韓曉慧,杜松懷,李 振,孫麗華
?
基于泄漏電流時頻奇異譜和模糊聚類的觸電故障診斷
韓曉慧1,杜松懷2,李 振2,孫麗華1
(1. 河北科技大學(xué)電氣工程學(xué)院,石家莊 050018;2. 中國農(nóng)業(yè)大學(xué)信息與電氣工程學(xué)院,北京 100083)
針對實測觸電故障信號具有非平穩(wěn)特性而不易被辨識問題,提出了一種基于泄漏電流時頻奇異譜和模糊聚類的觸電故障診斷方法。首先,利用平滑偽威格納-維爾分布(smoothed pseudo wigner-ville distribution, SPWVD)對觸電故障信號進(jìn)行時頻分析并依據(jù)信號的能量分布特征選擇時頻區(qū)域;然后對選擇的時頻區(qū)域進(jìn)行奇異譜分析,以獲取的局部時頻矩陣奇異值作為觸電信號的特征量輸入FCM,即可實現(xiàn)觸電信號的故障診斷。對剩余電流保護(hù)裝置試驗平臺上獲取的實測觸電故障信號的時頻矩陣奇異值進(jìn)行模糊C均值聚類,結(jié)果表明該方法識別準(zhǔn)確率為97.50%,平均識別時間為0.008 5 s,其中植物和動物觸電測試樣本識別準(zhǔn)確率分別為100%,95.00%,從而驗證了基于泄漏電流時頻奇異譜和模糊聚類的觸電故障診斷方法的有效性,該研究可為研發(fā)新一代基于觸電故障診斷的剩余電流保護(hù)裝置提供理論依據(jù)和方法參考。
電流檢測;電力系統(tǒng);診斷;觸電故障;時頻矩陣;奇異值分解;特征量提??;模糊C均值聚類
作為電力系統(tǒng)中重要的保護(hù)與控制設(shè)備之一,剩余電流保護(hù)裝置能夠及時對生物體觸電、電氣火災(zāi)及電氣設(shè)備損壞等情況做出防護(hù)動作[1]。但是,目前現(xiàn)有的剩余電流在線監(jiān)測保護(hù)裝置,其動作可靠性和正確投運(yùn)率都不太理想,因此學(xué)者在剩余電流保護(hù)裝置方面進(jìn)行了大量研究[2-7],筆者課題組成員也一直致力尋求剩余電流保護(hù)新原理和新算法以設(shè)計一種高精度、高可靠性的剩余電流保護(hù)裝置,如文獻(xiàn)[8-10]提出用神經(jīng)網(wǎng)絡(luò)建立了觸電信號檢測模型,能有效從總泄漏電流中檢測出觸電電流。文獻(xiàn)[11]針對神經(jīng)網(wǎng)絡(luò)容易陷入局部最優(yōu)、隱層單元數(shù)難以確定等缺點,用最小二乘支持向量機(jī)建立了觸電信號的檢測模型。文獻(xiàn)[12]針對如何準(zhǔn)確、快速提取生物體觸電故障暫態(tài)信號中電力參數(shù)問題,將局部均值分解用于總泄漏電流信號的瞬時幅值和瞬時頻率提取。這些方法雖然在一定程度上提高了剩余電流保護(hù)裝置的技術(shù)性能,但尚不具有對監(jiān)測到的電流信號自動識別和診斷能力。當(dāng)觸電故障出現(xiàn)時,僅依據(jù)人工經(jīng)驗判斷觸電故障類型,且所需時間長也容易出現(xiàn)誤判。為此,文獻(xiàn)[13]將量子遺傳計算和神經(jīng)計算有機(jī)結(jié)合,建立了一種量子遺傳模糊神經(jīng)網(wǎng)絡(luò)作為觸電故障模式分類歸屬的決策系統(tǒng)。但神經(jīng)網(wǎng)絡(luò)仍存在網(wǎng)絡(luò)訓(xùn)練速度慢等無法克服的缺點,針對這些缺點,本文通過對實時監(jiān)測總泄漏電流,提出一種新的特征提取及診斷算法,以實現(xiàn)觸電故障類型的準(zhǔn)確判斷。
