采用改进多分辨率快速S变换的电能质量扰动识别

Power Quality Disturbances Classification With Improved Multiresolution Fast S-transform

噪声干扰是影响电能质量暂态扰动识别准确率的最重要因素。经过S变换后获得的扰动信号的模时–频矩阵具有灰度图像特点。因此,可通过二维数学形态学方法,滤除噪声干扰,获得更高的识别准确率。首先,针对扰动信号时–频分布特点,设计具有不同时–频分辨率的多分辨率快速S变换方法以降低运算量、提高特征表现能力;之后,在阈值滤波基础上,根据信号时–频分布特点,选择线段型、零角度结构元进行灰度级形态学开运算,进一步滤除高频频域噪声;最后,从原始信号、信号傅里叶谱、多分辨率快速S变换模矩阵中提取5种特征建立决策树分类器,识别含噪声信号与6种复合扰动信号在内的12种电能质量信号。通过仿真对比实验发现,新方法具有更好的抗噪能力,更加适用于低信噪比环境下的电能质量信号识别。

The noise is the most important factor to affect the recognition accuracy of power quality disturbances. The time-frequency modular matrix obtained from S-transform has the characteristics of gray image. Therefore, the classification accuracy of disturbances can be improved by two-dimensional mathematical morphology de-noising method. Firstly, an improved multi-resolution fast S-transform with different time-frequency resolutions was constructed according to the time-frequency distribution characteristics of modular matrix. It was used to reduce the computation complexity and improve the ability of time-frequency feature presentation. Secondly, morphological open operator with a line type, zero angle structure element was used in the high frequency area of the modular matrix to immune noise affection after threshold filtering. Finally, a decision tree classifier was designed based on five features which were extracted from the original signals, Fourier spectrums of original signals and time-frequency modular matrix of multi-resolution fast S-transform. The new method can recognize the noise signal without disturbances and 12 types of disturbances including 6 types of complex disturbances. The comparison of simulation experiments shows that the new method has better noise immunity and more suitable for disturbances recognition in the noise environments.