sensors Article Validity Evaluation Method Based on Data Driving for On-Line Monitoring Data of Transformer under DC-Bias Yuanda He, Qi Zhou, Sheng Li

sensors Article Validity Evaluation Method Based on Data Driving for On-Line Monitoring Data Transformer under DC-Bias Yuanda He, Qi Zhou, Sheng Lin * and Liping Zhao Department, School Electrical Engineering, Southwest Jiaotong University, Chengdu 611756, China; he_yuanda@my.swjtu.edu.cn (Y.H.); zqi@my.swjtu.edu.cn (Q.Z.); lpzhao@swjtu.cn (L.Z.) * Correspondence: slin@swjtu.edu.cn Received: 1 July ; Accepted: 31 July ; Published: 3 August Abstract: The DC-bias monitoring device a transformer is easily affected by external noise interference, equipment aging, and communication failure, which makes it difficult to guarantee validity monitoring data and causes great problems for future data analysis. For this reason, this paper proposes a validity evaluation method based on data driving for on-line monitoring data a transformer under DC-bias. First, variation rule and threshold range monitoring data for neutral point DC, vibration, and noise transformer under different working conditions are obtained through statistical analysis. Then, data validity criterion DC bias monitoring data is proposed to achieve a comprehensive evaluation data validity based on data threshold, continuity, impact, and correlation. In addition, case studies are carried out on real measured data DC bias magnetic monitoring system a regional power grid by using this evaluation method. The results show that proposed method can systematically and comprehensively evaluate validity DC bias monitoring data and can judge wher monitoring device fails to a certain extent. Keywords: transformer; DC-bias; on-line monitoring; data validity evaluation; data driving 1. Introduction The main transformer urban power grid is affected by stray current subway, which produces phenomenon DC bias [1 3]. The concrete performance DC bias is that vibration transformer intensifies and noise increases [4]. Severe DC bias will affect working life transformer and even cause permanent damage to transformer [5,6]. In view this, on-line monitoring devices for DC bias have been installed in many urban power grids in China, such as Shanghai, Guangzhou, Guiyang, and so on [7,8]. The installation DC bias monitoring device realizes on-line monitoring neutral point DC, vibration, and noise transformer [9,1]. However, in practical application, DC bias monitoring sensors are easily affected by external noise interference, equipment aging, and communication failure [11 13]. Therefore, it is difficult to guarantee validity DC bias monitoring data which causes problems for DC bias state judgment and characteristic analysis main transformer. For example, audible noise measurement sensor is ten interfered with by background noise [11]. Most monitoring devices are installed outdoors, where operating environment is harsh. And service life monitoring devices is far lower than that power transformer itself [1]. In addition, re are problems unreliability in process remote monitoring data transmission [13]. Therefore, to solve above problems, it is great engineering application value to study validity evaluation method for on-line monitoring data transformer under DC-bias. Sensors,, 431; doi:1.339/s15431 www.mdpi.com/journal/sensors

Sensors,, 431 17 At present, some research has been carried out to solve problem that validity DC bias monitoring data is difficult to guarantee in practical application [13 15]. However, existing studies can only improve validity DC bias monitoring data partly. For example, a method has been proposed to solve problem unreliability in remote data transmission based on communication fault detection [13]. Liu et al. [14] and Tong et al. [15] proposed new monitoring methods to solve overload and anti-interference problems neutral DC monitoring sensors, respectively. To best our knowledge, re is no method to comprehensively solve problem poor validity monitoring data in field DC bias monitoring. However, in or fields, scholars have conducted many studies on evaluation method which can solve problem poor validity monitoring data [16 18]. Because valid data and invalid data can be clearly distinguished by using data validity evaluation method, through analyzing causes invalid data, problems existing in monitoring device can be solved. However, existing data validity evaluation methods are target specific data, such as land change data [17] and urban traffic data [18], and types and variation characteristics transformer DC bias monitoring data are quite different from above monitoring data. Therefore, it is necessary to propose a method focused on validity evaluation DC bias monitoring data. In addition, data-driven methods have been widely applied in fields status monitoring [19], anomaly detection [], and residual life estimation [1]. Unlike traditional methods based on physics models, data-driven methods do not need to know specific information objects mamatical model [], and data-driven methods can control and evaluate system only requiring monitoring data [3,4]. The objective this paper is to solve problem that validity DC bias monitoring data is difficult to guarantee in practical application. Thus, a data-driven and multi-criterion method is proposed to achieve validity evaluation DC bias monitoring data in this paper. On one hand, this paper evaluates validity monitoring data by using data-driven method, which does not need to know specific information transformer model and avoids complex electromagnetic analysis transformer. On or hand, multi-criterion method proposed in this paper considering data threshold, continuity, impact, and correlation is more comprehensive than single-criterion method in evaluation process, which can cover as many abnormal cases as possible. Finally, case studies have been carried out to verify correctness proposed method in this paper. The remainder this paper is organized as follows. The transformer DC bias monitoring devices and characteristics normal and abnormal monitoring data are presented in Section. Section 3 presents data validity criteria and evaluation process. The results case studies are presented in Section 4. Section 5 concludes whole paper.. Data Characteristics Analysis.1. Transformer DC Bias Monitoring Device When DC bias main transformer occurs, neutral point DC main transformer increases, vibration intensifies, and abnormal noise is obvious [7]. Therefore, to accurately identify DC bias hidden danger, transformer DC bias synchronous monitoring device based on neutral point DC, vibration, and noise is widely used [8,9]. Take transformer DC bias monitoring system a certain area power grid as an example. The composition and connection mode monitoring system are shown in Figure 1, and hardware DC bias synchronous monitoring device transformers is shown in Figure. The DC bias monitoring system transformer is mainly composed a synchronous monitoring device, neutral DC monitoring sensor, vibration speed sensor, and noise sensor. The synchronous monitoring device collects and displays real-time data from or three sensors. At same time,