觸電故障類型診斷實質(zhì)上是一個模式識別問題,包括三個環(huán)節(jié):信號采集、特征提取及狀態(tài)識別,其關(guān)鍵在于如何有效提取各觸電故障特征。若僅采用單一頻域或時域分析方法提取現(xiàn)場獲取的非平穩(wěn)總泄漏電流信號故障特征,并不能全面反映觸電信號所包含的頻率及其幅值的時變特征信息。而用時域和頻域的聯(lián)合時頻分析(time-frequency analysis,TFR)方法來揭示非平穩(wěn)信號中所包含信息得到了越來越多的應(yīng)用[14-15]。常用的聯(lián)合時頻分析方法按時頻聯(lián)合函數(shù)不同主要分為線性和非線性2種時頻分析方法[16]。以線性時頻分析方法為例的短時傅里葉變換方法[17]和小波變換方法[18-19]由于受Heisenberg不確定性原理的限制,難以獲得理想的時頻分辨率。以非線性時頻分析方法為例的威格納-維爾分布(Wigner-Ville distribution,WVD)由于具有較高的時頻聚集性被廣泛應(yīng)用于信號分析處理領(lǐng)域,但存在固有的交叉項干擾問題,影響了對WVD分析結(jié)果的正確評估[20]。平滑偽威格納-維爾分布(smoothed pseudo wigner-ville distribution,SPWVD)[21]是一種以WVD為基礎(chǔ)的非線性時頻分布,通過在時間軸和頻率軸方向加窗函數(shù)自適應(yīng)地調(diào)節(jié)窗口長度,抑制了WVD存在的交叉項干擾問題,能更準(zhǔn)確反映信號時頻特征,同時還具有時移和頻移不變性,因此是分析信號的有效工具。
鑒于此,本文將探索基于SPWVD的觸電故障信號時頻特征提取與診斷策略。首先,采用SPWVD對發(fā)生觸電故障時的總泄漏電流時頻特性進(jìn)行表征,并依據(jù)信號的能量分布特點選擇時頻區(qū)域;然后,采用奇異譜分析方法對選擇的時頻區(qū)域進(jìn)行奇異值分解,以獲取的該時頻矩陣奇異值構(gòu)成了觸電故障信號的特征量;最后,通過模糊C均值聚類算法實現(xiàn)了不同觸電故障類型的診斷,并利用該方法對剩余電流保護(hù)裝置觸電物理試驗系統(tǒng)平臺上獲取的觸電故障信號進(jìn)行了有效性驗證。
平滑偽威格納-維爾分布(smoothed pseudo wigner-ville distribution,SPWVD)是反映信號能量的時頻分布,其實質(zhì)是對信號的瞬時相關(guān)函數(shù)作傅里葉變換時在時域軸和頻率軸分別加窗函數(shù)進(jìn)行平滑處理[22]。
信號()經(jīng)SPWVD分析后的時頻矩陣中,其行向量為某一頻率隨時間變化的分布,列向量為某一時刻隨頻率變化的分布,某位置上元素的大小就是相應(yīng)時間和頻率處信號SPWVD分析的能量。利用SPWVD時頻矩陣可以用時頻等值線圖來表示信號的時頻分布。圖1所示為一仿真信號()及其經(jīng)STFT(short-time Fourier tranformation)、WVD、SPWVD分析后的時頻等值線圖。仿真信號包括2個頻率分量:在采樣序列=0~0.8 s之間為一200 Hz的余弦信號,在=0.2~1 s之間又疊加了一個50 Hz的余弦信號。
圖1 信號源s(t)及其經(jīng)STFT、WVD、SPWVD分析的時頻等值線圖
由圖1可以看出,圖1b中由于STFT采用固定的窗函數(shù)使其在整體上呈現(xiàn)較低的時頻分辨率;圖1c中WVD存在嚴(yán)重的交叉項干擾項,難以確定信號的頻率成分;圖1d中SPWVD通過時頻域窗函數(shù)的平滑作用,抑制了WVD的交叉項干擾,較好地反映了該信號頻率成分隨時間變化的分布情況。
綜上所述,本文選取的SPWVD時頻分析方法,具有較好的時頻聚焦性,可嘗試將其應(yīng)用到觸電故障信號處理中。
奇異譜分析是一種通過對信號進(jìn)行奇異值分解以獲取其內(nèi)在復(fù)雜特征的信號分析方法。依據(jù)奇異值分解理論[23],對于由SPWVD獲得時頻分布矩陣×n,求正交矩陣×m、×n和對角矩陣×n使其滿足
由于×n是一對角矩陣,奇異值分解也可表示為將矩陣×n分解為個秩為1的×階矩陣的加權(quán)和,各子矩陣由特征向量=(1,2,···,w)與=(1,2,···,h)及相應(yīng)的權(quán)值相乘得到
