趙鳳展,郝 帥,張 宇,杜松懷,單葆國,蘇 娟,井天軍,趙婷婷
基于變分模態(tài)分解-BA-LSSVM算法的配電網(wǎng)短期負(fù)荷預(yù)測
趙鳳展1,郝 帥1,張 宇2,杜松懷1,單葆國3,蘇 娟1,井天軍1,趙婷婷2
(1.中國農(nóng)業(yè)大學(xué)信息與電氣工程學(xué)院,北京 100083; 2. 國網(wǎng)北京市電力公司,北京 100031;3. 國網(wǎng)能源研究院,北京 102209)
配電臺(tái)區(qū)日負(fù)荷序列呈現(xiàn)為既包含變化趨勢、又含有波動(dòng)細(xì)節(jié)的不規(guī)則曲線,該文借助變分模態(tài)分解(variational mode decomposition,VMD)將包含這些信息的原始日負(fù)荷序列分解為不同頻率尺度的子序列,并結(jié)合一系列復(fù)雜的環(huán)境因素,分別利用不同的最小二乘支持向量機(jī)(least squares support vector machine,LSSVM)模型進(jìn)行負(fù)荷預(yù)測,最后將基于不同頻率分量的預(yù)測結(jié)果相加得到最終的日負(fù)荷預(yù)測結(jié)果。為了提高LSSVM預(yù)測能力,采用蝙蝠算法(bat algorithm,BA)對(duì)各LSSVM的參數(shù)進(jìn)行尋優(yōu),同時(shí),該文分析了影響負(fù)荷變化的環(huán)境因素,設(shè)計(jì)了一套因素歸一化方法,預(yù)測過程考慮了環(huán)境因素的影響。仿真結(jié)果表明,該文提出的考慮復(fù)雜環(huán)境因素的預(yù)測思想及對(duì)歷史日負(fù)荷進(jìn)行VMD分解、BA優(yōu)化、LSSVM預(yù)測的組合預(yù)測方法能有效提高短期日負(fù)荷預(yù)測的準(zhǔn)確性。
算法;電能;配電臺(tái)區(qū)負(fù)荷預(yù)測;變分模態(tài)分解;最小二乘支持向量機(jī);蝙蝠算法;復(fù)雜環(huán)境因素
隨著全球能源日益緊缺和污染加重,電能正逐漸替代化石能源,成為人們生產(chǎn)生活的主要能量來源。中華人民共和國國家發(fā)展和改革委員會(huì)于2017年6月發(fā)布了《電力發(fā)展“十三五”規(guī)劃》,“升級(jí)改造配電網(wǎng),推進(jìn)智能電網(wǎng)建設(shè)”已經(jīng)成為中國電力發(fā)展的重點(diǎn)任務(wù)[1]。電力需求增加,用電負(fù)荷迅速增長,將對(duì)配電網(wǎng)規(guī)劃和運(yùn)行可靠性帶來巨大沖擊[2-4]。因此,研究短期日負(fù)荷預(yù)測成為《規(guī)劃》中的重要一環(huán)。
現(xiàn)有的短期負(fù)荷預(yù)測方法主要是多種傳統(tǒng)預(yù)測方法[5]及以人工神經(jīng)網(wǎng)絡(luò)、支持向量機(jī)(support vector machine,SVM)等為代表的機(jī)器學(xué)習(xí)方法[6-7]。機(jī)器學(xué)習(xí)方法在處理多因素問題(如氣象因素)方面具有更強(qiáng)的學(xué)習(xí)和模擬能力,預(yù)測效果更好。SVM具有小樣本預(yù)測、泛化能力強(qiáng)等特征;最小二乘支持向量機(jī)(least squares support vector machine,LSSVM)是SVM的一種改進(jìn),繼承了SVM的優(yōu)點(diǎn),用平方差損失函數(shù)代替不敏感損失函數(shù),用等式約束代替不等式約束,將二次規(guī)劃問題轉(zhuǎn)為求解線性方程,降低了求解復(fù)雜性,更適用于短期快速預(yù)測[8-9]。
