Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Digital Twin of Distribution Power Transformer for Real-Time Monitoring of Medium Voltage from Low Voltage Measurements

Digital Twin of Distribution Power Transformer for Real-Time Monitoring of Medium Voltage from... Digital Twin of Distribution Power Transformer for Real-Time Monitoring of Medium Voltage from Low Voltage Measurements Panayiotis Moutis, Senior Member, IEEE, Omid Alizadeh-Mousavi heating and cooling, raises multiple technical concerns [1]. All Abstract—Real-time monitoring of distribution systems has be- actions to handle these concerns, both preventive and correc- come necessary, due to the deregulation of electricity markets and tive, are either limiting the deployment of more DER or curtail- the wide deployment of distributed energy resources. To monitor ing the energy that these may exchange with the grid [2]. Online voltage and current at sub-cycle detail, requires, typically, major monitoring of system behavior with high DER penetrations can investment undertaking and disruptions to the operation of the tackle efficiently most of the known technical concerns and oth- grid. In this work, measurements of the low voltage (LV) side of ers, not yet foreseen. Indicatively, system situational awareness distribution transformers (T/F) are used to calculate in real time the waveforms of their medium voltage (MV) sides, based on a (congestion, voltage stability, etc.) [3], use of innovative system mathematical model of said T/F. This model is, essentially, the dig- stability controls [4] as also traditional ones [5], assessment of ital twin of the MV side of the T/F. The method calculates T/F MV quality of supply, awareness of availability and timely activa- waveforms of voltage and current, and active and reactive power tion of flexibility services [6], and efficient fault identification as accurately as an instrument T/F, captures all harmonics con- and localization are few of the contributions of real-time system tent, is unaffected by asymmetrical loading and identifies most sys- monitoring in this context [7]. Secondly, online measurements tem faults on the MV side of the T/F. The digital twin method en- ables monitoring of distribution T/F that avoids MV instrumenta- of systems hosting DER and extracting metrics of average load- tion, does not suffer in accuracy and may be readily deployable. ing of the equipment will contribute positively in assessing in- Field data from an actual MV-LV T/F, agree with simulation re- vestment deference in grid infrastructure [8]. Thus, all available sults showcasing the efficacy of the digital twin method. grid hosting capacity can be made of use and avoid the tradi- tional oversizing design of power systems. In a similar sense, Index Terms—distribution, measurement, monitoring, trans- situations that affect equipment undesirably (e.g. inverse flows former, waveform. on relays) may be captured and handled accordingly. Thirdly, system operators seek to improve power quality and better serve NOMENCLATURE electric equipment according to performance standards [9]. A. Variables and Parameters A. Distribution System and Transformer Monitoring Review R Resistance. L Inductance. The above points have led to extensive research and innova- u Voltage. tive applications regarding monitoring medium (MV) & low i Current. voltage (LV) DSs either at the extent of whole feeders or as t Time. small as specific components in them. The methods and tech- n, N Discretized time steps and total (from sampling). nologies typically employed at transmission infrastructure are f Frequency. showing the way; transmission lines and substations are moni- tored either with Supervisory Control and Data Acquisition B. Scripts and indices (SCADA) systems [10] or, more recently, Phasor Measurement S, M Series and Shunt element/branch. Units (PMU), and operators monitor the grid at sub-second time 1, 2 Primary and Secondary transformer windings. scale [11]. At DSs, much research has focused on methods of s Sampling. state estimation [12, 13], PMU for this level specifically k Scenario numbering. (uPMU) [14], microgrid design and control and energy man- d, r Digital twin and real waveform. agement systems of various purposes and capacities [15]. In ac- A, B, C Three-phase phases. tual practice, there are different levels of deployments of data acquisition and managements systems at the distribution level I. INTRODUCTION of grids around the world. SCADA, distribution management any reasons have led to a growing interest in monitoring systems and Advanced Metering Infrastructure [16] – most re- M Distribution Systems (DSs) closely and in real time. cently with smart meters – have been those which have been Firstly, deployment of distributed energy resources (DER) in adopted more widely, while installations and use of uPMU have modern power systems, in order to electrify transportation, P. Moutis & O. Alizadeh-Mousavi are with DEPsys SA, Route du Verney 20B, 1070 Puidoux, Switzerland (email: panayiotis.moutis@depsys.ch; omid.mousavi@depsys.ch). This project has received funding from the European Union’s Horizon 2020 program under the Marie Sklodowska-Curie grant agree- ment No 797451. Preprint under Green Open Access policy based on monitoring of its LV side. The scope of this work spans DS monitoring in real time (for diagnosing faults and power quality issues) as also for long-term operating and planning assessments (feeder loading, hosting capacity, etc). The work contributes in the scoped field as follows: - Monitoring DSs at high time granularity, as anticipated by [12-16], is made possible, - The waveform monitoring allows to determine power qual- ity in real time [9], as it captures all harmonics content, - MV DS behavior under faults [7] can be captured fully to alert the system operator and logged for further analysis, - The MV-side waveforms outputted by the digital twin of the T/F are as accurate as the measurements of an instrument Fig. 1. Circuit models of a power T/F at various levels of detail. T/F [17] on the MV side of the actual T/F, been at earlier stages. Most of these works focus at the MV level - Technical personnel and system operator can assess imme- of DSs, which raises a concern of much practical interest – the diately any remedial actions to system events [6], use of voltage transformers for measurements [17]. Given the - The cost of measurement instrumentation at the LV side of extent of DS feeders, the costs, risks and network disruptions a power T/F is considerably lower than that of the MV side (e.g. replacement of cable headend) to deploy measurement or of monitoring both sides, thus, making the method less T/F, can gravely affect the goals for monitoring DS. costly for system operators with multiple feeders to cover, From the aspect of monitoring the operation and health of - Installation of LV side measurement devices requires the specific system components, power transformers (T/F) are at MV network to be interrupted under fewer circumstances, the center of attention of many studies. In [18], measurements thus, enabling a comparably seamless deployment, from both sides of the T/F help determine copper and iron losses - Works on uPMU and DS state estimation at the MV and the under harmonics-heavy conditions. In-rush and fault currents here proposed method may complement each other. are identified with wavelet methods in [19]. Harmonics analysis All the aforementioned points describe the digital twin as a is used in [20] to capture geomagnetic disturbance on T/F. Mon- method for monitoring the MV-side of a distribution T/F with itoring the higher-voltage side of the T/F in [21] assists in as- limited cost, no disruption to the DS for deployment, high ac- sessing the insulation performance under lightning impulses, curacy, and enabling fault diagnosis (previously unprecedented while frequency response analysis [22] can be an alternative to from monitoring the LV-side of feeders). locate faults in the equipment. Maintenance and ageing con- The innovation of this work is the real-time monitoring of cerns are also addressed more recently with greater attention the MV side of a distribution T/F without directly measuring [23, 24]. The value of measuring T/F performance, especially the MV side, but by rather measuring the LV-side of said T/F. for power quality matters, reflects also in the focus on instru- The remainder of this work is organized as follows. Section ment T/Fs accuracy [25]. As for using uPMUs to monitor T/F, II describes the methodology employed and the basic assump- their high costs and data requirements are prohibiting factors. tions both for single and three phase T/F. Section III describes the testing set-up, the phenomena that are assessed and the met- B. Innovation of the Power Transformer Digital Twin rics used. Section IV gathers all tests conducted and a discus- The following proposal is put forward in this paper. The MV sion ensues. Section V uses actual LV and MV measurements side of a distribution T/F may be monitored via its digital twin, from a DS T/F in Switzerland, to show that the digital twin instead of directly measuring its voltage and current waveforms method performs effectively and according to the statistical at its terminals. The digital twin [26] of the MV side will be testing results of Section IV. Section VI concludes this work informed by current and voltage waveform measurements of and proposes future considerations. the LV side of the distribution T/F. A digital twin is a digital replica of an actual entity, expected to behave and perform same II. METHODOLOGY AND REQUIRED ASSUMPTIONS to the actual entity. Hence, a digital twin offers insight to events A. Digital Twin of a Single-Phase Power Transformer that might occur to the actual entity or serve as the online model of it. The latter case is the one that this proposal is based on. The idea of the proposed methodology is based on taking The potential of digital twins in the power systems sector has measurements of voltage and current on the low-voltage side of only been recently acknowledged and explored [27]. the T/F and, by using the mathematical description of a typical To the authors’ knowledge there has been no effort like the model of single-phase T/F, calculate (as an estimate/projection) one presented in this manuscript. A digital twin of a power T/F what are the values of the voltage and current on the MV side has been proposed only as a means to assess its health status of the T/F. In this sense, the MV side of the T/F is emulated in [28], but not to monitor voltage, current and power in real time. silico; its digital twin is implemented. As well documented, T/F To monitor a DS MV-LV power T/F, voltage and current can be modeled as 2-port systems; either a pi- or a t-equivalent waveforms of the LV side are measured to calculate the circuit with the resistances, reactances lumped or split, accord- respective waveforms of the MV side. Hence, the proposed ing to the degree of detail sought for (see Fig. 1) [29]. Circuit work develops the digital twin of the MV side of the power T/F, model (a) is the full T/F model, circuit model (b) assumes that Fig. 3. Block diagram of simulation testing topology as set up in MATLAB. a sampling rate of f (sampling interval 1/f s). With the use of s s Fig. 2. Overall topology of the digital twin method for the calculation of the MV side waveforms of a DS T/F based on LV waveform measurements. the last two measurement samples, i.e. n and n-1, of voltage and current at every sampling interval n, and via formulation (1) the all series resistance and reactance is lumped on one side of the voltage and current of the MV side are retrieved. Measurements T/F as R =R +R and L =L +L , while circuit model (c) simpli- s 1 2 s 1 2 n and n-1 are required at every interval n, because of the effect fies (b) by assuming infinite impedance from L and R . Due to m m of the derivative and the integral in (1). The discretized its simplicity, circuit model (b) is the preferred digital twin of formulation of (1) is given, as the formal definition of the digital the MV side of the single-phase T/F for the proposed method- twin of the MV side of the T/F as by LV side measurements: ology. Additional detail will only improve the results that fol- ′ ′ ′ ′ 𝑢 [𝑛 ] = 𝑢 [𝑛 ] + 𝑅 𝑖 [𝑛 ] + 𝐿 (𝑖 [𝑛 ] − 𝑖 [𝑛 − 1])𝑓 low. As of circuit (b) and assuming that the T/F is sized and 2 1 𝑆 1 𝑆 1 1 𝑠 𝑢 [𝑛 ] 𝑢 [𝑛 ] − 𝑢 [𝑛 − 1] operated according to standard [30] (i.e. T/F core saturation is } (2) 2 2 2 𝑖 [𝑛 ] = + + 𝑖 [𝑛 ] 2 1 avoided, otherwise a piece-wise formulation [31] may comple- 𝑅 𝐿 ∙ 𝑓 𝑀 𝑀 𝑠 ment the following set-up), the MV calculations are as: With regards to harmonics content, formula set (2) is different 𝑑 𝑖 ′ (𝑡 ) 1 to the continuous set of (1), as the LV-side measurements must 𝑢 (𝑡 )= 𝑢 (𝑡 )+ 𝑅 𝑖 ′ (𝑡 )+ 𝐿 2 1 𝑆 1 𝑆 be properly sampled at high rates, so that no harmonic content (1) 𝑢 (𝑡 ) 1 is missed. This will be explored with assessment of multiple () () () 𝑖 𝑡 = + ∫𝑢 𝑡 + 𝑖 ′ 𝑡 2 2 1 sampling rates in Section IV and consequent analysis. 