由于矩陣奇異值還具有相對穩(wěn)定性、比例不變性、位移不變性及旋轉(zhuǎn)不變性[24-25],當(dāng)矩陣中有一定的干擾和分散性存在時,矩陣的奇異值是相對穩(wěn)定的代數(shù)特征參量,故矩陣奇異值在模式識別中常用于信號特征量的提取。
模糊C均值(fuzzy C-means,F(xiàn)CM)聚類作為一種非監(jiān)督動態(tài)聚類,利用隸屬度表征數(shù)據(jù)的相對歸屬性,對數(shù)據(jù)實現(xiàn)柔性模糊劃分。與硬分類K-means聚類相比,F(xiàn)CM對初始聚類中心要求較低,當(dāng)數(shù)據(jù)維數(shù)較大時FCM可以得到更合理的分類結(jié)果[27-28]。
FCM聚類實質(zhì)是通過若干次迭代求取各樣本到聚類中心的距離平方和最小值得到給定分類數(shù)下的聚類結(jié)果。對于給定數(shù)據(jù)集={1,2,···,x},每個樣本為維向量,即=(x1,x2,···,x)T,其中=1,2,…,,F(xiàn)CM算法的數(shù)學(xué)模型為[29]
式中fcm為FCM的目標(biāo)函數(shù),使得樣本與聚類中心之間的差異度最??;為隸屬度矩陣;為聚類中心;為聚類數(shù);為控制分類矩陣的模糊權(quán)重指數(shù)(>1,一般取值范圍為1.5~2.5);u表示第個樣本隸屬于第類的程度;d=||?||表示第個樣本與第類中心的歐氏距離。FCM的詳細(xì)算法流程詳見文獻(xiàn)[30],本文不予重述。
本文所使用的觸電故障信號來自于剩余電流保護(hù)裝置試驗平臺上獲取的總泄漏電流信號,試驗原理詳見文獻(xiàn)[11]。在10 kHz的采樣頻率下,采集了400組動植物觸電總泄漏電流數(shù)據(jù),其中包含了200組植物觸電和200組動物觸電數(shù)據(jù)。對采集的信號進(jìn)一步分析發(fā)現(xiàn),可以用觸電前一周期和觸電后一周期的數(shù)據(jù)共0.04 s時長的信號作為待分析觸電故障信號,圖2、圖3所示為上述2種觸電類型場景下的具有代表性的總泄漏電流時域波形及SPWVD時頻等值線圖和三維圖譜。
圖2 植物觸電總泄漏電流及SPWVD時頻等值線圖和三維圖譜
圖3 動物觸電總泄漏電流及SPWVD時頻等值線圖和三維圖譜
由圖2和圖3的SPWVD分析結(jié)果可以看出,動植物觸電總泄漏電流能量主要集中在頻率0~150 Hz之間。因此,考慮利用總泄漏電流在頻率0~150 Hz之間的時頻區(qū)域奇異值作為動物觸電與植物觸電模式識別特征量。
對上述2種觸電情況下總泄漏信號各取140組作為已知觸電故障類型樣本數(shù)據(jù),再各取60組數(shù)據(jù)作為待驗觸電故障類型樣本數(shù)據(jù)。其中第1~140個采樣樣本為植物觸電故障樣本;第141~280為動物觸電故障樣本;第281~340為待驗植物觸電故障樣本,第341~400為待驗動物觸電故障樣本。
依據(jù)4.2節(jié)觸電故障信號特征提取步驟,求取這400組觸電樣本數(shù)據(jù)的時頻矩陣奇異值,構(gòu)造出總泄漏電流特征向量矩陣,將特征向量矩陣作為FCM聚類的輸入,求得植物觸電模式和動物觸電模式的聚類中心分別為
式中的行號與列號分別與觸電故障類型與樣本編號對應(yīng),第1、2行分別對應(yīng)植物觸電故障和動物觸電故障,第1~5列分別對應(yīng)樣本編號1~5。由隸屬度矩陣可得到植物觸電模式和動物觸電模式的隸屬度劃分矩陣分別如圖4a、4b所示。
注: 隸屬度值越大,代表隸屬于對應(yīng)的觸電類型程度越高。
Note: The greater the membership value, the higher the degree of representation of the samples attached to belonging to the corresponding type of electric shock.
