基于KPLS和LS-SVM的过程参数预测及其应用

针对有色冶炼净化过程流程长、影响因素多以及非线性强等特点,提出了一种结合最小二乘支持向量机和核偏最小二乘回归的过程参数预测方法。考虑到过程参数时间序列具有高噪声和非平稳等特性,首先基于小波多分辨率分析方法将参数时间序列分解成具有不同频率特征的子序列,然后根据分解后各自序列的特点,利用最小二乘支持向量机和核偏最小二乘回归法对各子序列进行建模,最后对各分量预测信号重构合成得到最终的预测结果。将该建模方法应用于锌湿法冶炼净化除钴过程钴离子浓度的预测,并应用工业现场的数据进行了实验验证,结果表明,该预测模型性能优于最小二乘支持向量机模型,具有较好的泛化能力和较强的鲁棒性。

Prediction Model of Process Parameter Based on KPLS and LS-SVM

To the characteristics of strong non linearity,multiple influence factors and large delay of the nonferrous metallurgical process,a hybrid prediction model of the process parameter is proposed by using least square support vector machine(LS-SVM)and kernel partial least squares(KPLS).The original data series of process parameter is decomposed into the sub-series with different frequency characteristics based on wavelet multi-resolution analysis.According to the features of these decomposed components,KPLS mo...