基于单形进化算法优化支持向量机的运动想象脑电分类研究

Research on the classification of motor imagery EEG by optimized SVM based surface⁃simplex swarm evolution

由于支持向量机(support vector machine,SVM)优化算法存在易陷入局部最优解、控制参数较多的问题,提出一种基于单形进化(surface⁃simplex swarm evolution,SSSE)算法优化的SVM并对运动想象(motor imagery,MI)脑电信号的分类进行了研究。提取MI脑电信号模糊熵和AR(auto regressive)模型参数作为输入特征,然后将SSSE应用在SVM的参数寻优中,实现对MI脑电信号的分类。测试实验中,对2003国际BCI竞赛Data setⅢ和2008国际BCI竞赛Data sets 2b进行左右手分类,结果表明,所提方法的平均分类正确率和Kappa值分别为82.47%和0.88,单形进化算法减少了控制参数且有效避免粒子陷入局部最优,验证了该方法在MI脑电信号分类的有效性。

Because the optimization algorithm of support vector machine(SVM)falls into local optimum easily and has many control parameters,a SVM optimized by surface⁃simplex swarm evolution(SSSE)algorithm is proposed and the classification of Motor imagery EEG signals is studied.The fuzzy entropy and AR model parameters of MI EEG signals were extracted as input features,and then SSSE is applied to parameters optimization of SVM to classify MI EEG signals.In the test experiment,which classified the 2003 international brain⁃computer interface(BCI)competition Data setsⅢand the 2008 BCI competition Data sets 2b by left⁃hand and right⁃hand.The results showed that the average classification accuracy and Kappa value of the proposed method were 82.47%and 0.88 respectively.SSSE reduced the control parameters and effectively avoided the particles falling into the local optimum.The validity of this method in the classification of MI EEG signals was verified.