摘要
针对时变系统的在线辨识问题,提出了一种加权支持向量回归方法,根据时间信息给予历史数据不同的加权,实现了精确在线训练算法,在保持精度的同时避免了采集到新样本时重复训练,大大加快了训练速度。研究了该算法的复杂度并加以改进。将该方法应用于氯气投加系统过程模型的在线辨识,在训练速度和精度上都较为满意,这一结果说明了该算法的有效性。
Aiming at the problem of on-line identification of time-varying system, a weighted SVR （support vector regression） method was proposed which assigned different weighting factors to samples according to the time information. To avoid repetitious training when new sample arrived, an accurate on-line training algorithm was developed to implement the method by which training speed was increased greatly while accuracy was kept same. The complexity of the algorithm was explored and performance was improved. The proposed algorithm was applied to the on-line identification of chlorine dosing system. The result shows that the training rate and accuracy are satisfactory and the algorithm is effective.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2007年第17期3970-3973,共4页
Journal of System Simulation
