Phasor measurement Unit—s for system diagnosis and load identification in Australia

Abstract--In 1993 QUT developed phasor measurement andc ommunic ation systems whic h were initially deployed on the Queensland Network. This development and deployment was extended in 2001 to enc ompass the major load c enters in the Australian National Grid. The system has been used for eigenvalues-eigenvector estimation, for diagnosis of major system disturbances and for load identification.Index Terms—Phasor Measurement, Power System Dynamics, Power System identification,I. I NTRODUCTIONHE phasor measurement unit (PMU) system developed by QUT achieves a good performance through high sample rates, precise sampling and most importantly, a high precision in the phase locked loop to the GPS signal. The system of units installed enables live transfer of data across the Australian network, which offers significant benefits for on-line monitoring. The current system transfers up to 3 phasor measuremts at one site with a transfer rate of 10Hz using TCP/IP, and also allows for the local storage of data at both 50Hz and 10Hz sampling rates. Typical hard drive storage provides local storage for 1 year with automatic overwrite as the storage becomes full.Fig 1: Front view of the mobile unitThis work was supported in part by Australian research Council grant with assistance from Powerlink, Transgrid, ElectraNET SA VenCorp.G. Ledwich and P. O’Shea are with the School of Engineering Systems, Queensland University of Technology (e-mail g.ledwich@.au pj.oshea@.au)D. Geddey is with Transgrid (e-mail don.geddey@.au)II. E IGENVECTOR AND EIGENVALUE IDENTIFICATIONIt is critical to obtain reliable and accurate estimates of the eigenvalues and eigenvectors for the dominant modes in the power system. A number of strategies have been used to enhance the quality and reliability of the estimates. These strategies are described in the ensuing paragraphs.The most straightforward way to identify key modes in a power system is to stage brake resistor tests or on-line switching events, and then using Prony analysis of the system response so as to estimate the damping factors and frequencies. Early attempts to use this approach, however, yielded unreliable results because of the poor conditioning of the Prony method. Subsequent use of singular value decomposition in conjunction with Prony analysis improved the conditioning of the method, yielding a corresponding improvement in estimation quality. [7].Regular disruption of power systems via braking resistor tests and staged on-line switching events is problematical. To avert the need for these types of major disruptions, ‘ambient’ testing methods were investigated. During these investigations it was observed that during normal operation of power systems the normal customer driven load changes which stimulate voltage angle changes can be modelled well as the integral of white noise [8]. This led to the further realisation that one could use these load changes in normal power system operation as the ‘test stimulus’ and then used statistical analysis of the power system response to infer the damping and frequencies of the modes of interest. Given that the stimulus was assumed to be integrated white noise, a suitable pre-processing stage for the estimation process was differentiation followed by creation of the autocorrelation function. It is not hard to show that the autocorrelation function so formed contains the dominant modes in the power system, along with some additive noise [1]. It is therefore well suited to analysis with Prony methods enhanced with singular value decomposition [7]. The use of ambient power system data and autocorrelation function techniques for modal parameter estimation has the advantage that long data records can be used with relatively high signal to noise ratio (SNR). One of the problems encountered when using Prony type methods to estimate the parameters of decaying sinusoids isPhasor Measurement Unit’s for system diagnosis and load identification in AustraliaG. Ledwich, Senior Member, IEEE, D. Geddey non member and P. O’Shea, Senior Member, IEEE TCurrent probesGPSAntenna©2008 IEEE.knowing how much of the data record to use. If one uses too little of the data then the estimates have limited accuracy because of the small amount of data; if one uses too much of the data then (because of the decay of the modes) one is using a significant portion of data which has low SNR – as a result the results are unreliable. A partial solution to this problem was provided by analysis at Transgird, who proposed weighting the data with a window which is matched (at least approximately) to the mode of interest. This is tantamount to using a matched filter approach, which is known to provide good noise performance.Because the PMU system has multiple sites measuring the same modes, multisite algorithms were used to gain improved identification. In many cases when there are multiple measurements of the one event (each corrupted by noise), combining the signals will enhance the detection of the signal. The enhancement, however, depends on at least partial de-correlation of the noise across measurement sites as well aslong records, conditions which are not always the case atmultiple sites in interconnected power systems.A key benefit of multi-site is that the relative contribution of the different modes are different at different sites. The approach that has been found most effective in combining the data from multiple sites is to make linear combinations of measurements to enhance the mode of interest. Ideally if we have three measurements of three modes then there is a linear mapping to combine measurements to yield decoupledmeasurements of the modes. In the iterative decoupling of the modes in [3] this approach was shown to yield better resultsthan the methods described in [6] and [7].Another benefit of multisite