圖4 隸屬度劃分矩陣
Fig.4 Division matrix of membership degree
圖4a、4b中每個元素分別代表第(=1,2,3,…,400)個采樣樣本隸屬于植物觸電和動物觸電的程度,第個采樣樣本的最大值所在的類即為該樣本對應(yīng)的觸電類型狀態(tài)。因此利用隸屬度矩陣及其劃分矩陣可識別觸電類型狀態(tài)。
由隸屬度矩陣可知,第281~340列(待驗樣本)的隸屬度最大值分別為0.999 7,0.999 9,0.999 9,…,0.999 7,0.999 8出現(xiàn)在第1行,判定待驗樣本與樣本1~140屬于同一觸電類型樣本,即待驗樣本為植物觸電故障類型樣本,與實際類型一致;另外,從圖4a中也可明顯看出第281~340采樣樣本的隸屬度最大值均出現(xiàn)在植物觸電劃分矩陣中,由此也可判定待驗樣本為植物觸電故障類型樣本。第341~373、375~379、381~386、388~389列(待驗樣本)的隸屬度最大值分別為0.951 5,0.912 1,0.954 4,…,0.923 6,0.864 1出現(xiàn)在第2行,判定待驗樣本與樣本141~280屬于同一觸電類型樣本,即待驗樣本為動物觸電故障類型樣本,與實際類型一致,但第374、380、387列的最大值分別為0.551 8、0.605 6、0.594 6出現(xiàn)在第1行,判定待驗樣本為植物觸電故障類型樣本,與實際類型不一致;同樣,從圖4a、4b中也可看出第281~340采樣樣本中除有3個采樣樣本的隸屬度值大于0.5出現(xiàn)在植物觸電劃分矩陣中,其余采樣樣本的隸屬度最大值均出現(xiàn)在動物觸電劃分矩陣中,由此判定有3個待驗樣本為植物觸電故障類型樣本與實際故障類型不一致,其余待驗樣本為動物觸電故障類型樣本。
由以上識別結(jié)果可以看出,120組測試樣本中有3組樣本識別錯誤,識別準(zhǔn)確率為97.50%,其中植物觸電測試樣本識別準(zhǔn)確率為100%,動物觸電測試樣本中有3組樣本識別錯誤,識別準(zhǔn)確率為95.00%,取得了較好的檢測效果。平均識別時間為0.008 5 s,少于文獻(xiàn)[13]中量子遺傳模糊神經(jīng)網(wǎng)絡(luò)觸電故障診斷算法所需迭代1 156次的訓(xùn)練時間,克服了神經(jīng)網(wǎng)絡(luò)訓(xùn)練速度慢的問題,提高了識別效率,從而驗證了用所提取的總泄漏電流信號的特征量診斷觸電故障信號類型狀態(tài)的正確性和有效性。
本文針對觸電故障信號的診斷問題,提出了一種基于平滑偽威格納-維爾分布(smoothed pseudo wigner-ville distribution,SPWVD)時頻奇異譜特征提取和模糊C均值(fuzzy C-means,F(xiàn)CM)聚類的觸電故障信號的診斷方法。
1)采用SPWVD對觸電故障時刻的總泄漏電流進(jìn)行時頻分析,時頻等值線圖和三維圖譜表明,不同觸電故障類型信號具有相互區(qū)別的時頻分布特征,說明了利用SPWVD分析觸電故障信號的可行性;
2)引入奇異譜分析方法對觸電故障信號0~150 Hz頻率范圍內(nèi)13個子頻帶形成的局部時頻矩陣進(jìn)行奇異值分解,得到各子頻帶奇異值構(gòu)成的13維向量作為觸電故障信號的特征量;
3)通過對120組觸電故障信號特征量進(jìn)行FCM聚類測試,結(jié)果表明該方法識別準(zhǔn)確率為97.50%,平均識別時間為0.008 5 s,其中植物和動物觸電測試樣本識別準(zhǔn)確率分別為100%,95.00%,由此驗證了SPWVD時頻矩陣奇異譜特征提取和觸電信號故障診斷的有效性,為觸電故障信號的診斷提供了有效手段。
[1] 杜松懷,張筱慧.電力系統(tǒng)接地技術(shù)[M]. 北京:中國電力出版社,2011.
[2] 曹國臣,蔡國偉,王海軍. 繼電保護(hù)整定計算方法存在的問題與解決對策[J]. 中國電機(jī)工程學(xué)報,2003,23(10):51-57.
