基于经验模态分解和形态学的风电并网电压故障检测

针对风电场并网点电压故障引起的风机大规模脱网问题,提出了基于柔性形态算子和经验模态分解(empirical mode decomposition,EMD)去噪技术的电网电压故障检测方法。首先,利用EMD对采样信号进行时频自适应预处理,从而确定噪声主导模态;然后,通过柔性形态学变换加阈值输出,有效放大信号奇异点,避免了因电网电压信号周期性变化和噪声引起的背景梯度对检测结果的影响,实现故障定位检测。通过对不同噪声强度的电压暂降故障信号进行检测对比分析发现,随着信号信噪比下降,标准形态学方法的检测误差进一步增大,当信噪比达到25db时,甚至出现了误检现象,而柔性形态EMD检测方法仍然可以有效检测故障扰动的起止时间,表明该方法与标准形态学和小波阈值方法相比,在简化运算过程的同时可以获得更高的检测精度。最后,对某风电场并网点故障电压的分析结果与实测数据的一致性,验证了该方法可以有效检测电网电压的瞬态故障信息,从而为风电场无功补偿装置的投切控制提供了依据。

EMD and morphology based voltage disturbance detection method for power system connected with wind turbine generation

As the penetration of wind power has become significant, one of the important challenges of power distribution network with wind power integration is the risk of large-scale wind turbine tripping accidents caused by over/under voltage faults in farms and systems, which also leads to unexpected variations in frequency parameters and thereby power quality issues. Aiming at the difficulty to extract early weak fault feature for the voltage influenced by white noise and transient disturbance noise, a method combining empirical mode decomposition (EMD) with soft mathematical morphology (MM) was put forward in this paper. It was crucial for the requirements of fault ride-through devices, fault component extraction and reclose scheme on voltage detection accuracy and real-time performance. Firstly, the dominant mode of noise was identified by EMD preprocessing. For the fault transients producing non-stationary signals with large frequency spectrum, the present mainstream techniques such as windowed Fourier transform (WFT) and discrete wavelet transform (DWT) were unsatisfactory. The heavy calculation burden of DWT made this methodology prohibitive in real-time detection. Moreover, DWT had oscillations around singularities. EMD had a high extent of adaptation to process various non-stationary signals without imposing any serious restriction on the harmonic nature of basis functions. After decomposing the voltage signals into a series of intrinsic mode functions (IMFs), the spectrum available was used for discovering the hidden amplitude and frequency modulations in voltage signals and finding out the domains of energy concentration. The fault characteristic signal was restructured by accumulating the selected IMF components which characterized the fault characteristic frequencies. Then the voltage signal singularities were amplified by MM transform and threshold detection, avoiding the negative effects of background gradient on the results caused by cyclic variation of grid voltage and noise effects. The soft opening operation of dominant noise mode was performed to filtrate the spike noise, and the dilation of flat structuring element, whose length was one half of the period of power frequency, was operated to extract the magnitude characteristic of the signal; meanwhile, the gradient operation of short flat structuring element to differential signal was performed to detect and locate the singular point, and thus the defect of possibly omitting singular point by traditional methods was remedied. It was expected to offer better sensitivity and selectivity for voltage faults. A soft morphological edge detection based scheme was proposed to locate transient disturbance of voltage. To solve the problem of voltage detection inaccuracy caused by background gradient due to periodic variation of power signals and existing interferences during sampling process, a quantitative assessment method based on soft threshold was induced to improve detection accuracy. A standard to assess the filtering effect was put forward to choose the size of structuring elements adaptively to perform morphological filtering of original signals. And also the dilation-erosion transform was applied to morphological gradient by flat structuring elements to suppress background gradient to achieve location result preliminarily. Finally combining with the processing of soft threshold, the location of transient disturbance of power quality was implemented. Based on the noise ratio, correlation coefficient and mean-variance analysis, the MM-EMD could get better accuracy accompanied with simplified calculation process, compared with the standard morphology and wavelet threshold method. It was concluded according to the simulation analysis under different voltage fault scenarios, the detection error from standard morphology method increased as the signal to noise ratio (SNR) was degraded; particularly when SNR arrived to 25 db, standard morphology method was failed to locate voltage faults, while MM-EMD was still operative. The experimental results from the on-site survey in the northwest wind farm verified that the MM-EMD was effective in noise suppression and transient voltage detection, which was essential to the development of wind farm reactive power compensation devices.