当前位置:首页 >> 电力/水利 >>

minute-to-minute variations on power system operation


150

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

Impacts of Wind Power Minute-to-Minute Variations on Power System Operation
Hadi Banakar, Memb

er, IEEE, Changling Luo, Member, IEEE, and Boon Teck Ooi, Life Fellow, IEEE
Abstract—In this paper, minute-to-minute wind power variations are decomposed into slow, fast, and ramp components to assess the in?uence of each component on power system operation. Using detailed, long-term simulation models, this paper con?rms that most power systems can absorb the impacts of wind power variations with little dif?culty. Yet, since ramps in wind power and system demand could coincide, systems with limited ramping capabilities are at risk. It is shown that extending simulation models to include load dynamics and automatic generation control (AGC) time delays do not alter these conclusions. This paper also discusses wind penetration approaches and control area performance measures, linking the latter to the placement of the wind farms within the interconnection. Index Terms—AGC model, automatic generation control (AGC), long-term dynamic model, penetration limit, system dynamic response, wind events, wind farm model, wind power.
Fig. 1. Bode diagram for transfer function of the power system in [3].

I. INTRODUCTION HIS PAPER is a continuation of the research carried out by the authors on issues related to the integration of wind power facilities into power systems and achieving high wind penetration [1]–[4]. The earlier reports focus on the power system response to wind power ?uctuations. In [2] and [3], it is shown that these ?uctuations are attenuated on entering into the power grid, and the transfer function of the power system can be used to quantify contributions of each device category to the attenuation. Fig. 1 displays the transfer function Bode diagram for the power system of [3]. Regions C, B, and A, respectively, identify frequency ranges where attenuation is largely due to generator inertias (rotor shafts), primary controls (speed governors and steam turbines), and secondary control (AGC). The minute-tominute operation roughly corresponds to the frequency range of , which is largely in region A. In this range, units’ governor systems and AGC attenuate slow wind power variations, and the system remains stable for typical AGC time delays [3]. The goal of this study is twofold: 1) to validate the results of [3] by detailed, long-term digital simulations; and 2) to
Manuscript received April 11, 2007; revised September 12, 2007. This work was supported by the Natural Sciences and Engineering Research Council of Canada under a strategic grant on reducing greenhouse gas emission. Paper no. TPWRS-00236-2007. H. Banakar and B. T. Ooi are with the Electrical and Computer Engineering Department, McGill University, Montreal, QC H3A 2A7 Canada (e-mail: h_banakar@hotmail.com; ooi@ece.mcgill.ca). C. Luo is with SIMSMART Technologies, Inc., Brossard, QC J4W 3J9 Canada (e-mail: cluo@simsmart.com). Color versions of one or more of the ?gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identi?er 10.1109/TPWRS.2007.913298

T

complement the earlier works by examining system response to minute-to-minute wind power variations in conjunction with the requirements for reliable system operation. Operational reliability is de?ned here in terms of the system’s dynamic response alone, requiring system frequency to stay within the limits of the protection system for a set of wind-related events. This de?nition makes “acceptable system dynamics” the sole criterion for increasing wind penetration. Wind-related events, or simply wind events, include wind die outs, wind rises, wind lulls, and wind gusts, as well as sudden loss of a wind farm (WF). These events are treated as minute-tominute variations, since their multi-minute durations do not ?t into the hourly or second-by-second time frames. The rest of this paper is organized as follows. Section II provides some background material, while Section III examines wind power characteristics. Section IV presents the study’s approach, and Section V discusses simulation models. Section VI introduces the simulation environment, the power system, and the wind events used for the study, setting the stage for examining system response to wind power fast ?uctuations in Section VII and wind events in Section VIII. Section IX analyzes the system response following the loss of a wind farm, which provides an opportunity to discuss a number of operational issues in Section X. The study results and key conclusions are summarized in Section XI. II. BACKGROUND A. AGC Feedback Loop Time Delay The AGC system enables generators to track automatically system demand variations. Using a SCADA system, AGC periodically receives frequency and real power measurements and

0885-8950/$25.00 ? 2008 IEEE

BANAKAR et al.: IMPACTS OF WIND POWER MINUTE-TO-MINUTE VARIATIONS ON POWER SYSTEM OPERATION

151

Adopting any one of these approaches can have broad implications for a power system, as they in?uence issues as diverse as “long-term fuel contracts” and “capacity credits” [20]. The interest here is primarily in their penetration effect.
Fig. 2. Block representation of the AGC feedback loop.