Cao Guochen, Cai Guowei, Wang Haijun. Problems and solutions in relay setting and coordination[J]. Proceedings of the CSEE, 2003, 23(10): 51?57. (in Chinese with English abstract)
[3] Luo Xiang, Du Y, Wang X H, et al. Tripping characteristics of residual current devices under nonsinusoidal currents[J].IEEETransactions on Industry Applications2011, 47(3):1515-1521.
[4] Czapp S. Elimination of the negative effect of earth fault current higher frequency on tripping of residual current devices[J]. Elektronika Ir Elektrotechnika, 2009(3): 85-88.
[5] Czapp S. The impact of DC earth fault current shape on tripping of residual current devices[J]. Elektronika Ir Elektrotechnika, 2015(4): 9-12.
[6] Liew A C. Nuisance trippings of residual-current circuit breakers or ground fault protectors of power sources connected to computer and electronic loads[J]. Electric Power Systems Research, 1990, 20(1): 23-30.
[7] 陳義剛,李浩,范康林,等. 考慮相角特征的無死區(qū)新型觸/漏電保護(hù)技術(shù)[J]. 四川電力技術(shù),2013(2):35-39,72.
Chen Yigang, Li Hao, Fan Kanglin, et al. The new touch/leakage current protection technology eliminating dead-zone that considering the phase angle characteristics[J]. Sichuan Electric Power Technology, 2013(2): 35-39, 72. (in Chinese with English abstract)
[8] 李春蘭,蘇娟,杜松懷,等. 基于小波分析和 BP 神經(jīng)網(wǎng)絡(luò)的觸電信號檢測模型[J]. 農(nóng)業(yè)工程學(xué)報,2010,26(2):130-134.
Li Chunlan, Su Juan, Du Songhuai, et al. Detecting model of electric shock signal based on wavelet analysis and BP neural network[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2010, 26(2): 130-134. (in Chinese with English abstract)
[9] 關(guān)海鷗,杜松懷,李春蘭,等. 基于有限脈沖反應(yīng)和徑向基神經(jīng)網(wǎng)絡(luò)的觸電信號識別[J]. 農(nóng)業(yè)工程學(xué)報,2013,29(8):187-194.
