楊童亮 胡 東 唐 超 方 云 謝菊芳
基于SMA-VMD-GRU模型的變壓器油中溶解氣體含量預(yù)測(cè)
楊童亮1,2胡 東1唐 超1方 云1謝菊芳1,2
(1. 西南大學(xué)工程技術(shù)學(xué)院 重慶 400715 2. 西南大學(xué)智能電網(wǎng)及裝備新技術(shù)國(guó)際研發(fā)中心 重慶 400715)
針對(duì)電力變壓器油中溶解氣體濃度序列非線性、非平穩(wěn)特性影響預(yù)測(cè)精度問(wèn)題,該文基于黏菌算法(SMA)和變分模態(tài)分解(VMD)構(gòu)成黏菌算法優(yōu)化的變分模態(tài)分解(SMA-VMD),結(jié)合門(mén)控循環(huán)單元(GRU)組成分解-預(yù)測(cè)-重構(gòu)的變壓器油中溶解氣體含量預(yù)測(cè)模型。首先,采用差分法提取原始序列趨勢(shì)項(xiàng);然后利用SMA-VMD對(duì)剩余序列進(jìn)行分解,得到一組平穩(wěn)的模態(tài)分量;之后通過(guò)GRU對(duì)分解所得各模態(tài)分量分別進(jìn)行預(yù)測(cè);最后對(duì)預(yù)測(cè)結(jié)果進(jìn)行重構(gòu)。該研究通過(guò)對(duì)變壓器油中溶解氣體H2進(jìn)行仿真實(shí)驗(yàn),并與另外五種預(yù)測(cè)模型對(duì)比,得出SMA-VMD-GRU模型預(yù)測(cè)結(jié)果平均絕對(duì)百分比誤差為0.36%,方均根誤差為1.76mL/L,有效地提高了變壓器油中溶解氣體含量含量預(yù)測(cè)精度。通過(guò)對(duì)變壓器油中溶解氣體成分CH4、CO、總烴進(jìn)行仿真實(shí)驗(yàn),證明了該研究所提預(yù)測(cè)模型的有效性。
差分法 黏菌算法 變分模態(tài)分解 油中溶解氣體預(yù)測(cè) 門(mén)控循環(huán)單元
變壓器作為電力系統(tǒng)的關(guān)鍵樞紐設(shè)備,其安全穩(wěn)定運(yùn)行是優(yōu)質(zhì)電能正常供應(yīng)的必要基礎(chǔ)[1]。變壓器主要采用油紙絕緣結(jié)構(gòu),在變壓器運(yùn)行過(guò)程中,油紙絕緣材料受熱應(yīng)力、電應(yīng)力、催化劑等因素的影響,會(huì)逐漸老化分解,在油中產(chǎn)生少量的溶解氣體(H2、CH4、C2H6、C2H2、C2H4、CO、CO2等)[2]。油中溶解氣體分析(Dissolved Gas Analysis, DGA)技術(shù)依據(jù)氣體組分含量、比值、相對(duì)百分比的變化,能夠有效地發(fā)現(xiàn)變壓器內(nèi)部潛在故障,追蹤故障發(fā)展趨勢(shì)[2-5]。因此,油中溶解氣體含量預(yù)測(cè)有著重要的意義[6-7]。
變壓器油中溶解氣體所測(cè)數(shù)據(jù)一般為時(shí)間序列[8]。目前,變壓器油中溶解氣體預(yù)測(cè)研究方法一般有統(tǒng)計(jì)預(yù)測(cè)與機(jī)器學(xué)習(xí)的智能預(yù)測(cè)。統(tǒng)計(jì)預(yù)測(cè)主要為灰色理論[8]和卡爾曼濾波[9]等預(yù)測(cè)方法,這類預(yù)測(cè)模型所得預(yù)測(cè)結(jié)果對(duì)實(shí)驗(yàn)數(shù)據(jù)分布依賴性較大,即非線性擬合能力有限。機(jī)器學(xué)習(xí)智能預(yù)測(cè)主要是指利用機(jī)器學(xué)習(xí)方法構(gòu)建預(yù)測(cè)模型,包括支持向量機(jī)[10]、隨機(jī)森林[11]等。機(jī)器學(xué)習(xí)在處理非線性問(wèn)題時(shí)有著明顯的優(yōu)勢(shì)[9,11],但是傳統(tǒng)的機(jī)器學(xué)習(xí)模型在預(yù)測(cè)過(guò)程中沒(méi)有考慮其數(shù)據(jù)在時(shí)間上的聯(lián)系,因此對(duì)時(shí)間序列數(shù)據(jù)預(yù)測(cè)存在一定的不足。近些年,由于人工智能飛速發(fā)展,相應(yīng)的深度學(xué)習(xí)模型也被很多國(guó)內(nèi)外學(xué)者應(yīng)用到變壓器油中溶解氣體含量預(yù)測(cè)中。文獻(xiàn)[12]利用深度學(xué)習(xí)模型中的長(zhǎng)短期記憶神經(jīng)網(wǎng)絡(luò)(Long Short-Term Memory, LSTM)模型對(duì)變壓器油中溶解氣體含量進(jìn)行預(yù)測(cè),且從實(shí)驗(yàn)中可以看出在變壓器溶解氣體含量預(yù)測(cè)方面,深度學(xué)習(xí)模型預(yù)測(cè)效果相較于傳統(tǒng)機(jī)器學(xué)習(xí)預(yù)測(cè)結(jié)果精度更高,且能更準(zhǔn)確預(yù)測(cè)變化趨勢(shì)。文獻(xiàn)[13]利用深度信念網(wǎng)絡(luò)(Deep Recurrent BeliefNetwork, DRBN)模型對(duì)油中溶解氣體進(jìn)行預(yù)測(cè),很好地解決了時(shí)序數(shù)據(jù)時(shí)移特征所帶來(lái)的預(yù)測(cè)誤差問(wèn)題。文獻(xiàn)[14]利用元啟式優(yōu)化算法與深度神經(jīng)網(wǎng)絡(luò)相結(jié)合的思想,提出基于粒子群優(yōu)化長(zhǎng)短期記憶網(wǎng)絡(luò)的組合預(yù)測(cè)模型,有效地解決了由于LSTM在預(yù)測(cè)過(guò)程中因參數(shù)較多而無(wú)法確定其最優(yōu)參數(shù)所導(dǎo)致的預(yù)測(cè)精度誤差較大的問(wèn)題。