Sensors,, 431 3 17 real-time synchronous data will be transmitted to background monitoring system through Sensors,, x FOR PEER REVIEW 3 19 global position system (GPS). Sensors,, x FOR PEER REVIEW 3 19 Transformer Transformer Vibration sensor Vibration sensor GPS GPS Neutral DC monitoring Neutral sensor DC monitoring sensor Noise sensor Noise sensor Synchronous monitoring device Synchronous monitoring device Figure 1. The composition and connection mode monitoring system. Figure 1. The composition and connection mode monitoring system. Figure 1. The composition and connection mode monitoring system. (a) (b) (c) (a) (b) (c) Figure. The hardware DCbias biasmonitoring monitoring device (a) device appearance; Figure. The hardware DC device transformers: transformers: (a) device appearance; Figure. The hardware DC bias monitoring device transformers: (a) device appearance; (b) synchronous monitoring device; (c) physical picture internal hardware. (b) synchronous monitoring device; (c) physical picture internal hardware. (b) synchronous monitoring device; (c) physical picture internal hardware. installation mode neutral DC monitoring sensor, vibration sensor, and noise sensor is The The installation mode neutral sensor,vibration vibration sensor, noise sensor is The installation mode neutraldc DCmonitoring monitoring sensor, sensor, andand noise sensor is shown in Figure 3. The objects monitoring device are kv and 5 kv transformers in shown in Figure 3. The objects monitoring device are kv and 5 kv transformers in shown in Figure 3. The objects monitoring device are kv and 5 kv transformers in power grid. The basic information main transformer being monitored is shown in Table 1. Each power grid.grid. The The basic information beingmonitored monitored is shown in Table 1. Each power basic information main maintransformer transformer being is shown in Table 1. Each transformer is provided with a set monitoring devices including three sensors. transformer is provided with a set monitoringdevices devices including sensors. transformer is provided with a set monitoring includingthree three sensors. Table 1. Characteristic rules normal data. Transformer Type Rated Power (MVA) ODFPSZ9-5/55/ 3 SFSZ9-4/ SFSZ9-18/ SFSZ9-15/ 5/5/53.5 4/4/8 18/18/ 15/15/75 Rated Voltage (kv) 55/ 3/4/ 3/34.5 /115/1.5 /11/11 /11/11

Sensors,, 431 4 17 (a) (b) (c) Figure 3. The installation mode sensors: (a) neutral DC monitoring sensor; (b) vibration sensor; (c) noise sensor. Volume 6.. Characteristics Normal Monitoring Data B.N. In this part, characteristics neutral point DC, vibration, and noise monitoring data transformer under different working conditions are analyzed, which lays a foundation for obtaining criterion 前面地址逗号隔开 data validity evaluation...1. 公式斜体一致 The Classification Working Conditions To distinguish characteristics normal data under different working conditions, specific 括号, 逗号不斜体 classification method is shown in Figure 4. Firstly, according to occurrence DC bias, working conditions Equation were (1) (3) divided into two categories. Secondly, according to wher DC bias suppression device was turned on, secondary classification was carried out. When DC bias did not occur, wher L. suppression device was opened or not had no effect on monitoring data. Therefore, a secondary classification this working condition was not made. Thirdly, third classification <mml:mo> </mml:mo> was made according to wher transformer was an autotransformer. Similarly, it did not matter wher 应该是 transformer was an autotransformer or not when suppression device was f. Therefore, a third <mml:mtext> </mml:mtext> classification this working而且多个 condition was not 时同时放一个标签里面 made. To sum up, four working conditions were obtained, C 1 C 4. Sensors,, x FOR PEER REVIEW 5 19 gov. extra <named-content content-type="color:#131413"> missing Working author id Condition <break/> No DC Bias Occurs The Suppression Device is f C1 C Q1 Figures The DC Bias Occurs <supplement>(suppl. 3)</supplement> The Suppression Device is on <mml:munder> should be <mml:msub> in eq. (3) 1 Non-Autotransformer, Successfully suppressed Autotransformer, Only DC is suppressed C3 C4 Q Q3... Data Characteristics without DC Bias Figure Figure4. 4. The Theclassification classification working workingconditions. To obtain data characteristics transformer without DC bias, tests were carried out. There was no subway stray current interference at : 4: at night, and no DC bias occurred to transformer. The monitoring data neutral point DC, vibration, and noise a transformer are shown in Figure 5. The neutral point DC changed within ± A. The vibration velocity was within

Successfully suppressed The Suppression Device is on Q3 Autotransformer, Only C4 DC is suppressed Sensors,, 431 5 17 Figure 4. The classification working conditions.... Data Characteristics without DC Bias... Data Characteristics without DC Bias To obtain data characteristics transformer without DC bias, tests were carried out. To obtain data characteristics transformer without DC bias, tests were carried out. There was no subway stray current interference at : 4: at night, and no DC bias occurred to There was no subway stray current interference at : 4: at night, and no DC bias occurred to transformer. The monitoring data neutral point DC, vibration, and noise transformer are transformer. The monitoring data neutral point DC, vibration, and noise a transformer are shown in Figure 5. The neutral point DC changed within ± A. The vibration velocity was within and shown in Figure 5. The neutral point DC changed within ± A. The vibration velocity was within.7 mm/s, and amplitude change did not exceed.3 mm/s. The noise intensity was within 5 and and.7 mm/s, and amplitude change did not exceed.3 mm/s. The noise intensity was within 9 db, and amplitude change did not exceed 1 db. In addition, average noise intensity 5 and 9 db, and amplitude change did not exceed 1 db. In addition, average noise different transformers without DC bias was different. As shown in Figure 6, average noise intensity intensity different transformers without DC bias was different. As shown in Figure 6, average transformer A during period : 4: was 76.3 db, while average noise intensity noise intensity transformer A during period : 4: was 76.3 db, while average transformer B was only 63.36 db. noise intensity transformer B was only 63.36 db. Neutral DC (A) Vibration velocity (mm/s) Noise intensity (db) 1-1 - : : :4 3: 3: 3:4 4:.8.7.6.4.3mm/s. : : :4 3: 3: 3:4 4: 75 7 1dB 65 6 55 5 : : :4 3: 3: 3:4 4: Figure 5. Data characteristics a transformer without DC bias. Figure 5. Data characteristics a transformer without DC bias. Sensors,, x FOR PEER REVIEW 6 19 85 Noise intensity (db) 8 75 7 65 6 Transformer A Transformer B 55 5 : : :4 3: 3: 3:4 4: Figure Figure 6. The 6. The noise noise intensity transformer A and transformer B Bin in period period from from : : to 4:. to 4:...3...3. Data Data Characteristics When When DC DC Bias Occurs and Suppression Device Is f Is f When When DC bias DC bias transformer occurred, use DC bias suppression device device had had a great a great influence influence on on monitoring monitoring data [5 7]. data [5 7]. To obtain To obtain data characteristics data characteristics when when DC bias DC occurred bias and occurred suppression and device suppression was f, device tests was were f, carried tests were out. carried Duringout. During operation operation subway from subway from 14: to 16:, DC bias transformer occurred, and suppression device was 14: to 16:, DC bias transformer occurred, and suppression device was f. The monitoring f. The monitoring data neutral point DC, vibration, and noise a transformer are shown in Figure 7. The neutral point DC fluctuated within ±9 A. Vibration velocity was within and mm/s, and maximum variation amplitude did not exceed 1.5 mm/s. The noise intensity was within 5 and 11 db, and maximum variation range did not exceed 35 db. C 3

