A quantitative research on the level of disturbance to secondary signal ports of electronic voltage transformers under the operation of gas‐insulated switchgear

National Natural Science Foundation of China, Grant/Award Number: U1866201 Abstract The collectors of electronic transformers has the problem of electromagnetic incompatibility under the operation of disconnecting switches, which limits their wide application in smart grids. To evaluate the maximum level of disturbance that collectors are able to withstand in the stochastic process of switching operation, this work introduces a quantitative approach based on limited test measurements, which can quantify the level of disturbance in the signal acquisition ports of electronic voltage transformers (EVT) in different gas‐insulated switchgears (GIS). Based on large numbers of experimental statistical data of the true type simulation experimental platform, the Spearman rank correlation coefficient is used to prove that the amplitude of the disturbance voltage pulse at the secondary side port and the breakdown voltage between the primary side fracture approximately satisfy a linear relationship. Subsequently, the quantile regression model is used to quantify and verify the maximum disturbance level of the secondary side port. Moreover, a solution is proposed to tackle the difficulty in obtaining the breakdown voltage at breaks in the GIS substation field applicatvion, under which the quantified maximum disturbance voltage deviation of secondary ports is only −4.34% by comparison. The research work can further explore the practical value of the measured data in a few tests under GIS substation field conditions and provide references for the subsequent revision of EVT‐related standards.


| INTRODUCTION
With the popularisation and application of intelligent substation technology, an increasing number of secondary equipment is being installed on the casing of gas-insulated switchgear (GIS) pipelines or in the nearby control cabinet in the switchyard [1,2]. This application method makes electronic equipment closer to the sources of transient disturbance, which makes electronic equipment more susceptible to electromagnetic disturbance [3]. Taking the engineering application of the electronic transformer in the state grid as an example, electronic transformers were exposed to extremely serious electromagnetic compatibility problems in the promotion and application of an intelligent substation, resulting in successive operation failures, breakdowns, and even explosions [4]. The investigation found that the electromagnetic environment at the substation site is complicated. Electromagnetic disturbance phenomena such as switch operation, lightning impact, short-circuit current, insulation breakdown, spark discharge, power frequency electric field, and power frequency magnetic field may affect secondary electronic equipment. Among them, the fault case of the signal acquisition unit on the secondary side of the electronic transformer under strong transient electromagnetic disturbance of the switch operation is the most prominent, and relevant accident cases are reported from time to time [5][6][7][8]. Therefore, it is urgent to grasp the transient disturbance level of the signal acquisition port of the electronic transformer under the operation of the isolation switch to ensure safe and reliable operation of the smart substation.
So far, related researchers have carried out a lot of research from test-based measurements on the level of electromagnetic disturbance to secondary equipment in GIS substations during switching operations. In the very fast transient overvoltage (VFTO) studies at the Electric Power Research Institute (EPRI) of the United States, the disturbance to the secondary ports of electromagnetic transformers in 500 and 230 kV substations was measured. The measured maximum disturbance amplitude was 2.3 kV, with the main frequency distributing within the range of 1-12 MHz [9]. M.M. RAO and others measured the disturbance voltage on the cable in the secondary control circuits under switching operations in a 245 kV GIS substation. The influence of cable length and the inductance of the secondary winding on the disturbance voltage was also analysed [10]. Limin FENG and others measured the electromagnetic disturbance to electronic transformers under the operation of a GIS isolator and circuit breaker in a 500 kV intelligent substation in southern Changchun. The maximum disturbance amplitude measured at the input end of the acquisition card was as high as 9.9 kV [11]. The result of the measurement, which was done by Hengtian WU and others, on the sensor ports in the 1100 kV GIS tested circuit showed the maximum disturbance voltage of intelligent component ports on the sensor exceeded 7 kV. The distribution of the main frequency of disturbance voltage of these ports concentrated on three frequency points of 5, 10 and 20 MHz [12]. Subject to the restriction of field conditions of the substations in question, however, only a small number of measurement samples are obtained in the above studies. This resulted in a lack of effective statistical analysis. There is also no correlation study on the relationship of the level of disturbance to secondary sensitive equipment and the characteristic parameters of the primary disturbance source.
Furthermore, the actual physical structures of GIS substations are exceedingly complex. Therefore, the accurate acquisition of relevant parameters is extremely difficult, resulting in a considerable disparity between the results of the simulation calculation and the measured figures which are usually used as an auxiliary means for the prediction and estimation of the level of electromagnetic disturbance in GIS substations [13][14][15]. Therefore, in this work, lots of experiments and statistical analyses are carried out under GIS isolator switching operations on the true-type simulation experimental platform using EVT as the specific research object. The correlation between the pulse amplitude of the disturbance voltage at secondary ports and the breakdown voltage at primary breaks was analysed. Then, the method of quantifying the maximum amplitude of the disturbance voltage at the secondary ports was proposed. Moreover, a solution for the quantification method in the field application of GIS substations was provided. The work done here can further explore the potential application value of only a small amount of test results under the field conditions of GIS substations and provide reference for subsequent revision of related standards.
This work unfolds as follows: Section 2 introduces the 220 kV real simulation test circuits and contents of the tests. In Section 3, the test results are statistically analysed, and the level of disturbance in the ports tested is quantified and verified according to the correlation between the breakdown voltage of primary breaks and the pulse amplitude of the disturbance voltage on secondary ports. In Section 4, discussion and analysis are provided with the problem of difficulty in obtaining the breakdown voltage in GIS substation field application using a quantitative method, and the feasibility of this method in substation field application is demonstrated. Section 5 summarises the findings of this work.