由于負(fù)荷運(yùn)行的不確定性及負(fù)荷影響因素的復(fù)雜性,單一的負(fù)荷預(yù)測方法很難做到準(zhǔn)確地預(yù)測短期負(fù)荷,因此,組合方法已成為近年來短期負(fù)荷預(yù)測主流方法[10]。文獻(xiàn)[11]采用灰色模型及LSSVM組合預(yù)測方法及歷史負(fù)荷數(shù)據(jù)學(xué)習(xí)、日內(nèi)預(yù)測的思路,文獻(xiàn)[12]采用LSSVM預(yù)測、改進(jìn)并行粒子群算法優(yōu)化LSSVM參數(shù)的方法,都取得了較好的預(yù)測效果。
在預(yù)測前對(duì)數(shù)據(jù)進(jìn)行預(yù)處理可有效降低數(shù)據(jù)的不規(guī)律性帶來的干擾[10]。文獻(xiàn)[13]采用集成經(jīng)驗(yàn)?zāi)B(tài)分解(ensemble empirical mode decomposition,EEMD)將原始非平穩(wěn)負(fù)荷序列分解成一系列具有不同特征的子序列。變分模態(tài)分解(variational mode decomposition,VMD)是一種非遞歸、變模式的分解方法,克服了EEMD遞歸求解的缺點(diǎn),諧波分離效果更好[14]。文獻(xiàn)[15]采用VMD-模態(tài)重構(gòu)方法得到了信號(hào)的3個(gè)分量,各分量在不同頻率尺度上特點(diǎn)明顯;但對(duì)于短期日負(fù)荷預(yù)測來說,還需著重分析各分量在一日內(nèi)的變化特性。
另外一些最新的負(fù)荷預(yù)測文獻(xiàn)考慮利用人工智能算法對(duì)預(yù)測模型進(jìn)行參數(shù)優(yōu)化[16]。文獻(xiàn)[17]和文獻(xiàn)[18]分別采用粒子群算法(particle swarm optimization,PSO)和BBFMA(bare bones fireworks algorithm)進(jìn)行LSSVM參數(shù)優(yōu)化。文獻(xiàn)[19]采用蝙蝠算法(bat algorithm,BA)對(duì)SVM參數(shù)進(jìn)行優(yōu)化,結(jié)果表明借助參數(shù)優(yōu)化可以降低預(yù)測誤差;而LSSVM與SVM相比待優(yōu)化參數(shù)更少,所以,采用LSSVM預(yù)測可加快參數(shù)優(yōu)化及預(yù)測速度。文獻(xiàn)[20]采用BA-LSSVM優(yōu)化最小二乘支持向量機(jī)的懲罰參數(shù)和核參數(shù),與PSO-LSSVM相比具有更高的精度。
本文從分析短期日負(fù)荷的復(fù)雜環(huán)境因素入手,依次介紹了時(shí)序信號(hào)的VMD序列分解方法及BA優(yōu)化LSSVM的原理,進(jìn)而提出了基于VMD-BA-LSSVM模型的日負(fù)荷預(yù)測方法,最后以北京近郊某臺(tái)區(qū)配電變壓器一段時(shí)間的負(fù)荷數(shù)據(jù)及當(dāng)?shù)貧庀髷?shù)據(jù)為基礎(chǔ),利用所提模型進(jìn)行日負(fù)荷預(yù)測,并與其他幾種典型方法進(jìn)行比較,驗(yàn)證了所提預(yù)測方法的有效性。
用戶的用電行為受環(huán)境影響,在日負(fù)荷預(yù)測過程中考慮影響負(fù)荷的環(huán)境因素可以更真實(shí)預(yù)測該日實(shí)際用電情況。以北京某臺(tái)區(qū)配電變壓器一段時(shí)間的負(fù)荷數(shù)據(jù)及當(dāng)?shù)貧庀髷?shù)據(jù)為例,分析得到影響負(fù)荷變化的復(fù)雜環(huán)境因素主要包括日最低溫度、日平均溫度、天氣情況、天氣變化情況、負(fù)荷日類型以及季節(jié)情況,其中天氣情況包括:晴、多云、陰、小雨、中雨、大雨、暴雨、雨夾雪、小雪、中雪、大雪、暴雪、霜凍、霧、微風(fēng)、大風(fēng)、冰雹;天氣變化情況包括正常天氣和突變天氣;負(fù)荷日類型包括工作日和休息日。