Figure 6. The noise intensity transformer A and transformer B in period from : to 4:...3. Data Characteristics When DC Bias Occurs and Suppression Device Is f Sensors, When, 431 DC bias transformer occurred, use DC bias suppression device had a great 6 17 influence on monitoring data [5 7]. To obtain data characteristics when DC bias occurred and suppression device was f, tests were carried out. During operation data subway neutral from point 14: DC, to 16:, vibration, DC bias and noise transformer a transformer occurred, are and shown suppression in Figuredevice 7. Thewas neutral pointf. DCThe fluctuated monitoring within data ±9 neutral A. Vibration point DC, velocity vibration, wasand within noise and a transformer mm/s, and are shown maximum in variation Figure amplitude 7. The neutral did point not exceed DC fluctuated 1.5 mm/s. within The±9 noise A. Vibration intensityvelocity was within was within 5 and and 11 db, mm/s, and maximum and variation maximum range variation did not amplitude exceeddid 35 db. not exceed 1.5 mm/s. The noise intensity was within 5 and 11 db, and maximum variation range did not exceed 35 db. Neutral DC (A) Vibration velocity (mm/s) Noise intensity (db) 3 1-1 - -3 14: 14: 14:4 15: 15: 15:4 16: 1.6 1..8.4 14: 14: 14:4 15: 15: 15:4 16: 1 9 8 7 6 35dB 1.5mm/s 5 14: 14: 14:4 15: 15: 15:4 16: Figure Figure 7. The 7. The data data characteristics a transformer with DC bias and and suppression device device f. f...4...4. Data Data Characteristics When DC DC Bias Occurs and Suppression Device Device Is on Is on To obtain To obtain data data characteristics when DC DCbias bias occurs occurs and and suppression suppression device is device on, is on, tests were carried out. During operation subway subway from from 14: 14: to 16:, to 16:, DC bias DC bias transformer transformer occurred, occurred, and and suppression device was on. on. The Themonitoring data data neutral neutral point point DC, DC, vibration, and noise a transformer are shown in Figure 8. The data characteristics are like those vibration, and noise a transformer are shown in Figure 8. The data characteristics are like those without DC bias. In addition, considering particularity autotransformer, DC bias without DC bias. In addition, considering particularity autotransformer, DC bias autotransformer cannot be effectively suppressed by adding a single suppression device [8]. In this autotransformer cannot be effectively suppressed by adding a single suppression device [8]. In this Sensors,, x FOR PEER REVIEW 7 19 case, neutral point DC varied within ± A, indicating that DC was indeed limited. However, vibration case, neutral velocitypoint varied DC within varied within and ± mm/s, A, indicating and that noise DC intensity was indeed varied limited. within However, 5 and 11 db, indicating vibration thatvelocity DC bias varied was within not successfully and mm/s, suppressed, and noise as shown intensity varied Figurewithin 9. 5 and 11 db, indicating that DC bias was not successfully suppressed, as shown in Figure 9. Neutral DC (A) 1-1 - 14: 14: 14:4 15: 15: 15:4 16: Vibration velocity (mm/s) Noise intensity (db).7.6.4. 14: 14: 14:4 15: 15: 15:4 16: 7 65 6 55.3mm/s 5 14: 14: 14:4 15: 15: 15:4 16: Figure Figure 8. The 8. The data data characteristics aa transformer with DCbias bias and and suppression device device on. on. 1dB Neutral DC (A) 1-1 - 14: 14: 14:4 15: 15: 15:4 16:

Noise intensit (db) 65 6 55 5 14: 14: 14:4 15: 15: 15:4 16: Sensors,, 431 7 17 Figure 8. The data characteristics a transformer with DC bias and suppression device on. 1dB Neutral DC (A) Vibration velocity (mm/s) 1-1 - 14: 14: 14:4 15: 15: 15:4 16: 1.8.6.4..8mm/s 14: 14: 14:4 15: 15: 15:4 16: Noise intensity (db) 11 9 7 5 14: 14: 14:4 15: 15: 15:4 16: Figure 9. 9. Data characteristics an autotransformer an when when suppression suppression device device is on but is fails on to but suppress fails to DC suppress bias. DC bias...5...5. Summary Summary Characteristics Characteristics Normal Normal Monitoring Monitoring Data Data By By comparing comparing characteristics characteristics above above transformer transformer DC DC bias bias monitoring monitoring data, data, it can it be can seen be that seen that data characteristics data characteristics working working conditions conditions C 1 and C C1 3 were and C3 were same. Therefore, same. Therefore, we classified we classified two conditions two conditions into same into class. same The class. final classification The final classification result, Q 1 Q result, 3, is shown Q1 Q3, in is shown Figure in 4. The Figure first 4. type The Q first 1 was type that Q1 was transformer that transformer had DC-bias had and DC-bias suppression and suppression device was f. device The was second f. type The Q second consisted type two Q consisted cases: two transformer cases: had transformer no DC-bias, had no non-autotransformer DC-bias, non-autotransformer had DC-bias, and had DC-bias, suppression and device suppression was on. device The third was type on. The Q 3 was third that type autotransformer Q3 was that autotransformer had DC-bias and had DC-bias suppression and device suppression was on. device The data was characteristic on. The data rules characteristic under different rules under working different conditions working are shown conditions in Table are shown. in Table. Table. Characteristic rules normal data. Working Condition Neutral DC Noise Intensity Vibration Velocity Q 1 9 9 A 5 11 db mm/s Q A 5 9 db.7 mm/s Q 3 A 5 11 db mm/s 3dB.3. Characteristics Abnormal Monitoring Data Abnormal data ten correspond to various fault situations; refore, evaluation data validity can be completed by analyzing characteristics abnormal data caused by various fault situations. Abnormal data types [9 31] are mainly: 1. Blank data: monitoring value is always empty or zero. It is caused by inductive damage monitoring device or interruption system transmission.. Over range data: monitoring value exceeds allowable measurement range sensor. It is caused by strong external interference to sensor or system communication failure. 3. Offset data: re is a certain deviation between monitored value and real value. It is caused by aging sensor detection unit. 4. Abnormal zero drift data: monitoring value is deviated abnormally with change time. It is caused by aging sensor.