D. Protection System Operation Power system devices are protected by the circuit breakers that connect them to the power grid. Each breaker is attended by a protection relay and operates upon receiving an open/close signal from the relay. When a device is online, its associated protection relay monitors those device variables that de?ne its safe operation. The monitoring is done by periodic acquisition of relevant values and checking them against their prede?ned limits. Once a variable violates its limit, the device breaker would be signaled to open, if the violation lasts longer than a preset period. Under/over frequency relays protect the system against large frequency swings and sustained frequency errors. Their limits and “timings” are carefully chosen to rule out relay actions due to routine operational activities. III. WIND POWER CHARACTERISTICS A. Wind Turbine Generator (WTG) Output Power The real power output of a WTG, , is given by (1) where air density, area swept by turbine blades , gear ratio, of radius , wind speed, rotor speed, pitch angle, and power ef?ciency curve [13]. The output of a wind farm can be estimated from its WTG powers, considering the WF layout and wind speed time delays in reaching individual WTGs [4]. B. Wind Power Components As shown in Fig. 3, in the minute-to-minute time frame, the , can be decomposed into total wind power in a system, ; 2) a fast three components: 1) a slow moving average, ; and 3) a ramp event, ; that is ?uctuating part, (2) As in Fig. 3(a), plotting allows , if present, to [22] yields be easily spotted. Then the trend of , giving by . Here, the rolling or trending window is 1-min long, as it suits the time components are further described frame of interest. The below. : As shown in Fig. 3(c), has zero mean and 1) high-frequency parts. These parts, originating from wind speed despite being ?ltered by turbulences, remain partially in WTGs’ spatial distribution and blades inertia [4]. is indistinguishable from the For low wind penetration, demand noise and has no operational impact. For high wind penetration, the coalescence of many uncorrelated WF outputs leads , reducing the total harmonic to statistical smoothing of distortion of [4]. So, is a concern only when the penetration level is amid these two extremes.

computes the area control error to adjust set-points of AGC-controlled online units. The adjustments are transmitted by the SCADA system to power plants where they are realized via remote terminal units (RTUs) [16]. Subsequent to the loss of a generator or load, AGC forces the resulting ACE to zero, correcting the frequency error while moving unit outputs towards their secure/economic targets. These targets are usually set by the applications running on the utility Energy Management System (EMS). As shown in Fig. 2, an AGC feedback loop often contains several processing points and communication links, causing a time delay between an event occurring in the system and its AGC commands arriving at the RTUs. In conventional AGC simulations, this time delay, which ranges from a few to many seconds, is lumped with the units’ response times. Here, it is separately modeled to study its impacts on the system response, in the presence of wind power variations. B. Online Reserves and Operational Reliability To improve their operational reliability, power systems keep suf?cient online reserves to cover potential loss of their largest in-feeds. This requirement is typically enforced in three stages: 1) in the generation scheduling stage, where reserve constraints are imposed on the schedules; 2) in the approval stage, where system operation is simulated based on the computed schedules; and 3) in online operation, by keeping generator outputs close to their approved schedules. The detailed simulations of the approval stage are needed, as scheduling packages do not fully model many aspects of the system and its operation. Besides, suf?cient online reserves do not ensure survival of the system following an outage, as they are slow in in?uencing the system initial response. C. Wind Penetration Approaches Wind penetration limit is de?ned as the extent to which wind power can be added to a power system without compromising its operational reliability. Since WFs are often not equipped to provide reserve or regulation and, when they are, they cannot be fully relied on, such capabilities must be supplied primarily by the conventional units. Successful wind penetration approaches, therefore, are built on effective leveraging of the existing units’ capabilities. Two basic approaches for increasing wind penetration are: 1) coordinated and 2) expedited. The coordinated approach requires adding new WFs to the system in step with the system demand growth. Thus, the wind power is added to meet the new demands only and not to displace the existing units. Conversely, the expedited approach aims at displacing the conventional units with the new WFs and presumes that all WFs will become operational within a short period.

152

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

Fig. 4. (a) Wind die out. (b) Wind rise. (c) Wind lull. (d) Wind gust.

