Phasor-based wide area monitoring in the South African power system

Transcription

1 Phasor-based wide area monitoring in the South African power system by D H Wilson, R A Folkes, Psymetrix, UK; A Edwards B Berry Eskom; N Mbuli, Tshwane University of Technology; Brian van Rensburg, Actom This paper outlines the progress in South Africa on using synchrophasor-based wide area monitoring systems (WAMS) as a means of improving system security and increasing the use of the transmission network. A pilot project was carried out to investigate the dynamic performance of the system, and the results of this project are summarised in this paper. The South African power system covers a very large geographic area, and operates with a relatively small generation capacity margin. The transmission system must therefore be operated close to its limits, while not compromising the security of the system. In view of these challenges, Eskom chose to embark on a project to commission one of the world's largest phasor-based wide-area monitoring system (WAMS) to improve the observability of the system. A pilot WAMS project was commissioned in August 2009, principally to investigate the dynamic characteristics of the power system. The project used three phasor measurement units (PMUs) located at the Muldersvlei, Grootvlei and Pegasus substations. The siting of PMUs at both ends of the long transmission corridor between Johannesburg and Cape Town areas provides valuable information of the wide-area dynamic performance of the system. The monitoring locations and circuits are shown in Fig. 1. This article reviews data that was continuously acquired over a six month period, during which the power system experienced a wide range of conditions, spanning summer and winter operation. Continuous dynamics analysis of this data allows the dynamics of the system to be characterised through all of these operating conditions. Phasor measurement data is useful for understanding the system performance during network disturbances. During the course of the pilot project, a disturbance was captured in which there were three generation unit outages and an interconnection trip. This event, captured with high-speed phasor measurement, revealed aspects of system performance that would not have been clear from conventional monitoring. Some of the characteristics of the PMUs used in the project were reviewed, identifying connection issues that may not be identified in specifications and bench testing. Eskom went on to issue a competitive tender, and selected a consortium including the authors to deliver a large-scale WAMS implementation. Some of the factors important to Eskom in the current and future use of WAMS are described in this paper. Oscillatory stability Normal oscillatory behaviour A selection of modes of oscillation commonly present in the system is shown in Fig. 2. Of these, the 0,7 Hz mode is the most strongly observable in time-domain signals monitored, and is the dominant electromechanical oscillation seen in the corridor between Muldersvlei and Grootvlei. The 0,7 Hz mode occurs at two distinct frequencies 0,66 Hz and 0,73 Hz. The frequencies were not observed simultaneously, and are variations of the same mode in different system conditions. Normally, this mode is well damped, with an average decay time constant of around 3,4 seconds. Excursions to about 12 seconds decay time are quite regularly observed, but this is not a significant security concern. However, much poorer damping was observed. The 0,7 Hz oscillations are in opposing phase at Muldersvlei and Grootvlei, and are larger at Muldersvlei. This is characteristic of an inter-area mode where the path is the long corridor connecting the Cape area with the Johannesburg area. Since the Cape area has lower inertia, it is expected that the angle swings are larger. A 0,3 Hz mode was observed in the Eskom system. This mode was normally observed at Grootvlei, but when unusually large or poorly damped, it is observed at Fig. 1: Phasor measurement unit locations. all locations with very similar amplitude and phase. Previous modal analysis of the southern African grid suggests that this mode is a regional inter-area mode with generators in Zimbabwe, Zambia and Mozambique swinging in opposite phase to the generators in South Africa. The measurements are consistent with this interpretation, but it would be useful to confirm this through measurements in the AC corridor connecting South Africa to Zimbabwe, and if possible using phasor measurements from the other Southern African countries involved. Higher frequency modes between about 1 and 3 Hz were observed, which are normally related to local plant modes. There are several local modes, and the behaviour of two of these was explored in more detail. The two modes were at 1,1 Hz and 1,4 Hz, and both were observed in power in the Pegasus-Drakensberg line. Examining the relative size of the oscillations and the daily pattern, the modes could be identified with a pumped storage plant and a coal plant in the region. It was noted that the amplitude of the mode related to the coal generation increased at night when the plant would normally run at reduced output. A very low frequency common mode at 0,05 Hz was seen in frequency throughout the system, illustrated in Fig. 4. This is likely to be related to turbine governors participating in load-frequency control. energize - June Page 32