3.1. Data Validity Criterion 1. Blank data: monitoring value is always empty or zero. It is caused by inductive damage monitoring device or interruption system transmission.. Over range data: monitoring value exceeds allowable measurement range sensor. It is caused by strong external interference to sensor or system communication failure. 3. Sensors Offset,, data: 431 re is a certain deviation between monitored value and real value. 8 It 17 is caused by aging sensor detection unit. 4. Abnormal zero drift data: monitoring value is deviated abnormally with change time. 5. It Variable is caused ratio by deviation aging data: re sensor. is a certain proportion relationship between monitored 5. Variable value and ratio deviation real value. data: Itre is caused is a certain by proportion change relationship external between environment monitored or value misconfiguration and real value. sensor. It is caused by change external environment or 6. misconfiguration Abnormal mutation data: sensor. abnormal mutation monitoring value. It is caused by strong 6. Abnormal external interference mutation data: to abnormal sensor ormutation failure monitoring sensor itself. value. It is caused by strong 7. external Abnormal interference step data: to monitoring sensor or value failure has an unreasonable sensor itself. and sudden change. It is caused 7. Abnormal by change step data: external monitoring environment value has or an unreasonable strong interference and sudden to sensor. change. It is caused by change external environment or strong interference to sensor. Through analysis existing data, it can be found that above seven kinds abnormal Through analysis existing data, it can be found that above seven kinds abnormal data DC bias monitoring do not occur in isolation, but ten occur at same time. Therefore, it was data DC bias monitoring do not occur in isolation, but ten occur at same time. Therefore, it necessary to analyze data with multiple criteria at same time, so as to evaluate validity was necessary to analyze data with multiple criteria at same time, so as to evaluate validity data more accurately. Several typical abnormal data are shown in Figure 1. data more accurately. Several typical abnormal data are shown in Figure 1. Noise intensity (db) 14 13 1 1 8 6 14:3 14:5 15:1 15:3 15:5 16:1 16:3 (a) 6 Neutral DC (A) 4 - -4-6 14:3 14:5 15:1 15:3 15:5 16:1 16:3 Sensors,, x FOR PEER REVIEW 9 18 (b) 11 Noise intensity (db) 1 9 8 7 6 5 14:3 14:5 15:1 15:3 15:5 16:1 16:3 (c) Figure 1. Several typical abnormal data: (a) (a) noise noise overrange data data caused caused by by abnormal migration; (b) (b) abnormal zero zero drift drift data data neutral neutral DC; DC; (c) noise (c) noise abnormal mutation data. data. 3. Methods Data validity Evaluation In this section, according to characteristics normal data and abnormal data, this paper proposes data validity criteria based on data threshold, continuity, impact, and correlation. Each criterion reflects validity data in one respect. Therefore, a method for evaluating validity data with multiple criteria is presented in this paper.

Sensors,, 431 9 17 3. Methods Data Validity Evaluation In this section, according to characteristics normal data and abnormal data, this paper proposes data validity criteria based on data threshold, continuity, impact, and correlation. Each criterion reflects validity data in one respect. Therefore, a method for evaluating validity data with multiple criteria is presented in this paper. 3.1. Data Validity Criterion 3.1.1. Criterion Based on Threshold Value When evaluating validity monitoring data DC bias transformers, it is not enough to evaluate validity only based on information a single data point. Therefore, a certain data analysis period S needs to be selected. By analyzing monitoring data, it was found that re were data points beyond measuring range sensor in some time period. Since each sensor has its measuring range, if re are data points exceeding range or equal to boundary value in monitoring data, it means that data points this fraction are abnormal data points. The threshold criterion based on sensor measurement range is presented as follows. Criterion P 1 : according to measuring range corresponding sensor, set upper limit H and lower limit L monitoring data. This criterion is used to evaluate all data points in selected time period S. If value data point x i exceeds upper and lower limits measurement, that is, Equation (1) is satisfied, n data point x i is judged to be an invalid data point. x i H or x i L. (1) According to analysis characteristics normal monitoring data under different working conditions in Section, it can be seen that wher transformer is an auto-transformer or wher transformer DC bias suppression device is turned on has a great impact on characteristics monitoring data. The threshold criterion based on operating condition transformer is presented as follows. Criterion P : according to different transformer working conditions Q 1 Q 3, set effective upper limit H 1 H 3 and effective lower limit L 1 L 3 monitoring data. This criterion is used to evaluate all data points in selected period S. First, working condition Q transformer is determined, and n data point x i is compared with corresponding upper and lower limits monitoring data. If upper and lower limits are exceeded, which satisfies Equation (), n data point x i is judged to be an invalid data point. Q 1 : x i H 1 or x i L 1, (a) Q : x i H or x i L, (b) 3.1.. Criterion Based on Data Continuity Q 3 : x i H 3 or x i L 3. (c) According to characteristics sensor and time-varying characteristics measurement parameters, continuous and identical data points directly reflect fault sensors induction part or system s communication. Therefore, continuous and identical data points are abnormal data points. The criteria based on continuous sameness data are presented as follows. Criterion P3: According to tolerance number continuously identical data, set continuously identical data tolerance values N.

