基于正交信号校正和稳健回归的带钢酸洗浓度预测模型

Acid concentration prediction model of steel pickling process based on orthogonal signal correction and robust regression

为了实时获得冷轧带钢酸洗溶液的浓度值,便于进行酸浓度控制,采用软测量方法实时预测酸浓度.由于酸浓度建模数据中无关成分和特异点会影响模型精度,利用正交信号校正和稳健回归相结合的方法来建立酸浓度预测模型首先利用正交信号校正对建模数据进行预处理,去除自变量中与因变量无关的成分;然后采用基于迭代加权最小二乘的稳健回归算法进行建模,降低特异点对模型的影响;最后将预测结果和多元线性回归、传统稳健回归方法和正交信号校正多元线性回归进行比较.实验结果表明:采用正交信号校正-稳健回归方法后,模型预测能力得到提高,与多元线性回归结果相比,亚铁离子质量浓度和氢离子质量浓度的相对预测误差分别从1.82%降低到1.17%、从5.87%降低到4.73%.本文提出的方法具有更好的模型预测精度,可以满足工业应用要求.

In order to get and control acid concentration values in cold-rolled strip steel pickling, a soft measurement method was proposed for real-time predicting the acid concentration. Because of the influence of irrelevant components and outliers in acid concentration data on the accuracy of the acid concentration prediction model, orthog- onal signal correction （OSC） and iterative weighted least squares （IRLS） regression were combined to build the model. Firstly, orthogonal signal correction was used to remove irrelevant components which have nothing to do with tile mea- sured variables. Then robust regression based on the iteratively reweighted least squares algorithm was applied in the model to reduce the influence of outliers. Finally, the prediction results were compared with multiple linear regression （MLR）, IRLS, and OSC-MLR. It is found that OSC-IRLS has the best prediction accuracy. In comparison with MLR, the relative error of OSC-IRLS decrease from 1.82% to 1.17% in predicting the concentration of ferrous ions and from 5.87% to 4.73% in predicting the concentration of hydrogen ions. The proposed method has a better model prediction accuracy to meet the requirements of industrial applications.