文獻(xiàn)[15]考慮到在預(yù)測(cè)油中溶解氣體含量時(shí)其他變量帶來(lái)的影響,提出在預(yù)測(cè)模型中添加注意力機(jī)制的方法,有效地提高了預(yù)測(cè)精度。文獻(xiàn)[16]在參數(shù)優(yōu)化的預(yù)測(cè)模型中添加注意力機(jī)制,提出的預(yù)測(cè)模型有效解決了環(huán)境溫度、濕度及其他溶解氣體等多變量影響預(yù)測(cè)精度的問(wèn)題。以上變壓器油中溶解氣體預(yù)測(cè)方法有效地提高了氣體含量預(yù)測(cè)精度,但以增加多變量注意力機(jī)制權(quán)重來(lái)提高氣體預(yù)測(cè)精度,對(duì)多變量歷史數(shù)據(jù)采集的準(zhǔn)確性要求很高,沒(méi)有考慮到當(dāng)多變量歷史數(shù)據(jù)采集出現(xiàn)誤差時(shí),反而會(huì)給變壓器油中溶解氣體含量預(yù)測(cè)精度帶來(lái)誤差,且不可能將所有的變量都考慮到。另外,變壓器油中溶解氣體序列由于其他變量及自身歷史含量的影響,存在非線性、非平穩(wěn)特性,上述方法沒(méi)有考慮其對(duì)模型預(yù)測(cè)精度產(chǎn)生的影響,使得上述方法在進(jìn)一步提升油中溶解氣體含量預(yù)測(cè)精度上受到限制。
因此,既需要考慮多變量對(duì)預(yù)測(cè)精度的影響效果,也需要對(duì)氣體含量序列進(jìn)行平穩(wěn)化處理。通過(guò)對(duì)原始序列進(jìn)行分解,可以將多變量對(duì)氣體含量變化產(chǎn)生的影響及自身含量變化規(guī)律等多種成分分解為一組規(guī)律性較強(qiáng)的平穩(wěn)序列。目前對(duì)序列分解的手段多種多樣,例如文獻(xiàn)[17]中的小波分解、文獻(xiàn)[18]中的經(jīng)驗(yàn)?zāi)B(tài)分解(Empirical Mode Decomposition, EMD)、文獻(xiàn)[19]中的變分模態(tài)分解(Variational Mode Decomposition, VMD)等。其中,小波分解方法由于較難確定分解層數(shù)和小波基,從而導(dǎo)致模型效果的不確定性。經(jīng)驗(yàn)?zāi)B(tài)分解往往會(huì)出現(xiàn)嚴(yán)重的模態(tài)混疊而導(dǎo)致分解效果不佳,影響預(yù)測(cè)精度[20]。文獻(xiàn)[21]是在EMD分解的基礎(chǔ)上增加白噪聲來(lái)抑制其模態(tài)混疊,但是會(huì)造成模型計(jì)算量大且所得模態(tài)分量超過(guò)真實(shí)序列量。文獻(xiàn)[19]中的VMD方法是一種自適應(yīng)濾波器,能夠解決EMD在分解過(guò)程中所存在的問(wèn)題,但VMD在處理同時(shí)包含趨勢(shì)項(xiàng)與波動(dòng)項(xiàng)的序列時(shí),會(huì)將趨勢(shì)項(xiàng)誤分解為周期項(xiàng),從而影響預(yù)測(cè)精度,并且其參數(shù)的確定對(duì)分解結(jié)果至關(guān)重要[19]。
鑒于上述分析,本文首先采用差分法[22]對(duì)油中溶解氣體序列進(jìn)行提取趨勢(shì)預(yù)處理;其次,基于黏菌算法(Slime Mold Algorithm, SMA)與VMD方法構(gòu)造SMA-VMD序列分解模型,將去趨勢(shì)化預(yù)處理后的油中溶解氣體序列分解為平穩(wěn)、周期性強(qiáng)的模態(tài)分量;再次,對(duì)分解所得各模態(tài)分量分別建立預(yù)測(cè)模型,由于門(mén)控循環(huán)單元(Gate Recurrent Unit, GRU)相較于LSTM具有參數(shù)量少、訓(xùn)練速度快、預(yù)測(cè)精度高等優(yōu)點(diǎn)[23-24],所以本研究以GRU為變壓器油中溶解氣體含量預(yù)測(cè)模型,并基于以上方法,提出了基于SMA-VMD-GRU的油中溶解氣體含量組合預(yù)測(cè)模型;最后,分別與LSTM、GRU、EMD-LSTM、EMD-GRU、SMA-VMD-LSTM五種預(yù)測(cè)模型進(jìn)行對(duì)比仿真實(shí)驗(yàn)[18,25-26],驗(yàn)證SMA-VMD-GRU模型的有效性。
變分模態(tài)分解的實(shí)質(zhì)是將輸入序列分解為個(gè)具有中心頻率的有限帶寬,主要包括構(gòu)造變分問(wèn)題及其變分問(wèn)題的求解[27-28]。
1.1.1 變分問(wèn)題的構(gòu)造
1.1.2 變分問(wèn)題的求解
(5)對(duì)于給定判定精度>0,有
(6)滿足式(6)則停止迭代,否則返回步驟(2)。
黏菌算法是Li Shimin等于2020年提出的一種元啟式優(yōu)化算法[29],其思想是根據(jù)黏菌多頭絨泡菌在尋找食物過(guò)程中發(fā)生的一連串的動(dòng)作和身體變化來(lái)構(gòu)建數(shù)學(xué)模型。從文獻(xiàn)[30]可以看出該算法在解決實(shí)際尋找最優(yōu)解問(wèn)題時(shí)能夠快速收斂并求解。因?yàn)轲ぞ惴ㄏ噍^于常用的算法優(yōu)化算法(Arithmetic Opthmetic Algorithm, AOA)、海洋捕食者算法(Marine Predators Algorithm, MPA)、樽海鞘優(yōu)化算法(Salp Swarm Algorithm, SSA)等具有更優(yōu)的精度和穩(wěn)定性,尋優(yōu)能力強(qiáng)[30],所以本研究選擇SMA對(duì)VMD進(jìn)行優(yōu)化。