Sensors,, 431 1 17 This criterion is used to evaluate all data points in selected time period S. Starting from initial data point x 1, continuous N data points are compared. If Equation (3) is satisfied, data points x i, x i+1,..., x i+n-1 are all invalid data points, and n compare subsequent data points x i+n, x i+n+1,... with x i. If y are same as x i, y will be judged as invalid data points, until a data point different from x i appears or last data point is judged. x i = x i+1 =... = x i+n 1. (3) Due to limited measurement accuracy sensor, it is impossible to identify small changes monitoring data below its accuracy. Therefore, when measured data are very small but not zero, a large number continuous identical values will also appear in neutral DC and vibration data. In order to avoid data validity misjudgment caused by this, supplementary criterion P3 should be added. Supplementary criterion criterion P 3 : according to range and measurement accuracy sensor, set appropriate minimum applicable value M. Using this supplementary criterion, if value x i continuous identical data satisfies Equation (4), this part data will not be judged as invalid data. 3.1.3. Criterion Based on Impact Data x i and x i M. (4) The peak value index is quotient peak value and effective value, and pulse index is quotient peak value and mean value, both which are used to judge wher re are impact data in monitoring data. When monitoring device is strongly interfered with by outside, impact data will appear in monitoring data, but this part data has nothing to do with DC bias. Thus, criteria based on peak value and pulse value are proposed as follows: Criterion P 4 : according to tolerance degree abnormal impact data, set appropriate tolerance value C abnormal impact data. For neutral DC data, tolerance value C 1 based on peak index was selected, while for noise and vibration data, tolerance value C based on pulse index was selected. This criterion is used to evaluate all data points within selected time period S corresponding type data. First, according to Equations (5) (7), peak value X p, effective value X r or mean X av data in period S can be obtained. Then, I p or C f, peak value data in period S, are calculated and compared with corresponding abnormal impact tolerance value C. If Equation (8) or (9) are satisfied, abnormal impact data can be judged to exist in time period S. X p = E[max x i ], (5) X r = 1 n X av = 1 n n x i, (6) i=1 n X i, (7) i=1 I p = X p X r and I p > C 1, (8) C f = X p X av and C f > C. (9)

Sensors,, 431 11 17 Because criterion is proposed from perspective statistical analysis, criterion can only be used to judge wher re are abnormal impact data in time period S, rar than directly find abnormal data points. Therefore, it is necessary to supplement criterion data points analysis. Supplementary criterion criterion P 4 : according to tolerance degree data singularity, set appropriate change rate K data singularity tolerance and corresponding change minimum index K. This criterion is used to evaluate all data points in selected time period S. Starting from second data point x, data point x i is compared with two before and after data points. If Equations (1) (1) are satisfied at same time, data point x i is judged to be an invalid data point. The purpose setting change minimum index is to avoid miscalculation data point singularity caused by a data point that is too small. 3.1.4. Criterion Based on Data Correlation (x i 1 x i )(x i+1 x i ) >, (1) x i 1 x i x i > K and x i+1 x i x i > K, (11) x i 1 x i > K and xi+1 x i > K. (1) Transformer noise is mainly caused by transformer vibration; refore, re is a strong correlation between noise intensity and vibration speed in monitoring data. If noise intensity in monitoring data changes greatly and vibration velocity does not change correspondingly, n noise data in this part change are not caused by vibration, but by background noise independent DC bias transformer. The criteria based on data correlation are presented as follows. Criterion P 5 : according to tolerance degree data asynchronous changes, appropriate data asynchronous tolerance rate G is set. This criterion is used to make synchronous judgments on all noise intensity data points x and vibration velocity data point y within selected time period S. The rate change is calculated from initial data points x 1 and y 1. If difference between two rates change satisfies Equation (13), noise intensity data points x i and x i+1 are judged to be invalid data points. 3.. Data Validity Evaluation Process x i+1 x i x i y i+1 y i y i > G. (13) The validity evaluation process DC bias data includes four steps: selecting objects, setting criteria parameters, evaluating, and obtaining evaluation results. First all, monitoring data time period S to be analyzed should be selected. period S contains n data points. Then, according to rules historical data, sensor range parameters and actual evaluation requirements set criterion parameters H, L, H 1, L 1, etc. Next, each data point x i in selected time period S is evaluated by using five criteria P 1 P 5 mentioned above. Each criterion reflects validity data point x i in a certain aspect. As long as any criterion is satisfied, it is classified as an invalid data point, and rest are valid data points. Finally, data validity evaluation results are obtained, that is, set valid data points S 1 and set invalid data points S, number valid data points N 1, and number invalid data points N. The data validity evaluation process is shown in Figure 11. According to statistical analysis, invalid data account for more than 1% when monitoring device fails. Therefore, if number invalid data points exceeds 1% total number data points in evaluation results, monitoring device transformer can be speculated to have a fault.

certain aspect. As long as any criterion is satisfied, it is classified as an invalid data point, and rest are valid data points. Finally, data validity evaluation results are obtained, that is, set valid data points S1 and set invalid data points S, number valid data points N1, and number invalid data points N. The data validity evaluation process is shown in Figure 11. According to statistical analysis, invalid data account for more than 1% when monitoring Sensors, device, 431 fails. Therefore, if number invalid data points exceeds 1% total number data 1 17 points in evaluation results, monitoring device transformer can be speculated to have a fault. Start Select time period S, which contains n data points:x1, x,..., xi,..., xn. According to rules historical data, sensor range parameters and actual evaluation requirements, set criterion parameters H, L, H1, L1, etc. All data points xi in selected time period S were evaluated using criteria P1-P5 in turn. The data point xi that satisfies eir criterion is judged to be an invalid data point, and rest are valid data point. Get evaluation results: The set valid data points is S1, and set invalid data points is S; The number valid data points is N 1, and number invalid data points is N. End Figure 11. Data validity evaluation process. Figure 11. Data validity evaluation process. 4. Case Studies and Results 4. Case Studies and Results In order to verify validity proposed method in this paper, case studies were carried In order out on to verify data validity DC bias magnetic proposed monitoring method system ina this regional paper, power case grid. studies Using this were carried out on evaluation data method, DC bias magnetic validity monitoring data system was evaluated a regional systematically power grid. and Using this comprehensively. By manual reexamination, judgment invalid data was confirmed, and evaluation method, validity monitoring data was evaluated systematically and comprehensively. validity method was verified. By manual reexamination, judgment invalid data was confirmed, and validity method was verified. 4.1. Parameter Setting Criterion According to main technical parameters measurement sensor, parameters data validity evaluation Criterion 1 were set. According to statistical rules DC bias monitoring data obtained by many tests and general requirements for DC bias parameters transformer, parameters data validity evaluation Criterion 5 were set. The specific parameter setting data validity evaluation criteria are shown in Table 3. Table 3. Parameter setting data validity evaluation criterion. Criterion Parameter Neutral DC Noise Intensity Vibration Velocity Criterion 1 Criterion Criterion 3 Criterion 4 H 1 A 13 db mm/s L 1 A 3 db mm/s H 1 9 A 11 db mm/s L 1 9 A 5 db mm/s H A 9 db.7 mm/s L A 5 db mm/s H 3 A 11 db mm/s L 3 A 5 db mm/s N 1 1 1 M.3 A \.15 mm/s C 1.5 1.1 1.1 K % 1% 1% K 5 A 5 db. mm/s Criterion 5 G \ 1%