日負(fù)荷序列看似波動(dòng)且無規(guī)律,但是經(jīng)過變分模態(tài)分解(variational mode decomposition,VMD),便可得到由不同頻率表征的趨勢分量及波動(dòng)分量。與EEMD的遞歸篩選原理不同,VMD采用非遞歸、變模態(tài)原理將信號(hào)分解成一系列有限帶寬子序列;VMD具有更好的諧波分離能力,并且每個(gè)分序列具有更好的規(guī)律性[21-22]。
VMD包括3個(gè)步驟,分別為建立約束變分模型、拉格朗日變換和交替更新:
式中{u}為分解所得到的個(gè)模態(tài)分量,為模態(tài)函數(shù)總個(gè)數(shù);{}為各模態(tài)分量的頻率中心;()為狄拉克分布;()為一個(gè)序列,是采樣時(shí)刻。
2)拉格朗日變換。以上約束變分問題通過引入增廣Lagrange函數(shù)消除約束變分模型的約束性,得到Lagrange函數(shù)表示的變分約束模型
式中為Lagrange乘法算子,用以確保嚴(yán)格執(zhí)行約束條件;為二次懲罰因子,用以確保轉(zhuǎn)換的準(zhǔn)確性。
3)交替更新。步驟2)中的優(yōu)化問題公式(2)可以根據(jù)下面的2個(gè)更新方程來求解。
VMD算法流程總結(jié)如下:
1)輸入待分解的序列()。
5)重復(fù)步驟3)、4)進(jìn)行迭代,直到滿足
由此得到分解后的個(gè)子序列,其模態(tài)函數(shù)為u,中心頻率為。
采用蝙蝠算法確定最小二乘支持向量機(jī)的預(yù)測參數(shù),在結(jié)合蝙蝠算法的良好收斂性的同時(shí),保留了最小二乘支持向量機(jī)的小樣本和計(jì)算快速的預(yù)測特點(diǎn)[23-24]。
最小二乘支持向量機(jī)(least squares support vector machine,LSSVM)是一種成熟的機(jī)器預(yù)測方法,作為SVM的擴(kuò)展,LSSVM將最小二乘損失函數(shù)作為損失函數(shù),并用等式約束條件替代SVM中的不等式約束條件;LSSVM保留了結(jié)構(gòu)風(fēng)險(xiǎn)最小化、小樣本等特點(diǎn),大大降低了計(jì)算復(fù)雜度[24]。
LSSVM的回歸過程如下:
2)根據(jù)結(jié)構(gòu)風(fēng)險(xiǎn)最小化準(zhǔn)則,式(7)對(duì)應(yīng)的LSSVM優(yōu)化問題可以表示為
3)求解上述優(yōu)化問題,構(gòu)建Lagrange函數(shù)
式中為Lagrange乘法算子。
式中為核函數(shù)寬度。
蝙蝠算法(bat algorithm,BA)是一種新興的尋優(yōu)算法,BA克服了遺傳算法、粒子群算法(particle swarm optimization,PSO)等算法執(zhí)行時(shí)間長,性能與初始值有關(guān)及參數(shù)敏感等缺點(diǎn)[24-25];BA可以在局部搜索和全局搜索之間動(dòng)態(tài)轉(zhuǎn)換,搜索過程具有更好的收斂性[26]。
2)隨機(jī)初始化蝙蝠搜索位置x,其中包含LSSVM中和2個(gè)參數(shù)信息。
3)對(duì)比所有個(gè)體的適應(yīng)度,尋找當(dāng)前全局最優(yōu)解*。
4)根據(jù)式(13)至式(15)更新每輪蝙蝠的搜索速度、搜索脈沖頻率和搜索位置
式中(,)、(,)分別為預(yù)測日的負(fù)荷真實(shí)值、負(fù)荷預(yù)測值;為預(yù)測點(diǎn)數(shù),本文=24。
基于VMD-BA-LSSVM的短期日負(fù)荷組合預(yù)測流程圖如圖2所示。
1)數(shù)據(jù)來源:北京某配變臺(tái)區(qū)變壓器低壓側(cè)有功功率數(shù)據(jù),并已經(jīng)過不良數(shù)據(jù)處理。
2)輸入數(shù)據(jù):分別為2017年1月10日至2017年1月23日、2017年2月2日至2017年2月15日、2017年2月17日至2017年3月2日的24 h負(fù)荷數(shù)據(jù)及日環(huán)境數(shù)據(jù),見表2。