| 220 kV GIS true-type simulation experimental platform
The structural size of the 220 kV GIS true type simulation experimental platform is shown in Figure 1, in which ES stands for earthing switch, DS for disconnecting switch, EVT for electronic voltage transformer to be tested, and ECVT for electronic current-voltage transformer to be tested. The test platform allows the tested EVTs to be connected and installed in locations 2 and 3. The primary high frequency voltage test sensors are built in Position 1 and Position 4, respectively, to monitor the primary transient overvoltage in the process of disconnecting and connecting the operations of the GIS. Meanwhile, in order to be compatible with samples from different manufacturing units, the length of the GIS pipeline is adjustable, as shown at the bottom right side of the figure. It adopts a telescopic slide rail which can slide left and right for variable measurements.
In the scenario under IEC 62271-102-2018, the circuits for the GIS connecting and disconnecting tests are equivalent to three test modes: switching capacity test of a very short bus (duct) section, out-of-phase switching capacity test, and current switching capacity test [16]. When the disconnecting switch DS 1 on the right side in Figure 1 is in the disconnected state, the platform can simulate the switching of the very short bus (duct) section, which is denoted as test circuit A. When DS 1 is in the closed state, a centralised load capacitor C L is connected to the right bushing to simulate the current switching capacity test, which is denoted as test circuit B. In this work, the capacitor C L is 5000 pF, and all the tested samples are installed at Position 3.

| Test contents
A total of five EVTs were tested and measured under test circuits A and B, recorded as samples I, II, III, IV and V, respectively. Among them, samples I, II and III are resistive-capacitive divider voltage transformers, with a high side capacitance C 1 = 20 pF, a low voltage capacitance C 2 = 800 pF, and a resistance R = 3 kΩ; samples IV and V are capacitive divider voltage transformers, with a high side capacitance C 1 = 20 pF and a low voltage capacitance C 2 = 1.2 μF. As shown in Figure 2, the test and measurement contents under each test circuit are mainly divided into the following two forms: the signal ports from the sensor's secondary input to the collector are denoted as L and N, respectively, and the differential mode disturbance voltage U d between L and N, as well as the common mode disturbance voltage U c at the common connecting point O between the collector and the casing at the L end that are measured In addition, as shown in Figure 3, the measurement of the primary VFTO waveform is fulfiled with the hand-hole sensor pre-installed at Position 4 along with an impedance matcher, of which the measuring bandwidth is 2Hz-230 MHz and the divider ratio is 224506:1 [17]. The disturbance voltage waveform of the secondary port is fulfiled by the transient measurement system assembled by a high voltage probe TEK P6015 A, of which the measurement bandwidth is from quasi-DC to 75 MHz and the divider ratio is 1000:1 [18]. The abovementioned measurement system has been calibrated and checked in the laboratory to ensure its on-site measurement performance.

