基于BP神经网络的VD过程温降预报模型

VD temperature-drop forecast model based on the BP neural network

建立了基于BP神经网络的VD过程温降预报模型,利用五数总括值法和聚类分析法进行了BP神经网络输入数据的预处理,采用MINITAB软件确定了影响VD过程温降的主要因素为抽真空时间、保压时间、吹氩时间、非真空时间和钢水进VD过热度。利用245罐数据作为训练数据、50罐作为验证数据对模型进行了验证,结果表明:模型计算偏差在±5℃范围内的比例达到88%。

The BP neural network based VD temperature-drop forecast model is built up and the input data of the BP neural network is pretreated by five number method and cluster analysis method and principal factors affecting the VD processing temperature are determined to be the vacuumization time,vacuum duration,argon bubbling time,time of normal condition and the original superheat degree of VD by MINITAB software.The model is verified by data collected from 245 heats as the training data and by data acquired from 50 heats as the verification data.Results show that deviation of calculation within 5 ℃ by the model reaches 88% of the total.