𝑅 𝐿 } 𝑀 𝑀 Where u, i, R and L are voltage, current, resistance and induct- B. Digital Twin of a Three-Phase Power Transformer ance, respectively. Resistances R and R may also be ex- S M For the digital twin of a three-phase T/F, the approach builds pressed as functions of temperature or of T/F loading (i.e. infer- on that of the single-phase as follows: the digital twins of three ring temperature), to allude to the effect of temperature on re- single-phase T/F, each taking separate single-phase voltage and sistances at different loads. Voltage and current are given as current measurements from the LV side of a three-phase T/F, time variables, since waveforms are measured. Subscripts 1, 2, are appropriately integrated to emulate the three-phase voltage S and M denote the LV and MV sides, series and shunt (leakage and current of the MV side of the T/F. Practically, voltage and and magnetizing impedances) parts of the single-phase T/F, re- current measurements of each phase on the LV side are used to spectively. Voltage and current measurements of the LV side calculate the corresponding values of one of the phases on the are referenced to the MV side (i.e. multiplied by the T/F ratio). MV side through (2). The overall digital twin topology is shown Any tap-changing action in the T/F is considered an input to the in Fig. 2. Following, the calculated values are elaborated ac- digital twin model. Alternatively, tap-changing can be moni- cording to the vector group of the T/F; i.e. the connection of the tored electrically by the digital twin and, thus, adjust values in three phase windings. For the most commonly T/F vector (1). Let it be stressed that (1) may also calculate any harmonics groups in DSs, the phase voltages and the line currents of the content either in the voltage or the current of the MV-side of the MV side for one of the phases are as follows: T/F, provided it is present in the LV-side measurements. The Yy0: 𝑢 = 𝑢 and 𝑖 = 𝑖 (3) 𝐴 2𝐴 𝐴 2𝐴 only concern with regards to this calculation stems from any Dy1: 𝑢 ∙ 3 = 𝑢 − 𝑢 and 𝑖 = 𝑖 − 𝑖 (4) 𝐴𝐵 2𝐴 2C 𝐴 2𝐴 2C filtering effects probably caused by the T/F impedance. These Dy11: 𝑢 ∙ √3 = 𝑢 − 𝑢 and 𝑖 = 𝑖 − 𝑖 (5) 𝐴𝐵 2𝐴 2𝐵 𝐴 2A 2B concerns will be thoroughly assessed in Section IV. Where subscripts A, B, C denote the three phases of the MV From (1) monitoring of the voltage and current waveforms side of the digital twin of the three-phase T/F, and subscript 2X of the LV side can yield voltage and current waveforms of the (where X = A, B, C) denotes the calculated values of the single MV side of the T/F. In detail, an AC voltage and current phase MV side digital twin from LV measurements via (2). measurement device is connected to the LV side of the single- Grounding either T/F side does not alter formulations (3-5). The phase T/F and measures the respective values. The voltage and digital twin of a T/F connected as Dy will not be able to calcu- current sensors provide the measurements in analog form and late MV-side current harmonics of orders multiples of the third, the measurement device converts the analog signals to digital at since such T/F topologies eliminate said harmonics [32]. 𝑑𝑡 𝑑𝑡 III. TESTING FRAMEWORK AND ASSESSMENT METRICS every scenario about 17000 simulations with random initial and final conditions are conducted, to ensure statistical validity of Retrieving the MV side waveforms from monitoring the the assessment metrics (see Subsection III.D) in the 99% confi- ones of the LV side of a triphase T/F has to be tested for various dence interval at ±1% error. The statistical analysis, according levels of loads connected to the T/F, changing load (increas- to [35] ensures confidence in the performance metrics collected. ing/decreasing), presence of harmonics in the system, non-bal- anced three-phase loads, non-balanced voltage supply and sys- B. Testing of Digital Twin Methodology under System Faults, tem faults. All these situations will affect the LV side wave- Unbalanced Loading and Tap-Changing Operations forms differently and, by that, the performance of the method- These tests will show if the accuracy of the digital twin is ology, too. A last parameter that needs to be considered is that affected by system faults, asymmetrical behavior either at the of the parameters of the measurement device used for the appli- MV or LV side of the T/F, and tap-changing action of the T/F. cation (i.e. sampling rate f ). As from the above, sets of testing The variations that need to be assessed are as follows: scenarios are defined in Subsection A. iv. All types of system faults (i.e. line-to-ground (LG), LL and All simulation tests are conducted in MATLAB. For most of LLG – occurring at 0.2s of 0.4s simulation), asymmetrical the tests, a three-phase T/F is tested, while a single-phase T/F is voltage magnitudes fed to the T/F, asymmetrical load con- used for the study of some specific concerns. The test topology nected to the LV side of the T/F and tap-changing action. (see Fig. 3) comprises a voltage source, feeding a load imped- v. Presence/absence of a Peterson coil at the nearest substation ance via a MV-LV T/F – all components are modeled precisely (electrically); the said coil is commonly used to reduce with ratings given in the Appendix. The described topology is ground fault currents and sustain three-phase LV operation. not unique and represents, practically, any typical DS T/F con- vi. Grounding or not of the Y connected side of the T/F (Dy1 nection. This means that any of the typical IEEE or other test or Dy11 vector groups might be assessed interchangeably, systems could be implemented around the described test topol- grounding of both sides of Yy needs to be considered). ogy of the DS T/F and the digital twin. The inclusion of any vii. Location of the fault at the MV or LV side of the T/F. DER in the test topology would also not affect the set-up, as the This set includes 72 scenarios. For each scenario a case of load direction of flow of currents does not affect the digital twin cal- about equal to the rated of the T/F is assumed to be served at culations. This T/F will be monitored through the proposed the time. The waveforms of the digital twin and of the actual methodology. This T/F is the full model as found in MATLAB MV side of the T/F will be compared on similarity of behavior libraries and has not been simplified, i.e. it represents the actual without the use of metrics. The reason for not using statistical T/F of the simulation tests with synthetic data. The methodol- metrics and as it will be noted in Section IV, is because the dig- ogy, as described in Section II, will sample the waveforms of ital twin of the MV side is either correctly calculating the actual voltage and current on the LV side of the T/F. A GridEye meas- waveforms of the T/F or completely miscalculates one or two urement device (class-A power quality monitoring certified) is phases by magnitude or phase shifting. In this sense, the statis- assumed to sample the waveforms [33]. GridEye is assumed to tical metrics as extracted from scenarios (i-iii) will either be have a random error in the interval of its accuracy (see Appen- valid for a correct calculation by the method in scenarios (iv- dix). It will then calculate the respective voltage and current vii) or the method fails substantially. All testing for the second waveforms and active and reactive power values of the MV side set will be conducted at the presence of harmonics (see details – i.e. its digital twin behavior. The digital twin behavior will be in (ii) above) as this represents the worst-case scenario for the compared for accuracy to the respective waveforms of the MV monitoring methodology here proposed. side of the actual T/F model (oscilloscope assumed connected). The metrics of accuracy used are defined in Subsection D. C. Filtering Effect of the Digital Twin on Harmonics The last set of tests aims to assess the effect of the T/F model A. Statistical Testing of Digital Twin Methodology used (see (1),(2) and Fig. 1) to extract the digital twin of the The first set of tests is that of normal operation varied ac- MV side of the T/F, with regards to whether it captures all volt- cording to the following conditions: age and current components. The tests will be conducted on i. Constant load, increasing load and decreasing load; the load MATLAB, but for a single-phase T/F, due to simplicity. Both increase/decrease (step change) will occur at the 0.02s mark the actual T/F and its models, as circuits, are, essentially, low- of a 0.04s simulation interval. pass filters, which damp the magnitude of higher-order harmon- ii. Presence or absence of harmonics in the system at the max- ics. That been said, the model of the T/F used for the digital imum allowed level described by standards [34]. twin of its MV side might be filtering higher harmonics differ- iii. Sampling rates of the measurement device at f = [5kHz, ently than the actual T/F, hence overestimating or underestimat- 10kHz, 30kHz, 52kHz]. ing the harmonic distortion at the MV level of the grid. To ad- For every option in (i-iii), a different scenario is described, thus, dress this concern, firstly, the Fourier transformation of voltage a total of 24 scenarios is tested. The above described conditions and current waveforms of the digital twin and the actual MV and the values assessed are non-linear. Hence, statistical analy- side of the T/F will be compared for simulations of changing sis will be employed [35]. This means that multiple similar load at the presence of harmonics (see details in (b) above). Sec- tests/simulations will be conducted for every scenario. From ondly, the Bode diagram of the model used for the digital twin these tests, averages, minima and maxima of the differences be- and that of the circuit of the actual T/F will be compared for tween actual and calculated waveforms will be calculated. For high and low loading of the T/F. TABLE I TABLE II STATISTICS OF ΔX FOR NORMAL OPERATION WITHOUT HARMONICS, GIVEN STATISTICS OF ΔX FOR NORMAL OPERATION WITHOUT HARMONICS, GIVEN 𝑡 ,𝑀 IN % OF RMS VALUE OF CORRESPONDING VARIABLE, FOR IN % OF RMS VALUE OF CORRESPONDING VARIABLE, FOR CONSTANT (0), CONSTANT (0), INCREASING (+) AND DECREASING (-) LOAD IMPEDANCE INCREASING (+) AND DECREASING (-) LOAD IMPEDANCE ΔL 0 + - ΔL 0 + - f f s s (kHz) (kHz) V 3.7 7.4 3.4 3.8 5.4 3.5 3.8 5.4 3.5 V 8.8 14.4 8.3 9.9 11.5 8.4 9.1 11.5 8.4 I 6.2 10.6 4.2 8.2 11.7 6.2 10.3 35.3 6.0 I 2.7 5.6 1.8 2.6 4.4 2.0 2.6 4.4 2.0 P 4.9 36.7 0.1 2.6 8.9 0.2 2.6 8.9 0.2 P 4.9 36.4 0.01 2.4 7.6 0.1 2.4 8.5 0.1 Q 8.6 >100 1.5 4.2 77.3 1.9 4.2 77.3 1.9 Q 8.6 >100 1.5 3.1 29.2 1.5 3.1 60.7 1.6 V 4.4 8.5 3.9 4.8 7.4 4.1 4.8 7.0 4.1 f 0.0 0.0 0.0 0.004 0.05 0.0 0.004 0.05 0.0 I 4.9 8.3 2.0 3.5 7.1 2.3 3.6 7.4 2.3 fI 0.0 0.0 0.0 0.02 0.08 0.0 0.02 0.08 0.0 10 P 4.8 39.0 0.1 2.5 7.9 0.1 2.7 10.6 0.2 V 1.9 4.7 1.6 2.1 3.8 1.7 2.1 3.5 1.7 Q 7.0 >100 1.5 4.0 40.4 1.9 4.0 57.8 1.8 I 2.5 4.9 0.9 1.6 3.3 0.9 1.6 3.4 0.9 V 1.9 7.6 1.4 2.3 4.7 1.6 2.3 4.9 1.7 P 4.8 38.7 0.01 2.4 7.6 0.1 2.3 9.8 0.1 I 4.7 8.4 0.8 3.0 4.8 1.0 3.0 5.0 0.9 Q 7.0 >100 1.5 3.0 29.2 1.5 3.5 50.4 1.5 P 5.0 36.7 0.1 2.6 8.9 0.1 2.6 8.3 0.1 f 0.0 0.0 0.0 0.002 0.03 0.0 0.002 0.03 0.0 Q 8.5 >100 1.5 3.9 76.5 1.8 3.9 71.2 1.8 f 0.0 0.0 0.0 0.02 0.06 0.0 0.02 0.2 0.0 I V 1.3 5.4 0.9 2.0 4.8 1.2 1.8 4.2 1.2 V 0.9 4.9 0.6 1.1 2.7 0.8 1.1 2.7 0.8 I 4.7 8.3 0.6 2.9 5.1 0.8 3.0 4.9 0.8 I 2.9 5.5 0.3 1.7 3.0 0.4 1.7 3.2 0.5 P 4.8 39.0 0.1 2.5 7.9 0.1 2.6 10.4 0.2 Q 7.0 >100 1.5 3.9 40.2 1.8 3.9 57.7 1.8 P 4.9 36.4 0.01 2.4 8.5 0.1 2.3 7.6 0.1 Q 8.5 >100 1.5 3.0 60.3 1.5 3.5 57.2 1.6 f 0.003 0.005 0.0 0.003 0.02 0.0 0.003 0.02 0.0 (𝑥 − 𝑥 ) 𝑑 ,𝑘 ,𝑛 𝑟 ,𝑘 ,𝑛 Δx = 𝑚𝑎𝑥 ( ) (7) 𝑡 ,𝑀 f 0.002 0.005 0.0 0.01 0.04 0.0 0.01 0.1 0.0 2 𝑟 ,𝑘 ,𝑀𝑆𝑅 V 0.7 3.5 0.4 0.9 2.6 0.6 0.9 2.4 0.6 I 3.1 5.6 0.3 1.8 3.4 0.4 1.9 3.2 0.5 The metrics describe the error between the digital twin and the P 4.8 38.7 0.01 2.4 7.6 0.1 2.3 9.7 0.1 actual MV side T/F waveforms for one simulation of a scenario. Q 7.0 >100 1.5 3.0 29.2 1.5 3.5 50.3 1.6 For the case of a load change ΔL at the point of 0.2s of the test- f 0.003 0.008 0.0 0.003 0.02 0.0 0.004 0.02 0.0 f 0.002 0.008 0.0 0.01 0.04 0.0 0.01 0.2 0.0 I ing interval, metrics Δx and Δx are adjusted for samples n 𝑡 𝑡 ,𝑀 in the time interval of t=0.16-0.24s, i.e. two electric cycles be- D. Assessment Metrics fore and after the load change. For the first set of scenarios, regarding all options in (i-iii), From all simulations of each scenario statistical measures the waveforms of voltage and current, values of active and re- will be extracted as here defined. The average, the maximum active powers, and frequency of the digital twin of the MV side and minimum of metrics Δx and Δx for all simulations of 𝑡 𝑡 ,𝑀 of the T/F (as calculated by the proposed methodology) will be each scenario are extracted. Hence, the metrics presented fol- compared to those of the actual T/F. Specifically for frequency, lowing are the metrics over the entirety of simulations of each the zero crossing of ten consecutive periods of waveforms of scenario tested, so as to ensure statistical validity. The analysis current and voltage (compared separately) will be averaged represents realistically the whole range of operating situations. over the time between the first and last crossing and the respec- tive frequency will be calculated and compared via a similar IV. RESULTS AND DISCUSSION elaboration of the actual waveform of voltage or current of the The Section is organized in two subsections, the first one MV side. For the said comparisons, metrics of average and summarizing the results per sets of scenarios with common point errors between the actual waveforms and that of the digi- characteristics and the second one discussing the findings in tal twin will be used, and are defined as follows. light of the efficacy of the proposed methodology. For one of the simulations in scenario k, assuming that x is d,k the waveform (voltage or current) of the digital twin of the MV A. Results side of the T/F and x the respective waveform of the actual r,k 1) Statistical Testing under Normal Operation: In Tables I- T/F on its MV side, the two will be compared point-to-point for II the statistics of metrics 𝛥 𝑥 and 𝛥 𝑥 for voltage, current, 𝑡 𝑡 ,𝑀 every time sample interval n. Firstly, the average error of a active and reactive power, and system frequency are presented waveform outputted by the digital twin normalized over the for the case that no harmonics are present in the grid, while Ta- RMS of the actual waveform of the T/F is defined as: bles III-IV the same metrics at the presence of all harmonics ∑ within the limits of [34]. The statistics given are average (Avg), (𝑥 − 𝑥 ) 𝑛 =1 𝑑 ,𝑘 ,𝑛 𝑟 ,𝑘 ,𝑛 ̃ (6) Δx = maximum (max) and minimum (min) for metrics 𝛥 𝑥 and 𝑥 ∙ 𝑁 𝑡 𝑟 ,𝑘 , 𝛥 𝑥 , accordingly. “0” denotes a scenario of constant load, “+” 𝑡 ,𝑀 where N is the sample points taken in the simulated testing time of