基于递推最小二乘支持向量回归估计的建模与预报

Modeling and Prediction Based on Recursive Least Square Support Vector Regression

提出一种新的递推最小二乘支持向量回归估计算法(RLS-SVR),该算法具有实时性高、更新速度快的特点.充分应用RLS-SVR在线学习和训练的实时性,避免辨识模型的维数过高而降低估计精度,本文进一步提出了基于RLS-SVR的混合训练—估计辨识结构.TE过程的组分软测量建模和预报验证了该方法的有效性和优越性.*

A new reeursive algorithm for least square support vector regression （RLS-SVR） is proposed. This algorithm has the characteristics of improving the real-time property of LS-SVR and updating rapidly. Moreover＇ a hybrid training-regression framework based on the algorithm of RLS-SVR is also presented. The method takes advantage of the speed of online learning and training of RLS-SVR effeetively, and avoids high dimension model that will reduce the prediction precision. A soft sensor model is set up with which the composition in Tennessee Eastman （TE） process is predicted. The validity and feasibility of file presented method are illustrated.