3)預(yù)測模型:本文所提VMD-BA-LSSVM及其他5種組合預(yù)測模型:EEMD-LSSVM、VMD-LSSVM、VMD-BA-SVM、VMD-PSO-LSSVM、不含有天氣變化類型的VMD-BA-LSSVM(即原VMD-BA-LSSVM模型中前14天的變量()設(shè)為0),該方法簡稱為“VMD-BA-LSSVM(0)”。
4)預(yù)測目標(biāo):分別預(yù)測2017年1月24日、2017年2月16日、2017年3月3日這3 d的24 h負(fù)荷值。
采用以上6種組合預(yù)測模型得到的2017年1月24日、2017年2月16日、2017年3月3日24個(gè)時(shí)刻的負(fù)荷預(yù)測結(jié)果,前2天的預(yù)測結(jié)果如圖3所示。由圖3可得,利用VMD-BA-LSSVM的預(yù)測結(jié)果與實(shí)際值最為貼合;在實(shí)際值波動(dòng)較大的時(shí)刻(如2017年1月24日22:00),VMD-BA-LSSVM預(yù)測值最接近實(shí)際值,說明VMD-BA-LSSVM對(duì)負(fù)荷波動(dòng)預(yù)測最準(zhǔn)確。
圖2 基于VMD-BA-LSSVM的短期日負(fù)荷組合預(yù)測流程圖
表2 2017年1月10日至23日的負(fù)荷及環(huán)境數(shù)據(jù)
注:()為負(fù)荷日類型變量;()為天氣變化類型變量。 Note:() is load day type variable;() is weather change type variable.
圖3 原始負(fù)荷序列及各模型預(yù)測序列
以上6種組合預(yù)測模型3日的預(yù)測誤差平均值和平均預(yù)測用時(shí)及其優(yōu)化模型的平均優(yōu)化用時(shí)如表3所示。
表3 各組合模型3次預(yù)測的平均預(yù)測誤差及計(jì)算用時(shí)
注:VMD-BA-LSSVM(0)不考慮天氣變化類型。
Note: VMD-BA-LSSVM(0) is without considering weather change type.
通過比較6種組合預(yù)測方法的預(yù)測誤差可以反映各方法預(yù)測精度的優(yōu)劣性,同時(shí)比較各方法的運(yùn)行程序時(shí)間可以反映各方法預(yù)測效率的優(yōu)劣性。由表3比較得知,VMD-BA-LSSVM的預(yù)測誤差MAPE、max最小,說明該方法預(yù)測精準(zhǔn)度最高。同時(shí),由表3比較得知,對(duì)于同一LSSVM預(yù)測模型,BA的平均優(yōu)化用時(shí)比PSO的平均優(yōu)化用時(shí)少,說明BA的優(yōu)化效率更高;對(duì)于同一BA優(yōu)化模型,LSSVM的平均預(yù)測用時(shí)比SVM的平均預(yù)測用時(shí)少,是因?yàn)長SSVM待優(yōu)化參數(shù)比SVM少,所以參數(shù)優(yōu)化速度更快。
因此,由表3可得到以下結(jié)論:
1)VMD-LSSVM比EEMD-LSSVM誤差更低,預(yù)測速度更快,這表明相比EEMD而言,VMD有更好的序列分解能力,分解得到的分序列具有更好的規(guī)律。
2)VMD-BA-LSSVM比VMD-BA-SVM預(yù)測誤差更低,預(yù)測速度及優(yōu)化速度更快,這表明相比SVM而言,LSSVM有更好的負(fù)荷預(yù)測能力,預(yù)測精度更高,預(yù)測速度更快;LSSVM待優(yōu)化參數(shù)比SVM少,參數(shù)優(yōu)化速度更快。
3)VMD-BA-LSSVM比VMD-LSSVM、VMD-PSO- LSSVM誤差更低,優(yōu)化速度更快,這表明相比不優(yōu)化或PSO優(yōu)化,BA的優(yōu)化結(jié)果更優(yōu),且參數(shù)優(yōu)化速度更快。