increasing, and “-” of decreasing. V, I, P, Q, f and f are volt- V I interval of 0.4s. The maximum point error of the waveform out- age, current, active and reactive power, and frequency calcu- putted by the digital twin normalized over the RMS of the actual lated by the voltage and current waveforms, respectively. In waveform of the T/F is calculated by: Avg Max Min Avg Max Min Avg Max Min Avg Max Min Avg Max Min Avg Max Min 𝑅𝑀𝑆 TABLE III TABLE IV STATISTICS OF ΔX FOR NORMAL OPERATION WITH HARMONICS, GIVEN IN STATISTICS OF ΔX FOR NORMAL OPERATION WITH HARMONICS, GIVEN IN % 𝑡 ,𝑀 OF RMS VALUE OF CORRESPONDING VARIABLE, FOR % OF RMS VALUE OF CORRESPONDING VARIABLE, FOR CONSTANT (0), INCREASING (+) AND DECREASING (-) LOAD IMPEDANCE CONSTANT (0), INCREASING (+) AND DECREASING (-) LOAD IMPEDANCE ΔL 0 + - ΔL 0 + - f f s s (kHz) (kHz) V 5.0 8.1 4.8 5.1 6.3 4.8 3.8 5.1 3.4 V 37.5 43.2 37.0 37.9 40.2 37.2 9.0 11.3 8.3 I 8.7 26.0 5.4 15.3 50.0 8.4 44.5 >100 8.0 I 2.8 5.7 1.8 2.5 4.7 2.0 9.4 25.3 2.6 P 5.0 36.7 0.1 2.6 8.9 0.2 5.1 6.2 4.8 P 4.9 36.4 0.01 2.4 8.4 0.1 26.1 28.4 25.2 Q 8.7 >100 1.5 4.2 78.5 1.9 2.8 4.8 1.2 Q 8.6 >100 1.5 3.1 61.5 1.6 11.9 18.7 8.2 V 18.2 22.3 17.7 18.8 22.2 18.0 18.8 21.5 17.0 f 0.03 0.05 0.02 0.03 0.05 0.02 0.05 0.2 0.004 I 6.5 18.0 2.4 6.0 26.2 3.1 4.9 12.2 2.4 fI 0.02 0.05 0.0 0.02 0.08 0.0 0.07 1.4 0.04 10 P 4.8 39.0 0.1 2.5 7.9 0.1 2.7 10.6 0.2 V 2.5 5.0 2.3 2.6 4.1 2.4 2.6 3.9 2.3 Q 7.1 >100 1.5 4.0 40.8 1.9 4.0 57.6 1.8 I 2.5 5.1 0.9 1.6 3.4 0.9 1.7 3.5 1.0 V 7.3 13.3 6.8 7.7 10.2 6.8 7.6 10.7 5.8 P 4.8 38.7 0.01 2.4 7.6 0.1 2.3 9.8 0.1 I 6.2 11.2 1.0 3.9 6.4 1.1 3.9 6.6 1.0 Q 7.0 >100 1.5 3.0 29.4 1.5 3.5 50.3 1.6 P 5.0 36.7 0.1 2.6 8.9 0.1 2.6 8.3 0.1 f 0.01 0.02 0.003 0.01 0.03 0.003 0.02 0.05 0.003 Q 8.6 >100 1.5 3.9 77.5 1.8 3.9 70.9 1.8 f 0.01 0.03 0.0 0.02 0.07 0.0 0.02 0.2 0.0 I V 4.4 8.7 3.9 4.9 8.8 3.8 4.7 7.6 3.0 V 1.1 4.9 0.8 1.2 2.8 1.0 1.3 2.7 1.0 I 6.3 11.0 0.8 3.0 6.8 1.0 4.0 6.5 0.9 I 2.9 5.5 0.3 1.8 3.0 0.4 1.7 3.2 0.5 P 4.8 39.0 0.1 2.5 7.9 0.1 2.6 10.4 0.2 Q 7.0 >100 1.5 3.9 40.6 1.8 3.9 57.6 1.8 P 4.9 36.4 0.01 2.5 8.5 0.1 2.3 7.6 0.1 Q 8.5 >100 1.5 3.0 61.0 1.5 3.5 57.0 1.6 shifts) if: f 0.003 0.02 0.0 0.004 0.02 0.0 0.01 0.03 0.0 - the fault is LLG or LG and f 0.004 0.04 0.0 0.01 0.05 0.0 0.01 0.2 0.0 - the T/F is connected as Dy (regardless of grounding) or Yy V 0.8 3.5 0.3 1.0 2.7 0.7 1.0 2.4 0.7 with only one or neither of the two sides grounded and I 3.1 5.6 0.5 1.8 3.4 0.4 1.9 3.2 0.5 P 4.8 38.7 0.01 2.4 7.6 0.1 2.3 9.7 0.1 - the substation is grounded at its MV side. Q 7.0 >100 1.5 3.0 29.4 1.5 3.5 50.2 1.6 Phase voltages of the MV side of the T/F are also the only ones f 0.003 0.01 0.0 0.004 0.02 0.003 0.004 0.04 0.003 miscalculated at the occurrence of any type of fault on the LV f 0.002 0.04 0.003 0.01 0.04 0.0 0.01 0.06 0.0 side of the T/F. For the ease of the reader the rest of these plots are omitted. Tables II and IV, no statistics of metric 𝛥 𝑥 are noted for fre- 𝑡 ,𝑀 The phase voltages and line currents, at the occurrence of a quency. The reason is that frequency is averaged over multiple LG fault on the MV side of a Dy11 T/F grounded on its LV side electric cycles (See Section III.B), hence there is no point error and without grounding of the neutral of the voltage source are (and for that matter maximum point error). presented in Fig. 6. The sampling rate of the digital twin method Indicatively, some statistics are here read for the ease of the is at f =30kHz. The absence of grounding of the voltage source reader. From Table I, for the scenario of increasing load at s is equivalent to the substation upstream of the distribution T/F f =10kHz of the methodology with no harmonics in the system, the maximum of the average errors Δx of the active power of the digital twin over all simulations for this scenario is 7.6% with regards to the RMS of the actual active power of the T/F. From Table IV, for the scenario of constant load at f =30kHz of the method with harmonics in the system, the average of the maximum errors Δx of the current of the digital twin over all 𝑡 ,𝑀 simulations for this scenario is 6.2% with regards to the RMS of the actual active power of the T/F. For a load increase, the MV side waveforms of the digital twin at f =52kHz and the ac- tual T/F are given in Fig. 4, to depict the performance of the method. 2) System Faults, Unbalanced Loading and Tap-Changing Operations: The scenarios of system faults are presented first. In Fig. 5 the phase voltages and line currents, at the occurrence of a LG fault on the MV side of a Dy11 T/F grounded on its LV side and a coil grounding the neutral of the voltage source are presented. The sampling rate of the digital twin methodology is at f =30kHz. The coil emulates the Peterson coil connected to the substation upstream from distribution T/F. The digital twin methodology fails to calculate only phase voltages of the MV side of the T/F by phase and magnitude. The method fails also Fig. 4. MV side current and line-to-line voltage waveforms of the digital twin only for phase voltages (but for different magnitudes and phase (subscript e, f =52kHz) and of the actual T/F (subscript r) at a load increase. Avg Max Min Avg Max Min Avg Max Min Avg Max Min Avg Max Min Avg Max Min Fig. 5. MV side current and phase voltage waveforms of the digital twin (sub- Fig. 6. MV side current and phase voltage waveforms of the digital twin (sub- script e, f =30kHz) and of the actual T/F (subscript r) at a LG fault with a script e, f =30kHz) and of the actual T/F (subscript r) at a LG fault with a non- grounded voltage source supply. grounded voltage source supply. not being grounded on its MV side. In this case, the digital twin In Fig. 8 the Fourier transform of voltage and current wave- phase voltages of the MV side of the T/F and the actual ones forms of the MV side of the digital twin (f = 30kHz) and the compare successfully. The methodology succeeds similarly, if: actual T/F are given. A load increase from 10% to 100% load- - the fault is LL and regardless of any topology specifics for ing has been tested. Given how the suppression of the higher either the T/F or the substation or harmonics between the actual and the digital twin of the MV - the fault is either LG or LLG, the substation is not grounded th sides of the T/F is substantially different beyond the 20 har- on its LV side and regardless of the topology specifics for monic (see Fig. 7), the detail in the Fourier transform of voltage the distribution T/F or th th is focused on the interval of 20 -25 harmonics. - the fault is LG or LLG and the T/F is of the Yy topology grounded on both its sides and regardless of the topology specifics of the substation. For the ease of the reader the rest of these plots are omitted. It is stressed again that the digital twin method calculates correctly all line-to-line voltages, line currents, active and reac- tive powers of the MV side of the T/F under any type of fault, regardless of T/F and DS grounding topologies. Following, various cases of asymmetrical voltage magnitude of the source and asymmetrical load impedances were tested. In all cases, the digital twin methodology calculated correctly all voltage and current waveforms of the T/F MV side. As for tap- changing actions, the accuracy of the digital twin remains unaf- fected. This was expected as any tap-changing action only af- fects the parameters of the digital twin model, but not its math- ematical behavior, and, thus, its accuracy. For the ease of the reader all plots of these tests are omitted. 3) Filtering Effect of T/F and Digital Twin on Harmonics: The Bode diagrams of the circuits representing the digital twin and the actual T/F are first plotted. The respective transfer func- tions of circuit (b) (digital twin) and circuit (a) (actual T/F) in Fig. 1 are, first, calculated. Load equal to 10% and 100% of the rated apparent power of the T/F at 0.8 lag power factor has been tested, to assess how the T/F circuit models (acting as filters) Fig. 7. Bode diagrams of load voltage and current magnitude response to elec- suppress harmonic components (even beyond those mentioned tric frequency components up to 10kHz for nominal loading of the T/F for the in the standard [34]). Both loading tests had similar results and actual and the digital twin circuit models. the case of 100% loading is shown in Fig. 7. Fig. 9. Statistics of average error metric as of Δx for voltage of the MV digital twin of the T/F at the absence of harmonics. Fig. 8. Fourier transform of voltage and current waveforms of MV side for ac- (a) tual and digital twin (f =30kHz) of a T/F under a load increase. B. Discussion 1) Statistical Testing under Normal Operation: To assess the performance of the methodology under normal operation (Tables I-IV) both at the presence and absence of harmonics, Fig. 9 is here presented. The figure shows the envelope of max- imum and minimum of the average error metric 𝛥 𝑥 around the average of the said metric of the voltage waveform of the digital twin from that of the actual T/F MV sides for decreasing load (as taken from Table I). It is clear that the average is consider- ably closer to its minimum value than its maximum. This can be noticed for all values, for both metrics (𝛥 𝑥 and 𝛥 𝑥 ) in 𝑡 𝑡 ,𝑀 Tables I-IV. This observation translates as that the average be- (b) havior of the here proposed digital twin methodology is more Fig. 10. Expected error interval of waveform calculations by the digital twin of likely to be represented by its best performance rather than cal- the MV side of T/F compared to the actual ones at the absence (a) and presence (b) of harmonics and with constant load connected to the T/F. culate waveforms that deviate considerably from the ones of the actual T/F. This observation is even more substantial for the by the average metric at f ≥ 10kHz. Fig. 10 confirms this ob- statistics of the maximum error metric in 𝛥 𝑥 . The averages 𝑡 ,𝑀 servation as error tends to decrease as f increases. With limited of metric 𝛥 𝑥 can be used to define an interval around the 𝑡 ,𝑀 calculation errors of all MV waveforms as f increases, the com- averages of error metric 𝛥 𝑥 instead of the standard deviation plexity introduced by expressing T/F resistances as functions of of the latter, representing, thus, a more realistic error interval temperature (see Section II.A) can be avoided. for the waveform calculations of the digital twin methodology. Some particularly high maximum errors for both metrics for Fig. 10 shows exactly that, as the expected error interval for all the active and reactive powers may be noted in Tables I-IV. waveforms of the digital twin of the MV side of the T/F for the Those errors are noted to occur only at loading of the T/F less case of constant load. For either increasing or decreasing load than 5% of its nominal, thus, are not of concerning nature. Also, the respective graphs are similar (see Tables I-IV for the statis- it is reminded that Δx and Δx are normalized over the RMS 𝑡 𝑡 ,𝑀 tics) and have been omitted. of the respective waveform. Hence, if the normalization of Δx In terms of absolute values, let it be first reminded that in- and Δx was over the nameplate power instead of the RMS of 𝑡 ,𝑀 strument voltage transformers are classified according to their the waveform, they would be negligibly small, i.e. less than 4%. accuracy errors which span the interval 0.1-3% over the voltage As for the frequency calculated by the digital twin, the aver- RMS value [17]. It may be seen that at the absence of harmonics ages of the average error metric Δx is about the same regard- and for f ≥ 30kHz even the average maximum errors fall in this less on whether the frequency is calculated by the voltage or the interval for the proposed digital twin methodology. At the pres- current waveform (Tables I and III), with a slight favorite of the ence of harmonics, the digital twin methodology can have ac- voltage. Nevertheless, from the maximum errors of the metric curacy performance comparable to that of the voltage T/F only TABLE V METRICS OF ΔX AND ΔX BETWEEN THE DIGITAL TWIN (AT 13.2 KHZ 𝑡 𝑡 ,𝑀 SAMPLING RATE) AND THE ACTUAL WAVEFORMS OF THE MV-SIDE OF A DISTRIBUTION TRANSFORMER IN SWITZERLAND, GIVEN IN % OF RMS VALUE OF CORRESPONDING VARIABLE Dataset 1 Dataset 2 Dataset 3 Dataset 4 ̃ ̃ ̃ ̃ Δx Δx Δx Δx Δx Δx Δx Δx 𝑡 ,𝑀 𝑡 ,𝑀 𝑡 ,𝑀 𝑡 ,𝑀 𝑡 𝑡 𝑡 𝑡 V 3.8 7.1 2.9 5.4 2.9 5.0 2.9 6.1 V 4.7 8.7 3.6 7.0 3.7 6.1 3.9 8.2 V 3.9 7.3 2.9 5.5 3.0 5.5 3.1 6.4 I 12.7 38.5 14.6 27.9 12.5 25.4 15.7 55.4 I 15.8 38.0 16.3 29.8 13.7 26.2 18.9 51.7 I 19.7 61.6 19.2 34.7 16.8 32.2 26.6 80.8 P 20.2 24.6 20.3 21.9 18.2 19.9 24.7 28.0 Q 5.8 21.3 6.6 13.3 4.3 11.4 5.5 12.3 it is made clear that the voltage waveform should be preferred Fig. 11. Topology of the collection of field data from the actual T/F in Switzer- to estimate system frequency with better accuracy. land to assess the validity of the digital twin methodology. Harmonics affect the accuracy of the voltage waveform out- (regardless of grounding), YGy and Yyg T/F at the occurrence putted by the digital twin much more than any other value (cur- of faults on the MV side of such T/F [29]. Hence, if LG or LLG rent, powers, frequency). Comparing Tables I and III, for the faults occur and the MV side of the DS is grounded (with same type of load change and f , one can notice that statistics Peterson coil or other), the zero sequence circuit at the MV side for all values but for voltage are similar. Especially for voltage, of the said T/F is subject to voltage and current that are not the metrics of the average errors are greater in Table III that propagated to the LV side, hence the method cannot calculate tested for voltage supply with harmonics. This remark can be phase voltages correctly. explained by the analysis of transfer function responses with the The digital twin performs similarly for faults on the LV side Bode diagrams in Fig. 7, as also the circuit itself. The shunt el- of the T/F, too, because the voltage measurements of the ements of the T/F pose practically