Guan Haiou, Du Songhuai, Li Chunlan, et al. Recognition of electric shock signal based on FIR filtering and RBF neural network[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2013, 29(8): 187-194. (in Chinese with English abstract)
[10] 關(guān)海鷗,杜松懷,蘇娟,等. 一種觸電信號的自動快速檢測模型[J]. 電網(wǎng)技術(shù),2013,37(8):2328-2335.
Guan Haiou, Du Songhuai, Su Juan, et al. An automatic and quick detection model of electric shock signals[J]. Power System Technology, 2013, 37(8): 2328-2335. (in Chinese with English abstract)
[11] 韓曉慧,杜松懷,蘇娟,等. 基于參數(shù)優(yōu)化的最小二乘支持向量機(jī)觸電電流檢測方法[J]. 農(nóng)業(yè)工程學(xué)報,2014,30(23):238-245.
Han Xiaohui, Du Songhuai, Su Juan, et al. Determination method of electric shock current based on parameter-optimized least squares support vector machine[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2014, 30(23): 238-245. (in Chinese with English abstract)
[12] 韓曉慧,杜松懷,蘇娟,等. 基于局部均值分解的觸電故障信號瞬時參數(shù)提取[J]. 農(nóng)業(yè)工程學(xué)報,2015,31(17):221-227.
Han Xiaohui, Du Songhuai, Su Juan, et al. Extraction of biological electric shock signal instantaneous amplitude and frequency based on local mean decomposition[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2015, 31(17): 221-227. (in Chinese with English abstract)
[13] 王金麗,劉永梅,杜松壞,等. 基于剩余電流固有模態(tài)能量特征的生物觸電故障診斷模型[J]. 農(nóng)業(yè)工程學(xué)報,2016,32(21):202-208.
Wang Jinli, Liu Yongmei, Du Songhuai, et al. Fault diagnosis model for biological electric shock based on residual current intrinsic mode function energy features[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2016, 32(21): 202-208. (in Chinese with English abstract)
[14] Jardine A K S, Lin D, Banjevic D. A review on machinery diagnostics and prognostics implementing condition-based maintenance[J]. Mechanical Systems and Signal Processing, 2006, 20(7): 1483-1510.
[15] Heng A, Zhang S, Tan A C, et al. Rotating machinery prognostics: State of the art, challenges and opportunities[J]. Mechanical Systems and Signal Processing, 2009, 23(3): 724-739.
[16] 章立軍,劉博,張彬,等. 基于時頻圖像融合的軸承性能退化特征提取方法[J]. 機(jī)械工程學(xué)報,2013,49(22):53-58.
Zhang Lijun, Liu Bo, Zhang Bin, et al.Feature extraction method of bearing performance degradation based on time-frequency image fusion[J]. Journal of Mechanical Engineering, 2013, 49(22): 53-58. (in Chinese with English abstract)
[17] 徐永海,趙燕. 基于短時傅里葉變換的電能質(zhì)量擾動識別與采用奇異值分解的擾動時間定位[J]. 電網(wǎng)技術(shù),2011,35(8):174-179.
Xu Yonghai, Zhao Yan. Identification of power quality disturbance based on short-term Fourier transform and disturbance time orientation by singular value decomposition[J]. Power System Technology, 2011, 35(8): 174-179. (in Chinese with English abstract)
[18] Pillay P, Bhattacharjee A. Application of wavelets to model short-term power system disturbances[J]. IEEE Trans on Power System, 1996, 11(4): 2031-2037.
[19] 馮浩,周雒維,劉毅. 基于復(fù)小波變換的暫態(tài)電能質(zhì)量擾動檢測與分類[J]. 電網(wǎng)技術(shù),2010,34(3):91-95.
Feng Hao, Zhou Luowei, Liu Yi. Detection and classification of transient power quality disturbances based on complex wavelet transform[J]. Power System Technology, 2010, 34(3): 91-95. (in Chinese with English abstract)
[20] 張賢達(dá). 現(xiàn)代信號處理[M]. 北京:清華大學(xué)出版社,2002.
[21] 林勇,周曉軍,楊先勇,等. 基于 SPWVD 識別的滾動軸承智能檢測方法[J]. 振動與沖擊,2009,28(9):86-90.
