高炉铁水硅含量自组织预测中的模式量化

研究了高炉铁水硅含量自组织经验进化预测模型中的模式量化问题。在模式量化方案中，采用高炉过程变量时间序列数据的均值、梯度值和波动值作为数据的特征最来进行特征提取，将可预测率、命中率、趋势命中率等判据用于评判预测效果，并用天津铁厂1号高炉的过程数据进行了离线检验。结果表明：基于过程变量的特征提取方法可用于具有均匀时间间隔的高炉过程数据的特征提取。正确运用该方法可使铁水硅含量自组织预测模型的预测命中率提高10％左右。

Pattern Classification of Model for Predicting Silicon Content in Hot Metal Based on Self-Organized Experience Evolution

The pattern classification of prediction model for silicon content in hot metal based on self-organized experience evolution was presented. The mean, gradient and variance of process variables on blast furnace were used for feature extraction. A series of off-line prediction for silicon content in hot metal were conducted based on the data taken from the blast furnace at Tianjin Iron Plant and criteria of prediction, such as predictability, hit ratio, hit ratio in trend and so, were used for evaluation of the model and pattern classification. It is shown that the designated feature extraction works well for process data with regular time interval. In this case, hit-ratio of prediction for silicon content in hot metal is increased by about 10 %.