infinite impedance to the cur- method taken on the LV side of the T/F see towards the fault. rents through it. Hence, the voltage drops caused by the series I.e. due to the, typically, much smaller fault impedance, those elements (impedances with S, 1, 2 subscripts in Fig. 1) are those measurements see close to zero voltage on the affected phases, that affect the accuracy of the methodology, while there is no hence, via (2) fail on MV side calculations. Nevertheless, this current divider to affect the accuracy from the perspective of concern is not critical, since the priority of system operators in currents. In Fig. 7 it may be noted that higher frequency com- such occasions would be the LV fault (captured by the method). ponents of voltage are damped more substantially by the digital twin model compared to the actual of the T/F. This, in turn, V. FIELD TEST VALIDATION OF DIGITAL TWIN OF means that the digital twin methodology model will amplify any DISTRIBUTION POWER TRANSFORMER METHODOLOGY measurement errors (due to sampling in the context of this To assess the validity of the proposed method, current and study) more than what would happen if the actual T/F model voltage waveform measurements from an actual DS MV-LV was used as the digital twin. However, as from the Fourier T/F in Switzerland, are used (for T/F characteristics, see Ap- transform analysis in Fig. 8, it is clear that the energy contents pendix). The waveform measurements were logged with of voltage of the actual and digital twin models of the MV side GridEye devices [33] on both its LV and MV sides in sync (see of the T/F for the frequencies that are most affected (as shown Fig. 11). GridEye, although not designed to operate as a PMU, by the Bode diagrams in Fig. 7) are almost identical for a f of it largely abides by the measurement requirements (see Appen- 30kHz. This also confirms findings noted in the previous para- dix) of the respective standard [36]. graph. In other words, preferring the actual T/F model (circuit LV waveform measurements of the actual T/F are inputted (a) in Fig. 1) instead of the one used (circuit (b) in Fig. 1) will to the digital twin method, and the output of calculated MV not improve substantially the accuracy of the digital twin, if the waveforms of the digital twin is compared to the respective MV f is adequately high. To conclude these observations, the accu- waveform measurements of the said T/F. Four sets of waveform racy of the digital twin output is unaffected by the presence of data, of about 10 minutes each, sampled at 13.2kHz are used. harmonics (i.e. calculates MV waveforms properly), provided To assess the performance of the method Table V presents the that the sampling rate of the input waveforms is high enough. metrics defined by Δx and Δx as the average and maximum 2) System Faults, Unbalanced Loading and Tap-Changing 𝑡 𝑡 ,𝑀 Operations: With regards to the performance of the digital twin errors of the waveform outputted by the digital twin method method for the case of system faults, the following are noted. normalized over the RMS of the actual waveform of the T/F. The method relies on measurements of the LV side of the T/F, As the sampling rate of the input LV waveforms is 13.2kHz hence, unless the LV side is implicated in a phenomenon, the and there is ample harmonic content as the data sets show (see method will fail to calculate phase voltages. In this sense, it is Fig. 12), the results in Table V will be compared to those of here reminded that no currents flow or voltages are induced to Tables III and IV for sampling rates of 10kHz and 30kHz. the LV side of the zero sequence circuit component of Dy actual T/F in Fig. 12. The field data clearly shows that the dig- ital twin methodology can accurately calculate the MV-side be- havior of a T/F based on LV-side waveform input. VI. CONCLUSION This work proposes a method to monitor the MV-side wave- forms of DS T/F, to enhance the visibility of operators into DS. The method is the digital twin of the MV side of a DS T/F. It has high calculation accuracy and captures most system faults and harmonics content. The accuracy of the digital twin in- creases as the sampling rate of the input LV waveforms in- creases, and is comparable to that of an instrument T/F. Next step in the assessment of the proposed technique is to deploy the method on a DS monitoring device [33] to a MV-LV T/F and assess its field performance. Thus, the technique will also be realistically assessed in terms of installation costs and disruption. Practically, the digital twin method is expected to offer much visibility of the MV DS to DS operators. Moreover, the health of the T/F can be monitored through its performance. A first research extension that can be explored on the T/F digital twin method, is its improvement in cases of compensated MV DS ground faults, if not both the T/F primary and second- ary windings are grounded. The proposed digital twin T/F set- up can calculate accurately MV current flows and line-to-line voltages under all MV system fault conditions, but fails to cal- culate the phase voltages (indicator of nature of fault) in the aforementioned specific cases. Another extension can be the ap- plication of the method to MV to high voltage T/F to monitor the transmission system with reduce costs and grid disruptions. VII. APPENDIX A. Simulation Testing on MATLAB Fig. 12. MV side current and phase voltage waveforms, and active and reactive power of the digital twin (subscript e, f =13.2 kHz) and of the MV measurement MV-LV DS 3-phase T/F parameters: data of the actual T/F in Switzerland (subscript r). S = 50 kVA, V /V = 400V/20kV, R = 0.0075 pu, L = 0.02 pu, 1 2 1 1 As it may be noted all voltage, active and reactive power met- R = 0.0075 pu, L = 0.02 pu, R = 500 pu, L = 500 pu. 2 2 m m rics are within the intervals of the statistical testing with the LV 3-phase lumped varying load parameters: synthetic/simulation data. The only exception is with the met- R = [0.75-5.25] Ω, L = [1.5-17] mH. L L rics of the currents, which are greater than the maxima of both MV 3-phase line parameters: Δx and Δx . Since the metrics of voltages and powers are 𝑡 𝑡 ,𝑀 R = 2 Ω, L = 1 mH. LN LN within the intervals of the statistical results collected from the B. Digital Twin Tests with Field Data from T/F in Switzerland simulations with synthetic data and the current is a function of them, the noted error metrics of current are a result of how these LV-MV DS 3-phase T/F parameters: metrics are normalized over the RMS of the respective wave- S = 630 kVA, V /V = 400V/20.5kV, R = 0.0035 pu, 1 2 1 form. In fact, the T/F loading across all four datasets was par- L = 0.0233 pu, R = 0.0035 pu, L = 0.0233 pu, R = 500 pu, 1 2 2 m ticularly low (less than 5% of the nameplate 630kVA); i.e. if L = 500 pu. the normalization of the error metrics was over the nominal cur- GridEye measurement accuracy characteristics: rent instead the RMS of the waveform, the errors would be neg- Voltage magnitude = 0.1%, ligibly small, i.e. no more than 4%. Similar observations were Total Vector Error (see [36]) < 1%, made for the high error metrics of the active and reactive power Current Magnitude = 1%, of the digital twin for the statistical testing with synthetic data Frequency = ±1 mHz. (see Section IV.B). The generally positive behavior of the digital twin method, VIII. REFERENCES as shown by the above error metrics analysis, is corroborated [1] N. D. Hatziargyriou, E. I. Zountouridou, A. Vassilakis, P. Moutis, C. N. by the comparison of the waveforms of the MV-side of the T/F Papadimitriou, and A. G. Anastasiadis, "Overview of Distributed Energy between the output of the digital twin and the datasets of the Resources," Smart Grid Handbook, pp. 1-44, 2016. [2] L. Bird et al., "Wind and solar energy curtailment: A review of [19] A. Gaouda and M. Salama, "DSP wavelet-based tool for monitoring international experience," Renewable and Sustainable Energy Reviews, transformer inrush currents and internal faults," IEEE Transactions on vol. 65, pp. 577-586, 2016. Power Delivery, vol. 25, no. 3, pp. 1258-1267, 2010. [3] M. M. Esfahani and G. R. Yousefi, "Real time congestion management in [20] E. E. Bernabeu, "Single-phase transformer harmonics produced during power systems considering quasi-dynamic thermal rating and congestion geomagnetic disturbances: theory, modeling, and monitoring," IEEE clearing time," IEEE Transactions on Industrial Informatics, vol. 12, no. Transactions on Power Delivery, vol. 30, no. 3, pp. 1323-1330, 2014. 2, pp. 745-754, 2016. [21] P. Sun, W. Sima, M. Yang, X. Lan, and J. Wu, "Study on voltage-number [4] M. Tarafdar Hagh and T. Khalili, "A review of fault ride through of PV characteristics of transformer insulation under transformer invading non- and wind renewable energies in grid codes," International Journal of standard lightning impulses," IEEE transactions on dielectrics and Energy Research, vol. 43, no. 4, pp. 1342-1356, 2019. electrical insulation, vol. 22, no. 6, pp. 3582-3591, 2015. [5] A. Jafari, H. G. Ganjehlou, T. Khalili, B. Mohammadi-Ivatloo, A. Bidram, [22] R. K. Senobari, J. Sadeh, and H. Borsi, "Frequency response analysis and P. Siano, "A two-loop hybrid method for optimal placement and (FRA) of transformers as a tool for fault detection and location: A scheduling of switched capacitors in distribution networks," IEEE Access, review," Electric Power Systems Research, vol. 155, pp. 172-183, 2018. vol. 8, pp. 38892-38906, 2020. [23] R. M. A. Velásquez, J. V. M. Lara, and A. Melgar, "Converting data into [6] F. Hu, K. Sun, A. Del Rosso, E. Farantatos, and N. Bhatt, "Measurement- knowledge for preventing failures in power transformers," Engineering based real-time voltage stability monitoring for load areas," IEEE Failure Analysis, vol. 101, pp. 215-229, 2019. Transactions on Power Systems, vol. 31, no. 4, pp. 2787-2798, 2015. [24] A. Bracale, G. Carpinelli, and P. De Falco, "Probabilistic risk-based [7] W. Santos, F. Lopes, N. Brito, and B. Souza, "High-impedance fault management of distribution transformers by dynamic transformer rating," identification on distribution networks," IEEE Transactions on Power International Journal of Electrical Power & Energy Systems, vol. 113, Delivery, vol. 32, no. 1, pp. 23-32, 2016. pp. 229-243, 2019. [8] H. L. Willis, Distributed power generation: planning and evaluation. Crc [25] M. Faifer, C. Laurano, R. Ottoboni, S. Toscani, and M. Zanoni, Press, 2000. "Harmonic Distortion Compensation in Voltage Transformers for [9] M. M. Albu, M. Sănduleac, and C. Stănescu, "Syncretic use of smart Improved Power Quality Measurements," IEEE Transactions on meters for power quality monitoring in emerging networks," IEEE Instrumentation and Measurement, 2019. Transactions on Smart Grid, vol. 8, no. 1, pp. 485-492, 2016. [26] S. Boschert and R. Rosen, "Digital twin—the simulation aspect," in [10] M. S. Thomas and J. D. McDonald, Power system SCADA and smart Mechatronic Futures: Springer, 2016, pp. 59-74. grids. CRC press, 2017. [27] A. Ebrahimi, "Challenges of developing a digital twin model of [11] A. Monticelli, "Electric power system state estimation," Proceedings of renewable energy generators," in 2019 IEEE 28th International the IEEE, vol. 88, no. 2, pp. 262-282, 2000. Symposium on Industrial Electronics (ISIE), 2019: IEEE, pp. 1059-1066. [12] P. P. Moutis and O. Alizadeh-Mousavi, "A Practical Proposal for State [28] Y. Yang et al., "State Evaluation of Power Transformer Based on Digital Estimation at Balanced, Radial Distribution Systems," in 2019 IEEE PES Twin," in 2019 IEEE International Conference on Service Operations and Innovative Smart Grid Technologies Europe (ISGT-Europe): IEEE, pp. 1- Logistics, and Informatics (SOLI), 2019: IEEE, pp. 230-235. 5. [29] J. J. Grainger and W. D. Stevenson, Power system analysis. 2003. [13] A. Primadianto and C.-N. Lu, "A review on distribution system state [30] I. Std, "60076-8,“," Power transformers. Part, vol. 8. estimation," IEEE Transactions on Power Systems, vol. 32, no. 5, pp. [31] P. Kundur, N. J. Balu, and M. G. Lauby, Power system stability and 3875-3883, 2017. control. McGraw-hill New York, 1994. [14] A. Von Meier, E. Stewart, A. McEachern, M. Andersen, and L. [32] A. A. Sallam and O. P. Malik, Electric distribution systems. John Wiley Mehrmanesh, "Precision micro-synchrophasors for distribution systems: & Sons, 2018. A summary of applications," IEEE Transactions on Smart Grid, vol. 8, [33] M. Carpita, A. Dassatti, M. Bozorg, J. Jaton, S. Reynaud, and O. no. 6, pp. 2926-2936, 2017. Mousavi, "Low Voltage Grid Monitoring and Control Enhancement: The [15] C. Chen, J. Wang, F. Qiu, and D. Zhao, "Resilient distribution system by GridEye Solution," in 2019 International Conference on Clean Electrical microgrids formation after natural disasters," IEEE Transactions on smart Power (ICCEP), 2019: IEEE, pp. 94-99. grid, vol. 7, no. 2, pp. 958-966, 2015. [34] I.-I. E. Commission, "IEC TR 61000-2-5," 2011. [16] J. Peppanen, M. J. Reno, R. J. Broderick, and S. Grijalva, "Distribution [35] W. A. Fuller, Sampling statistics. John Wiley & Sons, 2011. system model calibration with big data from AMI and PV inverters," [36] K. E. Martin and I. P. S. R. Comm, "Synchrophasor Measurements Under IEEE Transactions on Smart Grid, vol. 7, no. 5, pp. 2497-2506, 2016. the IEEE Standard C37.118.1-2011 With Amendment C37.118.1a," Ieee [17] I. Standard, "61869-3, 2011,“," Instrument transformers–Part, vol. 3. Transactions on Power Delivery, vol. 30, no. 3, pp. 1514-1522, Jun 2015, [18] D. Lin and E. F. Fuchs, "Real-time monitoring of iron-core and copper doi: 10.1109/tpwrd.2015.2403591. losses of transformers under (non) sinusoidal operation," IEEE Transactions on Power Delivery, vol. 21, no. 3, pp. 1333-1341, 2006. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Electrical Engineering and Systems Science arXiv (Cornell University)

Digital Twin of Distribution Power Transformer for Real-Time Monitoring of Medium Voltage from Low Voltage Measurements

Loading next page...
