《電子技術(shù)應(yīng)用》
您所在的位置:首頁(yè) > 通信與網(wǎng)絡(luò) > 設(shè)計(jì)應(yīng)用 > 基于自適應(yīng)線(xiàn)性神經(jīng)元網(wǎng)絡(luò)的諧波檢測(cè)算法
基于自適應(yīng)線(xiàn)性神經(jīng)元網(wǎng)絡(luò)的諧波檢測(cè)算法
2017年電子技術(shù)應(yīng)用第6期
方 樹(shù)1,,韓 楊2,,羅 飛3,,徐 琳4
1.國(guó)網(wǎng)四川省電力公司,,四川 成都610041,;2.電子科技大學(xué),,四川 成都611731,; 3.國(guó)網(wǎng)涼山供電公司,四川 西昌615050,;4.國(guó)網(wǎng)四川省電力公司電力科學(xué)研究院,,四川 成都610072
摘要: 為了克服基于傅里葉變換(FFT)諧波檢測(cè)算法運(yùn)算量大、實(shí)時(shí)性不強(qiáng),、易受噪聲影響的缺點(diǎn),,提出了基于自適應(yīng)線(xiàn)性神經(jīng)元網(wǎng)絡(luò)(ADALINE)的諧波檢測(cè)算法,建立了基于最小二乘法(LMS)的最優(yōu)解求解過(guò)程的數(shù)學(xué)模型,,根據(jù)LMS誤差與各次諧波傅里葉系數(shù)之間的三維流形的幾何形狀選擇算法的步長(zhǎng)因子。采用時(shí)域迭代的方法準(zhǔn)確地提取基波有功,、無(wú)功和各次諧波分量,,為實(shí)現(xiàn)APF可選擇性諧波補(bǔ)償?shù)於嘶A(chǔ)。
中圖分類(lèi)號(hào): TN911,;TM46
文獻(xiàn)標(biāo)識(shí)碼: A
DOI:10.16157/j.issn.0258-7998.2017.06.021
中文引用格式: 方樹(shù),,韓楊,羅飛,,等. 基于自適應(yīng)線(xiàn)性神經(jīng)元網(wǎng)絡(luò)的諧波檢測(cè)算法[J].電子技術(shù)應(yīng)用,,2017,43(6):83-86.
英文引用格式: Fang Shu,,Han Yang,,Luo Fei,et al. A novel harmonic estimation algorithm based on adaptive linear neural network[J].Application of Electronic Technique,,2017,,43(6):83-86.
A novel harmonic estimation algorithm based on adaptive linear neural network
Fang Shu1,Han Yang2,,Luo Fei3,,Xu Lin4
1.State Grid Sichuan Electric Power Company,Chengdu 610041,,China,; 2.University of Electronic Science and Technology of China,Chengdu 611731,,China,; 3.State Grid LiangShan Electric Power Supply Company,Xichang 615050,China,; 4.State Grid Sichuan Electric Power Research Institute,,Chengdu 610072,China
Abstract: To overcome the shortcomings of the conventional FFT-based harmonic estimation algorithms due to extensive computational load, poor real-time capabilities and poor noise immunities, a novel harmonic estimation algorithm based on adaptive linear neural network(ADALINE) is proposed. The mathematical model of the algorithm is presented using the least mean square(LMS) algorithm, and the step-size selection of the algorithm is discussed using three-dimensional manifold based on the LMS error and the Fourier series coefficients. The fundamental active, reactive component and the individual harmonic component are extracted from the load current using time-domain iterations, which lays the basis for the selective harmonic compensation purposes.
Key words : Harmonic detection,;least mean square algorithm,;ADALINE;step-size

