資源描述:
《基于小波和改進(jìn)神經(jīng)樹的電能質(zhì)量擾動(dòng)分類.pdf》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在行業(yè)資料-天天文庫(kù)。
1、第42卷第24期電力系統(tǒng)保護(hù)與控制Vo1.42NO.242014年12月16日PowerSystemProtectionandControlDec.16.2014基于小波和改進(jìn)神經(jīng)樹的電能質(zhì)量擾動(dòng)分類吳兆剛,李唐兵,姚建剛,龔文龍,陳強(qiáng)(1.湖南大學(xué)電氣與信息工程學(xué)院,湖南長(zhǎng)沙410082;2.江西省電力科學(xué)研究院,江西南昌3300963.湖南湖大華龍電氣與信息技術(shù)有限公司,湖南長(zhǎng)沙410082)摘要:準(zhǔn)確地識(shí)別和分類電能質(zhì)量擾動(dòng)對(duì)分析和綜合治理電能質(zhì)量問題具有重要意義。提出了一種基于小波和改進(jìn)神經(jīng)樹
2、的電能質(zhì)量擾動(dòng)分類方法。該方法利用小波分解擾動(dòng)信號(hào)到各個(gè)頻帶,在基頻頻帶、諧波頻帶和高頻帶上分別計(jì)算其能量值和小波系數(shù)熵作為特征值,另計(jì)算基波頻帶擾動(dòng)過程的均方根作為特征的補(bǔ)充,融合能量值、熵和均方根值作為擾動(dòng)判斷的特征向量,規(guī)范化后輸入到改進(jìn)神經(jīng)樹分類器進(jìn)行訓(xùn)練和分類。改進(jìn)神經(jīng)樹分類器是由神經(jīng)網(wǎng)絡(luò)和決策樹及其分類規(guī)則構(gòu)成。仿真表明,該方法提取特征值的計(jì)算量小且融合后的特征向量能夠很好地體現(xiàn)不同擾動(dòng)信號(hào)之間的差異信息,構(gòu)造的改進(jìn)神經(jīng)樹分類器結(jié)合了神經(jīng)網(wǎng)絡(luò)和決策樹在模式分類中各自的優(yōu)點(diǎn),結(jié)構(gòu)簡(jiǎn)單且表現(xiàn)出
3、良好的收斂性、全局最優(yōu)性和泛化性,分類準(zhǔn)確率較高,能夠有效地識(shí)別七種常見的電能質(zhì)量擾動(dòng)。關(guān)鍵詞:電能質(zhì)量;擾動(dòng)分類;小波變換;特征向量;改進(jìn)神經(jīng)樹PowerqualitydisturbanceelassiticationbasedonawaveletandimprovedneuraltreeWUZhao—gang,LITang—bing~,YAOJian·gang,GONGWen-long,CHENQiang(1.CollegeofElectricalandInformationEngineering
4、,HunanUniversity,Changsha410082,China;2.JiangxiElectricPowerResearchInstitute,Nanchang330096,China;3.HunanHDHLElectrical&InformmionTechnologyCo.,Ltd.,Changsha410082,China)Abstract:Preciseidentificationandclassificationforpowerqualitydisturbancesissignif
5、icantlyimpo~anttoanalyzeandcomprehensivelycopewithpowerqualityproblems.Basedonwaveletandimprovedneuraltreetechniques,anewclassificationmethodologyforpowerqualitydisturbancesisproposed.Inthemethod,thedisturbancesignalisdecomposedintodifferent~equencyband
6、s,whilstenergyvaluesandwaveletcoeficiententropiesofthebase,harmonicandhighfrequencybandsarecalculatedaseigenvaluesrespectively.Therootmeanproducedinthedisturbanceprocessofthebasewavebandiscalculatedasasupplement,whichisthencombinedwiththeenergyvaluesand
7、waveletcoeficiententropiesaseigenvectorsforjudgingthedisturbances.Thereaftertheeigenvectorsarenormalizedandinputintotheimprovedneuraltreeclassifier,composedofneuralnetwork,decisiontreesandclassificationrules,fortrainingandclassifying.Simulationresultsde
8、monstratethemethodhasasmallamountofcalculationtoextracteigenvaluesandtheobtainedeigenvectorsCanadequatelyreflectthediferenceinformationfordifferentdisturbancesignals.Theimprovedneuraltreeclassifiercombinesrespectivesuperioritieso