基于混合模型的Kappa值软测量研究与应用

Study and application of soft sensing for kappa number based on hybrid model

由于技术和工艺原因,制浆蒸煮过程纸浆Kappa 值难于在线测量,而纸浆Kappa 值又是蒸煮过程的一个重要质量指标,需要严格控制。在纸浆蒸煮Kappa 值软测量技术的研究与应用中,发现基于经验模型的预测精度仍须进一步提高,本文分析了精度不够高的可能原因,提出了利用过程工况特征信息来进行经验模型残差补偿的方法,并给出了这种混合模型的结构框图,对于混合模型的校正,同时采用长期校正和短期校正两种机制。然后针对制浆蒸煮过程纸浆Kappa 值软测量建模,分析了两种残差补偿模型的建模方法。最后,以某造纸厂化浆车间的130组样本数据为对象进行分析,证明了该方法的有效性。

In the cooking process to stabilize the Kappa number is the key to stabilize the quality of paper pulp.Regrettably,until nowthe Kappa number online measurement instrument,which is suitable in our country,has not been developed.Therefore,to make theKappa number soft sensing possible is the key to control the quality of pulp.During the work of research and application of soft sensingtechnology of Kappa number,we found that the empirical model,which had good performance in laboratory,got worse effect when usedin industrial condition.To improve the prediction effect,a new hybrid model is constructed which utilizing process condition informa-tion to compensate the prediction error of the empirical model.The hybrid model is composed by two-part:classical empirical modeland error compensation model.Each part has different input variables.For the empirical model,technical parameters are main inputvariables.While for the compensation model,it is character information from production process.The correction methods for the hybridmodel,including long-term and short-term methods are also presented.When applied in the Kappa number soft sensing problem,twokinds of error compensation models are discussed in detail:Principle Component Regression(PCR)and Radical Basis Function NeuralNetwork(RBFNN).The hybrid model is proved being of higher performance by data analyzing fi'om actual factory cooking process.