基于参数优化的最小二乘支持向量机状态估计方法  

State estimate based on parameter-optimized least square support vector machines

作　　者：陈刚[1] 闫飞[1] 龚啸[1] 王烨 

机构地区：[1]重庆大学电气工程学院,重庆400044 [2]河南省登封市电业局,河南登封452470

出　　处：《电力系统保护与控制》2011年第19期83-88,共6页Power System Protection and Control

摘　　要：考虑到最小二乘支持向量机的参数选择会对状态估计的结果产生较大影响,提出了一种基于参数优化的最小二乘支持向量机的状态估计方法。首先,在非线性回归估计模型的训练过程中,采用两层网格搜索策略和交叉验证法来动态地调整LS-SVM的参数,从而更好地反映估计模型的复杂度,以此提高状态估计的精度。其次,由于估计模型在应对不良数据时可能出现误差增大的问题,通过对核参数的特性进行分析,加入了一种鲁棒优化方法,增强了估计模型的抗差能力。仿真分析发现,该方法具有估计精度高,面对不良数据时鲁棒性好的特点。In consideration of that parameter selection of least square support vector machines exerts a major influence on state estimate, this paper presents a state estimation algorithm on the basis of parameter-optimized least square support vector machines. Firstly, in the training process of nonlinear regression estimated model, it adopts two-layer grid search strategy and cross validation to dynamically adjust parameters of the LS-SVM algorithm for better reflecting the complexity of the estimated model, thus increases the estimated accuracy. Secondly, owing to that the estimated model maybe produce big errors, this paper appends a robust method to enhance robustness of the estimated model by means of analyzing characteristic of kernel parameter. Numerical simulation indicates that the proposed method possessses higher estimated accuracy and excellent robustness coping with bad data.

关 键 词：状态估计 最小二乘支持向量机 网格搜索 交叉验证 参数优化 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]