| ANALYSIS AND QUANTITATIVE RESEARCH OF TEST RESULTS
During the tests, the supply voltage was raised to the rated operating voltage of 127.0 kV and 10 connecting operations and 10 disconnecting operations of the disconnecting switch DS 2 were performed with each tested set under two test circuits and each measurement mode. More than 400 groups of transient waveforms were measured and recorded throughout the process. For selecting characteristic parameters, this work refers to the research conclusions on the correlation between  the breakdown voltage at TEV and VFTO breaks set out in the cited studies [19]. In a similar manner, it focussed on the breakdown voltage at the primary break and the disturbance voltage at the secondary port, with a view to quantify the disturbance level at secondary ports through the application of existing research conclusions on VFTO.

| Extraction of typical waveform characteristics and parameters
In the tests and measurements described herein, both primary and secondary voltages are triggered by the same synchronous triggering device at the same triggering time and setting. Therefore, in the same disconnecting switch operation process, the n breakdown of VFTO presents a good one-to-one correspondence with the n micro-pulse of the disturbance voltage, as shown in Figures 4 and 5.
From the typical waveforms of the disturbance voltage of VFTO and the port obtained from a given measurement, as shown in Figures 4 and 5, respectively, it can be seen that the modalities of the measured whole-process load waveforms of VFTO include the power-frequency voltage waveform, the high-frequency oscillation and the quasi-DC step. For the n th breakdown, the breakdown voltage can be expressed by the difference between the voltage value at the end of this high-frequency oscillation and that at the end of the n − 1 th high-frequency oscillation, and its absolute value is denoted as U 1n . The port disturbance voltage waveform is composed of a series of discrete micro-pulses. For the n th micro-pulse, its amplitude is represented by the maximum absolute value in the process of micro-pulse oscillation, denoted as U 2n .
In the connecting process, both breakdown voltage and port disturbance voltage pulses in the VFTO waveform gradually decrease, while the disconnecting process goes exactly the opposite. For more intuitive comparison, the extraction of characteristic parameters is specified as follows: When connecting, the first breakdown and the first pulse are set as n = 1, and the first 20 breakdowns in the whole process of transient waveforms are extracted in chronological order. When disconnecting, take the last breakdown and the last pulse as n, the penultimate breakdown and the penultimate pulse as n − 1, and so on, to extract the last 20 breakdowns in the whole process of transient waveforms.

| Analysis of amplitude correlation
The breakdown voltage U 1n at the primary VFTO break and the disturbance voltage U 2n at the secondary port are regarded as an event pair (U 1n , U 2n ). Different measuring schemes are applied to the test circuits, secondary differential mode disturbance voltage U d and common mode disturbance voltage U c . Based on that, the event pair (U 1n , U 2n ) formed in the test circuit A is expressed by scatters (U 1Adn , U 2Adn ) and (U 1Acn , U 2Acn ), respectively, and the event pair formed in the test circuit B is expressed by scatters (U 1Bdn , U 2Bdn ) and (U 1Bcn , U 2Bcn ), respectively. It can be seen from Figures 6-10 that the data pairs of the fracture breakdown voltage and the port disturbance voltage pulse amplitude in the statistics of 8000 tests have good linear relationship. According to the intuitive linear assumption of the scatter diagram, the correlation coefficient is further used to characterise the linear relationship of data pairs (U 1n , U 2n ) obtained under different test conditions. Considering that it is difficult to ensure that the data pairs obtained from the actual measurement can meet the normal distribution sampling or equal spacing within the logical range, Pearson correlation coefficient analysis may get an unreliable result. Therefore, this work selects the Spearman rank correlation coefficient, which is not required for data distribution, for calculation and analysis [20]. The calculating method is as follows: Assume the original data pair (U 1n , U 2n ) to be a sample of (X,Y ) = {(X 1 , Y 1 ), (X 2 , Y 2 ), …, (X n , Y n )} let R i represent the rank of X i in (X 1 , X 2 , …, X n ) and Q i represent the rank of Y i in (Y 1 , Y 2 , …, Y n ). Then the equation for calculating the Spearman rank correlation coefficient can be expressed as follows: The Spearman rank correlation coefficient r S was shown in Table 1 after calculating the data pairs obtained using various measurement methods.
It is noticeable from the data in Table 1 that all the values of correlation coefficient r S between the pulse amplitude of the disturbance voltage under differential and common modes and the breakdown voltage at the breaks of secondary signal ports with the 5 tested samples under the operation of the disconnecting switch are close to 1, which is within the range of very strong correlation, almost approximating perfect correlation. Therefore, the relationship between two variables can be expressed as a monotone function.

