基于RPLS的造纸废水处理过程软测量建模

Soft Sensor Modeling of Papermaking Effluent Treatment Processes Using RPLS

关键词:  递归偏最小二乘  偏最小二乘  软测量建模  造纸废水处理

Key Words:recursive partial least squares  partial least squares  soft sensor modeling  papermaking effluent treatment

摘要:偏最小二乘(PLS)软测量预测模型在预测造纸废水处理过程中的出水化学需氧量(CODCr)和固体悬浮物(SS)时,易受过程非线性特性和系统外部干扰等因素的影响而失效。针对以上问题,研究了递归偏最小二乘(RPLS)算法的造纸废水处理过程软测量建模。计算结果表明,采用PLS模型预测出水CODCr时,平均绝对百分比误差(MAPE)、均方根误差(RMSE)和相关系数(R2)分别为5.3832%、4.6878和0.5892；采用RPLS模型预测时，MAPE、RMSE、R2分别为1.3861%、1.8792和0.9221。采用PLS模型预测SS时,MAPE、RMSE和R2分别为2.5962%、0.7412和0.6651；采用RPLS模型时MAPE、RMSE、R2分别为0.6795%、0.2198和0.9627。以上结果表明,RPLS预测模型比PLS预测模型具有更好的预测性能和更高的精度。

Abstract:Soft sensor modeling methods based on partial least squares (PLS) and recursive PLS （RPLS）were used to predict effluent chemical oxygen demand(COD_Cr) and effluent suspended solids(SS) in a papermaking wastewater treatment process. PLS is unsuitable for the systems with non-linear characteristics and external disturbances. The results showed that the mean absolute percentage error(MAPE),root mean square error(RMSE), and squared correlation coefficient(R2) for COD_Cr using PLS were 5.3832%, 4.6878, and 0.5892, respectively, and they were 1.3861%, 1.8792, and 0.9221, respectively using RPLS. In terms of SS, the MAPE, RMSE, and R2 were 2.5962%,0.7412, and 0.6651, respectively when using PLS, and the three indices using RPLS were 0.6795%, 0.2198, and 0.9627, respectively. These results indicated that the RPLS model had better prediction performance and higher accuracy compared to the PLS model.