摘要
为实现铜转炉吹炼过程中的关键操作参数的准确预测,构造一种基于核偏最小二乘法的动态预测模型,并提出一种适用于动态建模的在线式异常样本剔除方法。该动态预测模型使用滑动窗方法不断更新建模数据,再利用核偏最小二乘法对动态模型的参数进行辨识,最后根据反馈的前次计算误差对本次预测值进行修正。仿真研究结果表明:该动态预估模型具有较好的泛化能力和较强的鲁棒性,并具有较好预测精度(风量预测的相对均方根误差小于10%,氧量预测的相对均方根误差小于19%)。目前,该预测模型被用于某转炉的吹炼辅助决策系统中。
In order to predict accurately the key operational parameters in copper converting process, a dynamical prediction model based on kernel partial least squares was constructed, and a method of online eliminating abnormal samples for dynamical model was presented. Firstly, moving widow method was utilized to update samples continuously in dynamical prediction model. Then, kernel partial least squares was used to identify parameters of dynamical model. Lastly, the prediction values were modified according to the last feedback computing errors. The simulation result shows that this dynamical prediction model has the performances like, better generalization, stronger robust, and preferable accuracy （the relative root mean square error of air is lower than 10%, and the relative root mean square error of oxygen is lower than 19%）. Now, the prediction model is applied in the assistant decision-making system for a conner convener.
出处
《中国有色金属学报》
EI
CAS
CSCD
北大核心
2007年第7期1201-1206,共6页
The Chinese Journal of Nonferrous Metals
基金
国家重点基础研究发展规划资助项目(2002cb312200)
国家自然科学基金重点资助项目(60634020
60574030
50374079)
博士点基金(20050533016)
作者简介
通讯作者；宋海鹰，博士研究生；电话：0731—8830394；E-majl：songhaiying1975@163.com