| Quantification and verification of the level of disturbance to ports
Available results show that the maximum breakdown voltage is 2.0 pu [21]. If a function expression is set up with the breakdown voltage as an independent variable and the pulse amplitude of the port disturbance voltage as a dependent variable, the maximum level of disturbance to secondary signal ports can be quantified by using this research conclusion.
Considering that the distribution of the actual measured data pair is unknown to the function, this work applies a quantile regression model based on the idea of data-driven modelling, which does not require any specific assumptions about the random error to obtain the function expression.
The expression of the unitary linear quantile regression model is as follows: where yðτ|xÞ is a dependent variable, x is an independent variable, β 0 ðτÞ and β 1 ðτÞ are unknown parameters to be estimated, and τ is the quantile, τ ∈ ð0; 1Þ.
As for the solution of the parameters β 0 ðτÞ and β 1 ðτÞ to be estimated in Equation (3), it can be transformed into an optimisation problem of the objective function Q½β 1 ðτÞ; β 0 ðτÞ�. See Equation (4) for the expression. In the formula, x i and y i represent the i th independent variable and the dependent variable, respectively.
Scatter event pairs (U 1n , U 2n ) under different test circuits and measurement contents are divided into two categories: sample set S and test set T. Among them, 380 scattered points extracted from 19 groups of test waveforms were randomly selected to form a sample set S to be used to train and obtain the function expression of the quantile regression model. The 20 scattered points extracted from the remaining 1 group of measurement waveforms were selected to form test set T to be used to test the predictive ability of the model.
According to Equation (4), the quantile regression equation under different quantiles can be calculated based on the data in the sample set S. Take sample set S of sample I (U 1Adn , U 2Adn ) extracted under test circuit A as an example, the calculated 0.05 quantile regression equation y ¼ 0:005x þ 63:98, factoring coefficient of determination R 2 ¼ 0:9861, the calculated 0.95 quantile regression equation y ¼ 0:00554x þ 88:34, and the factoring coefficient of determination R 2 ¼ 0:9861. The coefficients of determination of the fitting equations were all close to 1, indicating that the model could well fit the measured data and pass the significance test. Meanwhile, in order to further verify that the model can well predict the pulse amplitude of disturbance voltage at the secondary signal port obtained from a number of tests, the measured scatters (U 1Adn , U 2Adn ) in test set T and the quantile regression equation were presented as shown in Figure 11.
As can be seen in Figure 11, most of the measured scatters (>90% scatters) in test set T are distributed within the interval contained by the two straight lines of the 0.05 quantile regression equation and the 0.95 quantile regression equation. Likewise, the remaining 19 sample sets S and test sets T are processed accordingly. The distribution of the 0.05 quantile regression equation, 0.95 quantile regression equation and the test set T obtained by the training in the same diagram is similar to Figure 11, and this article will not show them one by one. In light of the above observations, it can be concluded that

| METHOD FOR QUANTIFYING THE LEVEL OF DISTURBANCE TO FIELD PORTS OF THE GIS SUBSTATION
For practical application in the GIS substation, the above method will face the following challenges: First, in the field circuit of the GIS substation, specialised sensors are usually not installed and it is difficult to obtain the breakdown voltage at breaks. Second, the VFTO sensor is constructed with the help of the existing components and the structure on GIS. In addition to the difficulty in construction, such sensor construction is limited to measure VFTO and other relevant parameters of the circuit, which is not conducive to its extended application to other circuits. Third, in actual substations, the number of tests allowed is usually greatly reduced, and further verification will be required to determine whether a small number of tested and measured samples can accurately quantify the level of disturbance to ports. For these reasons, this section mainly focusses on studying the level of disturbance to the ports in GIS substations using a quantitative approach.

| Analysis of the method for obtaining breakdown voltage at breaks
In GIS substations, disconnecting switches are generally used in conjunction with circuit breakers. Under normal circumstances, a disconnecting switch can only be operated after a circuit breaker is disconnected from the high load and the short bus in the section from the circuit breaker to the disconnecting switch is in suspension. Therefore, the operation of the disconnecting switch is, in essence, to disconnect and connect operations to the short bus between the disconnecting switch and circuit breaker, and its wiring pattern is shown in Figure 12.

