基于改进LS-SVR的超磁致伸缩作动器模型辨识

摘要

A LS - SVR method based on rectangle window algorithm with forgetting factor is proposed. First, this method combines rectangle windows with the conventional LS - SVR method, uses a finite number of past data to form the sample of LS - SVR method, continuously replaces old data with new data over time and keeps the total number of data unchanged. Then the method combines forgetting factor to rectangle windows in order to consider the effect of old data when discarding them. The calculating speed and the precision of model identification have been improved after ameliorating the method. Simulation example demonstrates the effectiveness and the feasibility of the presented method. Good result has been obtained when doing online modeling identification of CMM actuator with the presented method.%提出了基于遗忘因子矩形窗的最小二乘支持向量机回归( LS - SVR)方法.该方法先将矩形窗方法与传统的LSSVR相结合,用有限个过去的数据做LS - SVR的样本,随着时间增加不断在样本中丢弃旧数据加入新数据,保持样本总数不变；然后在此基础上添加了遗忘因子,在丢弃旧样本的同时兼顾了历史数据的影响.改进后的LS - SVR方法提高了运算速度,同时也提高了辨识的精度.通过仿真实例验证了该方法的有效性和可行性.将此方法应用到超磁致伸缩作动器的在线模型辨识上,取得了较好的辨识结果.