4)含有天氣變化類型的VMD-BA-LSSVM模型比不含有天氣變化類型的同種模型誤差更低,這表明本文設(shè)計(jì)的考慮復(fù)雜環(huán)境因素的預(yù)測方法提高了預(yù)測精度。
針對(duì)短期日負(fù)荷精準(zhǔn)、高效預(yù)測方法的迫切需求,本文提出了一種基于變分模態(tài)分解和蝙蝠算法優(yōu)化最小二乘支持向量機(jī)的短期日負(fù)荷組合預(yù)測方法,主要結(jié)論如下:
1)采用VMD對(duì)非線性、非平穩(wěn)的日負(fù)荷序列進(jìn)行分解,并將得到的不同頻率尺度的負(fù)荷矩陣輸入不同的LSSVM中進(jìn)行預(yù)測。VMD可以更細(xì)致表征日負(fù)荷在不同頻率尺度上的變化特性,分解結(jié)果具有更好的規(guī)律性。
2)采用LSSVM進(jìn)行負(fù)荷預(yù)測,LSSVM的參數(shù)由BA進(jìn)行尋優(yōu)。LSSVM可以有效地預(yù)測短期負(fù)荷序列,與SVM相比預(yù)測精度更高,預(yù)測速度更快,并且待優(yōu)化參數(shù)更少,尋優(yōu)速度更快;同時(shí)BA參數(shù)優(yōu)選方法具有比PSO參數(shù)優(yōu)選方法更優(yōu)秀的全局尋優(yōu)能力。
3)考慮VMD、LSSVM、BA各自優(yōu)點(diǎn),綜合設(shè)計(jì)了VMD-BA-LSSVM組合預(yù)測方法,該方法與算例中的其他5種組合預(yù)測方法相比預(yù)測精度最高,預(yù)測速度最快,適用于短期日負(fù)荷預(yù)測。
4)考慮了復(fù)雜環(huán)境因素對(duì)日負(fù)荷變化的影響,并將復(fù)雜環(huán)境因素?cái)?shù)據(jù)量化輸入到預(yù)測模型中,使預(yù)測結(jié)果更加準(zhǔn)確。
[1] 中華人民共和國國家發(fā)展和改革委員會(huì).電力發(fā)展“十三五”規(guī)劃(2016-2020年)[EB/OL]. http://www.ndrc.gov.cn/ fzgggz/fzgh/ghwb/gjjgh/201706/t20170605_849994.html.
[2] Ziari I, Ledwich G, Ghosh A, et al. Optimal distribution network reinforcement considering load growth, line loss, and reliability[J]. IEEE Transactions on Power Systems, 2013, 28(2): 587-597.
[3] Ziari I, Ledwich G, Ghosh A, et al. Integrated distribution systems planning to improve reliability under load growth[J]. IEEE Transactions on Power Delivery, 2012, 27(2): 757-765.
[4] 竇震海,楊仁剛,焦嬌. 基于均生函數(shù)-最優(yōu)子集回歸模型的短期電力負(fù)荷預(yù)測方法[J]. 農(nóng)業(yè)工程學(xué)報(bào),2013,29(14):178-184.
Dou Zhenhai, Yang Rengang, Jiao Jiao. Method of short-term load forecasting based on mean generating function-optimal subset regression[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2013, 29(14): 178-184. (in Chinese with English abstract)
[5] 康重慶,夏清,劉梅. 電力系統(tǒng)負(fù)荷預(yù)測[M]. 北京:中國電力出版社,2007.