神經(jīng)網(wǎng)絡(luò)模型由許多神經(jīng)元組成,GRU神經(jīng)元的內(nèi)部結(jié)構(gòu)如圖1所示。
圖1 GRU神經(jīng)元內(nèi)部結(jié)構(gòu)
GRU神經(jīng)元由以下幾個(gè)部分組成:
(3)一個(gè)狀態(tài),即時(shí)刻神經(jīng)元的內(nèi)部狀態(tài),可以表達(dá)為
(4)一個(gè)輸出,即為最終輸出,可以表達(dá)為
文獻(xiàn)[31]使用控制變量法討論了VMD參數(shù)選取對(duì)分解結(jié)果的影響,說(shuō)明了分解數(shù)與懲罰因子對(duì)分解結(jié)果有較大的影響,因此,有必要研究與的優(yōu)化取值方法。其他參數(shù)則選取經(jīng)驗(yàn)值。
本文采用信號(hào)排列熵[31-32]與能量追蹤法[33]作為適應(yīng)度函數(shù)的構(gòu)造原理。排列熵可以很好地反映時(shí)間序列的變化規(guī)律,而且序列排列熵能很好地評(píng)估時(shí)間序列的復(fù)雜程度;而能量追蹤法能夠很好地體現(xiàn)各子序列與原始時(shí)間序列的關(guān)系。
對(duì)各固有模態(tài)函數(shù)(Intrinsic Mode Function, IMF)分量進(jìn)行相空間重構(gòu)得到矩陣。構(gòu)造向量為
此時(shí)矩陣每一行均可得到一組符號(hào)序列矩陣,有
SMA對(duì)VMD進(jìn)行參數(shù)尋優(yōu)的步驟見(jiàn)文獻(xiàn)[29],尋優(yōu)流程如圖2所示。
圖2 SMA尋優(yōu)流程
SMA-VMD-GRU預(yù)測(cè)模型流程如圖3所示。
圖3 SMA-VMD-GRU預(yù)測(cè)模型流程
本文油中溶解氣體預(yù)測(cè)模型分為四個(gè)步驟:
(1)首先,對(duì)原始序列進(jìn)行去除病態(tài)值等預(yù)處理,處理方法見(jiàn)文獻(xiàn)[32,35-36];其次,使用差分法[28]對(duì)序列進(jìn)行去趨勢(shì)化處理。
(2)使用SMA對(duì)VMD進(jìn)行參數(shù)自適應(yīng)尋優(yōu),接著使用最優(yōu)化VMD對(duì)去趨勢(shì)序列進(jìn)行分解得到頻段不同的各模態(tài)分量。
(3)歸一化分解數(shù)據(jù)后,對(duì)各分量分別建立GRU預(yù)測(cè)模型,并選用Adam優(yōu)化器更新網(wǎng)絡(luò)參數(shù)。
(4)反歸一化預(yù)測(cè)分量,重構(gòu)各分量預(yù)測(cè)結(jié)果,輸出最終預(yù)測(cè)結(jié)果。
本文選用重慶市某220kV在運(yùn)變壓器2020年7月23日~2021年10月5日在線采集的450組溶解氣體時(shí)序數(shù)據(jù)作為實(shí)驗(yàn)數(shù)據(jù),時(shí)序數(shù)據(jù)為等間隔采樣,采樣周期為1天。其中,將單一氣體含量前390天數(shù)據(jù)作為預(yù)測(cè)模型訓(xùn)練集,后60天數(shù)據(jù)作為預(yù)測(cè)模型測(cè)試集,即訓(xùn)練好的模型使用前10天歷史數(shù)據(jù)預(yù)測(cè)第11天氣體含量,再結(jié)合第2天~第10天數(shù)據(jù)與上一步預(yù)測(cè)所得的第11天數(shù)據(jù)來(lái)預(yù)測(cè)第12天的含量,依此類推,將會(huì)得到第11天~第60天的預(yù)測(cè)數(shù)據(jù),即未來(lái)50天氣體含量預(yù)測(cè)數(shù)據(jù)。選用Keras框架構(gòu)建預(yù)測(cè)模型,對(duì)比實(shí)驗(yàn)預(yù)測(cè)模型與本文所提SMA-VMD-GRU預(yù)測(cè)模型均設(shè)計(jì)為80-100-1雙隱藏層結(jié)構(gòu),即雙隱藏層神經(jīng)元分別設(shè)定為80個(gè)與100個(gè),并包含1個(gè)輸出。訓(xùn)練過(guò)程中選用Adam優(yōu)化更新模型參數(shù),迭代周期設(shè)定為100,學(xué)習(xí)率設(shè)定為0.001。本文以H2含量為例對(duì)所提模型進(jìn)行對(duì)比分析。H2含量序列曲線如圖4所示。
圖4 H2含量序列曲線
首先,直接利用差分法對(duì)序列進(jìn)行去趨勢(shì)化處理,圖4a為含趨勢(shì)項(xiàng)原始H2含量序列,圖4b為去趨勢(shì)后H2含量變化曲線。
其次,將SMA優(yōu)化算法初始種群數(shù)量設(shè)為30,迭代次數(shù)為50次,進(jìn)而得到最小適應(yīng)度值EW為0.265 963,最優(yōu)值為7,值為1 572。其所對(duì)應(yīng)的SMA迭代曲線與最優(yōu)值所對(duì)應(yīng)的分解模態(tài)中心頻率分別如圖5a和圖5b所示。由圖5b可以看出,使用最優(yōu)參數(shù)分解H2序列得到的各模態(tài)分量中心頻率沒(méi)有出現(xiàn)混疊現(xiàn)象。