2 two modes. In another event, although the damping was not poor, the amplitude of the mode increased over a period of about 10 h. During this particular day, several transmission lines faulted due to bush fires. The 0,7 Hz mode was also significant during a disturbance described in the next section. With the system lightly loaded, the inertia was reduced by a series of generation trips and loss of interconnection. In this disturbed condition, the 0,7 Hz mode was poorly damped. 0,25 0,35 Hz mode Measured mostly in frequency at GRO, but when larger, it is seen at MUL and PEG in frequency as a common-mode oscillation. Normally well damped (2,7 s) but typically poorer (6 s) when observed at all locations. Small, but very poorly damped oscillations (50 s) observed at 0,35 Hz, coincident with 0,7 Hz mode changes. Significant amplitude observed in event on 8th Jan'10. 0,7 Hz mode Strongly observed in power between MUL and GRO, and in frequency. Interarea mode, with opposite phase oscillations at MUL and GRO, with larger oscillations at MUL. Normally well damped (3,4 s), occasionally up to 12 s. Very lightly damped oscillations (50 s) observed, linked to 0,35 Hz mode. This is the dominant mode seen in power at MUL. 1,4 and 1,8 Hz modes These modes are observable in frequency only at Pegasus. The 1,4 Hz mode is also observable in power in the PEG-DRAK line. Note: The timetdamping values given in the decay time constant, for an oscillation to decay to 1/e (37%) of its starting value. Fig. 2: Mode shape of selected modes observed in frequency signals. These oscillations are seldom more than 15 mhz in amplitude, and therefore not a threat. In other power systems, stability problems have been observed around the same frequency, but with much larger amplitude and causing significant disturbances. It is useful to monitor these low frequency oscillations to ensure that they continue to be small and well damped. Poorly damped oscillations In a number of cases, poorly damped oscillations occurred at various frequencies. The 0,3 Hz and 0,7 Hz oscillations were of particular interest, although there were several other cases. During the period reviewed, there were no cases of largeswing oscillatory instability observed. In Fig. 3, a 0,3 Hz oscillation is shown that occurred on 8th January The oscillation was about 30 mhz in amplitude, and in phase at all three monitoring points. Power oscillations were also observed at all three locations reaching about 20 MW peak-to-peak. However, given that the 0,3 Hz mode is believed to be an oscillation between South Africa and its north-eastern neighbours, the power oscillations could be significantly larger in the international AC interconnection. It may be noted that the 0,3 Hz event follows a 300 mhz slide in frequency, crossing the boundary of normal operation at 49,85 Hz. The oscillation ends with another frequency slide of 80 mhz, possibly due to generation tripping outside South Africa or interconnection tripping. This second frequency slide crosses the 49,7 Hz boundary at which load is shed. The 0,7 Hz mode also has periods of poor damping, in some cases coinciding with poor damping of the 0,3 Hz mode. There appears to be a linkage between these Disturbances A number of disturbances were detected and studied, including both network tripping events and frequency disturbances. One event of particular significance, involving a cascade of generation and interconnection tripping is described in this section. On 27 Dec'09, three coal-fired units tripped, separated by seconds. Each unit was producing approximately 600 MW. These events are clearly seen in the frequency as three sharp steps. As the third unit tripped, a further 335 MW was lost in the Southern Africa Power Pool (SAPP). Frequency Type MW Shed 49,7 DMP ,2 UFLS Stage ,1 UFLS Stage 2517 Total 1448 Table 1: Load shedding following 27th December 17:06 SAST event. energize - June Page 34

