论文部分内容阅读
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 view of the complexity of the influencing factors of LF smelting temperature, the GMDH neural network with self-organizing data mining as the core is proposed to predict the temperature of the molten steel terminal. The hits of error within ± 5 ℃ are 78.31% and the error of ± 7.5 ℃ hits Rate was 92.77%. The thermal effect model of material was established. Through the calculation of thermal effect of adding different materials into molten steel, the material added into LF refining process was converted into an input factor. The improved GMDH neural network was used to predict the temperature of molten steel, Hit rate was 88.72%, and the error rate within ± 7.5 ℃ was 98.44%. The GMDH neural network based on material model not only improved the hit rate but also improved the material structure change caused by smelting more steel adaptability.