圖5 SMA迭代曲線與最優(yōu)分解中心頻率
最后,將最優(yōu)分解后的子序列進(jìn)行預(yù)測(cè)。
使用GRU和LSTM方法的兩組單一預(yù)測(cè)模型預(yù)測(cè)結(jié)果如圖6所示,預(yù)測(cè)誤差見(jiàn)表1。由表1可得,GRU模型的預(yù)測(cè)結(jié)果MAPE為2.03%,RMSE為43.30mL/L,相較于LSTM模型分別降低了0.30%和6.66mL/L。結(jié)合圖6可以得出,GRU模型在解決變壓器油中溶解氣體含量預(yù)測(cè)問(wèn)題時(shí)相較于LSTM預(yù)測(cè)模型更優(yōu)。
圖6 單一模型預(yù)測(cè)結(jié)果對(duì)比
表1 GRU與LSTM預(yù)測(cè)誤差對(duì)比
Tab.1 Comparison of GRU and LSTM prediction errors
LSTM和GRU模型的網(wǎng)絡(luò)總參數(shù)、可訓(xùn)練參數(shù)及運(yùn)行時(shí)間見(jiàn)表2。GRU預(yù)測(cè)模型總參數(shù)和可訓(xùn)練參數(shù)都為74 621,相比于LSTM預(yù)測(cè)模型參數(shù)量減少了24120,在運(yùn)行時(shí)間上可以看出GRU模型運(yùn)行時(shí)間為9.71s,而LSTM模型為11.18s?;诖?,從考慮變壓器油中溶解氣體含量預(yù)測(cè)時(shí)效性角度出發(fā),GRU也更加具有優(yōu)勢(shì)。
表2 GRU與LSTM性能對(duì)比
Tab.2 Performance comparison between GRU and LSTM
EMD對(duì)序列分解后的效果如圖7所示,SMA-VMD對(duì)序列分解后的效果如圖8所示。其中圖7a和圖8a為所對(duì)應(yīng)的IMF,圖7b和圖8b代表各自的IMF頻譜圖。圖9為GRU對(duì)SMA-VMD分解分量預(yù)測(cè)效果圖。本文將LSTM、GRU分別與EMD、SMA-VMD結(jié)合,組成四種預(yù)測(cè)模型,對(duì)H2含量的預(yù)測(cè)結(jié)果如圖10所示。
圖7 EMD分解分量與頻譜
圖9 SMA-VMD-GRU分量預(yù)測(cè)結(jié)果
圖10 組合模型預(yù)測(cè)結(jié)果對(duì)比
組合預(yù)測(cè)模型對(duì)油中H2含量的預(yù)測(cè)誤差見(jiàn)表3。表1中LSTM與GRU預(yù)測(cè)結(jié)果MAPE分別為2.33%與2.03%,而表3中EMD-LSTM組合預(yù)測(cè)模型的MAPE最大為1.35%。再結(jié)合圖10和圖6可以明顯看出,含有模態(tài)分解的變壓器油中溶解氣體含量組合預(yù)測(cè)模型相比單一預(yù)測(cè)模型對(duì)變壓器油中溶解氣體H2預(yù)測(cè)效果更加貼近真實(shí)值。
表3 組合預(yù)測(cè)模型預(yù)測(cè)誤差對(duì)比
Tab.3 Comparison of combined prediction models
由圖7a與圖8a可以看出,通過(guò)模態(tài)分解的手段將原始含量序列分解為平穩(wěn)的低頻高幅成分與規(guī)律性較強(qiáng)的高頻低幅成分,很好地將變壓器油中溶解氣體序列中所包含的環(huán)境溫度、濕度等多變量影響因素及自身變化規(guī)律信息進(jìn)行了提取,有效地解決了因其他變量采集誤差影響變壓器油中溶解氣體預(yù)測(cè)精度的問(wèn)題。
由圖7b中可以看出EMD分解出現(xiàn)了嚴(yán)重的模態(tài)混疊,而從圖8b可看出SMA-VMD分解得到的各IMF基本沒(méi)有出現(xiàn)頻譜混疊現(xiàn)象。此外,從圖8a中后3個(gè)IMF分量看出分解得到的高頻分量的變化更加規(guī)律??梢缘贸?,SMA-VMD相較于EMD性能較優(yōu),并且從圖9分量預(yù)測(cè)結(jié)果圖可以看出GRU能對(duì)平穩(wěn)的低頻高幅分量和規(guī)律性較強(qiáng)的高頻低幅分量進(jìn)行較好的預(yù)測(cè),解決了變壓器油中溶解氣體含量非線性、非平穩(wěn)特性以及多變量影響預(yù)測(cè)精度問(wèn)題。
表3則定量地評(píng)價(jià)了四種組合預(yù)測(cè)模型,EMD-LSTM和EMD-GRU兩組基于EMD的組合預(yù)測(cè)模型的MAPE分別為1.35%與1.05%,相較于SMA-VMD-LSTM和SMA-VMD-GRU兩組基于VMD的組合預(yù)測(cè)模型的MAPE分別增加0.58%與0.69%。其中,SMA-VMD-GRU組合預(yù)測(cè)模型的MAPE與RMSE分別為0.36%與1.76mL/L,相較于其他三組組合預(yù)測(cè)模型預(yù)測(cè)誤差最小。根據(jù)圖10的最終預(yù)測(cè)結(jié)果所示,SMA-VMD-GRU模型在預(yù)測(cè)含有非線性、非平穩(wěn)特性的變壓器油中溶解氣體含量時(shí),相較于未含模態(tài)分解的模型與分解較差模型都有較優(yōu)的擬合效果。
本文以3.2節(jié)所使用變壓器中另外三種油中溶解氣體CH4、CO、總烴含量為例,驗(yàn)證所提SMA-VMD-GRU組合預(yù)測(cè)模型的有效性。驗(yàn)證實(shí)驗(yàn)中訓(xùn)練集與測(cè)試集劃分、模型構(gòu)建框架、模型結(jié)構(gòu)設(shè)計(jì)、參數(shù)更新方法、學(xué)習(xí)率設(shè)定均與3.2節(jié)中H2實(shí)驗(yàn)相同。