3 Due to the large frequency excursion, under frequency load shedding stages were activated. These are noted in Table 1. Thus, a total of 2135 MW generation was lost in the space of a minute. The load was very low around 27th December because it is summer and industries are typically shut-down over the holiday period. The total load for South Africa during the time of this event was approximately MW meaning that this generation trip was around 9% of the total load. Since most of the loss was coal generation with relatively high inertia, the total reduction in inertia of the system could be significantly greater than 9%. The rate of frequency drop is related to the inertia of the system, and can be a useful indicator for validating the system model. This value can change quite significantly. On December, Fig. 3: 0,3 Hz frequency oscillation, 8th Jan'10. Fig. 4: 0,05 Hz oscillations preceding second disturbance on 27t Dec there were two significant disturbances in frequency. The event at 17:06 SAST, shown in Fig. 5, includes three frequency drops in succession. The first two show a rate of change of frequency of 0,05 4 Hz/sec, while the gradient of the third event is much steeper, around 0,20 Hz/ sec. Another event just under 8 h later (Fig. 4) showed a frequency gradient of about 0,069 Hz/sec, which is noticeably steeper than the initial two events of the previous disturbance. The gradient relates to the change in power and the system inertia. A steep gradient occurs with low inertia in the system, which in this case relates to fewer generators and less load in the network. After the two-stages of under-frequency load shedding (UFLS 1&2) occurred, the frequency recovered to approximately 49,25 Hz at which point a total of eight gas and pumped storage units autostarted. During the start-up phase of these generators, approximately 30 MW per machine is drawn until the unit can be synchronised to the system. This explains the dip in frequency after the UFLS action. Once synchronised, the units ramp up to full power which is evident in both the frequency and angle difference data from the PMUs. By observing the angle difference data for both Grootvlei-Muldersvlei as well as Grootvlei- Pegasus, it can be seen that the generators that tripped are situated in a region that supplies Pegasus. The Grootvlei-Pegasus angle increases to compensate for the power lost near Pegasus whilst the Grootvlei-Muldersvlei angle stays the constant. The Grootvlei- Muldersvlei angle decreases only when the UFLS is activated, as the load was shed nationally. The Grootvlei- Muldersvlei angle decreases further as the pumped storage and gas units in the Cape are auto-started to recover the system frequency. The auto-starting of the Drakensberg pumped storage units can be seen in both the Grootvlei-Pegasus angle difference as well as the Pegasus- Drakensberg line power. This means that the machine ramp response can be observed regionally in the angle difference data and this can be particularly useful for modelling. In terms of angle deviation, the 27/12/09 event shows a sharp drop in the angle separation between Grootvlei and Muldersvlei (see Fig. 7). Together with frequency, the relative changes in angle indicate where generation and load has been lost. In this case it shows that mainly generation is lost (frequency drops), and that the main loss is at the Grootvlei end of the corridor, rather than the Muldersvlei end of the corridor, as the angle reduces. Fig. 7 shows that the angle difference is significantly different from the normal behaviour, indicating that the system is operating well outside its normal operating conditions. If angle differences were to widen, this would indicate a greater likelihood of transient instability or thermal overload. The angle differences across the network are potentially valuable information for wide area defence schemes that intelligently respond to multiple contingencies with a locational response. Lessons learnt The following lessons were learnt about the power system from this event: The large difference between the initial load shedding (DMP) and UFLS1 levels meant that there was no load shedding to counter the second energize - June Page 35

4 contingency, leading to a continuing frequency slide. With reduced inertia because of the initial condition and the disturbance, the rate of change of frequency can be unusually large, leading to an unexpectedly deep depression of frequency. Emergency generator start-up depresses the frequency, and this must be accounted for in restoration plans. The frequency overshoots to 50,2 Hz because of the generator response. This is outside the boundary of normal operation (50,1 Hz). The inter-area damping is degraded following the disturbance. The event was also instructive in the use of WAMS during and after a severe network disturbance: Phasor measurements help to regionalise the source of the problem. In this case, the cascading disturbance events occurred over a period of 1 min, which is easily within the timeframe that an automated defence scheme could operate, but challenging for an operator. Angle information can be a key input to wide-area defence schemes. The precision of the phasor measurement timestamping and the high sampling rate provide a very valuable resource to understand the disturbance. In contrast with other data sources for investigation, the information is immediately available and time-aligned. Operational warnings could be