 
/lp/arxiv-cornell-university/digital-twin-of-distribution-power-transformer-for-real-time-V0RKaVr9Th
ISSN
0885-8977
eISSN
ARCH-3348
DOI
10.1109/TPWRD.2020.3017355
Publisher site
See Article on Publisher Site

Abstract

Digital Twin of Distribution Power Transformer for Real-Time Monitoring of Medium Voltage from Low Voltage Measurements Panayiotis Moutis, Senior Member, IEEE, Omid Alizadeh-Mousavi heating and cooling, raises multiple technical concerns [1]. All Abstract—Real-time monitoring of distribution systems has be- actions to handle these concerns, both preventive and correc- come necessary, due to the deregulation of electricity markets and tive, are either limiting the deployment of more DER or curtail- the wide deployment of distributed energy resources. To monitor ing the energy that these may exchange with the grid [2]. Online voltage and current at sub-cycle detail, requires, typically, major monitoring of system behavior with high DER penetrations can investment undertaking and disruptions to the operation of the tackle efficiently most of the known technical concerns and oth- grid. In this work, measurements of the low voltage (LV) side of ers, not yet foreseen. Indicatively, system situational awareness distribution transformers (T/F) are used to calculate in real time the waveforms of their medium voltage (MV) sides, based on a (congestion, voltage stability, etc.) [3], use of innovative system mathematical model of said T/F. This model is, essentially, the dig- stability controls [4] as also traditional ones [5], assessment of ital twin of the MV side of the T/F. The method calculates T/F MV quality of supply, awareness of availability and timely activa- waveforms of voltage and current, and active and reactive power tion of flexibility services [6], and efficient fault identification as accurately as an instrument T/F, captures all harmonics con- and localization are few of the contributions of real-time system tent, is unaffected by asymmetrical loading and identifies most sys- monitoring in this context [7]. Secondly, online measurements tem faults on the MV side of the T/F. The digital twin method en- ables monitoring of distribution T/F that avoids MV instrumenta- of systems hosting DER and extracting metrics of average load- tion, does not suffer in accuracy and may be readily deployable. ing of the equipment will contribute positively in assessing in- Field data from an actual MV-LV T/F, agree with simulation re- vestment deference in grid infrastructure [8]. Thus, all available sults showcasing the efficacy of the digital twin method. grid hosting capacity can be made of use and avoid the tradi- tional oversizing design of power systems. In a similar sense, Index Terms—distribution, measurement, monitoring, trans- situations that affect equipment undesirably (e.g. inverse flows former, waveform. on relays) may be captured and handled accordingly. Thirdly, system operators seek to improve power quality and better serve NOMENCLATURE electric equipment according to performance standards [9]. A. Variables and Parameters A. Distribution System and Transformer Monitoring Review R Resistance. L Inductance. The above points have led to extensive research and innova- u Voltage. tive applications regarding monitoring medium (MV) & low i Current. voltage (LV) DSs either at the extent of whole feeders or as t Time. small as specific components in them. The methods and tech- n, N Discretized time steps and total (from sampling). nologies typically employed at transmission infrastructure are f Frequency. showing the way; transmission lines and substations are moni- tored either with Supervisory Control and Data Acquisition B. Scripts and indices (SCADA) systems [10] or, more recently, Phasor Measurement S, M Series and Shunt element/branch. Units (PMU), and operators monitor the grid at sub-second time 1, 2 Primary and Secondary transformer windings. scale [11]. At DSs, much research has focused on methods of s Sampling. state estimation [12, 13], PMU for this level specifically k Scenario numbering. (uPMU) [14], microgrid design and control and energy man- d, r Digital twin and real waveform. agement systems of various purposes and capacities [15]. In ac- A, B, C Three-phase phases. tual practice, there are different levels of deployments of data acquisition and managements systems at the distribution level I. INTRODUCTION of grids around the world. SCADA, distribution management any reasons have led to a growing interest in monitoring systems and Advanced Metering Infrastructure [16] – most re- M Distribution Systems (DSs) closely and in real time. cently with smart meters – have been those which have been Firstly, deployment of distributed energy resources (DER) in adopted more widely, while installations and use of uPMU have modern power systems, in order to electrify transportation, P. Moutis & O. Alizadeh-Mousavi are with DEPsys SA, Route du Verney 20B, 1070 Puidoux, Switzerland (email: panayiotis.moutis@depsys.ch; omid.mousavi@depsys.ch). This project has received funding from the European Union’s Horizon 2020 program under the Marie Sklodowska-Curie grant agree- ment No 797451. Preprint under Green Open Access policy based on monitoring of its LV side. The scope of this work spans DS monitoring in real time (for diagnosing faults and power quality issues) as also for long-term operating and planning assessments (feeder loading, hosting capacity, etc). The work contributes in the scoped field as follows: - Monitoring DSs at high time granularity, as anticipated by [12-16], is made possible, - The waveform monitoring allows to determine power qual- ity in real time [9], as it captures all harmonics content, - MV DS behavior under faults [7] can be captured fully to alert the system operator and logged for further analysis, - The MV-side waveforms outputted by the digital twin of the T/F are as accurate as the measurements of an instrument Fig. 1. Circuit models of a power T/F at various levels of detail. T/F [17] on the MV side of the actual T/F, been at earlier stages. Most of these works focus at the MV level - Technical personnel and system operator can assess imme- of DSs, which raises a concern of much practical interest – the diately any remedial actions to system events [6], use of voltage transformers for measurements [17]. Given the - The cost of measurement instrumentation at the LV side of extent of DS feeders, the costs, risks and network disruptions a power T/F is considerably lower than that of the MV side (e.g. replacement of cable headend) to deploy measurement or of monitoring both sides, thus, making the method less T/F, can gravely affect the goals for monitoring DS. costly for system operators with multiple feeders to cover, From the aspect of monitoring the operation and health of - Installation of LV side measurement devices requires the specific system components, power transformers (T/F) are at MV network to be interrupted under fewer circumstances, the center of attention of many studies. In [18], measurements thus, enabling a comparably seamless deployment, from both sides of the T/F help determine copper and iron losses - Works on uPMU and DS state estimation at the MV and the under harmonics-heavy conditions. In-rush and fault currents here proposed method may complement each other. are identified with wavelet methods in [19]. Harmonics analysis All the aforementioned points describe the digital twin as a is used in [20] to capture geomagnetic disturbance on T/F. Mon- method for monitoring the MV-side of a distribution T/F with itoring the higher-voltage side of the T/F in [21] assists in as- limited cost, no disruption to the DS for deployment, high ac- sessing the insulation performance under lightning impulses, curacy, and enabling fault diagnosis (previously unprecedented while frequency response analysis [22] can be an alternative to from monitoring the LV-side of feeders). locate faults in the equipment. Maintenance and ageing con- The innovation of this work is the real-time monitoring of cerns are also addressed more recently with greater attention the MV side of a distribution T/F without directly measuring [23, 24]. The value of measuring T/F performance, especially the MV side, but by rather measuring the LV-side of said T/F. for power quality matters, reflects also in the focus on instru- The remainder of this work is organized as follows. Section ment T/Fs accuracy [25]. As for using uPMUs to monitor T/F, II describes the methodology employed and the basic assump- their high costs and data requirements are prohibiting factors. tions both for single and three phase T/F. Section III describes the testing set-up, the phenomena that are assessed and the met- B. Innovation of the Power Transformer Digital Twin rics used. Section IV gathers all tests conducted and a discus- The following proposal is put forward in this paper. The MV sion ensues. Section V uses actual LV and MV measurements side of a distribution T/F may be monitored via its digital twin, from a DS T/F in Switzerland, to show that the digital twin instead of directly measuring its voltage and current waveforms method performs effectively and according to the statistical at its terminals. The digital twin [26] of the MV side will be testing results of Section IV. Section VI concludes this work informed by current and voltage waveform measurements of and proposes future considerations. the LV side of the distribution T/F. A digital twin is a digital replica of an actual entity, expected to behave and perform same II. METHODOLOGY AND REQUIRED ASSUMPTIONS to the actual entity. Hence, a digital twin offers insight to events A. Digital Twin of a Single-Phase Power Transformer that might occur to the actual entity or serve as the online model of it. The latter case is the one that this proposal is based on. The idea of the proposed methodology is based on taking The potential of digital twins in the power systems sector has measurements of voltage and current on the low-voltage side of only been recently acknowledged and explored [27]. the T/F and, by using the mathematical description of a typical To the authors’ knowledge there has been no effort like the model of single-phase T/F, calculate (as an estimate/projection) one presented in this manuscript. A digital twin of a power T/F what are the values of the voltage and current on the MV side has been proposed only as a means to assess its health status of the T/F. In this sense, the MV side of the T/F is emulated in [28], but not to monitor voltage, current and power in real time. silico; its digital twin is implemented. As well documented, T/F To monitor a DS MV-LV power T/F, voltage and current can be modeled as 2-port systems; either a pi- or a t-equivalent waveforms of the LV side are measured to calculate the circuit with the resistances, reactances lumped or split, accord- respective waveforms of the MV side. Hence, the proposed ing to the degree of detail sought for (see Fig. 1) [29]. Circuit work develops the digital twin of the MV side of the power T/F, model (a) is the full T/F model, circuit model (b) assumes that Fig. 3. Block diagram of simulation testing topology as set up in MATLAB. a sampling rate of f (sampling interval 1/f s). With the use of s s Fig. 2. Overall topology of the digital twin method for the calculation of the MV side waveforms of a DS T/F based on LV waveform measurements. the last two measurement samples, i.e. n and n-1, of voltage and current at every sampling interval n, and via formulation (1) the all series resistance and reactance is lumped on one side of the voltage and current of the MV side are retrieved. Measurements T/F as R =R +R and L =L +L , while circuit model (c) simpli- s 1 2 s 1 2 n and n-1 are required at every interval n, because of the effect fies (b) by assuming infinite impedance from L and R . Due to m m of the derivative and the integral in (1). The discretized its simplicity, circuit model (b) is the preferred digital twin of formulation of (1) is given, as the formal definition of the digital the MV side of the single-phase T/F for the proposed method- twin of the MV side of the T/F as by LV side measurements: ology. Additional detail will only improve the results that fol- ′ ′ ′ ′ 𝑢 [𝑛 ] = 𝑢 [𝑛 ] + 𝑅 𝑖 [𝑛 ] + 𝐿 (𝑖 [𝑛 ] − 𝑖 [𝑛 − 1])𝑓 low. As of circuit (b) and assuming that the T/F is sized and 2 1 𝑆 1 𝑆 1 1 𝑠 𝑢 [𝑛 ] 𝑢 [𝑛 ] − 𝑢 [𝑛 − 1] operated according to standard [30] (i.e. T/F core saturation is } (2) 2 2 2 𝑖 [𝑛 ] = + + 𝑖 [𝑛 ] 2 1 avoided, otherwise a piece-wise formulation [31] may comple- 𝑅 𝐿 ∙ 𝑓 𝑀 𝑀 𝑠 ment the following set-up), the MV calculations are as: With regards to harmonics content, formula set (2) is different 𝑑 𝑖 ′ (𝑡 ) 1 to the continuous set of (1), as the LV-side measurements must 𝑢 (𝑡 )= 𝑢 (𝑡 )+ 𝑅 𝑖 ′ (𝑡 )+ 𝐿 2 1 𝑆 1 𝑆 be properly sampled at high rates, so that no harmonic content (1) 𝑢 (𝑡 ) 1 is missed. This will be explored with assessment of multiple () () () 𝑖 𝑡 = + ∫𝑢 𝑡 + 𝑖 ′ 𝑡 2 2 1 sampling rates in Section IV and consequent analysis. 