LS-SVM模型在线校正的替代法及其软测量应用

Substituting Method for LS-SVM Online Updating and Its Application to Soft Sensing

软测量技术是现代化工过程控制系统的重要组成部分,化工过程的时变性要求软测量模型及时进行在线校正,以适应新工况,保持模型的预测精度。基于矩阵计算理论,提出了最小二乘支持向量机模型在线校正的替代法,它无需重新训练,校正速度快。通过分析还提出以距离标准选择被替代个体,可提高预测精度。将该法用于某炼油厂常压塔塔顶汽油干点的软测量,效果良好,表明该法可以克服工况变化对模型预测精度的影响。

Soft sensing is one of the important parts in modern chemical process control system,and time-varying of chemical process requires the soft senor can be updated online according to new work condition to keep its modeling performance.Based on the matrix computation theory,substituting method for least squares support vector machine（LS-SVM） updating was proposed,the distance criterion for choosing substituted instance was also proposed by analyzing the reason for invalidation of LS-SVM model.The proposed method was applied to the LS-SVM based soft sensor for the gasoline end point of a crude distillation unit and the result shows that the method can avoid the influence of variation of work condition to prediction performance of the model.