摘要
针对在蒸煮过程纸浆Kappa值软测量中,基于经验的直接模型法在复杂工况情况下预测精度不高的问题,提出了一种分类模型软测量方法。该方法选择Daubechies小波作为分析工具,提取五维特征向量对升温过程曲线进行特征描述,并利用曲线特征对工况进行分类,对不同工况采用不同的软测量模型进行预测。用某造纸厂化浆车间的130组实际生产数据对该方法进行了检验,其中前100组数据用于训练,后30组数据用于测试。检验数据结果显示,分类模型预测标准偏差(3.87)比直接模型预测偏差(4.21)小,取得了更好的效果。
In the soft-sensing application for kappa number of cooking process, the prediction result based on direct experience model was not satisfying in complex production situations. Aim at this problem, a classified-model prediction method was presented to improve the prediction precision. In this method, Daubechies wavelet worked as an analyzer and extracted a five-dimension character vector to describe temperature-rising curve. And then the character vector was utilized to classify the complex production situations and each class of situation was assigned a suitable model to predict kappa number. The method was tested by 130 groups of production data from certain paper mill. Among them, the first 100 groups of learning data were training data and last 30 were testing data. The testing result shows that classified-model prediction method gets a lower standard error (3.87) than direct-model method (4.21). The former method gets a better result.
出处
《吉林大学学报（信息科学版）》
CAS
2004年第4期388-391,共4页
Journal of Jilin University（Information Science Edition）
基金
国家863计划经费项目(2001AA413110)
国家自然科学基金资助项目(60274033)