𝑅 𝐿 } 𝑀 𝑀 Where u, i, R and L are voltage, current, resistance and induct- B. Digital Twin of a Three-Phase Power Transformer ance, respectively. Resistances R and R may also be ex- S M For the digital twin of a three-phase T/F, the approach builds pressed as functions of temperature or of T/F loading (i.e. infer- on that of the single-phase as follows: the digital twins of three ring temperature), to allude to the effect of temperature on re- single-phase T/F, each taking separate single-phase voltage and sistances at different loads. Voltage and current are given as current measurements from the LV side of a three-phase T/F, time variables, since waveforms are measured. Subscripts 1, 2, are appropriately integrated to emulate the three-phase voltage S and M denote the LV and MV sides, series and shunt (leakage and current of the MV side of the T/F. Practically, voltage and and magnetizing impedances) parts of the single-phase T/F, re- current measurements of each phase on the LV side are used to spectively. Voltage and current measurements of the LV side calculate the corresponding values of one of the phases on the are referenced to the MV side (i.e. multiplied by the T/F ratio). MV side through (2). The overall digital twin topology is shown Any tap-changing action in the T/F is considered an input to the in Fig. 2. Following, the calculated values are elaborated ac- digital twin model. Alternatively, tap-changing can be moni- cording to the vector group of the T/F; i.e. the connection of the tored electrically by the digital twin and, thus, adjust values in three phase windings. For the most commonly T/F vector (1). Let it be stressed that (1) may also calculate any harmonics groups in DSs, the phase voltages and the line currents of the content either in the voltage or the current of the MV-side of the MV side for one of the phases are as follows: T/F, provided it is present in the LV-side measurements. The Yy0: 𝑢 = 𝑢 and 𝑖 = 𝑖 (3) 𝐴 2𝐴 𝐴 2𝐴 only concern with regards to this calculation stems from any Dy1: 𝑢 ∙ 3 = 𝑢 − 𝑢 and 𝑖 = 𝑖 − 𝑖 (4) 𝐴𝐵 2𝐴 2C 𝐴 2𝐴 2C filtering effects probably caused by the T/F impedance. These Dy11: 𝑢 ∙ √3 = 𝑢 − 𝑢 and 𝑖 = 𝑖 − 𝑖 (5) 𝐴𝐵 2𝐴 2𝐵 𝐴 2A 2B concerns will be thoroughly assessed in Section IV. Where subscripts A, B, C denote the three phases of the MV From (1) monitoring of the voltage and current waveforms side of the digital twin of the three-phase T/F, and subscript 2X of the LV side can yield voltage and current waveforms of the (where X = A, B, C) denotes the calculated values of the single MV side of the T/F. In detail, an AC voltage and current phase MV side digital twin from LV measurements via (2). measurement device is connected to the LV side of the single- Grounding either T/F side does not alter formulations (3-5). The phase T/F and measures the respective values. The voltage and digital twin of a T/F connected as Dy will not be able to calcu- current sensors provide the measurements in analog form and late MV-side current harmonics of orders multiples of the third, the measurement device converts the analog signals to digital at since such T/F topologies eliminate said harmonics [32]. 𝑑𝑡 𝑑𝑡 III. TESTING FRAMEWORK AND ASSESSMENT METRICS every scenario about 17000 simulations with random initial and final conditions are conducted, to ensure statistical validity of Retrieving the MV side waveforms from monitoring the the assessment metrics (see Subsection III.D) in the 99% confi- ones of the LV side of a triphase T/F has to be tested for various dence interval at ±1% error. The statistical analysis, according levels of loads connected to the T/F, changing load (increas- to [35] ensures confidence in the performance metrics collected. ing/decreasing), presence of harmonics in the system, non-bal- anced three-phase loads, non-balanced voltage supply and sys- B. Testing of Digital Twin Methodology under System Faults, tem faults. All these situations will affect the LV side wave- Unbalanced Loading and Tap-Changing Operations forms differently and, by that, the performance of the method- These tests will show if the accuracy of the digital twin is ology, too. A last parameter that needs to be considered is that affected by system faults, asymmetrical behavior either at the of the parameters of the measurement device used for the appli- MV or LV side of the T/F, and tap-changing action of the T/F. cation (i.e. sampling rate f ). As from the above, sets of testing The variations that need to be assessed are as follows: scenarios are defined in Subsection A. iv. All types of system faults (i.e. line-to-ground (LG), LL and All simulation tests are conducted in MATLAB. For most of LLG – occurring at 0.2s of 0.4s simulation), asymmetrical the tests, a three-phase T/F is tested, while a single-phase T/F is voltage magnitudes fed to the T/F, asymmetrical load con- used for the study of some specific concerns. The test topology nected to the LV side of the T/F and tap-changing action. (see Fig. 3) comprises a voltage source, feeding a load imped- v. Presence/absence of a Peterson coil at the nearest substation ance via a MV-LV T/F – all components are modeled precisely (electrically); the said coil is commonly used to reduce with ratings given in the Appendix. The described topology is ground fault currents and sustain three-phase LV operation. not unique and represents, practically, any typical DS T/F con- vi. Grounding or not of the Y connected side of the T/F (Dy1 nection. This means that any of the typical IEEE or other test or Dy11 vector groups might be assessed interchangeably, systems could be implemented around the described test topol- grounding of both sides of Yy needs to be considered). ogy of the DS T/F and the digital twin. The inclusion of any vii. Location of the fault at the MV or LV side of the T/F. DER in the test topology would also not affect the set-up, as the This set includes 72 scenarios. For each scenario a case of load direction of flow of currents does not affect the digital twin cal- about equal to the rated of the T/F is assumed to be served at culations. This T/F will be monitored through the proposed the time. The waveforms of the digital twin and of the actual methodology. This T/F is the full model as found in MATLAB MV side of the T/F will be compared on similarity of behavior libraries and has not been simplified, i.e. it represents the actual without the use of metrics. The reason for not using statistical T/F of the simulation tests with synthetic data. The methodol- metrics and as it will be noted in Section IV, is because the dig- ogy, as described in Section II, will sample the waveforms of ital twin of the MV side is either correctly calculating the actual voltage and current on the LV side of the T/F. A GridEye meas- waveforms of the T/F or completely miscalculates one or two urement device (class-A power quality monitoring certified) is phases by magnitude or phase shifting. In this sense, the statis- assumed to sample the waveforms [33]. GridEye is assumed to tical metrics as extracted from scenarios (i-iii) will either be have a random error in the interval of its accuracy (see Appen- valid for a correct calculation by the method in scenarios (iv- dix). It will then calculate the respective voltage and current vii) or the method fails substantially. All testing for the second waveforms and active and reactive power values of the MV side set will be conducted at the presence of harmonics (see details – i.e. its digital twin behavior. The digital twin behavior will be in (ii) above) as this represents the worst-case scenario for the compared for accuracy to the respective waveforms of the MV monitoring methodology here proposed. side of the actual T/F model (oscilloscope assumed connected). The metrics of accuracy used are defined in Subsection D. C. Filtering Effect of the Digital Twin on Harmonics The last set of tests aims to assess the effect of the T/F model A. Statistical Testing of Digital Twin Methodology used (see (1),(2) and Fig. 1) to extract the digital twin of the The first set of tests is that of normal operation varied ac- MV side of the T/F, with regards to whether it captures all volt- cording to the following conditions: age and current components. The tests will be conducted on i. Constant load, increasing load and decreasing load; the load MATLAB, but for a single-phase T/F, due to simplicity. Both increase/decrease (step change) will occur at the 0.02s mark the actual T/F and its models, as circuits, are, essentially, low- of a 0.04s simulation interval. pass filters, which damp the magnitude of higher-order harmon- ii. Presence or absence of harmonics in the system at the max- ics. That been said, the model of the T/F used for the digital imum allowed level described by standards [34]. twin of its MV side might be filtering higher harmonics differ- iii. Sampling rates of the measurement device at f = [5kHz, ently than the actual T/F, hence overestimating or underestimat- 10kHz, 30kHz, 52kHz]. ing the harmonic distortion at the MV level of the grid. To ad- For every option in (i-iii), a different scenario is described, thus, dress this concern, firstly, the Fourier transformation of voltage a total of 24 scenarios is tested. The above described conditions and current waveforms of the digital twin and the actual MV and the values assessed are non-linear. Hence, statistical analy- side of the T/F will be compared for simulations of changing sis will be employed [35]. This means that multiple similar load at the presence of harmonics (see details in (b) above). Sec- tests/simulations will be conducted for every scenario. From ondly, the Bode diagram of the model used for the digital twin these tests, averages, minima and maxima of the differences be- and that of the circuit of the actual T/F will be compared for tween actual and calculated waveforms will be calculated. For high and low loading of the T/F. TABLE I TABLE II STATISTICS OF ΔX FOR NORMAL OPERATION WITHOUT HARMONICS, GIVEN STATISTICS OF ΔX FOR NORMAL OPERATION WITHOUT HARMONICS, GIVEN 𝑡 ,𝑀 IN % OF RMS VALUE OF CORRESPONDING VARIABLE, FOR IN % OF RMS VALUE OF CORRESPONDING VARIABLE, FOR CONSTANT (0), CONSTANT (0), INCREASING (+) AND DECREASING (-) LOAD IMPEDANCE INCREASING (+) AND DECREASING (-) LOAD IMPEDANCE ΔL 0 + - ΔL 0 + - f f s s (kHz) (kHz) V 3.7 7.4 3.4 3.8 5.4 3.5 3.8 5.4 3.5 V 8.8 14.4 8.3 9.9 11.5 8.4 9.1 11.5 8.4 I 6.2 10.6 4.2 8.2 11.7 6.2 10.3 35.3 6.0 I 2.7 5.6 1.8 2.6 4.4 2.0 2.6 4.4 2.0 P 4.9 36.7 0.1 2.6 8.9 0.2 2.6 8.9 0.2 P 4.9 36.4 0.01 2.4 7.6 0.1 2.4 8.5 0.1 Q 8.6 >100 1.5 4.2 77.3 1.9 4.2 77.3 1.9 Q 8.6 >100 1.5 3.1 29.2 1.5 3.1 60.7 1.6 V 4.4 8.5 3.9 4.8 7.4 4.1 4.8 7.0 4.1 f 0.0 0.0 0.0 0.004 0.05 0.0 0.004 0.05 0.0 I 4.9 8.3 2.0 3.5 7.1 2.3 3.6 7.4 2.3 fI 0.0 0.0 0.0 0.02 0.08 0.0 0.02 0.08 0.0 10 P 4.8 39.0 0.1 2.5 7.9 0.1 2.7 10.6 0.2 V 1.9 4.7 1.6 2.1 3.8 1.7 2.1 3.5 1.7 Q 7.0 >100 1.5 4.0 40.4 1.9 4.0 57.8 1.8 I 2.5 4.9 0.9 1.6 3.3 0.9 1.6 3.4 0.9 V 1.9 7.6 1.4 2.3 4.7 1.6 2.3 4.9 1.7 P 4.8 38.7 0.01 2.4 7.6 0.1 2.3 9.8 0.1 I 4.7 8.4 0.8 3.0 4.8 1.0 3.0 5.0 0.9 Q 7.0 >100 1.5 3.0 29.2 1.5 3.5 50.4 1.5 P 5.0 36.7 0.1 2.6 8.9 0.1 2.6 8.3 0.1 f 0.0 0.0 0.0 0.002 0.03 0.0 0.002 0.03 0.0 Q 8.5 >100 1.5 3.9 76.5 1.8 3.9 71.2 1.8 f 0.0 0.0 0.0 0.02 0.06 0.0 0.02 0.2 0.0 I V 1.3 5.4 0.9 2.0 4.8 1.2 1.8 4.2 1.2 V 0.9 4.9 0.6 1.1 2.7 0.8 1.1 2.7 0.8 I 4.7 8.3 0.6 2.9 5.1 0.8 3.0 4.9 0.8 I 2.9 5.5 0.3 1.7 3.0 0.4 1.7 3.2 0.5 P 4.8 39.0 0.1 2.5 7.9 0.1 2.6 10.4 0.2 Q 7.0 >100 1.5 3.9 40.2 1.8 3.9 57.7 1.8 P 4.9 36.4 0.01 2.4 8.5 0.1 2.3 7.6 0.1 Q 8.5 >100 1.5 3.0 60.3 1.5 3.5 57.2 1.6 f 0.003 0.005 0.0 0.003 0.02 0.0 0.003 0.02 0.0 (𝑥 − 𝑥 ) 𝑑 ,𝑘 ,𝑛 𝑟 ,𝑘 ,𝑛 Δx = 𝑚𝑎𝑥 ( ) (7) 𝑡 ,𝑀 f 0.002 0.005 0.0 0.01 0.04 0.0 0.01 0.1 0.0 2 𝑟 ,𝑘 ,𝑀𝑆𝑅 V 0.7 3.5 0.4 0.9 2.6 0.6 0.9 2.4 0.6 I 3.1 5.6 0.3 1.8 3.4 0.4 1.9 3.2 0.5 The metrics describe the error between the digital twin and the P 4.8 38.7 0.01 2.4 7.6 0.1 2.3 9.7 0.1 actual MV side T/F waveforms for one simulation of a scenario. Q 7.0 >100 1.5 3.0 29.2 1.5 3.5 50.3 1.6 For the case of a load change ΔL at the point of 0.2s of the test- f 0.003 0.008 0.0 0.003 0.02 0.0 0.004 0.02 0.0 f 0.002 0.008 0.0 0.01 0.04 0.0 0.01 0.2 0.0 I ing interval, metrics Δx and Δx are adjusted for samples n 𝑡 𝑡 ,𝑀 in the time interval of t=0.16-0.24s, i.e. two electric cycles be- D. Assessment Metrics fore and after the load change. For the first set of scenarios, regarding all options in (i-iii), From all simulations of each scenario statistical measures the waveforms of voltage and current, values of active and re- will be extracted as here defined. The average, the maximum active powers, and frequency of the digital twin of the MV side and minimum of metrics Δx and Δx for all simulations of 𝑡 𝑡 ,𝑀 of the T/F (as calculated by the proposed methodology) will be each scenario are extracted. Hence, the metrics presented fol- compared to those of the actual T/F. Specifically for frequency, lowing are the metrics over the entirety of simulations of each the zero crossing of ten consecutive periods of waveforms of scenario tested, so as to ensure statistical validity. The analysis current and voltage (compared separately) will be averaged represents realistically the whole range of operating situations. over the time between the first and last crossing and the respec- tive frequency will be calculated and compared via a similar IV. RESULTS AND DISCUSSION elaboration of the actual waveform of voltage or current of the The Section is organized in two subsections, the first one MV side. For the said comparisons, metrics of average and summarizing the results per sets of scenarios with common point errors between the actual waveforms and that of the digi- characteristics and the second one discussing the findings in tal twin will be used, and are defined as follows. light of the efficacy of the proposed methodology. For one of the simulations in scenario k, assuming that x is d,k the waveform (voltage or current) of the digital twin of the MV A. Results side of the T/F and x the respective waveform of the actual r,k 1) Statistical Testing under Normal Operation: In Tables I- T/F on its MV side, the two will be compared point-to-point for II the statistics of metrics 𝛥 𝑥 and 𝛥 𝑥 for voltage, current, 𝑡 𝑡 ,𝑀 every time sample interval n. Firstly, the average error of a active and reactive power, and system frequency are presented waveform outputted by the digital twin normalized over the for the case that no harmonics are present in the grid, while Ta- RMS of the actual waveform of the T/F is defined as: bles III-IV the same metrics at the presence of all harmonics ∑ within the limits of [34]. The statistics given are average (Avg), (𝑥 − 𝑥 ) 𝑛 =1 𝑑 ,𝑘 ,𝑛 𝑟 ,𝑘 ,𝑛 ̃ (6) Δx = maximum (max) and minimum (min) for metrics 𝛥 𝑥 and 𝑥 ∙ 𝑁 𝑡 𝑟 ,𝑘 , 𝛥 𝑥 , accordingly. “0” denotes a scenario of constant load, “+” 𝑡 ,𝑀 where N is the sample points taken in the