一種基于SAE-RF算法的配電變壓器故障診斷方法
陳錦鋒1,張軍財1,盧思佳2,高偉2,范賢盛1,陳致遠3
(1 國網(wǎng)福建南平供電公司,福建 南平 353000;2 福州大學 電氣工程與自動化學院,福建 福州 350108;
3 上海宏力達信息技術(shù)股份有限公司,上海 200030)
摘 要:為有效解決配電變壓器故障診斷中面臨的數(shù)據(jù)特征人工提取、機器學習調(diào)參困難等問題,提出了一種基于堆棧自編碼器(SAE)和隨機森林(RF)組合的配電變壓器故障診斷方法。建立SAE配電變壓器故障特征自動挖掘模型,利用大量的無標簽數(shù)據(jù)對SAE模型中的每一個自編碼器進行逐層無監(jiān)督訓練,通過貝葉斯優(yōu)化算法自動選擇模型的最優(yōu)參數(shù);通過有標簽數(shù)據(jù)對模型參數(shù)進行有監(jiān)督細調(diào),挖掘出能夠代表各種故障本質(zhì)屬性的特征量;創(chuàng)建一個RF分類器對故障類型進行辨識,調(diào)參過程同樣實現(xiàn)參數(shù)的自動尋優(yōu)。試驗結(jié)果表明,所提方法對配電變壓器故障診斷準確率達96.67%,顯著優(yōu)于單獨使用SAE和RF的分類結(jié)果。
關(guān)鍵詞:配電變壓器;故障診斷;堆棧自編碼器;隨機森林;貝葉斯優(yōu)化
中圖分類號:TM407;TM421 文獻標識碼:A 文章編號:1007-3175(2021)02-0017-07
A Novel Fault Diagnosis Method for Distribution Transformer Via Automatic
Feature Mining and Automatic Parameter Optimization
CHEN Jin-feng1, ZHANG Jun-cai1, LU Si-jia2, GAO Wei2, FAN Xian-sheng1, CHEN Zhi-yuan3
(1 State Grid Nanping Power Supply Company, Nanping 353000, China;
2 College of Electrical Engineering and Automation, Fuzhou University, Fuzhou 350108, China;
3 Shanghai Holystar Information Technology Co., Ltd, Shanghai 200030, China)
Abstract: In order to effectively solve the problems of manual extraction of data features and difficulty of machine learning parameter adjustment in distribution transformer fault diagnosis, a fault diagnosis method for distribution transformer via the combination of stacked autoencoder (SAE) and random forest (RF) is proposed. First, a SAE model is established to realize automatic mining of distribution transformer fault characteristics, and a large number of unlabeled data is used to perform layer-by-layer unsupervised training of each auto-encoder in the model. After that, the optimal parameters of the model are automatically selected by Bayesian optimization algorithm. And then, fine-tune the model parameters through labeled data to mine features that can represent the essential attributes of various faults. Finally, an RF classifier is created to identify the fault type, and the parameter tuning process also realizes automatic parameter optimization. The test results show that the proposed method has an accuracy of 96.67% for distribution transformers fault diagnosis, which is significantly better than the results using SAE and RF alone.
Key words: distribution transformer; fault diagnosis; stacked auto-encoder (SAE); random forest (RF); Bayesian optimization
參考文獻
[1] 趙莉華,豐遙,謝榮斌,等. 基于交叉小波的變壓器振動信號幅頻特征量提取方法[J] . 高電壓技術(shù),2019,45(2):505-511.
[2] 王逸萍. 基于最小二乘支持向量機的變壓器故障診斷[J]. 電工電氣,2016(6):24-27.
[3] 孔德錢,張新燕,童濤,等. 基于差分進化算法與BP 神經(jīng)網(wǎng)絡(luò)的變壓器故障診斷[J] . 電測與儀表,2020,57(5):57-61.
[4] 叢日立,趙明宇,周洋,等. 基于參數(shù)優(yōu)化的電力變壓器故障診斷模型[J] . 電測與儀表,2019,56(22):84-88.
[5] 齊波,王一鳴,張鵬,等. 基于自決策主動糾偏的電力變壓器油色譜診斷模型[J] . 高電壓技術(shù),2020,46(1):23-32.
[6] 周光宇,馬松齡. 振動法在線監(jiān)測變壓器的研究現(xiàn)狀與發(fā)展[J]. 電工電氣,2019(3):1-6.
[7] 楊毅,王豐華,段若晨,等. 基于自適應篩選EMD和CFDC的變壓器繞組狀態(tài)檢測[J] . 振動與沖擊,2017,36(19):106-111.
[8] 武立平,馬維青,程胤璋,等. 基于振動信號指標能量的變壓器機械故障檢測[J] . 電測與儀表,2020,57(6):126-131.
[9] ZHAO Miaoying, XU Gang.Feature extraction of power transformer vibration signals based on empirical wavelet transform and multiscale entropy [J]. IET Science , Measurement & Technology, 2018, 12(1):63-71.
[10] 趙妙穎,許剛. 基于經(jīng)驗小波變換的變壓器振動信號特征提取[J]. 電力系統(tǒng)自動化,2017,41(20):63-69.
[11] 王杰,張曦煌. 基于圖卷積網(wǎng)絡(luò)和自編碼器的半監(jiān)督網(wǎng)絡(luò)表示學習模型[J] . 模式識別與人工智能,2019,32(4):317-325.
[12] 王懷遠,陳啟凡. 基于堆疊變分自動編碼器的電力系統(tǒng)暫態(tài)穩(wěn)定評估方法[J] . 電力自動化設(shè)備,2019,39(12):134-139.
[13] 崔廣新,李殿奎. 基于自編碼算法的深度學習綜述[J]. 計算機系統(tǒng)應用,2018,27(9):47-51.
[14] 劉振興,陳震. 一種基于堆棧自編碼器的水聲信道均衡算法[J]. 通信技術(shù),2019,52(11):2605-2610.
[15] 徐鋒,方彥軍. 基于貝葉斯優(yōu)化XGBoost 的現(xiàn)場校驗儀誤差預測[J] . 電測與儀表,2019,56(18):120-125.
[16] 葛陽,郭蘭中,牛曙光,等. 基于t-SNE和LSTM的旋轉(zhuǎn)機械剩余壽命預測[J] . 振動與沖擊,2020,39(7):223-231.