采用最优多分辨率快速S变换的电能质量分析

为兼顾电能质量信号分析的类型识别与参数估计需要,设计一种最优化多分辨率快速S变换(OMFST),用于电能质量信号识别与参数估计。首先,分析不同时-频分辨率下时间-幅值曲线与频率-幅值曲线中,扰动起、止处峭度与扰动参数估计误差间的关系;之后,根据离差最大化法,确定不同频率范围内最优窗宽调整因子,并通过3次样条插值法进行拟合,自适应调整不同扰动信号识别和参数估计所需最优窗宽;然后,针对扰动信号基频与扰动所在的中、高频频域范围进行OMFST处理;最后,从原始信号、原始信号傅里叶谱和OMFST变换结果中提取5条特征,构建基于模糊决策树的扰动分类器,识别13种电能质量信号,并估计电能质量信号参数。仿真实验和实测数据分析表明,新方法能够满足电能质量复合扰动参数估计需要,参数估计误差低于广义S变换等方法,同时保留了良好的分类能力。

Power quality analysis adopting optimal multi resolution fast S transform

In order to meet the requirement of type recognition and parameter estimation in power quality disturbance signal analysis, this paper proposes an Optimal Multi resolution Fast S transform (OMFST) method. Firstly, the relationship between the disturbance parameter estimation errors and the kurtosis at the start and end locations of the disturbances in the time amplitude curve and frequency amplitude curve under different time frequency resolutions is analyzed. The optimal window width adjustment factors in different frequency ranges are determined according to the deviation maximization method, and the cubic spline interpolation method is used for fitting; the optimal window width required for the type recognition and parameter estimation of different disturbance signals is adjusted automatically. Secondly, OMFST processing is conducted according to the fundamental frequency of the disturbance signal and the middle and high frequency ranges in which the disturbance signal locate. Finally, the disturbance classifier based on fuzzy decision tree is constructed based on the 5 features extracted from the original signal, the Fourier transform spectrum of the original signal and the time frequency modular matrix of OMFST. The new method can recognize 13 types of power quality signals and estimate the disturbance parameters. Simulation and actual test data analysis show that the new method can meet the requirement of the parameter estimation of compound power quality disturbance signal. The parameter estimation error is lower than those of the generalized S transform and etc. Meanwhile, the proposed new method remains good classification accuracy.