simulated testing time of increasing, and “-” of decreasing. V, I, P, Q, f and f are volt- V I interval of 0.4s. The maximum point error of the waveform out- age, current, active and reactive power, and frequency calcu- putted by the digital twin normalized over the RMS of the actual lated by the voltage and current waveforms, respectively. In waveform of the T/F is calculated by: Avg Max Min Avg Max Min Avg Max Min Avg Max Min Avg Max Min Avg Max Min 𝑅𝑀𝑆 TABLE III TABLE IV STATISTICS OF ΔX FOR NORMAL OPERATION WITH HARMONICS, GIVEN IN STATISTICS OF ΔX FOR NORMAL OPERATION WITH HARMONICS, GIVEN IN % 𝑡 ,𝑀 OF RMS VALUE OF CORRESPONDING VARIABLE, FOR % OF RMS VALUE OF CORRESPONDING VARIABLE, FOR CONSTANT (0), INCREASING (+) AND DECREASING (-) LOAD IMPEDANCE CONSTANT (0), INCREASING (+) AND DECREASING (-) LOAD IMPEDANCE ΔL 0 + - ΔL 0 + - f f s s (kHz) (kHz) V 5.0 8.1 4.8 5.1 6.3 4.8 3.8 5.1 3.4 V 37.5 43.2 37.0 37.9 40.2 37.2 9.0 11.3 8.3 I 8.7 26.0 5.4 15.3 50.0 8.4 44.5 >100 8.0 I 2.8 5.7 1.8 2.5 4.7 2.0 9.4 25.3 2.6 P 5.0 36.7 0.1 2.6 8.9 0.2 5.1 6.2 4.8 P 4.9 36.4 0.01 2.4 8.4 0.1 26.1 28.4 25.2 Q 8.7 >100 1.5 4.2 78.5 1.9 2.8 4.8 1.2 Q 8.6 >100 1.5 3.1 61.5 1.6 11.9 18.7 8.2 V 18.2 22.3 17.7 18.8 22.2 18.0 18.8 21.5 17.0 f 0.03 0.05 0.02 0.03 0.05 0.02 0.05 0.2 0.004 I 6.5 18.0 2.4 6.0 26.2 3.1 4.9 12.2 2.4 fI 0.02 0.05 0.0 0.02 0.08 0.0 0.07 1.4 0.04 10 P 4.8 39.0 0.1 2.5 7.9 0.1 2.7 10.6 0.2 V 2.5 5.0 2.3 2.6 4.1 2.4 2.6 3.9 2.3 Q 7.1 >100 1.5 4.0 40.8 1.9 4.0 57.6 1.8 I 2.5 5.1 0.9 1.6 3.4 0.9 1.7 3.5 1.0 V 7.3 13.3 6.8 7.7 10.2 6.8 7.6 10.7 5.8 P 4.8 38.7 0.01 2.4 7.6 0.1 2.3 9.8 0.1 I 6.2 11.2 1.0 3.9 6.4 1.1 3.9 6.6 1.0 Q 7.0 >100 1.5 3.0 29.4 1.5 3.5 50.3 1.6 P 5.0 36.7 0.1 2.6 8.9 0.1 2.6 8.3 0.1 f 0.01 0.02 0.003 0.01 0.03 0.003 0.02 0.05 0.003 Q 8.6 >100 1.5 3.9 77.5 1.8 3.9 70.9 1.8 f 0.01 0.03 0.0 0.02 0.07 0.0 0.02 0.2 0.0 I V 4.4 8.7 3.9 4.9 8.8 3.8 4.7 7.6 3.0 V 1.1 4.9 0.8 1.2 2.8 1.0 1.3 2.7 1.0 I 6.3 11.0 0.8 3.0 6.8 1.0 4.0 6.5 0.9 I 2.9 5.5 0.3 1.8 3.0 0.4 1.7 3.2 0.5 P 4.8 39.0 0.1 2.5 7.9 0.1 2.6 10.4 0.2 Q 7.0 >100 1.5 3.9 40.6 1.8 3.9 57.6 1.8 P 4.9 36.4 0.01 2.5 8.5 0.1 2.3 7.6 0.1 Q 8.5 >100 1.5 3.0 61.0 1.5 3.5 57.0 1.6 shifts) if: f 0.003 0.02 0.0 0.004 0.02 0.0 0.01 0.03 0.0 - the fault is LLG or LG and f 0.004 0.04 0.0 0.01 0.05 0.0 0.01 0.2 0.0 - the T/F is connected as Dy (regardless of grounding) or Yy V 0.8 3.5 0.3 1.0 2.7 0.7 1.0 2.4 0.7 with only one or neither of the two sides grounded and I 3.1 5.6 0.5 1.8 3.4 0.4 1.9 3.2 0.5 P 4.8 38.7 0.01 2.4 7.6 0.1 2.3 9.7 0.1 - the substation is grounded at its MV side. Q 7.0 >100 1.5 3.0 29.4 1.5 3.5 50.2 1.6 Phase voltages of the MV side of the T/F are also the only ones f 0.003 0.01 0.0 0.004 0.02 0.003 0.004 0.04 0.003 miscalculated at the occurrence of any type of fault on the LV f 0.002 0.04 0.003 0.01 0.04 0.0 0.01 0.06 0.0 side of the T/F. For the ease of the reader the rest of these plots are omitted. Tables II and IV, no statistics of metric 𝛥 𝑥 are noted for fre- 𝑡 ,𝑀 The phase voltages and line currents, at the occurrence of a quency. The reason is that frequency is averaged over multiple LG fault on the MV side of a Dy11 T/F grounded on its LV side electric cycles (See Section III.B), hence there is no point error and without grounding of the neutral of the voltage source are (and for that matter maximum point error). presented in Fig. 6. The sampling rate of the digital twin method Indicatively, some statistics are here read for the ease of the is at f =30kHz. The absence of grounding of the voltage source reader. From Table I, for the scenario of increasing load at s is equivalent to the substation upstream of the distribution T/F f =10kHz of the methodology with no harmonics in the system, the maximum of the average errors Δx of the active power of the digital twin over all simulations for this scenario is 7.6% with regards to the RMS of the actual active power of the T/F. From Table IV, for the scenario of constant load at f =30kHz of the method with harmonics in the system, the average of the maximum errors Δx of the current of the digital twin over all 𝑡 ,𝑀 simulations for this scenario is 6.2% with regards to the RMS of the actual active power of the T/F. For a load increase, the MV side waveforms of the digital twin at f =52kHz and the ac- tual T/F are given in Fig. 4, to depict the performance of the method. 2) System Faults, Unbalanced Loading and Tap-Changing Operations: The scenarios of system faults are presented first. In Fig. 5 the phase voltages and line currents, at the occurrence of a LG fault on the MV side of a Dy11 T/F grounded on its LV side and a coil grounding the neutral of the voltage source are presented. The sampling rate of the digital twin methodology is at f =30kHz. The coil emulates the Peterson coil connected to the substation upstream from distribution T/F. The digital twin methodology fails to calculate only phase voltages of the MV side of the T/F by phase and magnitude. The method fails also Fig. 4. MV side current and line-to-line voltage waveforms of the digital twin only for phase voltages (but for different magnitudes and phase (subscript e, f =52kHz) and of the actual T/F (subscript r) at a load increase. Avg Max Min Avg Max Min Avg Max Min Avg Max Min Avg Max Min Avg Max Min Fig. 5. MV side current and phase voltage waveforms of the digital twin (sub- Fig. 6. MV side current and phase voltage waveforms of the digital twin (sub- script e, f =30kHz) and of the actual T/F (subscript r) at a LG fault with a script e, f =30kHz) and of the actual T/F (subscript r) at a LG fault with a non- grounded voltage source supply. grounded voltage source supply. not being grounded on its MV side. In this case, the digital twin In Fig. 8 the Fourier transform of voltage and current wave- phase voltages of the MV side of the T/F and the actual ones forms of the MV side of the digital twin (f = 30kHz) and the compare successfully. The methodology succeeds similarly, if: actual T/F are given. A load increase from 10% to 100% load- - the fault is LL and regardless of any topology specifics for ing has been tested. Given how the suppression of the higher either the T/F or the substation or harmonics between the actual and the digital twin of the MV - the fault is either LG or LLG, the substation is not grounded th sides of the T/F is substantially different beyond the 20 har- on its LV side and regardless of the topology specifics for monic (see Fig. 7), the detail in the Fourier transform of voltage the distribution T/F or th th is focused on the interval of 20 -25 harmonics. - the fault is LG or LLG and the T/F is of the Yy topology grounded on both its sides and regardless of the topology specifics of the substation. For the ease of the reader the rest of these plots are omitted. It is stressed again that the digital twin method calculates correctly all line-to-line voltages, line currents, active and reac- tive powers of the MV side of the T/F under any type of fault, regardless of T/F and DS grounding topologies. Following, various cases of asymmetrical voltage magnitude of the source and asymmetrical load impedances were tested. In all cases, the digital twin methodology calculated correctly all voltage and current waveforms of the T/F MV side. As for tap- changing actions, the accuracy of the digital twin remains unaf- fected. This was expected as any tap-changing action only af- fects the parameters of the digital twin model, but not its math- ematical behavior, and, thus, its accuracy. For the ease of the reader all plots of these tests are omitted. 3) Filtering Effect of T/F and Digital Twin on Harmonics: The Bode diagrams of the circuits representing the digital twin and the actual T/F are first plotted. The respective transfer func- tions of circuit (b) (digital twin) and circuit (a) (actual T/F) in Fig. 1 are, first, calculated. Load equal to 10% and 100% of the rated apparent power of the T/F at 0.8 lag power factor has been tested, to assess how the T/F circuit models (acting as filters) Fig. 7. Bode diagrams of load voltage and current magnitude response to elec- suppress harmonic components (even beyond those mentioned tric frequency components up to 10kHz for nominal loading of the T/F for the in the standard [34]). Both loading tests had similar results and actual and the digital twin circuit models. the case of 100% loading is shown in Fig. 7. Fig. 9. Statistics of average error metric as of Δx for voltage of the MV digital twin of the T/F at the absence of harmonics. Fig. 8. Fourier transform of voltage and current waveforms of MV side for ac- (a) tual and digital twin (f =30kHz) of a T/F under a load increase. B. Discussion 1) Statistical Testing under Normal Operation: To assess the performance of the methodology under normal operation (Tables I-IV) both at the presence and absence of harmonics, Fig. 9 is here presented. The figure shows the envelope of max- imum and minimum of the average error metric 𝛥 𝑥 around the average of the said metric of the voltage waveform of the digital twin from that of the actual T/F MV sides for decreasing load (as taken from Table I). It is clear that the average is consider- ably closer to its minimum value than its maximum. This can be noticed for all values, for both metrics (𝛥 𝑥 and 𝛥 𝑥 ) in 𝑡 𝑡 ,𝑀 Tables I-IV. This observation translates as that the average be- (b) havior of the here proposed digital twin methodology is more Fig. 10. Expected error interval of waveform calculations by the digital twin of likely to be represented by its best performance rather than cal- the MV side of T/F compared to the actual ones at the absence (a) and presence (b) of harmonics and with constant load connected to the T/F. culate waveforms that deviate considerably from the ones of the actual T/F. This observation is even more substantial for the by the average metric at f ≥ 10kHz. Fig. 10 confirms this ob- statistics of the maximum error metric in 𝛥 𝑥 . The averages 𝑡 ,𝑀 servation as error tends to decrease as f increases. With limited of metric 𝛥 𝑥 can be used to define an interval around the 𝑡 ,𝑀 calculation errors of all MV waveforms as f increases, the com- averages of error metric 𝛥 𝑥 instead of the standard deviation plexity introduced by expressing T/F resistances as functions of of the latter, representing, thus, a more realistic error interval temperature (see Section II.A) can be avoided. for the waveform calculations of the digital twin methodology. Some particularly high maximum errors for both metrics for Fig. 10 shows exactly that, as the expected error interval for all the active and reactive powers may be noted in Tables I-IV. waveforms of the digital twin of the MV side of the T/F for the Those errors are noted to occur only at loading of the T/F less case of constant load. For either increasing or decreasing load than 5% of its nominal, thus, are not of concerning nature. Also, the respective graphs are similar (see Tables I-IV for the statis- it is reminded that Δx and Δx are normalized over the RMS 𝑡 𝑡 ,𝑀 tics) and have been omitted. of the respective waveform. Hence, if the normalization of Δx In terms of absolute values, let it be first reminded that in- and Δx was over the nameplate power instead of the RMS of 𝑡 ,𝑀 strument voltage transformers are classified according to their the waveform, they would be negligibly small, i.e. less than 4%. accuracy errors which span the interval 0.1-3% over the voltage As for the frequency calculated by the digital twin, the aver- RMS value [17]. It may be seen that at the absence of harmonics ages of the average error metric Δx is about the same regard- and for f ≥ 30kHz even the average maximum errors fall in this less on whether the frequency is calculated by the voltage or the interval for the proposed digital twin methodology. At the pres- current waveform (Tables I and III), with a slight favorite of the ence of harmonics, the digital twin methodology can have ac- voltage. Nevertheless, from the maximum errors of the metric curacy performance comparable to that of the voltage T/F only TABLE V METRICS OF ΔX AND ΔX BETWEEN THE DIGITAL TWIN (AT 13.2 KHZ 𝑡 𝑡 ,𝑀 SAMPLING RATE) AND THE ACTUAL WAVEFORMS OF THE MV-SIDE OF A DISTRIBUTION TRANSFORMER IN SWITZERLAND, GIVEN IN % OF RMS VALUE OF CORRESPONDING VARIABLE Dataset 1 Dataset 2 Dataset 3 Dataset 4 ̃ ̃ ̃ ̃ Δx Δx Δx Δx Δx Δx Δx Δx 𝑡 ,𝑀 𝑡 ,𝑀 𝑡 ,𝑀 𝑡 ,𝑀 𝑡 𝑡 𝑡 𝑡 V 3.8 7.1 2.9 5.4 2.9 5.0 2.9 6.1 V 4.7 8.7 3.6 7.0 3.7 6.1 3.9 8.2 V 3.9 7.3 2.9 5.5 3.0 5.5 3.1 6.4 I 12.7 38.5 14.6 27.9 12.5 25.4 15.7 55.4 I 15.8 38.0 16.3 29.8 13.7 26.2 18.9 51.7 I 19.7 61.6 19.2 34.7 16.8 32.2 26.6 80.8 P 20.2 24.6 20.3 21.9 18.2 19.9 24.7 28.0 Q 5.8 21.3 6.6 13.3 4.3 11.4 5.5 12.3 it is made clear that the voltage waveform should be preferred Fig. 11. Topology of the collection of field data from the actual T/F in Switzer- to estimate system frequency with better accuracy. land to assess the validity of the digital twin methodology. Harmonics affect the accuracy of the voltage waveform out- (regardless of grounding), YGy and Yyg T/F at the occurrence putted by the digital twin much more than any other value (cur- of faults on the MV side of such T/F [29]. Hence, if LG or LLG rent, powers, frequency). Comparing Tables I and III, for the faults occur and the MV side of the DS is grounded (with same type of load change and f , one can notice that statistics Peterson coil or other), the zero sequence circuit at the MV side for all values but for voltage are similar. Especially for voltage, of the said T/F is subject to voltage and current that are not the metrics of the average errors are greater in Table III that propagated to the LV side, hence the method cannot calculate tested for voltage supply with harmonics. This remark can be phase voltages correctly. explained by the analysis of transfer function responses with the The digital twin performs similarly for faults on the LV side Bode diagrams in Fig. 7, as also the circuit itself. The shunt el- of the T/F, too, because