Sensors,, 431 13 17 4.. Case Studies 4..1. Case 1: Data Validity Evaluation Transformer C The method proposed in this paper was used to evaluate data validity transformer C in regional power grid. There was a period continuous identical data lasting for 1 s from 14:4 to 15: on 8 April. The vibration velocity monitoring data transformer from 14:44:8 to 14:44:19 were.58 mm/s, as shown in Figure 1. According to setting continuous same tolerance value N in Criterion 3, continuous same number was greater than tolerance value N = 1. Therefore, data points this part were judged as invalid data points. After querying network communication records monitoring system, it was found that a temporary data communication failure occurred during this period, which led to abnormal situation shown in Figure 1. Through analysis this case, it can be seen that validity evaluation method DC bias data proposed in this paper can well identify data anomalies that are difficult to observe artificially. Sensors,, x FOR PEER REVIEW 15 19 1.5 Vibration velocity (mm/s) 1..9.6.3.65.6.58.55 14:4 14:45 14:5 14:55 15: Figure 1. Monitoring data vibration velocity transformer C on 8 April. 4... Case : Data Validity Evaluation Transformer D The method proposed in this paper was used to evaluate data validity transformerd. The monitoring data transformer D were abnormal on 3 October 19. The monitoring data neutral point DC at 8: 8:1 transformer are shown in Figure 13. When transformer was a non-autotransformer and suppression devicewas wason on (working conditionq), Q a ), large a large amount amount neutral neutral DC DC data data were were judged judged invalid invalid because because y y exceeded exceeded upper upper limit limit H (A) H (A) or or lower lower limit L limit ( A) L ( A) Criterion Criterion. Through. Through analysis analysis se se invalid invalid data, data, it was itfound was found that thatcause cause this part this part invalid invalid data data could could be attributed be attributed to toabnormal abnormal zero zero drift drift data. data. The monitoring The monitoring device device was was presumed presumed to have to have malfunctioned malfunctioned because because a large a large amount amount invalid invalid data. data. The The inspection inspection neutral neutral point point DC monitoring DC monitoring device device transformer transformer showed showed that re that re was indeed was indeed a serious a serious fault. After fault. After device device was replaced was replaced on 4 on November 4 November 19, 19, monitoring data data same period were analyzed again next day using same method. The results showed that re were no longer many invalid data, as shown in Figure 14. The results two data validity analyses transformer D are shown in Table 4. 6 4 Neutral DC (A) -

presumed to have malfunctioned because a large amount invalid data. The inspection neutral point DC monitoring device transformer showed that re was indeed a serious fault. After device was replaced on 4 November 19, monitoring data same period were analyzed again next day using same method. The results showed that re were no longer many Sensors invalid,, 431 data, as shown in Figure 14. The results two data validity analyses transformer 14 D 17 are shown in Table 4. 6 4 Neutral DC (A) - -4-6 8: 8: 8:4 8:6 8:8 8:1 Sensors,, x Figure FOR Figure PEER 13. 13. REVIEW Monitoring Monitoringdata data neutral neutraldc DC transformer Don on3 3October October19. 16 19 Neutral DC (A) 1-1 - 8: 8: 8:4 8:6 8:8 8:1 Figure 14. Monitoring data neutral DC transformer D on 5 November 19. Table 4. Transformer D data validity analysis results. Table 4. Transformer data validity analysis results. Date Period Total Number Number Number Normal Number Number Invalid Period Total Number Analysis Date Analysis Analysis Data Points Data Normal Points Data Invalid Data Points Data Analysis Data Points 3 October Points Points 8: 8:1 6 5 348 3 19 October 19 8: 8:1 6 5 348 5 5 November 19 8: 8:1 6 6 8: 8:1 6 6 19 4..3. Case 3: Data Validity Evaluation Transformer E 4..3. Case 3: Data Validity Evaluation Transformer E The method proposed in this paper was used to evaluate data validity transformer The method proposed in this paper was used to evaluate data validity transformer E. E. The monitoring data transformer E were abnormal on 18 October 19. The monitoring data The monitoring data transformer E were abnormal on 18 October 19. The monitoring data transformer noise intensity at 9:3 9:4 are shown in Figure 15. A large number intermittent transformer noise intensity at 9:3 9:4 are shown in Figure 15. A large number intermittent continuous anomalous hopping data occurred in noise intensity data. These data were judged invalid continuous anomalous hopping data occurred in noise intensity data. These data were judged because y exceeded upper limit H 1 (11 db) in Criterion and data singularity tolerance invalid because y exceeded upper limit H1 (11 db) in Criterion and data singularity value (1%) in Criterion 4. The monitoring device was presumed to have malfunctioned because tolerance value (1%) in Criterion 4. The monitoring device was presumed to have malfunctioned a large amount invalid data. The inspection noise monitoring device transformer because a large amount invalid data. The inspection noise monitoring device showed that re was indeed a serious fault. After device was replaced on 4 November 19, transformer showed that re was indeed a serious fault. After device was replaced on 4 monitoring data same period were analyzed again next day using same method. November 19, monitoring data same period were analyzed again next day using The results showed that re were no longer many invalid data, as shown in Figure 16. The results same method. The results showed that re were no longer many invalid data, as shown in Figure two data validity analyses transformer E are shown in Table 5. 16. The results two data validity analyses transformer E are shown in Table 5. 1 oise intensity (db) 11 1 9 8 3dB

