电能质量扰动分类中特征选择问题的研究

以电能质量扰动信号为研究对象,首先分别从时域和变换域两个角度初步选取24个特征值构成初始特征空间。然后采用包括最优和次优搜索法的几种常用特征选择方法对所得到的初始特征空间进行特征选择,并基于3种常见分类方法,利用分类准确率对所选特征向量的有效性进行验证。研究结果表明:通过特征选择可以明显地改善各种分类器的性能。同时也发现不同的分类器其最优特征空间也有所不同,所以在设计分类器的同时也应该合理考虑特征值的选择问题。

Feature selection in power quality event classification

The primal feature space is comprised with twenty-four features derived from the PQ disturbances signal based on the time domain and transform domain. In order to select the more useful features among the primal feature space, several general feature selection methods including the optimal and sub-optimal search approaches are applied. The efficiency of the selected features is tested by three well known classifier techniques in terms of the percentage of the accuracy. By inspecting the results of the experiments, it was observed that the performance of the classifier is improved evidently through the selection of the features. The simulation results also indicate that the efficiency of the selected features depends on not only the number of desired features but also the utilized classifier.It means that the feature selection should be considered in the process of the classifier design simultaneously.