[6] 肖白,劉慶永,牛強(qiáng),等. 基于元胞負(fù)荷特性分析的RBF神經(jīng)網(wǎng)絡(luò)空間負(fù)荷預(yù)測方法[J]. 電網(wǎng)技術(shù),2018,42(1):301-307. Xiao Bai, Liu Qingyong, Niu Qiang, et al. A spatial load forecasting method based on RBF neural network and cellular load characteristics analysis[J]. Power System Technology, 2018, 42(1): 301-307. (in Chinese with English abstract)
[7] 謝敏,鄧佳梁,吉祥,等. 基于信息熵和變精度粗糙集優(yōu)化的支持向量機(jī)降溫負(fù)荷預(yù)測方法[J]. 電網(wǎng)技術(shù),2017,41(1):210-214. Xie Min, Deng Jialiang, Ji Xiang, et al. Cooling load forecasting method based on support vector machine optimized with entropy and variable accuracy roughness set[J]. Power System Technology, 2017, 41(1): 210-214. (in Chinese with English abstract)
[8] 王新,孟玲玲. 基于EEMD-LSSVM的超短期負(fù)荷預(yù)測[J]. 電力系統(tǒng)保護(hù)與控制,2015,43(1):61-66. Wang Xin, Meng Lingling. Ultra-short-term load forecasting based on EEMD-LSSVM[J]. Power System Protection and Control, 2015, 43(1): 61-66. (in Chinese with English abstract)
[9] Zhang J, Lin Y, Lu P. Short-term electricity load forecasting based on ICA and LSSVM[C]//International Conference on Computational Intelligence & Software Engineering. IEEE, 2009.
[10] Lin Y, Luo H, Wang D, et al. An ensemble model based on machine learning methods and data preprocessing for short-term electric load forecasting[J]. Energies, 2017, 10(8): 1186.
[11] 唐杰明,劉俊勇,楊可,等. 基于灰色模型和最小二乘支持向量機(jī)的電力短期負(fù)荷組合預(yù)測[J]. 電網(wǎng)技術(shù),2009,33(3):63-68. Tang Jieming, Liu Junyong, Yang Ke, et al. Short-term load combination forecasting by grey model and least square support vector machine[J]. Power System Technology, 2009, 33(3): 63-68. (in Chinese with English abstract)
[12] 王保義,王冬陽,張少敏. 基于Spark和IPPSO_LSSVM的短期分布式電力負(fù)荷預(yù)測算法[J]. 電力自動(dòng)化設(shè)備,2016,36(1):117-122. Wang Baoyi, Wang Dongyang, Zhang Shaomin, et al. Distributed short-term load forecasting algorithm based on Spark and IPPSO_LSSVM[J]. Electric Power Automation Equipment, 2016, 36(1): 117-122. (in Chinese with English abstract)
[13] 陳艷平,毛弋,陳萍,等. 基于EEMD-樣本熵和Elman神經(jīng)網(wǎng)絡(luò)的短期電力負(fù)荷預(yù)測[J]. 電力系統(tǒng)及其自動(dòng)化學(xué)報(bào),2016,28(3):59-64. Chen Yanping, Mao Yi, Chen Ping, et al. Short-term power load forecasting based on ensemble empirical mode decomposition-sample entropy and elman neural network[J]. Proceedings of the CSU-EPSA, 2016, 28(3): 59-64. (in Chinese with English abstract)
[14] Lahmiri S. Comparing variational and empirical mode decomposition in forecasting day-ahead energy prices[J]. IEEE Systems Journal, 2017, 11(3): 1907-1910.
[15] 梁智,孫國強(qiáng),李虎成,等. 基于VMD與PSO優(yōu)化深度信念網(wǎng)絡(luò)的短期負(fù)荷預(yù)測[J]. 電網(wǎng)技術(shù),2018,42(2):598-606. Liang Zhi, Sun Guoqiang, Li Hucheng, et al. Short-term load forecasting based on VMD and PSO optimized deep belief network[J]. Power System Technology, 2018, 42(2): 598-606. (in Chinese with English abstract)
[16] Niu D, Dai S. A short-term load forecasting model with a modified particle swarm optimization algorithm and least squares support vector machine based on the denoising method of empirical mode decomposition and grey relational analysis[J]. Energies, 2017, 10(3): 598-606.