produced, for example on angle difference and damping, which distinguish a more critical operating condition from normal disturbances. Alarms can be selective for wide-area problems. Practical PMU performance In identifying the PMU models to be used in a large-scale WAMS project, it is important to identify the steady-state and dynamic performance of the devices, and also to gain practical experience of using the devices. The current IEEE standard for synchrophasor measurement, C37.118(2005), does not define the dynamic response of PMUs, and therefore compliance to the standard is not sufficient for use in dynamics applications. The focus in this project, however, was on the practical aspects of PMU performance that are not necessarily detailed in formal product testing. In the pilot project, two different models of PMU were used, and differences were noted that informed the eventual selection of the PMUs for the large-scale WAMS deployment. PMUs communicate with the central server (or Phasor Data Concentrator, PDC) by streaming data continuously in response to a request from the PDC. This data stream may be broken by a communications interruption, for example due to congestion in the communication channels. Since no special communications were installed for the pilot project, this occurred relatively frequently. The PDC detects the break in data and re-sends a request for data. As the WAMS system becomes missioncritical, absence of data could significantly degrade the reliability of the system. The quality of the data is confirmed by formal testing for a specified range of operating Fig. 5: Frequency disturbance on 27 Dec Fig. 6: Angle difference oscillations, low inertia condition, 27 Dec conditions. However, the data should not be significantly misleading in any credible system condition. It was noted that one of the PMU models reported appeared to report a zero voltage incorrectly, showing a volatile voltage and frequency, as illustrated in Fig. 8. One of the PMU models used did not exhibit the above issues, and the experience of extended operation of this PMU was good. This informed the choice of model for largescale deployment, although there were other performance criteria that had to be satisfied before the final choice was made. energize - June Page 36

5 Fig. 7: Muldersvlei-Grootvlei angle difference during 27 Dec 2009 compared with observed distribution. Fig. 8: Incorrect reporting of low voltage condition. Large-scale WAMS system The move to a large-scale deployment of WAMS is motivated by a need for more detailed and accurate information in order to improve the security and performance of the transmission system. The information is required in real-time for operational needs, and also for post-event analysis and planning requirements. The pilot project revealed interesting and useful observations on the operation of the system, but there is a limit to the depth of analysis that can be achieved with only three points in a large power system. The large-scale WAMS will therefore involve a major roll-out of PMUs across the entire system. The initial phase of the WAMS deployment is expected to include 28 substations with multiple circuits monitored at each substation. In this phase, around circuits may be instrumented. The system has been designed to be expandable in the future to 3500 phasors, which could provide full observability of the transmission system with phasor measurements. A deep penetration of PMU measurements throughout the system is beneficial in many ways, including: Consistent identification of dynamics problems across the whole system Ability to identify the sources of oscillation problems Operational facilities to drill down and see the detailed dynamic behaviour when problems occur Clear indication of exactly where system splitting occurs, and valuable information for reducing impact and improving restoration time Improved state estimation accuracy, robustness and solution time Phasor-based WAMS also provide a new approach to alarming on events affecting a wide area. In contrast to conventional EMS signal-by-signal alarming, the WAMS approach centralises and time-aligns incoming raw data before identifying alarm conditions, and there are therefore new opportunities to identify system conditions using several signals simultaneously. This can be useful, for example, in identifying threats to voltage stability. Also, the WAMS data has a sufficiently high time resolution to observe system dynamics. As the use of the system and experience with WAMS increases, it is expected that phasor-based control mechanisms will be developed in areas such as voltage stability control and emergency defence against multiple contingencies. A robust infrastructure and monitoring platform is a pre-requisite for development of control schemes. Contact Brian van Rensburg, Actom Transmission & Distribution, Tel , brian.van-rensburg@actom.co.za energize - June Page 38