基于棧式降噪自編碼器的輸變電設(shè)備狀態(tài)數(shù)據(jù)清洗方法
發(fā)布時間：2018-10-14 18:56
【摘要】：針對當(dāng)前輸變電設(shè)備狀態(tài)監(jiān)測數(shù)據(jù)清洗過程繁瑣,易造成信息丟失等問題,利用棧式降噪自編碼器對"臟"數(shù)據(jù)的還原解析能力及異常狀態(tài)特征提取能力,提出了一種基于棧式降噪自編碼器的數(shù)據(jù)清洗方法。對設(shè)備正常工況及異常運行狀態(tài)數(shù)據(jù)分別利用棧式降噪自編碼器進(jìn)行訓(xùn)練學(xué)習(xí),獲取損失函數(shù)向量,形成奇異點、缺失數(shù)據(jù)修復(fù)模型和設(shè)備異常運行狀態(tài)數(shù)據(jù)降噪模型。通過核密度估計確定訓(xùn)練樣本損失函數(shù)上限和容限時窗,根據(jù)測試數(shù)據(jù)重構(gòu)誤差和異常數(shù)據(jù)時長與損失函數(shù)上限和容限時窗間的關(guān)系,對"臟"數(shù)據(jù)進(jìn)行分類處理。對某變壓器油色譜中總烴含量及某導(dǎo)線溫度數(shù)據(jù)進(jìn)行清洗,結(jié)果表明所提方法能有效辨識奇異點、缺失信息及異常運行狀態(tài)數(shù)據(jù),并對奇異點、缺失值進(jìn)行修復(fù)重構(gòu)。在設(shè)備異常運行時刻,可以有效過濾干擾數(shù)據(jù)。
[Abstract]:Aiming at the problems of the current status monitoring data cleaning process of power transmission and transformation equipment, such as tedious cleaning process and easy to cause information loss, the ability of reducing and analyzing dirty data and extracting abnormal state features of stack noise reduction self-encoder are used. A data cleaning method based on stack noise reduction self-encoder is proposed. The data of normal working condition and abnormal operation state of equipment are trained and studied by stack noise reduction self-encoder to obtain loss function vector and form singularity data repair model and abnormal operation state data de-noising model. The upper limit and tolerance window of the loss function of training samples are determined by kernel density estimation. According to the relationship between the error of reconstruction of test data and the time of abnormal data and the upper limit and tolerance window of loss function, the dirty data are classified and processed. The results show that the proposed method can effectively identify the singularity point, the missing information and the abnormal operation state data, and repair and reconstruct the singular point and the missing value. Interference data can be filtered effectively at the abnormal operation time of the device.
【作者單位】： 上海交通大學(xué)電子信息與電氣工程學(xué)院;國網(wǎng)山東省電力公司電力科學(xué)研究院;
【基金】：國家自然科學(xué)基金資助項目(51477100) 國家高技術(shù)研究發(fā)展計劃(863計劃)資助項目(2015AA050204) 國家電網(wǎng)公司科技項目(520626150032)~~
【分類號】：TM507
本文編號：2271346
[Abstract]:Aiming at the problems of the current status monitoring data cleaning process of power transmission and transformation equipment, such as tedious cleaning process and easy to cause information loss, the ability of reducing and analyzing dirty data and extracting abnormal state features of stack noise reduction self-encoder are used. A data cleaning method based on stack noise reduction self-encoder is proposed. The data of normal working condition and abnormal operation state of equipment are trained and studied by stack noise reduction self-encoder to obtain loss function vector and form singularity data repair model and abnormal operation state data de-noising model. The upper limit and tolerance window of the loss function of training samples are determined by kernel density estimation. According to the relationship between the error of reconstruction of test data and the time of abnormal data and the upper limit and tolerance window of loss function, the dirty data are classified and processed. The results show that the proposed method can effectively identify the singularity point, the missing information and the abnormal operation state data, and repair and reconstruct the singular point and the missing value. Interference data can be filtered effectively at the abnormal operation time of the device.
【作者單位】： 上海交通大學(xué)電子信息與電氣工程學(xué)院;國網(wǎng)山東省電力公司電力科學(xué)研究院;
【基金】：國家自然科學(xué)基金資助項目(51477100) 國家高技術(shù)研究發(fā)展計劃(863計劃)資助項目(2015AA050204) 國家電網(wǎng)公司科技項目(520626150032)~~
【分類號】：TM507
【相似文獻(xiàn)】
相關(guān)期刊論文 前1條
1 朱六璋;;調(diào)度信息系統(tǒng)的數(shù)據(jù)清洗應(yīng)用[J];電力信息化;2007年04期
，本文編號：2271346
本文鏈接：http://sikaile.net/kejilunwen/dianlidianqilunwen/2271346.html
最近更新
教材專著