[17] 金璐,鐘鳴,張景霞,等. 基于PSO-LSSVM的海水淡化負(fù)荷預(yù)測[J]. 智慧電力,2019,47(2):55-61.
[18] Lei C, Fang B, Gao H, et al. Short-term power load forecasting based on least squares support vector machine optimized by bare bones fireworks algorithm[C]//2018 IEEE 3rd Advanced Information Technology, Electronic and Automation Control Conference (IAEAC), Chongqing, 2018. Doi: 10.1109/IAEA, 2018.8577212
[19] 沈淵彬,劉慶珍,李友軍,等. 基于模糊組合權(quán)重的BA-SVM短期負(fù)荷預(yù)測[J]. 南昌大學(xué)學(xué)報(bào):工科版,2015,37(3):300-306. Shen Yuanbin, Liu Qingzhen, Li Youjun, et al. Short-term load forecasting based on BA-SVM with fuzzy combined weight[J]. Journal of Nanchang University: Engineering & Technology, 2015, 37(3): 300-306. (in Chinese with English abstract)
[20] 公政,姜文,王來河,等. 基于BA-LSSVM的短期電力負(fù)荷預(yù)測研究[J]. 電子質(zhì)量,2017(3):1-4.
Gong Zheng, Jiang Wen, Wang Laihe, et al. Based on the BA-LSSVM short-term power load forecasting research[J]. Electronics Quality, 2017(3): 1-4. (in Chinese with English abstract)
[21] Abdoos A A. A new intelligent method based on combination of VMD and ELM for short term wind power forecasting[J]. Neurocomputing, 2016, 203: 111-120.
[22] 張亞超,劉開培,秦亮. 基于VMD-SE和機(jī)器學(xué)習(xí)算法的短期風(fēng)電功率多層級(jí)綜合預(yù)測模型[J]. 電網(wǎng)技術(shù),2016,40(5):1334-1340. Zhang Yachao, Liu Kaipei, Qin Liang. Short-term wind power multi-leveled combined forecasting model based on variational mode decomposition-sample entropy and machine learning algorithms[J]. Power System Technology, 2016, 40(5): 1334-1340. (in Chinese with English abstract)
[23] Zhang Q, Lai K K, Niu D, et al. A fuzzy group forecasting model based on least squares support vector machine (LS-SVM) for short-term wind power[J]. Energies, 2012, 5(9): 3329-3346.
[24] Khan K, Sahai A. A comparison of BA, GA, PSO, BP and LM for training feed forward neural networks in e-learning context[C]//International Journal of Intelligent Systems and Applications (IJISA), 2012, 7: 23-29.
[25] Sun W, Liu M, Liang Y. Wind speed forecasting based on FEEMD and LSSVM optimized by the bat algorithm[J]. Energies, 2015, 8(7): 6585-6607.
[26] Gandomi A H, Yang X S, Alavi A H, et al. Bat algorithm for constrained optimization tasks[J]. Neural Computing & Applications, 2013, 22(6): 1239-1255.