CH4、CO、總烴含量原始序列曲線如圖11所示,并對(duì)去趨勢(shì)預(yù)處理后的序列進(jìn)行分解。
圖11 三種氣體原始序列曲線
首先,利用SMA優(yōu)化算法對(duì)三種序列進(jìn)行VMD分解參數(shù)尋優(yōu),SMA初始種群數(shù)量設(shè)為30,迭代次數(shù)為50次,進(jìn)而得到最小適應(yīng)度值EW、最優(yōu)值、值見(jiàn)表4。三種序列尋優(yōu)所對(duì)應(yīng)的SMA迭代曲線與最優(yōu)值所對(duì)應(yīng)分解模態(tài)中心頻率分別如圖12、圖13所示。由圖13可以看出,使用最優(yōu)參數(shù)分解CH4、CO、總烴三種油中溶解氣體含量序列得到的各模態(tài)分量中心頻率沒(méi)有出現(xiàn)混疊現(xiàn)象。
表4 SMA-GRU尋優(yōu)參數(shù)
Tab.4 Optimization parameters of SMA-GRU
圖12 三種氣體SMA迭代曲線
CH4、CO和總烴含量分解分量預(yù)測(cè)結(jié)果分別如圖14、圖15和圖16所示。
圖13 三種氣體最優(yōu)SMA-VMD分解中心頻率
圖14a、圖15a、圖16a分別為CH4、CO、總烴序列最優(yōu)SMA-VMD分解圖。可以看出,相較于圖11中含有非線性、非平穩(wěn)特性的三種油中溶解氣體原始序列,三組分解序列已平穩(wěn)化且周期性較強(qiáng)。
圖14 CH4最優(yōu)分解與分量預(yù)測(cè)
圖14c、圖15c、圖16c分別為三組分解分量預(yù)測(cè)結(jié)果擬合圖。從預(yù)測(cè)效果可以看出,對(duì)于平穩(wěn)化且周期性較強(qiáng)的序列,預(yù)測(cè)效果較佳,證明了分解得到的平穩(wěn)化序列的可預(yù)測(cè)性。
綜上所述,證明了本文中所提預(yù)測(cè)模型分解部分對(duì)油中溶解氣體含量序列的有效性。
三種氣體分解序列預(yù)測(cè)結(jié)果重構(gòu)后得到的最終預(yù)測(cè)結(jié)果如圖17所示,預(yù)測(cè)誤差見(jiàn)表5。表5中三種氣體預(yù)測(cè)的MAPE僅為0.29%、0.15%、4.99%,RMSE僅為0.02mL/L、1.13mL/L、0.50mL/L,結(jié)合圖17,證明了本研究所提SMA-VMD-GRU組合預(yù)測(cè)模型對(duì)變壓器油中溶解氣體含量預(yù)測(cè)的有效性。
表5 SMA-VMD-GRU組合預(yù)測(cè)模型預(yù)測(cè)誤差
Tab.5 Prediction accuracy of SMA-VMD-GRU combined prediction mode
本文針對(duì)變壓器油中溶解氣體含量序列非線性、非平穩(wěn)特性影響預(yù)測(cè)精度問(wèn)題,提出了一種SMA-VMD結(jié)合GRU的變壓器油中溶解氣體組合預(yù)測(cè)模型,實(shí)驗(yàn)結(jié)果表明:
1)使用差分法對(duì)序列趨勢(shì)項(xiàng)提取,有效地解決了VMD無(wú)法準(zhǔn)確提取趨勢(shì)項(xiàng)的不足。再通過(guò)SMA優(yōu)化后的VMD分解處理后的序列,可以將復(fù)雜的油中溶解氣體序列分解,成為一組平穩(wěn)的、周期性強(qiáng)的模態(tài)分量,有效地解決了原始序列非線性、非平穩(wěn)特性對(duì)預(yù)測(cè)精度的影響。
2)在變壓器油中溶解氣體預(yù)測(cè)中,GRU網(wǎng)絡(luò)相比LSTM網(wǎng)絡(luò)收斂速度更快。因此,GRU網(wǎng)絡(luò)比LSTM更加具有優(yōu)勢(shì)。在前期通過(guò)差分法與SMA-VMD提升序列可預(yù)測(cè)性的前提下,進(jìn)一步將油中溶解氣體含量預(yù)測(cè)精度提高,有助于變壓器前期故障診斷。
3)通過(guò)對(duì)變壓器油中溶解氣體中多種氣體含量進(jìn)行仿真預(yù)測(cè)實(shí)驗(yàn),證明了本研究所提基于SMA-VMD-GRU的變壓器油中溶解氣體含量預(yù)測(cè)模型的有效性。
[1] 梁得亮, 柳軼彬, 寇鵬, 等. 智能配電變壓器發(fā)展趨勢(shì)分析[J]. 電力系統(tǒng)自動(dòng)化, 2020, 44(7): 1-14.
Liang Deliang, Liu Yibin, Kou Peng, et al. Analysis of development trend for intelligent distribution transformer[J]. Automation of Electric Power Systems, 2020, 44(7): 1-14.
[2] 李恩文, 王力農(nóng), 宋斌, 等. 基于改進(jìn)模糊聚類算法的變壓器油色譜分析[J]. 電工技術(shù)學(xué)報(bào), 2018, 33(19): 4594-4602.
Li Enwen, Wang Linong, Song Bin, et al. Analysis of transformer oil chromatography based on improved fuzzy clustering algorithm[J]. Transactions of China Electrotechnical Society, 2018, 33(19): 4594-4602.