| Voltage in the supply side
The frequency power U S in the power supply side is expressed as follows: where A 0 = Amplitude of supply voltage, w = Angular frequency at power frequency, t = Time, and φ = Initial phase of power supply.

| Voltage in load side
In the disconnected state, the suspended potential voltage of the short bus in the load side contains two components: 1.
Residual charge voltage component U Q retained by the short bus in the load side when the movable and static contacts of the disconnecting switch are separated and 2. Power frequency induced voltage component U C , which is determined by the DS 3 break capacitance C 3 and the no-load short bus capacitance C 4 voltage to the ground. Thus, the residual voltage U L of short bus can be expressed as follows: It should be noted that during the operation of the disconnecting switch, multiple breakdowns will occur in the break gaps and each breakdown will generate an arc, accompanied by a high-frequency transient oscillation process. When the arc is extinguished, the high-frequency transient process ends and the two ends of the break are at equal potential. At the same time, the short bus in the load side is in suspended potential, and its residual charge voltage is equal to the power frequency supply voltage in the supply side at the moment of arc extinguishment. Due to the high leak resistance of the insulator or short bus in GIS, the residual charge of the short bus attenuates slowly, usually consuming at least several hours [22]. Therefore, the residual charge voltage can be regarded as the DC component during switching operation. In addition, since the capacitance value C 4 of the no-load short bus to the ground is much higher than that of the disconnecting switch break capacitance C 3 , the instantaneous value of the induced voltage U C on the short bus C 4 is very small and is generally negligible. Thus, Equation (6) can be simplified as follows:

F I G U R E 1 1 Verification of Predictions Using Quantile Regression Model
The positional relationship between n th and n + 1 th breakdown high-frequency transients on the time axis during the DS 3 operation is shown in Figure 13. There are three time concepts in the figure: breakdown moments σ n and σ nþ1 , durations τ n and τ nþ1 of high-frequency transients, and interval of adjacent breakdown times. The duration of highfrequency transient is measured in microsecond, while the interval of the adjacent breakdown time is generally measured in millisecond [23]. So there is no overlap between two adjacent breakdowns. However, the residual charge of the n th breakdown bus determines that of the n + 1 th breakdown bus in the load side. The residual charge voltage U Qn at this moment of the n th breakdown of the short bus can be derived from Equation (8).
As the duration of high-frequency transient is generally measured in microsecond, which is far less than the cycle time of the power frequency supply voltage, that is τ n ≪ 20ms, the value of Equation (8) can be approximately represented by the value of power frequency voltage at any moment in the period of σ n ≤ t n ≤ σ n þ τ n , as follows:

| Breakdown voltage at break
The breakdown voltage of the gap between the DS' movable and static contacts is the difference between the voltage of movable and static contacts, namely the difference between the power frequency supply voltage and the residual voltage of the short bus. The breakdown voltage can be expressed as follows: Successively apply Equations (5) (6) (7) and (9) into Equation (10). Then the n th breakdown voltage can be derived as follows: It can be seen from Equation (11) that the breakdown voltage can be calculated as the absolute time t n and the initial phase φ under each breakdown are obtained.

