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

摘要: 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．

Abstract: A new recursive 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 effectively,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 the presented method are illustrated.