estimation is that the relative magnitude and phase of the modes in each of the differentmeasurements can be extracted. It has been fortunate that most of the sites chosen have shown a strong response to inter-area measurements and almost no response to localmodes. This supports the contention that the measurement sites represent a reasonable measurement of inter-area nodephasors. This means that the measurements can be useddirectly to form eigenvectors without further processing. One example which illustrates this point was found frommeasurements associated with the joining of Tasmania to mainland Australia with the HVDC link Basslink (see Fig 2). Here there is a single major mode that is clearly seen Fig 3 as a mode at 1Hz (i.e. 6.3 rad/s) as a North-South grouping in thecorresponding vector plot in Fig 4.Fig 2 Locations of generation groups within Tasmania’s networkSpectrum of autocorrelation (Basslink imports 450MW from VIC)Freuquency (rad/s)Post-Basslink with 450MW import from VICFig 3 Spectrum of autocorrelation from Creek Road, George Town and Farrell902700Phasor plot at W=7.5961 rad/s (Basslink export 510MW to VIC )Fig 4 Mode shape of mode 1 (Basslink with exportation power to VIC) III. A LARMING ON SUDDEN DAMPING DETERIORATIONTo obtain accurate damping factor estimates using ambient data, one needs long data sets – typically more than half an hour of data. If there is a fault in a system, however, one cannot in general afford to wait for half and hour before taking action. To deal with the case of sudden and potentially dangerous deterioration in damping one can monitor for sudden changes rather than try and wait to calculate exact damping values. Once sudden changes are detected alarms are created so that appropriate action is taken. This approach hasbeen tested on both simulated and real data and found to be effective – it relied on power system modelling and statistical monitoring of the Kalman innovation for that model [2].In [2] one sees the alarming algorithm demonstrated on a signal taken from a PMU measurement of an incident of poor connection of a generator in Victoria. This detection was based on 1 minute samples. While this case was not one of a sudden loss of mode damping such as from tripping of a SVC with power oscillation damper, it illustrates how a change of damping and the relevant frequency can be reliably alarmed.IV. R ECENT E VENTS IN A USTRALIAA summary of 4 major events in the Australian network as seen by the measurement network is presented in [5]. The 13/08/2004 event resulted in the 3 stage loss of 5 major generators in NSW. What was of interest was that the load shedding process was much more vigorous in Qld than NSW so the power transfer and thus the angle between Qld and NSW steps up to near 800 as seen in Fig 5 and Fig 6. It was fortunate that this incident started from an unusually low levelof transfer or the link would probably have been broken.Fig 5 Angle difference between Brisbane and SydneyFig 8 Allocation of power flow after eventThe relative strength of load shedding could be enhanced by use of the initial df/dt to indicate the region with the greatest need to load shed. The plot in Fig 7 shows that the loss of generation was clearly detectable to be in NSW. The overall pattern of post fault flows is seen in fig 8.V. L OAD IDENTIFICATIONRecent research projects have focused on the power system item with greatest uncertainty, namely the consumer load. The original measurement system yielded quite accurate measurements of system angle variations (of the order of 0.1 degree with measurement accuracy to 0.01 degrees). The variation of voltage magnitude in the bandwidth of electromagnetic oscillations is much smaller than this angle under normal operation. For load modelling the process adopted has been to seek correlations between voltage/frequency to P/Q. When the voltage magnitude measurement is poorly quantized then the power measurement formed from the voltage and current phasors is also going to suffer. This is a problem because the original measurement system was such that perurbations might onlycross 4 or 5 significant bits. Changes to the PMU system were therefore implemented to increase the transducers to 16 bit,perform data transfers using 32 bit files or use filtered enhancement of the variations in the 16 bit files. This latter process enhanced the high frequency terms before quantization, so that there were lesser quantization effects on the variations of interest.The benefit of the PMU concept is the ability to make high precision measurement of fundamental voltage/current magnitude and angle. This has provided an unprecedented opportunity for high precision angle/frequency measurements. It is the ability to get some idea of the response of the loads to a very small frequency variation that makes it possible to separate motor loads from impedance type loads.A model which has proved useful for analysing power system responses to load changes is provided in Fig 9 specifically for frequency to power relations. The plots in Fig 10 and 11 relate to the same feedback structure of Voltage affecting real and reactive power respectively. In figures 10 and 11 there are two types of excursions. There is one line of a slight droop in voltage with increase of power and other excursions which show a rise of P with a rise in V. The nearly horizontal line is interpreted as changes in the load P and Q respectively causing changes in the voltage at the substation. It is clear to see that this supply impedance effect is much more strongly associated with reactive loads. The other excursions are interpreted as external events causing voltage changes which then cause positive changes in Power for positive voltage changes.Fig 9 expanded model of feedback system Fig 10 Plot of Diff(V) against diff(P) Fig 11 Plot of Diff(V) against Diff(Q)Once the slope is known then the loop can