基于t-SNE的PQD特征提取可视化仿真分析

Visual Simulation of PQD Feature Extraction Method Based on t-SNE

针对电能质量扰动(power quality disturbance,PQD)的复杂性,提出了一种基于t分布随机近邻嵌入(t-distributed stochastic neighbor embedding,t-SNE)流行学习算法的PQD特征提取方法。首先,建立了常见的PQD信号数学模型,并考虑了扰动参数和噪声的影响;其次,采用小波分解得到信号的小波能量向量构造原始特征集;最后,通过t-SNE算法进行二次特征提取,得到保持样本高维空间结构的,敏感度高且聚类性好的低维特征。仿真实验结果证明了提出的基于t-SNE的特征提取方法在PQD分析中的有效性。

In view of the complexity of power quality disturbance(PQD),this paper proposed a PQD feature extraction method based on t-distributed stochastic neighbor embedding(t-SNE).Firstly,considering the influence of random parameters and noise,we constructed typical disturbances models.Then,the wavelet energy features of PQD were extracted by wavelet transform to construct the original feature set.Finally,we applied t-SNE method to compress the sample features into 3-dimension features,which keep low dimensional structure in high dimensional space and make a rough recognition of PQD.The simulation proved the effectiveness of the PQD feature extraction method based on t-SNE.