Abstract: Least Squares Support Vector Machine（LSSVM）’s predicting effect is worse in process modeling while it is employed directly in the presence of the process output measurement noise.In order to improve LSSVM’s predicting accurancy，an anti-noise online process modeling method based on non-bias LSSVM is presented.During per predicting step，plant output measuring error judgement is executed；the measuring value is revised，and the revised one is applied to formation of sample，if it differs from the predicting one seriously，consequently，the effects of noise overriding on the measuring value is reduced.The experimental results indicate that，the approach provides more accurate prediction than the direct LSSVM and existing weighted LSSVMs with the process output measurement in Gaussian white noise.

Key words: online process modeling, anti-noise, time-varying nonlinear process, Least Squares Support Vector Machine（LSSVM）

摘要： 当动态过程的输出含有测量噪声时，直接用最小二乘支持向量机（LSSVM）对过程建模预测效果较差，为了提高LSSVM模型的预测精度，提出了一种基于无偏LSSVM的抗噪在线过程建模方法。该方法在每一预测步期间对过程输出测量值进行误差判断，若输出测量值与预测值相差较大，就对测量值予以修正，然后用修正值构成样本在线建模，从而减少噪声影响。数字仿真显示，当过程输出测量值混有高斯白噪声时，该文方法比直接LSSVM和现有的加权LSSVM的预测精度要高。

关键词: 在线过程建模, 抗噪, 时变非线性过程, 最小二乘支持向量机