Lin Yong, Zhou Xiaojun, Yang Xianyong, et al. Intelligent fault diagnosis methods of rolling bearing based on SPWVD and AIN[J]. Journal of Vibration and Shock, 2009, 28(9): 86-90. (in Chinese with English abstract)
[22] 臧懷剛,王石云,李玉奎. EMD和平滑偽 Wigner-Ville 譜熵的軸承故障診斷[J]. 噪聲與振動控制,2014,34(5):145-149.
Zang Huaigang, Wang Shiyun, Li Yukui. Bearing fault diagnosis based on EMD and smoothed pseudo Wigner-Ville spectrum entropy[J]. Noise and Vibration Control, 2014, 34(5): 145-149. (in Chinese with English abstract)
[23] 張賢達(dá). 矩陣分析與應(yīng)用[M]. 北京:清華大學(xué)出版社,2004.
[24] 杜林,戴斌,陸國俊,等. 基于S變換局部奇異值分解的過電壓特征提取[J]. 電工技術(shù)學(xué)報,2010,25(12):147-153.
Du Lin, Dai Bin, Lu Guojun, et al. Overvoltage features extraction based on S transform and local singular value decomposition[J]. Transactions of China Electrotechnical Society, 2010, 25(12): 147-153. (in Chinese with English abstract)
[25] 郭謀發(fā),徐麗蘭,繆希仁,等. 采用時頻矩陣奇異值分解的配電開關(guān)振動信號特征量提取方法[J]. 中國電機(jī)工程學(xué)報,2014,34(28):4990-4997.
Guo Moufa, Xu Lilan, Miao Xiren, et al. A vibration signal feature extraction method for distribution switches based on singular value decomposition of time-frequency matrix[J]. Proceedings of the CSEE, 2014, 34(28): 4990-4997. (in Chinese with English abstract)
[26] Tian Y, Tang T, Wang Y H, et al. Do singular value contain adequate information for face recognition[J]. Pattern Recognition, 2003, 36(6): 649-655.
[27] 王淳,高元海. 采用最優(yōu)模糊C均值聚類和改進(jìn)化學(xué)反應(yīng)算法的配電網(wǎng)絡(luò)動態(tài)重構(gòu)[J]. 中國電機(jī)工程學(xué)報,2014,34(10):1682-1691.
Wang Chun, Gao Yuanhai. Dynamic reconfiguration of distribution network based on optimal fuzzy C-means clustering and Improved chemical reaction optimization[J]. Proceedings of the CSEE, 2014, 34(10): 1682-1691. (in Chinese with English abstract)
[28] 宋易陽,李存斌,祁之強(qiáng). 基于云模型和模糊聚類的電力負(fù)荷模式提取方法[J]. 電網(wǎng)技術(shù),2014,38(12):3378-3383.
Song Yiyang, Li Cunbin, Qi Zhiqiang. Extraction of power load patterns based on cloud model and fuzzy clustering[J]. Power System Technology, 2014, 38(12): 3378-3383. (in Chinese with English abstract)
[29] 汪可,廖瑞金,王季宇,等. 局部放電 UHF 脈沖的時頻特征提取與聚類分析[J]. 電工技術(shù)學(xué)報,2015,30(2):211-219.
Wang Ke, Liao Ruijin, Wang Jiyu, et al. Time-frequency features extraction and clustering analysis of partial discharge UHF pulses[J]. Transactions of China Electrotechnical Society, 2015, 30(2): 211-219. (in Chinese with English abstract)
[30] 高新波. 模糊聚類分析及其應(yīng)用[M]. 西安:西安電子科技大學(xué)出版社,2004.