[3] 國(guó)家能源局. DL/T 573—2021 電力變壓器檢修導(dǎo)則[S]. 北京: 中國(guó)電力出版社, 2021.
[4] 國(guó)家能源局. DL/T 722—2014 變壓器油中溶解氣體分析和判斷導(dǎo)則[S]. 北京: 中國(guó)電力出版社, 2015.
[5] 張燕, 方瑞明. 基于油中溶解氣體動(dòng)態(tài)網(wǎng)絡(luò)標(biāo)志物模型的變壓器缺陷預(yù)警與辨識(shí)[J]. 電工技術(shù)學(xué)報(bào), 2020, 35(9): 2032-2041.
Zhang Yan, Fang Ruiming. Fault detection and identification of transformer based on dynamical network marker model of dissolved gas in oil[J]. Transactions of China Electrotechnical Society, 2020, 35(9): 2032-2041.
[6] 江秀臣, 盛戈皞. 電力設(shè)備狀態(tài)大數(shù)據(jù)分析的研究和應(yīng)用[J]. 高電壓技術(shù), 2018, 44(4): 1041-1050.
Jiang Xiuchen, Sheng Gehao. Research and application of big data analysis of power equipment condition[J]. High Voltage Engineering, 2018, 44(4): 1041-1050.
[7] 蒲天驕, 喬驥, 韓笑, 等. 人工智能技術(shù)在電力設(shè)備運(yùn)維檢修中的研究及應(yīng)用[J]. 高電壓技術(shù), 2020, 46(2): 369-383.
Pu Tianjiao, Qiao Ji, Han Xiao, et al. Research and application of artificial intelligence in operation and maintenance for power equipment[J]. High Voltage Engineering, 2020, 46(2): 369-383.
[8] Liu Chang, Zhang Hongzhi, Xie Zhicheng, et al. Combined forecasting method of dissolved gases concentration and its application in condition-based maintenance[J]. IEEE Transactions on Power Delivery, 2019, 34(4): 1269-1279.
[9] 修春波, 任曉, 李艷晴, 等. 基于卡爾曼濾波的風(fēng)速序列短期預(yù)測(cè)方法[J]. 電工技術(shù)學(xué)報(bào), 2014, 29(2): 253-259.
Xiu Chunbo, Ren Xiao, Li Yanqing, et al. Short-term prediction method of wind speed series based on Kalman filtering fusion[J]. Transactions of China Electrotechnical Society, 2014, 29(2): 253-259.
[10] 張婷婷, 于明, 李賓, 等. 基于Wavelet降噪和支持向量機(jī)的鋰離子電池容量預(yù)測(cè)研究[J]. 電工技術(shù)學(xué)報(bào), 2020, 35(14): 3126-3136.
Zhang Tingting, Yu Ming, Li Bin, et al. Capacity prediction of lithium-ion batteries based on Wavelet noise reduction and support vector machine[J]. Transactions of China Electrotechnical Society, 2020, 35(14): 3126-3136.
[11] 徐肖偉, 李鶴健, 于虹, 等. 基于隨機(jī)森林的變壓器油中溶解氣體濃度預(yù)測(cè)[J]. 電子測(cè)量技術(shù), 2020, 43(3): 66-70.
Xu Xiaowei, Li Hejian, Yu Hong, et al. Concentration prediction of dissolved gases in transformer oil based on random forest[J]. Electronic Measurement Technology, 2020, 43(3): 66-70.
[12] 代杰杰, 宋輝, 盛戈皞, 等. 采用LSTM網(wǎng)絡(luò)的電力變壓器運(yùn)行狀態(tài)預(yù)測(cè)方法研究[J]. 高電壓技術(shù), 2018, 44(4): 1099-1106.
Dai Jiejie, Song Hui, Sheng Gehao, et al. Prediction method for power transformer running state based on LSTM network[J]. High Voltage Engineering, 2018, 44(4): 1099-1106.
[13] Qi Bo, Wang Yiming, Zhang Peng, et al. A novel deep recurrent belief network model for trend prediction of transformer DGA data[J]. IEEE Access, 2019, 7: 80069-80078.
[14] 劉可真, 茍家萁, 駱釗, 等. 基于粒子群優(yōu)化-長(zhǎng)短期記憶網(wǎng)絡(luò)模型的變壓器油中溶解氣體濃度預(yù)測(cè)方法[J]. 電網(wǎng)技術(shù), 2020, 44(7): 2778-2785.
Liu Kezhen, Gou Jiaqi, Luo Zhao, et al. Prediction of dissolved gas concentration in transformer oil based on PSO-LSTM model[J]. Power System Technology, 2020, 44(7): 2778-2785.
[15] 崔宇, 侯慧娟, 胥明凱, 等. 基于雙重注意力機(jī)制的變壓器油中溶解氣體預(yù)測(cè)模型[J]. 中國(guó)電機(jī)工程學(xué)報(bào), 2020, 40(1): 338-347, 400.
Cui Yu, Hou Huijuan, Xu Mingkai, et al. A prediction method for dissolved gas in power transformer oil based on dual-stage attention mechanism[J]. Proceedings of the CSEE, 2020, 40(1): 338-347, 400.
[16] 劉展程, 王爽, 唐波. 基于SSA-BiGRU-Attention模型的變壓器油中溶解氣體含量預(yù)測(cè)[J]. 高電壓技術(shù), 2022, 48(8): 2972-2981.
Liu Zhancheng, Wang Shuang, Tang Bo. Prediction of dissolved gas content in transformer oil based on SSA-BiGRU-attention model[J]. High Voltage Engineering, 2022, 48(8): 2972-2981.
[17] 馬星河, 張登奎. 基于改進(jìn)經(jīng)驗(yàn)小波變換的高壓電纜局部放電噪聲抑制研究[J]. 電工技術(shù)學(xué)報(bào), 2021, 36(增刊1): 353-361.
