基于物料模型的GMDH神经网络LF终点温度预测  

LF end-point temperature prediction of molten steel by GMDH neural network based on material model

作　　者：冯凯[1,2] 汪红兵[3] 徐安军[1,2] 贺东风[1,2] 田乃媛[1,2] 

机构地区：[1]北京科技大学冶金与生态工程学院,北京100083 [2]北京科技大学高效钢铁冶金国家重点实验室,北京100083 [3]北京科技大学计算机与通信工程学院,北京100083

出　　处：《炼钢》2013年第2期38-41,共4页Steelmaking

基　　金："十一五"国家科技支撑计划重大项目(2006BAE03A07);中央高校基本科研业务费(FRF-TP-12-086A)

摘　　要：针对LF冶炼终点温度影响因素的复杂性,提出以自组织数据挖掘原理为核心的GMDH神经网络对钢水终点温度进行预测,±5℃内误差的命中率为78.31%,±7.5℃内误差的命中率为92.77%;建立物料的热效应模型,通过不同物料加入钢水中的热效应计算,将LF精炼过程中加入的物料折算为一个输入因素,改进的GMDH神经网络对钢水温度预测,±5℃内误差的命中率为88.72%,±7.5℃内误差的命中率为98.44%,基于物料模型的GMDH神经网络不仅在命中率上有显著提高,而且对冶炼多钢种导致的物料结构改变有更好的适应能力。In light of the complexity of influencing factors of LF smelting endpoint tem- perature, the temperature prediction of molten steel is carried out by GMDH （Group Method of Data Handling） neural network with the self-organizing data mining principle as the core. The hit rate of temperature between ＋ 5 ℃ is 78.31%. The hit rate of temperature between ± 7.5℃ is 92. 77 %. The material thermal effect model is established. Through calculation of thermal molten steel, all materials added to the effects of different materials charging into the LF refining furnace are translated into an input factor. The temperature prediction of molten steel based on GMDH neural network is improved. The hit rate of temperature between ＋ 5 ℃ reaches 88.72 % and the hit rate of temperature between ＋ 7.5 ℃ is 98.44 %. The hit rate of temperature prediction based on GMDH neural network using material model is significantly increased and the temperature prediction has better adaptability for the changes of material structure caused by the smelting multiple steel.

关 键 词：物料模型 GMDH神经网络 钢水温度 LF 

分 类 号：TF796.2[冶金工程—钢铁冶金]