Diagnosis of electric shock fault based on time-frequency singular value spectrum of leakage current and fuzzy clustering
Han Xiaohui1, Du Songhuai2, Li Zhen2, Sun Lihua1
(1.,,050018,; 2.,,100083,)
Residual current devices (RCDs), a type of protective equipment in low-voltage systems, are widely used to prevent and avoid leakage accident of power grid and protect the safety of life and property. At present, the operation of an RCD is based on sensing the root mean square value of residual current in an electrical circuit. The circuit will be interrupted on the action of a relay when the residual current exceeds a predetermined level (30 mA for human safety), known as the tripping current. Although such devices offer a large degree of protection, they are prone to nuisance tripping or maloperation in the actual operation of the grid due to the lack of the ability to diagnose electric shock type and identify touch current, which reduces the reliability and the rate of proper commissioning for RCDs. Thus, aiming at the problem that the measured electric shock signals are non-stationary and difficult to diagnose the type of electric shock, a new method of fault diagnosis of electric shock signal based on time-frequency singular spectrum of leakage current and fuzzy clustering is proposed. First of all, a simulation signal is used to compare and analyze the time-frequency analysis performance of short-time Fourier transformation (STFT), wigner-ville distribution (WVD) and smoothed pseudo Wigner-Ville distribution (SPWVD). The simulation results show that the STFT presents a lower time-frequency resolution because of the fixed window function, the WVD has serious crosstalk terms and it is difficult to determine the frequency components of the signal, and the SPWVD suppresses the crosstalk of WVD and reflects the distribution of signal frequency components with time through the smoothing of time-frequency window function. Therefore, SPWVD is chosen as the time-frequency analysis method in this paper. Then, numerous groups of total leakage current signals were measured using a recorder on the electric shock experiment platform of RCDs. We select a total of 0.04 s of data (one cycle before the electric shock and one cycle after the electric shock) as electric shock sample data. The SPWVD is used to analyze the total leakage current signal to obtain the time-frequency matrix, and the frequency band width of the main spectrum energy is 0-150 Hz, which can be divided into 13 sub-bands. The singular value decomposition (SVD) is adopted to decompose the time-frequency matrix formed by 13 sub-bands, and the singular values corresponding to the respective sub-frequency band are obtained as the feature vectors of the electric shock signal. And then the fuzzy C means (FCM) algorithm is applied to perform the clustering of extracted feature vectors to get the electric shock signal type. Finally, a total of 400 groups of animals and plants shock data are used as the research object. Among them, 140 groups of animal electric shock samples and 140 groups of plant electric shock samples are taken as known samples, and 60 groups of animal electric shock samples and 60 groups of plant electric shock samples are used as test samples. The experimental results show that there are 3 groups of samples in 120 groups of test samples which are wrongly identified and the recognition accuracy rate is 97.50%. Among them, the accuracy rate of plant electric shock test sample is 100%, and there are 3 samples in animal electric shock test samples, which are identified incorrectly and the recognition accuracy rate is 95.00%. The above results verify the correctness and validity of diagnosing the type of the electric shock fault signal by the extracted characteristic value of the total leakage current, which can lay a solid theoretical and technical foundation for developing new generations of adaptive residual current protection devices.
electric current measurement; electric power systems; diagnosis; electric shock fault; time-frequency matrix; singular value decomposition (SVD); feature extraction; fuzzy C-mean (FCM) clustering
2017-07-16
2018-02-01
國家自然科學(xué)基金項目(51177165)
韓曉慧,河北石家莊人,講師,主要研究方向為電力系統(tǒng)繼電保護(hù)。Email:hanhui854201@126.com
10.11975/j.issn.1002-6819.2018.04.026
TM77
A
1002-6819(2018)-04-0217-06
韓曉慧,杜松懷,李 振,孫麗華. 基于泄漏電流時頻奇異譜和模糊聚類的觸電故障診斷[J]. 農(nóng)業(yè)工程學(xué)報,2018,34(4):217-222.doi:10.11975/j.issn.1002-6819.2018.04.026 http://www.tcsae.org
Han Xiaohui, Du Songhuai, Li Zhen, Sun Lihua. Diagnosis of electric shock fault based on time-frequency singular value spectrum of leakage current and fuzzy clustering[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(4): 217-222. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2018.04.026 http://www.tcsae.org