism model, and the prediction accuracy is low and the robustness is poor. In order to improve the accuracy of prediction of carbon content and temperature and the accuracy of end-point hit, a BP neural network was established based on the study of flue gas data, and the implicit law between flue gas data and carbon content and temperature was excavated, and the numerical prediction was carried out. At the same time, the causes of errors are analyzed and improved, and the back prediction model of errors is established by using neural network to improve the original model. The research and analysis show that the improved prediction model of carbon content and temperature based on error prediction has high prediction accuracy and robustness.