tolerance value (1%) in Criterion 4. The monitoring device was presumed to have malfunctioned because a large amount invalid data. The inspection noise monitoring device transformer showed that re was indeed a serious fault. After device was replaced on 4 November 19, monitoring data same period were analyzed again next day using same method. The results showed that re were no longer many invalid data, as shown in Figure 16. The results two data validity analyses transformer E are shown in Table 5. Sensors,, 431 15 17 1 Noise intensity (db) 11 1 9 8 7 3dB 6 5 9:3 9:3 9:34 9:36 9:38 9:4 Figure 15. Monitoring data noise intensity transformer E on 18 October 19. Figure 15. Monitoring data noise intensity transformer E on 18 October 19. Sensors,, x FOR PEER REVIEW 17 19 8 75 Noise intensity (db) 7 65 6 55 6dB 5 9:3 9:3 9:34 9:36 9:38 9:4 Figure 16. Monitoring data noise intensity transformer E on 5 November 19. Table Table 5. 5. Transformer Transformer E data data validity validity analysis analysis results. results. Period Period Total Total Number Number Number Normal Number Invalid Date Analysis Analysis Data Points Data Points Data Points Analysis Analysis Data Points Data Points Data Points 18 18 October October 19 9:3 9:4 6 54 96 5 November 19 9:3 9:4 9:3 9:4 6 6 54 6 96 19 5 November 9:3 9:4 6 6 5. Conclusions 19 This paper proposes a validity evaluation method based on data driving for on-line monitoring 5. Conclusions data transformer under DC-bias. First, variation rule and threshold range monitoring data for neutral This paper pointproposes DC, vibration, a validity andevaluation noise method transformer based under on data different driving working for on-line conditions monitoring are data obtained through transformer statistical under analysis. DC-bias. Then, First, according variation to rule characteristics and threshold normal range andmonitoring abnormal data, for neutral data validity point DC, criteria vibration, based on and data noise threshold, transformer continuity, impact, under different and correlation working areconditions proposed are in this obtained paper. through Using se statistical criteria, analysis. a comprehensive Then, according evaluation to system characteristics for data validity normal DC bias and abnormal is established. data, data validity criteria based on data threshold, continuity, impact, and correlation are proposed The method in this proposed paper. in Using this paper se criteria, is used toa evaluate comprehensive validity evaluation real system measured for data data validity DC DC bias bias magnetic is established. monitoring system a regional power grid. The results show that it can replace traditional The method manual proposed method toin evaluate this paper data is validity. used to evaluate In addition, when validity re is a large real measured amount data invalid datadc in bias evaluation magnetic monitoring results, can system be inferred a regional that power monitoring grid. The device results fails, show andthat correctness it can replace traditional fault judgment manual method monitoring evaluate device data canvalidity. be confirmed In addition, through when manual re reexamination. is a large amount invalid data in evaluation results, it can be inferred that monitoring device fails, and correctness fault judgment monitoring device can be confirmed through manual reexamination. One insufficiency this paper lies in classification working conditions. In this paper, working conditions are classified only from three perspectives DC bias generation, suppression device input, and autotransformer. The influence voltage class, saturation degree, and power factor

Sensors,, 431 16 17 One insufficiency this paper lies in classification working conditions. In this paper, working conditions are classified only from three perspectives DC bias generation, suppression device input, and autotransformer. The influence voltage class, saturation degree, and power factor transformer is ignored. In our future work, we will strive to find a better data classification method. All in all, data validity evaluation method proposed in this paper can systematically and comprehensively evaluate validity DC bias monitoring data, laying a foundation for subsequent analysis DC bias characteristics. Author Contributions: Conceptualization, Y.H. and S.L.; investigation, Y.H. and Q.Z.; methodology, Y.H., Q.Z. and S.L.; supervision, S.L. and L.Z.; writing original draft, Y.H.; writing review and editing, Y.H., Q.Z., S.L., and L.Z. All authors have read and agreed to published version manuscript. Funding: This research was funded by National Natural Science Foundation China grant number 51677153. Conflicts Interest: The authors declare no conflict interest. References 1. Yuan, P.; Mao, W.; Ye, H.; Liu, Y. Model Construction and Analysis Transformer DC Magnetic Bias Induced by Rail Transit Stray Current. In Proceedings 19 IEEE 3rd Conference on Energy Internet and Energy System Integration, Changsha, China, 8 1 November 19; pp. 171 1713.. Ni, Y.-R.; Zeng, X.-J.; Yu, K.; Leng, Y.; Peng, P. Research on Modeling Method Transformer DC Bias Caused by Metro Stray Current. In Proceedings 18 International Conference on Power System Technology, Guangzhou, China, 6 8 November 18; pp. 3834 3839. 3. Lin, S.; Zhou, Q.; Lin, X.-H.; Liu, M.-J.; Wang, A.-M. Infinitesimal method based calculation metro stray current in multiple power supply sections. IEEE Access, 8, 96581 96591. [CrossRef] 4. He, J.-L.; Yu, Z.-Q.; Rong, Z.; Zhang, B. Vibration and audible noise characteristics AC transformer caused by HVDC system under monopole operation. IEEE Trans. Power Del. 1, 7, 1835 184. [CrossRef] 5. Girgis, R.; Vedante, K. Effects GIC on power Transformers and Power Systems. In Proceedings IEEE Transmission and Distribution Conference and Exposition, Orlando, FL, USA, 7 1 May 1. 6. Jiang, W.; He, L.; Zhang, Z.-X. Monitoring and Suppression Measures Transformer DC Bias Current. In Proceedings International Conference on Condition Monitoring and Diagnosis, Xi an, China, 5 8 September 16. 7. Wang, A.-M.; Lin, S.; Hu, Z.-H.; Li, J.-Y.; Wang, F.; Wu, G.-X.; He, Z.-Y. Evaluation model DC current distribution in AC power systems caused by stray current DC metro systems. IEEE Trans. Power Del.. [CrossRef] 8. Wu, T.-Y.; Chen, L. Data Mining Based on DC Bias On-Line Monitoring System Shanghai Power Grid. In Proceedings IEEE Conference on Energy Internet and Energy System Integration, Beijing, China, 6 8 November 17. 9. Liu, B.-W.; Ma, H.; Zhou, L.-X. On-line Monitoring Transformer Vibration and Noise Based on DC Magnetic Bias. In Proceedings International Conference on Intelligent Systems Design and Engineering Applications, Zhangjiajie, China, 6 7 November 13. 1. Cui, M.-D.; Yang, Q.; Liu, L.-G. Application Virtual Instrument Technology in Monitoring System Transformer DC Magnetic Bias. In Proceedings International Conference on Computer Application and System Modeling, Taiyuan, China, 4 October 1. 11. Liu, Y.-Q.; Lu, J.-Y.; Zhang, Q.; Guo, J. Effectiveness determination method audible noise test data for high voltage dc transmission lines. High Volt. Eng. 14, 4, 78 733. 1. Jia, J.; Tao, F.-B.; Zhang, G.-J.; Shao, J.; Zhang, X.-H.; Wang, B. Validity evaluation transformer DGA online monitoring data in grid edge systems. IEEE Access, 8, 6759 6768. [CrossRef] 13. Qiao, J.; Liu, Q.; Zhang, Y.-F. Design Geomagnetic Induction Current Monitoring and Early Warning System Based on Cloud Server. In Proceedings IEEE Conference on Industrial Electronics and Applications, Xi an, China, 19 1 June 19. 14. Liu, C.; Zhou, X.-X.; Tian, H.-Y.; Zhao, Y.; Qu, T.-H.; Chen, W.-Z. Research on DC BIAS Current Monitoring Power Transformer Neutral Point. In Proceedings IEEE International Conference on High Voltage Engineering and Application, Chengdu, China, 19 September 16.