0 引言

    目前,,諧波電流檢測(cè)主要采用頻域法和時(shí)域檢測(cè)法,。文獻(xiàn)[1]提出基于快速傅里葉變換(FFT)的頻域諧波檢測(cè)方法,但容易出現(xiàn)頻譜泄漏等諸多問(wèn)題,。文獻(xiàn)[2]提出自適應(yīng)諧波檢測(cè)方法,,該方法根據(jù)自適應(yīng)干擾對(duì)消的原理,具有較高的檢測(cè)精度,,但是動(dòng)態(tài)響應(yīng)較慢,。文獻(xiàn)[3-5]提出采用人工神經(jīng)網(wǎng)絡(luò)的諧波檢測(cè)算法,但該方法計(jì)算量大,、實(shí)時(shí)性差,。文獻(xiàn)[6]提出基于神經(jīng)網(wǎng)絡(luò)的諧波辨識(shí)方法,描述了該方法的實(shí)現(xiàn)過(guò)程,,但計(jì)算量很大,。文獻(xiàn)[7]采用神經(jīng)網(wǎng)絡(luò)控制器實(shí)現(xiàn)對(duì)諧波電壓的抑制。文獻(xiàn)[8]比較了時(shí)域與頻域神經(jīng)網(wǎng)絡(luò)方法在有源濾波器中的應(yīng)用,。文獻(xiàn)[9]采用神經(jīng)網(wǎng)絡(luò)實(shí)現(xiàn)諧波檢測(cè),,并采用滑模變結(jié)構(gòu)控制實(shí)現(xiàn)對(duì)諧波的補(bǔ)償。文獻(xiàn)[10]采用神經(jīng)網(wǎng)絡(luò)從負(fù)荷電流中提取3,、5次諧波分量,,該方法采用10個(gè)隱含層神經(jīng)元,網(wǎng)絡(luò)規(guī)模小,,有較高的穩(wěn)態(tài)精度,。文獻(xiàn)[11]提出提升小波變換和變步長(zhǎng)LMS相結(jié)合的自適應(yīng)諧波檢測(cè)算法,對(duì)諧波電流進(jìn)行正交變換,,有效減少輸入數(shù)據(jù)的互相關(guān)性,,加快LMS的收斂速度,穩(wěn)態(tài)誤差較小,。

1 基于自適應(yīng)線(xiàn)性神經(jīng)元網(wǎng)絡(luò)的諧波檢測(cè)算法模型

    自適應(yīng)線(xiàn)性神經(jīng)元網(wǎng)絡(luò)(Adaptive Linear Neural Network,,ADALINE)算法是一種采用最小二乘(LMS)尋找最優(yōu)解的優(yōu)化數(shù)學(xué)方法,其原理圖如1所示,。

ck3-t1.gif

    根據(jù)ADALINE的定義,,任意信號(hào)Y(t)可表示為:

     ck3-gs1.gif

ck3-gs2-8.gif

其中矩陣R為實(shí)對(duì)稱(chēng)陣,。從式(6)看出,均方誤差ε是關(guān)于權(quán)系數(shù)向量的二次函數(shù),,對(duì)ε求偏導(dǎo)得:

ck3-gs9-10.gif

    式(10)稱(chēng)為Weiner解,,對(duì)應(yīng)于權(quán)向量空間最優(yōu)解的點(diǎn),該點(diǎn)的目標(biāo)函數(shù)取得最小值εmin,。采用這種方法計(jì)算最優(yōu)權(quán)向量涉及到對(duì)矩陣求逆,,當(dāng)輸入信號(hào)是諧波含量高的隨機(jī)信號(hào)流時(shí),難以實(shí)時(shí)計(jì)算準(zhǔn)確的R-1,。因此,,可對(duì)第k步平方誤差εk直接求偏導(dǎo)得:

ck3-gs11-16.gif

2 步長(zhǎng)因子對(duì)諧波檢測(cè)算法的影響

    權(quán)系數(shù)向量的維數(shù)與估計(jì)的諧波次數(shù)有關(guān)。當(dāng)估計(jì)的諧波次數(shù)遠(yuǎn)小于實(shí)際負(fù)載的諧波電流次數(shù)時(shí),,必然會(huì)引起較大的計(jì)算誤差,,反之又會(huì)引起計(jì)算量的增加。因此需要在兩者之間進(jìn)行折衷,,同時(shí)權(quán)系數(shù)的精度也受到學(xué)習(xí)因子的影響,,以下通過(guò)仿真進(jìn)行分析。

