基于自适应不完全S变换与LOO-KELM算法的复合电能质量扰动识别

针对电能质量复合扰动识别中特征提取效率低、分类器识别能力与学习速度无法同步提高的问题,提出一种基于自适应窗不完全S变换与留一交叉验证优化的核极限学习机(LOO-KELM)算法的复合电能质量扰动识别方法。首先根据选定的主频率点自适应调节S变换窗宽系数,提取具有高时频分辨率的59种电能质量(PQ)特征,再通过留一交叉验证寻找最小预测残差平方和,实现核极限学习机的输出权重优化,最后根据提取PQ特征集与优化后的核极限学习机实现复合扰动的识别与分类。仿真和实测结果表明,所提方法对不同噪声下的16类混合电能质量扰动均具有较高的分类精度,相较于现有复合电能质量识别方法,分类精度更高且训练时间更短。

Recognition of composite power quality disturbance based on improved incomplete S transform and LOO-KELM algorithm

In order to solve the problems of low efficiency of feature extraction, and inability of classifier recognition and learning speed in composite power quality disturbance classification, a composite power quality disturbance recognition method based on incomplete S transformation of adaptive window and LOO-KELM(Kernel Extreme Learning Machine optimized by Leave-One-Out cross validation) algorithm is proposed. Firstly, the window width coefficient of S transform is adaptively adjusted according to the selected main frequency, 59 kinds of PQ(Power Quality) characteristics with high time-frequency resolution are extracted. Then through LOO, the minimum prediction residual sum of squares is obtained for the kernel extreme learning machine output weight optimization. According to the extraction of PQ feature set and the optimized kernel extreme learning machine, the identification and classification of compound disturbance are realized. Results of simulation and measurement show that the proposed method has higher classification accuracy for 16 types of mixed power quality disturbances under different noises. Compared with the existing composite power quality identification methods, the proposed method has higher classification accuracy and shorter training time.