be resolved and the transfer function for the load derived. One aspect of interest is that recent studies are showing that the area under the real component of the f-P relationship on a log of frequency plot is related to the total power of induction motors in the load (see fig 12) . The feedback component in the model in Fig 9 with the power affecting frequency (fig 13) is an indication of system electromagnetic modes and overall governor responseVI. D AMPING SIGNAL VALIDATIONi) The availability of an accurate long-term modal-parameter identification system has changed the way system damping tests are performed. Previously, time-domain fitting the ring-down of power oscillations following the brief insertion of a braking resistor was used to estimate the parameters of inter-area modes. This method suffered from the relatively large contributions made by randomly-excited oscillations of these modes. Now, modal parameter estimates during testing are obtained from the continuous-monitoring system that analyses the measured voltage angles at four sites. Figure 14 shows the variation of the damping of one inter-area mode (estimated using a 3-hour analysis window) as a power-oscillation damper (POD) was switched on and off at 3-hour intervals over a 24-hour period.Fig. 14 Estimated damping of a 2 rad/s inter-area oscillation mode in the Australian system during SVC power oscillation damper (POD) on/off testing.The sawtooth variation of estimated damping is a consequence of the 3-hour analysis window and 3-hourswitching interval. It is clear from Fig. A that changes otherthan the POD contributed to the variation of the mode damping over the 24-hour period. The results from the testing and those obtained from system eigenvalue studies provideda basis for validating the model of the SVC POD.ii) The measurements can be used to determine the (complex) ratios of the contributions made by particular variables to a chosen mode. Of particular value is the ratio of a voltage magnitude at a chosen location to a chosen voltage-angle difference signal. The evaluation of this ratio can be used to assess the contributions of nearby controllers (e.g. generator PSSs and SVC PODs) that are designed tocontribute damping of an inter-area mode. These controllers aim to produce voltage-magnitude contributions to the mode with a fixed phase ratio to voltage-angle contributions. Therefore identifying this phasor ratio is a useful check on the functioning of particular controllers. Applications have found two separate deficiences in the model of an SVC POD, which have both been corrected.VII. C ONCLUSIONSThe high precision data available from PMU systems when applied across a power system provide for a wide range of analysis control and identification improvements. The examples in Australia show the range of uses of PMU data from eigenvalue/eigenvector analysis, post fault analysis, damping alarms, load modelling and damping signal evaluation.VIII. R EFERENCES[1] M. Glickman, P. O’Shea G. Ledwich “Estimation of Modal Damping inPower Networks” Volume 22, Issue 3, Aug. 2007 Page(s):1340 - 1350 [2] R. A. Wiltshire, G. Ledwich, P. O’Shea “A Kalman Filtering Approach toRapidly Detecting Modal Changes in Power Systems” Volume 22, Issue 4, Nov. 2007 Page(s):1698 - 1706[3] G Ledwich “Decoupling for Improved Modal Estimates “IEEE PES 2007-03-05 24-28 June 2007 Page(s):1 - 6[4] C. L. Zhang G. F. Ledwich M. E. Langfellner, S. Karri "Modal AnalysisOf Pre- And Post Basslink With Tasmanian Network" Australian Universities Power Engineering Conference AUPEC, 10-13 Dec 2006, Melbourne, Australia[5] C. L. Zhang G. F. Ledwich ‘Analysis Of Major Australian Events UsingAn Angle Measurement System” AUPEC Sept 2005 Hobart PP159-164 [6] P. O’Shea, “The use of sliding spectral windows for parameter estimationin power system disturbance monitoring,” IEEE Trans. Power Syst.,vol. 15, pp. 1261–1267, Nov. 2000.[7] R. Kumaresan and D. Tufts, “Parameter estimation of dampedexponentials and pole-zero modeling,” IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-30, pp. 833–890, 1982.[8] G.Ledwich E.Palmer “Modal Estimates from normal operation of powersystems” IEEE PES Winter meeting Singapore 2000.Gerard Ledwic h (M’69 SM’78) is a professor in Electrical Power Engineering at QUT and Fellow of the Institution of Engineers Australia. He has held positionsat University of Queensland, and in the Queensland Electricity Commission. He was the Pacific PowerProfessor at the University of Newcastle. The current position is Chair in Power Engineering at Queensland University of Technology. His research interests include control systems, powerelectronics, power systems and distributed generation. Donald Geddey received a B.Sc. degree, a B.E. (Hons) degree and a PhD. degree from the University of Sydney in 1968, 1970 and 1980 respectively. Since 1974 he has worked in the transmission utility in New South Wales. He is now the System Investigations Manager in the Transmission Development Groupof TransGrid.the Queensland University of Technology, Australia. Hereceived the BE, DipEd and PhD from the University ofQueensland. He worked as an engineer at the OverseasTelecommunications Commission for 3 years, atUniversity of Queensland’s Department of ElectricalEngineering for 4 years, at Queensland University ofTechnology’s School of Electrical and Electronic SystemsEngineering for 3 years, and at the School of Electrical and Computer Systems Engineering at RMIT, for 7 years. While at RMIT he received awards in Student Centred Teaching from the Faculty of Engineering and the University President. He re-joined the staff of the School of Electrical and Electronic Systems Engineering at Queensland University of Technology in 2002. His interests are in i) signal processing for communications and power systems, ii) reconfigurable computing, and iii) the use of multi-media inengineering education.。