Fig. 3. Decomposition of a typical WF output power into its components.

down-ramp. A similar event is encountered when a wind rise exceeds the WTGs’ cutout speed, causing them to shutdown. IV. METHODOLOGY The objectives of this paper are approached by performing detailed digital simulations on a test system having power plants, WFs, transmission system, loads, and AGC. Consistent with performing multi-minute simulations, long-term models have been deployed for the thermal units. The test system has been given ample fast primary reserves and VAR sources to restrict dynamic variables of interest to load and generation bus frequencies. Frequency relay actions are not simulated but deduced via comparing relay limits and durations against dynamics of the monitored variables. As the simulated events primarily occur at the WFs, it is important to have authentic WF responses. Therefore, WFs are modeled at their outputs, using measurements collected at three existing WFs (supplied by US NREL [18]). V. SIMULATION MODELS The generator models are those suggested by IEEE [6] and model parameters are those in [1], save for the values listed in Appendix A. The nonclassical models are described below. A. Wind Farms The WTGs simulated here are assumed to be made up of doubly-fed induction generators (DFIGs). By running asynchronously, DFIGs largely decouple the WF’s output from the system frequency. Yet, they provide voltage support to the WF’s bus via their fast controls. These features allow a WF to be modeled at its output level, as opposed to its WTG level. The model then needs a time series describing the WF’s real power output variations, a bus voltage target de?ning its reactive power obligation, and a means of converting these quantities into P and Q injections at the WF’s bus. , , and can be Based on (2), for each WF, . is a slow built separately and joined to obtain ramp that can be de?ned by the user, obtained from the WF’s

2) : As indicated by (1), WTGs outputs vary with the cube of the wind speed, indicating that a modest increase in wind speed can substantially raise their outputs. Such wind as ramps, expressed by . speed changes appear in resulting from a lull in the wind. Fig. 3(d) shows : is the trend or the moving average of 3) in the absence of [22]. As shown in Fig. 3(b), it holds dc . The rate of change of and low-frequency parts of is typically well within the range of thermal units’ ramp rates. C. Wind Events , vary with the WF Wind events, represented here by layout. Within a WF, WTGs are typically interspaced in arrays of rows facing the prevailing wind direction. This layout causes a wind front passing over the WF to encounter the WTGs “one row at a time,” initiating a ramp in the WF’s output. Such a ramp may last for minutes and, at times, can be steep [19]. Using linear representations of their associated ramps, four key wind events are shown in Fig. 4. 1) Wind Die Out: A wind die out refers to a persistent drop in the wind speed which, as shown in Fig. 4(a), results in the WF output to be ramped down and ultimately becomes zero, once the speed falls below the WTGs’ cut-in speed. 2) Wind Rise: Based on (1), every 1% change in the wind speed causes the WTG outputs to change roughly by 3%. Therefore, a sustained rise in the wind speed can create an up-ramp in the WF’s output, similar to the one depicted in Fig. 4(b). 3) Wind Lull: Wind die outs are inevitably followed by wind rises. When the two events happen in short succession, they form a “wind lull,” depicted in Fig. 4(c). Wind lulls are more common than wind die outs or wind rises, as spells of low or high wind speeds are normally ephemeral. 4) Wind Gusts: As shown in Fig. 4(d), a wind gust is opposite of a wind lull; i.e., it starts with an up-ramp and ends with a

BANAKAR et al.: IMPACTS OF WIND POWER MINUTE-TO-MINUTE VARIATIONS ON POWER SYSTEM OPERATION

153

Fig. 6. Block diagram of AGC main processing modules.

including the AGC time delay, , and assuming AGC actions is expressed in the s-domain by to be continuous in time,
Fig. 5. Block diagram for wind farm intermediate dynamic model.

(3) is passed through a ?rst-order ?lter, As shown in Fig. 6, , to remove its noise and to ensure that obcharacterized by are persistent. The result, when large served changes in enough to overcome the dead-band, is fed to a PI-Reguand gains, to obtain the desired lator, distinguished by its . Again, in the s-domain generation change,