Short-term load forecasting for distribution transformer based on VMD-BA-LSSVM algorithm
Zhao Fengzhan1, Hao Shuai1, Zhang Yu2, Du Songhuai1, Shan Baoguo3, Su Juan1, Jing Tianjun1, Zhao Tingting2
(1.,100083,; 2.,100031,; 3.,102209,)
With the wide application of all kinds of electrical equipment in the distribution system, the power load has increased in recent years, which has a great impact on distribution network. Thus, forecasting the short-term daily load is required. Combining the advantages of VMD, LSSVM and BA, a novel VMD-BA-LSSVM short-term daily power load forecasting method was designed, and the complex environmental factors were considered in this paper. Least squares support vector machine (LSSVM) is a classical machine prediction method, which has the advantages of small sample size, powerful generalization ability and fast solution. However, with the gradual improvement of forecasting accuracy requirements, simple LSSVM can’t guarantee the accuracy of the forecasting work. The daily load sequence of the distribution transformer presents an irregular curve containing variation currents and fluctuation details. These information can be separated and predicted respectively in the prediction process, thus better prediction results can be obtained. Although the daily load sequence seems to be fluctuant and irregular, the trend component and wave components in different frequency scales can be obtained by the variational mode decomposition method (VMD). Compared with the process of recursion and screening in EEMD, VMD is characterized by its non-recursive and variable mode. VMD decomposes the original load sequence into a series of specific band-limited subsequences, which aims to decrease instability. VMD has the better capability of harmonic separation, and each subsequence has a better regularity. In this paper, the VMD was used to decompose daily load sequence of a day and yield a series of subsequences with specific frequencies. Subsequences were put into four LSSVMs for the respective forecast. Different parameters in LSSVMs were optimized by the bat algorithm (BA). Meanwhile, the affection of the complex environmental factors was studied and the normalization approach of those factors was proposed. Thus, complex environmental factors were considered in forecasting. The procedures of this prediction method were as following: Firstly, the input data of the method was the daily load data with a one-hour interval and daily environmental data with a one-day interval of the previous 14 days. The daily load sequence (1 row and 24 columns, 1×24) was decomposed by the VMD method and yielded four low-to-high frequency subsequences. Secondly, the four subsequences of the previous 14 days were combined into four 14×24 matrices. Thirdly, the normalized data of the four matrices and environmental data were put into four LSSVMs to forecast the load of the 15th day. Meanwhile, the parameters of LSSVM were optimized by BA. The last, the four LSSVMs results were summed and yielded the final prediction result. In this paper, the VMD was used to decompose nonlinear, fluctuant daily load sequence and yield subsequences with different frequency scales. Subsequences were combined and put into LSSVMs for the respective forecast. Simulation results showed that the forecasting accuracy of VMD-based forecasting method was higher than EEMD-based method. At the same time, LSSVM was used to forecast, and BA was used to optimize the uncertain parameters. The simulation results showed that compared with SVM, LSSVM had a better capability to approximate the load sequence, and got higher prediction efficiency. LSSVM had less uncertain parameters than SVM, thus the efficiency of parameter optimization was higher. Furthermore, BA had excellent capability of global optimization and rapid convergence. Simulation results showed that the proposed method was the most accurate and efficient method, compared with other five forecasting methods.
algorithms; power; load forecasting for the distribution transformer; variational mode decomposition; least squares support vector machine; bat algorithm; complex environmental factor
2018-12-12
2019-06-25
國家電網(wǎng)公司科技項(xiàng)目(《市場交易環(huán)境下電力供需技術(shù)模型和應(yīng)用研究》);國家重點(diǎn)研發(fā)項(xiàng)目(2016YFB0900100)
趙鳳展,博士,副研究員,研究方向?yàn)橹悄芘潆娋W(wǎng)分析、規(guī)劃、評(píng)價(jià)與優(yōu)化運(yùn)行等。Email:zhaofz@cau.edu.cn
10.11975/j.issn.1002-6819.2019.14.024
TM 715
A
1002-6819(2019)-14-0190-08
趙鳳展,郝 帥,張 宇,杜松懷,單葆國,蘇 娟,井天軍,趙婷婷. 基于變分模態(tài)分解-BA-LSSVM算法的配電網(wǎng)短期負(fù)荷預(yù)測[J]. 農(nóng)業(yè)工程學(xué)報(bào),2019,35(14):190-197. doi:10.11975/j.issn.1002-6819.2019.14.024 http://www.tcsae.org
Zhao Fengzhan, Hao Shuai, Zhang Yu, Du Songhuai, Shan Baoguo, Su Juan, Jing Tianjun, Zhao Tingting. Short-term load forecasting for distribution transformer based on VMD-BA-LSSVM algorithm[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2019, 35(14): 190-197. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2019.14.024 http://www.tcsae.org