    不失一般性,,設(shè)被檢測(cè)的信號(hào)為:

ck3-gs17.gif

    圖2為在不同基波步長(zhǎng)因子μ1情況下,,信號(hào)的基波幅值由500突變?yōu)? 000時(shí),基波參數(shù)a1sin(ω0t),、b1cos(ω0t)和檢測(cè)誤差ierr的三維流形圖,其中諧波步長(zhǎng)因子μi=0.003 9(i=3,,5,,7,9),。圖2表明,,當(dāng)μ1=0.001 2時(shí),三維流形的軌跡從ierr=0平面以500為半徑的圓,,基波幅值突增后,,經(jīng)過(guò)一系列的振蕩,收斂到ierr=0平面以1 000為半徑的圓,;當(dāng)μ1=0.012,,信號(hào)突變后ierr經(jīng)過(guò)短暫的過(guò)渡過(guò)程收斂到ierr=0平面;當(dāng)μ1=0.12時(shí),,三維流形軌跡從初態(tài)到終態(tài)都嚴(yán)重地畸變,。

ck3-t2.gif

    仿真結(jié)果表明,μ1過(guò)小,,盡管檢測(cè)精度很高,,但動(dòng)態(tài)過(guò)程收斂緩慢;μ1過(guò)大,又會(huì)引起檢測(cè)值畸變嚴(yán)重,,導(dǎo)致整個(gè)檢測(cè)過(guò)程發(fā)散,,系統(tǒng)失穩(wěn)。因此μ1需要在檢測(cè)精度和動(dòng)態(tài)響應(yīng)速度兩方面作折衷選擇,,其取值范圍可以從三維流形的幾何形狀直觀地看出,。

3 基于CCS的算例分析

    為了驗(yàn)證上述諧波檢測(cè)算法的可行性,將該算法在浮點(diǎn)型DSP TMS320C6726硬件平臺(tái)中通過(guò)編程實(shí)現(xiàn),。下面通過(guò)兩個(gè)算例加以說(shuō)明,。圖3、圖4中橫坐標(biāo)表示采樣點(diǎn)數(shù),,采樣頻率為10 kHz,,縱坐標(biāo)單位為A。

ck3-t3.gif

ck3-t4.gif

    第一組算例中Y(t)=500sin(ω0t)+200sin(3ω0t),。圖3分別給出了不同3次諧波步長(zhǎng)因子μ3下的采樣電流和通過(guò)該檢測(cè)算法重構(gòu)的3次諧波波形,,其中μ3分別為0.000 4和0.003 9,基波因子μ1均為0.012,。當(dāng)μ3=0.000 4時(shí),,估計(jì)的3次諧波幅值為134 A,即3次諧波的估計(jì)誤差達(dá)到33%,,如圖3(a)所示,;當(dāng)μ3=0.003 9時(shí),3次諧波幅值為208 A,,誤差減小至4%,,如圖3(b)所示。

    第二組算例中Y(t)=500sin(ω0t)+200sin(3ω0t)+200sin(5ω0t)+sin(7ω0t)+sin(9ω0t),,其中μ1=0.012,、μi=0.003 9(i=3,5,,7,,9)。圖4(a)為負(fù)載電流和APAF算法的估計(jì)誤差波形圖,,其中估計(jì)誤差在±6A之間波動(dòng),,即估計(jì)誤差在1%左右,確保了整個(gè)算法的收斂性,;圖4(b)為通過(guò)APAF算法重構(gòu)的基波有功和基波無(wú)功電流分量,,其中有功電流的估計(jì)誤差為8 A(即1.6%),而無(wú)功分量的估計(jì)誤差為1.44 A(即0.28%),;圖4(c)為APAF算法重構(gòu)的3次,、5次諧波分量結(jié)果,,估計(jì)誤差與圖3非常類(lèi)似,也就是說(shuō),,步長(zhǎng)因子在很大范圍內(nèi)是適用的,,并不會(huì)隨著負(fù)載電流諧波次數(shù)增加而改變。