the voltage measurements of the ements of the T/F pose practically infinite impedance to the cur- method taken on the LV side of the T/F see towards the fault. rents through it. Hence, the voltage drops caused by the series I.e. due to the, typically, much smaller fault impedance, those elements (impedances with S, 1, 2 subscripts in Fig. 1) are those measurements see close to zero voltage on the affected phases, that affect the accuracy of the methodology, while there is no hence, via (2) fail on MV side calculations. Nevertheless, this current divider to affect the accuracy from the perspective of concern is not critical, since the priority of system operators in currents. In Fig. 7 it may be noted that higher frequency com- such occasions would be the LV fault (captured by the method). ponents of voltage are damped more substantially by the digital twin model compared to the actual of the T/F. This, in turn, V. FIELD TEST VALIDATION OF DIGITAL TWIN OF means that the digital twin methodology model will amplify any DISTRIBUTION POWER TRANSFORMER METHODOLOGY measurement errors (due to sampling in the context of this To assess the validity of the proposed method, current and study) more than what would happen if the actual T/F model voltage waveform measurements from an actual DS MV-LV was used as the digital twin. However, as from the Fourier T/F in Switzerland, are used (for T/F characteristics, see Ap- transform analysis in Fig. 8, it is clear that the energy contents pendix). The waveform measurements were logged with of voltage of the actual and digital twin models of the MV side GridEye devices [33] on both its LV and MV sides in sync (see of the T/F for the frequencies that are most affected (as shown Fig. 11). GridEye, although not designed to operate as a PMU, by the Bode diagrams in Fig. 7) are almost identical for a f of it largely abides by the measurement requirements (see Appen- 30kHz. This also confirms findings noted in the previous para- dix) of the respective standard [36]. graph. In other words, preferring the actual T/F model (circuit LV waveform measurements of the actual T/F are inputted (a) in Fig. 1) instead of the one used (circuit (b) in Fig. 1) will to the digital twin method, and the output of calculated MV not improve substantially the accuracy of the digital twin, if the waveforms of the digital twin is compared to the respective MV f is adequately high. To conclude these observations, the accu- waveform measurements of the said T/F. Four sets of waveform racy of the digital twin output is unaffected by the presence of data, of about 10 minutes each, sampled at 13.2kHz are used. harmonics (i.e. calculates MV waveforms properly), provided To assess the performance of the method Table V presents the that the sampling rate of the input waveforms is high enough. metrics defined by Δx and Δx as the average and maximum 2) System Faults, Unbalanced Loading and Tap-Changing 𝑡 𝑡 ,𝑀 Operations: With regards to the performance of the digital twin errors of the waveform outputted by the digital twin method method for the case of system faults, the following are noted. normalized over the RMS of the actual waveform of the T/F. The method relies on measurements of the LV side of the T/F, As the sampling rate of the input LV waveforms is 13.2kHz hence, unless the LV side is implicated in a phenomenon, the and there is ample harmonic content as the data sets show (see method will fail to calculate phase voltages. In this sense, it is Fig. 12), the results in Table V will be compared to those of here reminded that no currents flow or voltages are induced to Tables III and IV for sampling rates of 10kHz and 30kHz. the LV side of the zero sequence circuit component of Dy actual T/F in Fig. 12. The field data clearly shows that the dig- ital twin methodology can accurately calculate the MV-side be- havior of a T/F based on LV-side waveform input. VI. CONCLUSION This work proposes a method to monitor the MV-side wave- forms of DS T/F, to enhance the visibility of operators into DS. The method is the digital twin of the MV side of a DS T/F. It has high calculation accuracy and captures most system faults and harmonics content. The accuracy of the digital twin in- creases as the sampling rate of the input LV waveforms in- creases, and is comparable to that of an instrument T/F. Next step in the assessment of the proposed technique is to deploy the method on a DS monitoring device [33] to a MV-LV T/F and assess its field performance. Thus, the technique will also be realistically assessed in terms of installation costs and disruption. Practically, the digital twin method is expected to offer much visibility of the MV DS to DS operators. Moreover, the health of the T/F can be monitored through its performance. A first research extension that can be explored on the T/F digital twin method, is its improvement in cases of compensated MV DS ground faults, if not both the T/F primary and second- ary windings are grounded. The proposed digital twin T/F set- up can calculate accurately MV current flows and line-to-line voltages under all MV system fault conditions, but fails to cal- culate the phase voltages (indicator of nature of fault) in the aforementioned specific cases. Another extension can be the ap- plication of the method to MV to high voltage T/F to monitor the transmission system with reduce costs and grid disruptions. VII. APPENDIX A. Simulation Testing on MATLAB Fig. 12. MV side current and phase voltage waveforms, and active and reactive power of the digital twin (subscript e, f =13.2 kHz) and of the MV measurement MV-LV DS 3-phase T/F parameters: data of the actual T/F in Switzerland (subscript r). S = 50 kVA, V /V = 400V/20kV, R = 0.0075 pu, L = 0.02 pu, 1 2 1 1 As it may be noted all voltage, active and reactive power met- R = 0.0075 pu, L = 0.02 pu, R = 500 pu, L = 500 pu. 2 2 m m rics are within the intervals of the statistical testing with the LV 3-phase lumped varying load parameters: synthetic/simulation data. The only exception is with the met- R = [0.75-5.25] Ω, L = [1.5-17] mH. L L rics of the currents, which are greater than the maxima of both MV 3-phase line parameters: Δx and Δx . Since the metrics of voltages and powers are 𝑡 𝑡 ,𝑀 R = 2 Ω, L = 1 mH. LN LN within the intervals of the statistical results collected from the B. Digital Twin Tests with Field Data from T/F in Switzerland simulations with synthetic data and the current is a function of them, the noted error metrics of current are a result of how these LV-MV DS 3-phase T/F parameters: metrics are normalized over the RMS of the respective wave- S = 630 kVA, V /V = 400V/20.5kV, R = 0.0035 pu, 1 2 1 form. In fact, the T/F loading across all four datasets was par- L = 0.0233 pu, R = 0.0035 pu, L = 0.0233 pu, R = 500 pu, 1 2 2 m ticularly low (less than 5% of the nameplate 630kVA); i.e. if L = 500 pu. the normalization of the error metrics was over the nominal cur- GridEye measurement accuracy characteristics: rent instead the RMS of the waveform, the errors would be neg- Voltage magnitude = 0.1%, ligibly small, i.e. no more than 4%. Similar observations were Total Vector Error (see [36]) < 1%, made for the high error metrics of the active and reactive power Current Magnitude = 1%, of the digital twin for the statistical testing with synthetic data Frequency = ±1 mHz. (see Section IV.B). The generally positive behavior of the digital twin method, VIII. REFERENCES as shown by the above error metrics analysis, is corroborated [1] N. D. Hatziargyriou, E. I. Zountouridou, A. Vassilakis, P. Moutis, C. N. by the comparison of the waveforms of the MV-side of the T/F Papadimitriou, and A. G. Anastasiadis, "Overview of Distributed Energy between the output of the digital twin and the datasets of the Resources," Smart Grid Handbook, pp. 1-44, 2016. [2] L. Bird et al., "Wind and solar energy curtailment: A review of [19] A. Gaouda and M. Salama, "DSP wavelet-based tool for monitoring international experience," Renewable and Sustainable Energy Reviews, transformer inrush currents and internal faults," IEEE Transactions on vol. 65, pp. 577-586, 2016. Power Delivery, vol. 25, no. 3, pp. 1258-1267, 2010. [3] M. M. Esfahani and G. R. Yousefi, "Real time congestion management in [20] E. E. Bernabeu, "Single-phase transformer harmonics produced during power systems considering quasi-dynamic thermal rating and congestion geomagnetic disturbances: theory, modeling, and monitoring," IEEE clearing time," IEEE Transactions on Industrial Informatics, vol. 12, no. Transactions on Power Delivery, vol. 30, no. 3, pp. 1323-1330, 2014. 2, pp. 745-754, 2016. [21] P. Sun, W. Sima, M. Yang, X. Lan, and J. Wu, "Study on voltage-number [4] M. Tarafdar Hagh and T. Khalili, "A review of fault ride through of PV characteristics of transformer insulation under transformer invading non- and wind renewable energies in grid codes," International Journal of standard lightning impulses," IEEE transactions on dielectrics and Energy Research, vol. 43, no. 4, pp. 1342-1356, 2019. electrical insulation, vol. 22, no. 6, pp. 3582-3591, 2015. [5] A. Jafari, H. G. Ganjehlou, T. Khalili, B. Mohammadi-Ivatloo, A. Bidram, [22] R. K. Senobari, J. Sadeh, and H. Borsi, "Frequency response analysis and P. Siano, "A two-loop hybrid method for optimal placement and (FRA) of transformers as a tool for fault detection and location: A scheduling of switched capacitors in distribution networks," IEEE Access, review," Electric Power Systems Research, vol. 155, pp. 172-183, 2018. vol. 8, pp. 38892-38906, 2020. [23] R. M. A. Velásquez, J. V. M. Lara, and A. Melgar, "Converting data into [6] F. Hu, K. Sun, A. Del Rosso, E. Farantatos, and N. Bhatt, "Measurement- knowledge for preventing failures in power transformers," Engineering based real-time voltage stability monitoring for load areas," IEEE Failure Analysis, vol. 101, pp. 215-229, 2019. Transactions on Power Systems, vol. 31, no. 4, pp. 2787-2798, 2015. [24] A. Bracale, G. Carpinelli, and P. De Falco, "Probabilistic risk-based [7] W. Santos, F. Lopes, N. Brito, and B. Souza, "High-impedance fault management of distribution transformers by dynamic transformer rating," identification on distribution networks," IEEE Transactions on Power International Journal of Electrical Power & Energy Systems, vol. 113, Delivery, vol. 32, no. 1, pp. 23-32, 2016. pp. 229-243, 2019. [8] H. L. Willis, Distributed power generation: planning and evaluation. Crc [25] M. Faifer, C. Laurano, R. Ottoboni, S. Toscani, and M. Zanoni, Press, 2000. "Harmonic Distortion Compensation in Voltage Transformers for [9] M. M. Albu, M. Sănduleac, and C. Stănescu, "Syncretic use of smart Improved Power Quality Measurements," IEEE Transactions on meters for power quality monitoring in emerging networks," IEEE Instrumentation and Measurement, 2019. Transactions on Smart Grid, vol. 8, no. 1, pp. 485-492, 2016. [26] S. Boschert and R. Rosen, "Digital twin—the simulation aspect," in [10] M. S. Thomas and J. D. McDonald, Power system SCADA and smart Mechatronic Futures: Springer, 2016, pp. 59-74. grids. CRC press, 2017. [27] A. Ebrahimi, "Challenges of developing a digital twin model of [11] A. Monticelli, "Electric power system state estimation," Proceedings of renewable energy generators," in 2019 IEEE 28th International the IEEE, vol. 88, no. 2, pp. 262-282, 2000. Symposium on Industrial Electronics (ISIE), 2019: IEEE, pp. 1059-1066. [12] P. P. Moutis and O. Alizadeh-Mousavi, "A Practical Proposal for State [28] Y. Yang et al., "State Evaluation of Power Transformer Based on Digital Estimation at Balanced, Radial Distribution Systems," in 2019 IEEE PES Twin," in 2019 IEEE International Conference on Service Operations and Innovative Smart Grid Technologies Europe (ISGT-Europe): IEEE, pp. 1- Logistics, and Informatics (SOLI), 2019: IEEE, pp. 230-235. 5. [29] J. J. Grainger and W. D. Stevenson, Power system analysis. 2003. [13] A. Primadianto and C.-N. Lu, "A review on distribution system state [30] I. Std, "60076-8,“," Power transformers. Part, vol. 8. estimation," IEEE Transactions on Power Systems, vol. 32, no. 5, pp. [31] P. Kundur, N. J. Balu, and M. G. Lauby, Power system stability and 3875-3883, 2017. control. McGraw-hill New York, 1994. [14] A. Von Meier, E. Stewart, A. McEachern, M. Andersen, and L. [32] A. A. Sallam and O. P. Malik, Electric distribution systems. John Wiley Mehrmanesh, "Precision micro-synchrophasors for distribution systems: & Sons, 2018. A summary of applications," IEEE Transactions on Smart Grid, vol. 8, [33] M. Carpita, A. Dassatti, M. Bozorg, J. Jaton, S. Reynaud, and O. no. 6, pp. 2926-2936, 2017. Mousavi, "Low Voltage Grid Monitoring and Control Enhancement: The [15] C. Chen, J. Wang, F. Qiu, and D. Zhao, "Resilient distribution system by GridEye Solution," in 2019 International Conference on Clean Electrical microgrids formation after natural disasters," IEEE Transactions on smart Power (ICCEP), 2019: IEEE, pp. 94-99. grid, vol. 7, no. 2, pp. 958-966, 2015. [34] I.-I. E. Commission, "IEC TR 61000-2-5," 2011. [16] J. Peppanen, M. J. Reno, R. J. Broderick, and S. Grijalva, "Distribution [35] W. A. Fuller, Sampling statistics. John Wiley & Sons, 2011. system model calibration with big data from AMI and PV inverters," [36] K. E. Martin and I. P. S. R. Comm, "Synchrophasor Measurements Under IEEE Transactions on Smart Grid, vol. 7, no. 5, pp. 2497-2506, 2016. the IEEE Standard C37.118.1-2011 With Amendment C37.118.1a," Ieee [17] I. Standard, "61869-3, 2011,“," Instrument transformers–Part, vol. 3. Transactions on Power Delivery, vol. 30, no. 3, pp. 1514-1522, Jun 2015, [18] D. Lin and E. F. Fuchs, "Real-time monitoring of iron-core and copper doi: 10.1109/tpwrd.2015.2403591. losses of transformers under (non) sinusoidal operation," IEEE Transactions on Power Delivery, vol. 21, no. 3, pp. 1333-1341, 2006.

Journal

Electrical Engineering and Systems SciencearXiv (Cornell University)

Published: Dec 3, 2020

There are no references for this article.