基于RBFN—PCR方法的制浆蒸煮终点预测模型  

Prediction model of batch pulp cooking end point based on RBFN-PCR approach

作　　者：郑启富[1] 吾国强[1] 吕延文[1] 谢建伟[1] 

机构地区：[1]浙江工业大学浙西分校化工系,浙江衢州324006

出　　处：《计算机与应用化学》2003年第4期463-466,共4页Computers and Applied Chemistry

摘　　要：实现制浆蒸煮终点的精确预测是稳定纸浆质量的关键,也是制浆行业一直未能很好解决的难题。利用径向基函数网络(RBFN)的最佳逼近性能和主成分回归(PCR)的空间变换技术,提出了一种径向基函数网络-主成分回归(RBFN-PCR)的建模方法。RBFN-PCR方法在确定隐含层结构和参数时,将隐单元数取为训练样本数,径向基函数中心矢量取相应样本值,宽度参数采用尝试方法选取,隐含层到输出层的网络权系数运用PCR求解。RBFN-PCR方法既保留了径向基函数网络的结构,又用数学方法直接求解,免去了冗长的训练过程和其它诸多欠缺。将RBFN-PCR方法用于间歇制浆蒸煮终点预测的建模,并通过实际生产数据对模型进行验证,结果显示模型的平均预测相对误差为6.21％,与文献上的10.62％相比,预测精度提高了许多。Realizes the accurate prediction to batch pulp cooking end point is a key to stabilize pulp quality, and also is a difficult problem in pulp industry which cannot be commendably solved all along. Making use of the furthest match capability of radial basis function networks (RBFN) and the space transform technology of principal component regression (PCR) , a modeling approach of RBFN combining with PCR (RBFN -PCR) was proposed. When confirming the structure and parameter of hidden layer in RBFN, the RBFN - PCR approach took the hidden layer unit number to the training swatch number, and took the center vector of radial basis function to the corresponding swatch numerical value, and chose the extent parameter by tentative method, and solved the weigh coefficient between hidden layer and output layer by PCR. The RBFN - PCR approach had the merit, i. e. using a structure similar to that of radial basis function networks, getting solution by mathematical method directly, without the tedious training process and other evoking shortcoming. A prediction model of batch pulp cooking end point was established by the RBFN - PCR approach. Through actual production data testing, the result showed that the average prediction relative error of the model is 6.21% . Comparing with the average prediction relative error 10. 62% of the literature, the prediction precision of the model proposed in the paper was enhanced very much.

关 键 词：RBFN-PCR方法 间歇制浆 蒸煮 预测模型 径向基函数网络 主成分回归 纸浆 

分 类 号：TS743[轻工技术与工程—制浆造纸工程]