| Design of the mobile capacitive voltage divider and verification by quantitative comparison
Since the sampling rate of the existing voltage measuring equipment in the field is 4 kHz, it is obvious that it fails to meet the requirement of time accuracy under Equation (11). In order to accurately obtain the breakdown time t n and the initial phase φ during switching operation, without causing connection with additional equipment, this work designed and developed a mobile capacitive voltage divider measuring device based on the principle of capacitor divider, the structural schematic diagram of which is shown in Figure 14. From the diagram, it can be seen that F plate is circular, the E plate is used as a shielding box, and the assembled voltage probe is placed on the back of the F plate.
During the test and measurement, move the shielding enclosure to the lower part of the overhead line at the power side of the circuit to be tested, and connect the shielding case with the nearest earthing location reliably. The electrode plate F of the shielding enclosure forms capacitor C 5 with the overhead line, and the electrode plates E and F form capacitor C 6 . C 5 and C 6 form a capacitive voltage divider in which C 6 ≫ C 5 . When precision equipment, such as an oscilloscope, photoelectric conversion components and lithium batteries, are placed inside the shielded enclosure, they can capture the steady and transient voltage waveforms at overhead lines during switching operation in a strong transient environment. Again, taking the measurement method of sample I under test circuit A and measuring the content of secondary disturbance voltage U d in differential mode as an example, the measured waveform of the mobile capacitive voltage divider measuring device is as shown in Figure 15. The left rectangular part represents the time period before the first transient voltage waveform is triggered, which is a steady-state sinusoidal voltage waveform. The measured data can be fitted to the F I G U R E 1 3 Schematic positional relationship on the adjacent breakdown time axis F I G U R E 1 2 Electrical schematic diagram of the disconnecting switch in GIS. DS, disconnecting switch; GIS, gas-insulated switchgear equation y ¼ 1:641 sinð0:3139x þ 1:255Þ using cftool in MATLAB, and the initial phase of the measured waveform is determined to be 1.255 rad. The right rectangular part contains multiple transient voltage waveforms. After the single micropulse transient voltage waveform is decomposed, the absolute time under each breakdown is approximately represented by the moment of the extreme value of the micro-pulse transient voltage waveform. Combined with Equation (11), the breakdown voltage U 1n at break during the operation can be calculated and derived.
In this way, the disconnecting switch was switched on and off for a total of four times and 80 data pairs (U 1Adn , U 2Adn ) were extracted. The measured scatter histogram and regression equation are shown in Figure 16.
According to the quantile regression equation in Figure 16, the interval distribution of the secondary side differential mode disturbance voltage U d under a different fracture breakdown voltage U 1n can be calculated. From the perspective of strict consideration, usually more attention is paid to the upper limit of the disturbance voltage. The deviation rates of the upper limit disturbance voltage quantified under different breakdown voltages in this section and Section 3.3 are shown in Table 2. It can be seen from the table that at the theoretical maximum value of 2.0 pu, the deviation rate of the quantised maximum disturbance voltage pulse amplitude under the two methods is only −4.34%. The above quantified results show that the maximum level of disturbance to secondary ports can still be quantified more accurately with a smaller number of test samples.
To sum up, the flowchart of determining the level of disturbance to the ports in GIS substations using quantification approach can be summarised as follows, as shown in Figure 17. In this approach, a mobile capacitive voltage divider is used to obtain the breakdown voltage at breaks which can be easily moved to the lower part of the overhead line on the power side of different circuits to be tested, without causing connection of additional equipment to the GIS test circuit. This method greatly expands the feasibility of the quantification method for the maximum level of disturbance to secondary ports proposed in this work in the field application of a GIS substation and is of good general applicability and prospect.

| CONCLUSION
Based on large numbers of tests and measurements on the 220 kV real simulation test platform, statistical analysis has been conducted on the correlation between the disturbance voltage pulse amplitude of secondary ports and the breakdown voltage between the breaks of a disconnecting switch. Then, a quantitative approach is introduced to determine the maximum level of disturbance to EVT signal acquisition ports under a GIS isolator switching operation. The main conclusions are as follows: (1) The findings of the Spearman rank correlation coefficient and quantile regression analysis showed that the disturbance voltage pulse amplitude of secondary ports and the breakdown voltage between the breaks of the the quantile regression model based on the test results can be used to accurately predict the interval of distribution of the disturbance voltage pulse amplitude of secondary ports corresponding to the breakdown voltage at the primary breaks specified. Considering that the theoretical maximum breakdown voltage at breaks is 2.0 pu, the upper limit of the secondary maximum disturbance voltage pulse amplitude can be quantified. (3) This work proposes a solution to the problem that a quantitative method is difficult to obtain breakdown voltage in the field application of a GIS substation. The quantitative results of four groups of measured data show that the deviation of the maximum disturbance voltage pulse amplitude of secondary ports quantised under this method is only −4.34%, which proves the feasibility of this method in the field application of substations. (4) The quantitative method proposed in this work can afford effective exploration into the practical value of a small amount of measured data under GIS substation field conditions. In the future, data from the tests which are carried out at various voltage levels and in various test circuits can be accumulated according to this method.
These data can be used to further supplement and improve EVT-related assessment standards.
T A B L E 2 Two methods quantify the deviation rate of the upper limit disturbance voltage of the secondary side under different breakdown voltages