    算例分析結(jié)果表明,,要實(shí)現(xiàn)快速準(zhǔn)確地提取各次諧波分量,,必須在考慮檢測(cè)精度和動(dòng)態(tài)響應(yīng)速度兩方面的前提下,合理地選擇步長(zhǎng)因子,。

4 結(jié)論

    基于自適應(yīng)線(xiàn)性神經(jīng)元網(wǎng)絡(luò)ADALINE的諧波檢測(cè)算法,,能準(zhǔn)確地提取基波有功、無(wú)功和各次諧波分量,,克服了基于傅里葉變換(FFT)諧波檢測(cè)算法運(yùn)算量大,、實(shí)時(shí)性不強(qiáng)、易受噪聲影響的缺點(diǎn),,也避免了基于瞬時(shí)無(wú)功理論(IRPT)諧波檢測(cè)算法易受電壓畸變影響的不足,,為實(shí)現(xiàn)APF可選擇性諧波補(bǔ)償?shù)於嘶A(chǔ)。

參考文獻(xiàn)

[1] 潘文,,錢(qián)愈濤,,周鸚.基于加窗插值FFT的電力諧波測(cè)量理論窗函數(shù)研究[J].電工技術(shù)學(xué)報(bào),1994,,9(1):50-54.

[2] BARROS J,,PEREZ E.An adaptive method for determining the reference compensating current in single-phase shunt active power filters[J].IEEE Transactions on Power Delivery,2003,,18(4):1578-1580.

[3] HAN B M,,BAE B Y,OVASKA S J.Reference signal generator for active power filters using improved adaptive predictive filter[J].IEEE Transactions on Industrial Electronics,,2005,52(2):576-584.

[4] RUKONUZZAMAN M,,UEMATSU T,,NISHIDA K,et al.Harmonic and reactive currents estimation with adaptive neural network and their compensation with DSP control in single-phase shunt active power filter[C].Telecommunications Energy Conference,,2003:358-365.

[5] RUKONUZZAMAN M,,NAKAOKA M.Adaptive neural network based harmonic current compensation in active power filter[C].International Joint Conference on Neural Networks,2001,,3:2281-2286.

[6] OSOWSKI S.Neural network for estimation of harmonic components in a power system[J].IEE Proc.C,,Gener.Transm.Distrib.,1992,,139(2):129-135.

[7] KUMAMOTO A,,HIKIHARA T,,HIRANE Y,et al.Suppression of harmonic voltage distortion by neural network controller[C].Proc.IEEE Ind.Appl.Annual Meeting,,Houston,,Texas,USA,,1992:754-761.

[8] ROUND S,,MOHAN N.Comparison of frequency and time domain neural network controllers for an active power filter[C].Proc.IEEE IECON,Maui,,Hawaii,,USA,1993:1099-1104.

[9] KANDIL M,,ABDELKADER S,,ELMITWALLY A,et al.A novel three-phase active filter based on neural networks and sliding mode control[C].Proc.IEEE Industrial Elec.Conf.(IECON),,San Jose,,California,USA,,1999:867-872.

[10] PECHARANIN N,MITSUI H.Harmonic detection using neural network[C].Proc.IEEE Intl.Conf.On Neural Networks,,Perth,Australia,,1995:923-926.

[11] 高紅霞,,劉曉樂(lè).電力諧波檢測(cè)改進(jìn)算法在DSP上的應(yīng)用與實(shí)現(xiàn)[J].電子技術(shù)應(yīng)用,2015,,41(4):125-128.



作者信息:

方  樹(shù)1,,韓  楊2,羅  飛3,,徐  琳4

(1.國(guó)網(wǎng)四川省電力公司,,四川 成都610041;2.電子科技大學(xué),,四川 成都611731,;

3.國(guó)網(wǎng)涼山供電公司,四川 西昌615050,;4.國(guó)網(wǎng)四川省電力公司電力科學(xué)研究院,,四川 成都610072)

此內(nèi)容為AET網(wǎng)站原創(chuàng),未經(jīng)授權(quán)禁止轉(zhuǎn)載,。