Ma Xinghe, Zhang Dengkui. Research on suppression of partial discharge noise of high voltage cable based on improved empirical wavelet transform[J]. Transactions of China Electrotechnical Society, 2021, 36(S1): 353-361.
[18] 林琳, 陳志英. 基于粗糙集神經(jīng)網(wǎng)絡(luò)和振動(dòng)信號(hào)的高壓斷路器機(jī)械故障診斷[J]. 電工技術(shù)學(xué)報(bào), 2020, 35(增刊1): 277-283.
Lin Lin, Chen Zhiying. Mechanical fault diagnosis of high voltage circuit breakers based on rough set neural networks and vibration signals[J]. Transactions of China Electrotechnical Society, 2020, 35(S1): 277-283.
[19] Dragomiretskiy K, Zosso D. Variational mode decomposition[J]. IEEE Transactions on Signal Processing, 2014, 62(3): 531-544.
[20] 葉林, 劉鵬. 基于經(jīng)驗(yàn)?zāi)B(tài)分解和支持向量機(jī)的短期風(fēng)電功率組合預(yù)測(cè)模型[J]. 中國(guó)電機(jī)工程學(xué)報(bào), 2011, 31(31): 102-108.
Ye Lin, Liu Peng. Combined model based on EMD-SVM for short-term wind power prediction[J]. Proceedings of the CSEE, 2011, 31(31): 102-108.
[21] Wu Zhaohua, Huang N E. Ensemble empirical mode decomposition: a noise-assisted data analysis method[J]. Advances in Adaptive Data Analysis, 2009, 1(1): 1-41.
[22] 曹奇, 岳東杰, 高永攀, 等. 基于非平穩(wěn)時(shí)間序列的不同趨勢(shì)項(xiàng)提取方法對(duì)比研究[J]. 大地測(cè)量與地球動(dòng)力學(xué), 2013, 33(6): 150-154.
Cao Qi, Yue Dongjie, Gao Yongpan, et al. Contrast study on various methods extracting trend extraction based on non-stationary time series[J]. Journal of Geodesy and Geodynamics, 2013, 33(6): 150-154.
[23] 楊茂, 白玉瑩. 基于多位置NWP和門(mén)控循環(huán)單元的風(fēng)電功率超短期預(yù)測(cè)[J]. 電力系統(tǒng)自動(dòng)化, 2021, 45(1): 177-183.
Yang Mao, Bai Yuying. Ultra-short-term prediction of wind power based on multi-location numerical weather prediction and gated recurrent unit[J]. Automation of Electric Power Systems, 2021, 45(1): 177-183.
[24] 李超然, 肖飛, 樊亞翔, 等. 基于門(mén)控循環(huán)單元神經(jīng)網(wǎng)絡(luò)和Huber-M估計(jì)魯棒卡爾曼濾波融合方法的鋰離子電池荷電狀態(tài)估算方法[J]. 電工技術(shù)學(xué)報(bào), 2020, 35(9): 2051-2062.
Li Chaoran, Xiao Fei, Fan Yaxiang, et al. A hybrid approach to lithium-ion battery SOC estimation based on recurrent neural network with gated recurrent unit and Huber-M robust Kalman filter[J]. Transactions of China Electrotechnical Society, 2020, 35(9): 2051-2062.
[25] 王科, 茍家萁, 彭晶, 等. 基于LSTM網(wǎng)絡(luò)的變壓器油中溶解氣體濃度預(yù)測(cè)[J]. 電子測(cè)量技術(shù), 2020, 43(4): 81-87.
Wang Ke, Gou Jiaqi, Peng Jing, et al. Prediction of dissolved gas concentration in transformer oil based on LSTM network[J]. Electronic Measurement Technology, 2020, 43(4): 81-87.
[26] 衛(wèi)永鵬, 蘇益輝, 王勝利, 等. 基于改進(jìn)門(mén)控循環(huán)單元的變壓器油中氣體濃度預(yù)測(cè)[J]. 電氣技術(shù), 2022, 23(2): 55-60.
Wei Yongpeng, Su Yihui, Wang Shengli, et al. Prediction of gas concentration in transformer oil based on improved gated recurrent unit[J]. Electrical Engineering, 2022, 23(2): 55-60.
[27] Ali M, Khan A, Rehman N U. Hybrid multiscale wind speed forecasting based on variational mode decomposition[J]. International Transactions on Electrical Energy Systems, 2018, 28(1): e2466.
[28] 陳強(qiáng)偉, 蔡文皓, 牛春光, 等. 基于VMD的APF諧波檢測(cè)算法[J]. 電力科學(xué)與技術(shù)學(xué)報(bào), 2018, 33(1): 120-124.
Chen Qiangwei, Cai Wenhao, Niu Chunguang, et al. A APF harmonics detection method based on VMD[J]. Journal of Electric Power Science and Technology, 2018, 33(1): 120-124.
[29] Li Shimin, Chen Huiling, Wang Mingjing, et al. Slime mould algorithm: a new method for stochastic optimization[J]. Future Generation Computer Systems, 2020, 111: 300-323.
[30] Gürses D, Bureerat S, Sait S M, et al. Comparison of the arithmetic optimization algorithm, the slime mold optimization algorithm, the marine predators algorithm, the salp swarm algorithm for real-world engineering applications[J]. Materials Testing, 2021, 63(5): 448-452.
[31] 李青, 張新燕, 馬天嬌, 等. 基于ECBO-VMD-WKELM的風(fēng)電功率超短期多步預(yù)測(cè)[J]. 電網(wǎng)技術(shù), 2021, 45(8): 3070-3080.
Li Qing, Zhang Xinyan, Ma Tianjiao, et al. Multi-step ahead ultra-short term forecasting of wind power based on ECBO-VMD-WKELM[J]. Power System Technology, 2021, 45(8): 3070-3080.
[32] 劉樹(shù)鑫, 宋健, 劉洋, 等. 交流接觸器觸頭系統(tǒng)運(yùn)動(dòng)分析及故障診斷研究[J]. 電工技術(shù)學(xué)報(bào), 2021, 36(增刊2): 477-486.