Sensors,, 431 17 17 15. Tong, X.; Quan, J.-T.; Xia, T.; He, J.-H.; Wang, C.-Z. Design and Application Sensor Measuring Neutral DC Current Transformer Based on Resistance Sampling. In Proceedings IEEE Sustainable Power and Energy Conference, Beijing, China, 1 3 November 19. 16. Xu, S.-W.; Su, D.-H.; Wang, X. The Data Validity Evaluation in Land Change Survey Based on Remote Sensing. In Proceedings International Conference on Geoinformatics, Beijing, China, 18 June 1. 17. Zhou, N.-N.; Huang, G.-F.; Zhong, S.-Y. Big data validity evaluation based on MMTD. Math. Probl. Eng. 18, 9, 1 6. [CrossRef] 18. Xie, C.-C.; Shao, M.-H. An evaluation on integrity and validity traffic data collected from urban expressways. J. Trans. Inf. Saf. 18, 36, 6 65, 89. 19. Liu, S.; Zhao, Y.; Lin, Z.; Ding, Y.; Yan, Y.; Yang, L.; Wang, Q.; Zhou, H.; Wu, H. Data-driven condition monitoring data acquisition for consumers transformers in actual distribution systems using t-statistics. IEEE Trans. Power Del. 19, 34, 1578 1587. [CrossRef]. Ijaz, M.F.; Attique, M.; Son, Y. Data-driven cervical cancer prediction model with outlier detection and over-sampling methods. Sensors,, 89. [CrossRef] 1. Turrin, S.; Subbiah, S.; Leone, G.; Cristaldi, L. An Algorithm for Data-Driven Prognostics Based on Statistical Analysis Condition Monitoring Data on a Fleet Level. In Proceedings 15 IEEE International Instrumentation and Measurement Technology Conference (IMTC), Pisa, Italy, 11 14 May 15; pp. 69 634.. Yin, S.; Ding, S.X.; Xie, X.; Luo, H. A review on basic data-driven approaches for industrial process monitoring. IEEE Trans. Ind. Electron. 14, 61, 6418 648. [CrossRef] 3. Jayaraman, B.; Mamun, S.M.A.A. On data-driven sparse sensing and linear estimation fluid flows. Sensors,, 375. [CrossRef] 4. Azimi, M.; Eslamlou, A.D.; Pekcan, G. Data-driven structural health monitoring and damage detection through deep learning: State---art review. Sensors,, 778. [CrossRef] [PubMed] 5. Xie, Z.-C.; Lin, X.-N.; Zhang, Z.-Y.; Li, Z.; Xiong, W.; Hu, H.; Khalid, M.S.; Adio, O.S. Advanced DC bias suppression strategy based on finite dc blocking devices. IEEE Trans. Power Del. 17, 1, 5 59. [CrossRef] 6. Zeng, R.; Yu, Z.-Q.; He, J.-L.; Zhang, B.; Niu, B. Study on restraining DC neutral current transformer during HVDC monopolar operation. IEEE Trans. Power Del. 11, 6, 785 791. [CrossRef] 7. Bolduc, L.; Granger, M.; Pare, G.; Saintonge, J.; Brophy, L. Development a DC current-blocking device for transformer neutrals. IEEE Trans. Power Del. 5,, 163 168. [CrossRef] 8. Wu, G.-X.; Chen, L.; Shi, Y.-T.; Lin, S. Analysis on test and suppression DC bias 5 kv main transformer in shenzhen substation. Guangdong Electr. Power, 33, 118 14. 9. Wu, H.; Wang, H.-B.; Zhang, P.; Huang, M.; Qi, B.; Rong, Z.-H. The Assess Method Validity for Partial Discharge Sensor Based on Multiple Criterion. In Proceedings IEEE Conference on Electrical Insulation and Dielectric Phenomena, Toronto, ON, Canada, 17 October 16. 3. Zhou, Z.-Q.; Xiao, L.; Nie, D.-X.; Qi, B.; Zhang, P.; Gao, C.-J.; Li, C.-R. Validity Evaluation Method DGA Monitoring Sensor in Power Transformer Based on Chaos Theory. In Proceedings IEEE Conference on Electrical Insulation and Dielectric Phenomena, Cancun, Mexico, 1 4 October 18. 31. Zhang, P.; Qi, B.; Rong, Z.-H.; Li, C.-R.; Li, F. The Assess Method Validity for DGA Sensor Based on Multiple Criterion. In Proceedings IEEE Electrical Insulation Conference, Montreal, QC, Canada, 19 June 16. by authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under terms and conditions Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4./).