Liu Shuxin, Song Jian, Liu Yang, et al. Research on motion analysis and fault diagnosis of contact system of AC contactor[J]. Transactions of China Electrotechnical Society, 2021, 36(S2): 477-486.
[33] 李舒適, 王豐華, 耿俊秋, 等. 基于優(yōu)化VMD的高壓斷路器機(jī)械狀態(tài)檢測(cè)[J]. 電力自動(dòng)化設(shè)備, 2018, 38(11): 148-154.
Li Shushi, Wang Fenghua, Geng Junqiu, et al. Mechanical state detection of high voltage circuit breaker based on optimized VMD algorithm[J]. Electric Power Automation Equipment, 2018, 38(11): 148-154.
[34] Zhang Lijun, Zhang Bin, He Fei, et al. Impact analyzing based on new method of phase space reconstruction[C]//2013 IEEE International Conference on Mechatronics and Automation, Takamatsu, Japan, 2013: 587-592.
[35] 向玲, 鄧澤奇, 趙玥. 基于LPF-VMD和KELM的風(fēng)速多步預(yù)測(cè)模型[J]. 電網(wǎng)技術(shù), 2019, 43(12): 4461-4467.
Xiang Ling, Deng Zeqi, Zhao Yue. Multi-step wind speed prediction model based on LPF-VMD and KELM[J]. Power System Technology, 2019, 43(12): 4461-4467.
[36] He Jiangbiao, Yang Qichen, Wang Zheng. On-line fault diagnosis and fault-tolerant operation of modular multilevel converters—a comprehensive review[J]. CES Transactions on Electrical Machines and Systems, 2020, 4(4): 360-372.
Prediction of Dissolved Gas Content in Transformer Oil Based on SMA-VMD-GRU Model
Yang Tongliang1,2Hu Dong1Tang Chao1Fang Yun1Xie Jufang1,2
(1. School of Engineering and Technology Southwest University Chongqing 400715 China 2. International R&D Center for Smart Grid and New Equipment Technology Southwest University Chongqing 400715 China)
Dissolved gas analysis (DGA) in transformer oil is the most effective and convenient method for fault diagnosis of oil-immersed transformers. However, DGA only analyzes the real-time content of dissolved gases in transformer oil. Therefore, how to use effective historical data to accurately predict the content of dissolved gas in transformer oil for a period of time in the future is of great significance for transformer early fault diagnosis.
The content of dissolved gas in transformer oil is affected by external factors such as temperature and its own content, which will lead to nonlinear and non-stationary characteristics of the gas content sequence, leading to errors in the prediction accuracy. Aiming at the problem that the nonlinear and non-stationary characteristics of dissolved gas concentration series in power transformer oil affect the prediction accuracy, a prediction model of dissolved gas concentration in power transformer oil is proposed based on slime mold algorithm (SMA) to optimize the variated mode decomposition (VMD) and combined with gating cycle unit (GRU).
First, the preprocessed original sequence is detrended by the difference method. Secondly, based on the slime mold algorithm and the variational mode decomposition, a variational mode decomposition optimized by the slime mold algorithm is constructed, and the detrending sequence is decomposed into a set of stationary and regular mode components. Thirdly, the GRU with better prediction performance is used to predict the modal components obtained by decomposition. Finally, the final prediction result is obtained by superposition reconstruction.
The simulation results of 450 days historical data of an oil-carrying immersed transformer show that the absolute percentage error and root mean square error of the proposed prediction model for the H2content of dissolved gas in transformer oil in the next 50 days are 0.36% and 1.76mL/L, respectively. Compared with the prediction model composed of empirical mode decomposition (EMD) and long short-term memory neural network (LSTM), the SMA-VMD-GRU prediction model proposed in this study has the smallest error. And the same method was used to predict the dissolved gas CH4, CO and total hydrocarbon content in the same transformer oil. The absolute percentage error of the three gas prediction results was 0.29%, 0.15% and 4.99%, respectively, and the root mean square error was 0.02mL/L, 1.13mL/L and 0.50mL/L, respectively. The effectiveness of the proposed prediction model based on SMA-VMD-GRU was verified.
Through simulation analysis, the following conclusions can be drawn: ①Using the difference method to extract the sequence trend term effectively solves the deficiency of VMD that cannot accurately extract the trend term. Then, through VMD decomposition after SMA optimization, the complex dissolved gas sequence in oil can be decomposed into a group of stable and periodic mode components, which effectively solves the problem of the influence of nonlinear and non-stationary characteristics of the original sequence on the prediction accuracy. ②In the prediction of dissolved gas in transformer oil, the GRU network converges faster than the LSTM network. Therefore, GRU network has more advantages than LSTM. On the premise that the differential method and VMD lifting sequence can be predicted in the early stage, the prediction accuracy of dissolved gas concentration in oil is further improved, which is helpful to the early fault diagnosis of transformers. ③The effectiveness of the prediction model of dissolved gas content in transformer oil based on SMA-VMD-GRU is proved by simulation and prediction experiments of various gas concentrations in dissolved gas in transformer oil.
Difference method, slime mold algorithm, variational modal decomposition, dissolved gas in oil prediction, gate recurrent unit
10.19595/j.cnki.1000-6753.tces.221085
TM411
楊童亮 男,1996年生,碩士研究生,研究方向?yàn)殡娏υO(shè)備狀態(tài)監(jiān)測(cè)與故障診斷。E-mail:Yang__TL@163.com
謝菊芳 女,1975年生,博士,副教授,碩士生導(dǎo)師,研究方向?yàn)殡娏ο到y(tǒng)自動(dòng)化、電力設(shè)備狀態(tài)監(jiān)測(cè)與故障診斷。E-mail:375464831@qq.com(通信作者)
國(guó)家自然科學(xué)基金資助項(xiàng)目(51977179)。
2022-06-10
2022-08-11
(編輯 李冰)