past output measurements, or constructed from the WF’s output usually comes in the form of a forecasts. The forecasted step function and has to undergo a spline interpolation to exhibit the desired slow variation. is obtained by removing As described earlier, from a portion of measurements that does not contain . The resulting , however, needs to be suitably scaled in the simulation. to correspond to the value of characterizes the event’s worst-case For each event, scenario. When the available WF output records do not contain suitable data to represent , it can be formed using the WF’s historical wind speed data and physical layout. Fig. 5 shows the WF model made up of: 1) a Plant Controller; 2) a Power Pro?ler; and 3) an S-Modulator. The Plant Controller, based on its operation mode, decides on options and set points needed by the Power Pro?ler and S-Modulator. The Power Pro?ler generates the time series de?ning the WF output power, consistent with its speci?ed options. The S-Modulator, described in [1], converts the selected time series into controllable, instantaneous, complex power injections at the WF bus. It also moves the bus voltage to its target (typically set to 1.0 p.u.) by monitoring its value and adjusting the injected bus MVAR accordingly. B. Thermal Units In the context of power system dynamic simulation, one minute is a long time. As such, one needs to deploy turbine models that are suited for long-term simulations. When modeling steam turbines, drum pressure is often assumed to be constant [7], [8], implying that the drum is large enough to supply the required steam with little impact on its pressure. This leads to a fast unit response, acceptable when the initial pressure is known to be high and the simulation lasts only for a few seconds. However, when the simulation is carried out over minutes, it is crucial to represent the drum’s limited size and variable pressure, which entails modeling the unit’s fuel supply system and its links to the drum pressure [9]. Such a modeling captures the effects of both instantaneous and long-term unit ramp rates on the system response. C. AGC System In an isolated power system, power imbalance is estimated by evaluating , where is the frequency is the average of frequency errors measured bias factor and at several power plants. By ignoring adaptive changes to ,

(4)

to generating units, it is asTo simplify allocation of sumed that AGC-controlled units are both regulating and dis, do patchable and their participation factors, , not change during the simulation period. Recalling that and , unit set-point changes are given by (5) D. System Demand Dynamics Variations in system frequency set off-related changes in the system demand that can be expressed in the s-domain by (6) is the system demand when , represents In (6), motor inertias, and denotes the sensitivity of system demand is always positive, could to frequency changes. While take on positive or negative values [10]. VI. SIMULATION PARTICULARS A. Simulation Environment To perform the required digital simulations, the authors have access to Hypersim [8]. Except for using models that are introduced in Section V, simulations are based on the models recommended by the IEEE Committee [6], which are available as part of the Hypersim components (device objects) library. To perform the simulations, a time step of 0.2 ms is used. B. Simulated Power System The simulations are performed on the isolated system of Fig. 7, which contains six thermal units, six load centers, and two wind facilities, connected together by short transmission lines. Each unit is 162.3 MW in capacity while the wind facilities are rated for 200 MW (WF1) and 350 MW (WF2). The WFs are DFIG-based and assumed to be driven by uncorrelated

154

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

TABLE II SIMULATED WIND EVENTS (ALL OCCURRING AT WF2)

Fig. 7. Schematic representation of the simulated power system. TABLE I INITIAL CONDITIONS FOR THE SYSTEM OF FIG. 7

(Forecasts: P

= 753:0 MW; P

= 55 MW; P

= 100 MW).

wind régimes. This assumption is needed to limit the impact of a wind event to a single WF. All units are equipped with governor systems, but only three are controlled by AGC. Table I shows generation schedules for the peak hour, based and WF proon forecasted system demand of and . Also shown ductions of are unit primary reserves (PR) in both “Raise” and “Lower” directions, covering potential outage of the most heavily loaded unit (PPB_T2 at 120 MW) as well as the loss of both WFs. No secondary reserve is listed since all units are online. C. Simulated Events and Related Goals The performed simulations fall into three categories: 1) normal operation—to con?rm conclusions of earlier studies on ; 2) wind events—to assess impacts of on system operation; and 3) WF and conventional generator outages—to appraise their in?uences on wind penetration approaches. Load dynamics and AGC time delay impacts are studied as part of category 2 simulations. Details of the simulated events are given in Table II. In all cases, thermal unit outputs are initially set to those shown in Table I. Wind events are directed at WF2, while WF1 is assumed to have a constant wind speed. In each case, the value of is decided by the load-generation balance requirement at time . VII. IMPACTS OF WIND POWER FAST FLUCTUATIONS to have limited impact on Based on Fig. 1, one expects the minute-to-minute system dynamics. To verify that, system
Fig. 8. (a) Actual WF output. (b) Resulting system frequency error.

frequency dynamics is studied when is assigned the WF was kept output power measurements of Fig. 8(a), while at 12.0 MW. The AGC function was turned on and its 50-mHz at dead-band was enforced. Consistent with and thermal units’ schedules, the system demand was , making the total wind power 9.5% ?xed at . of Simulation of frequency dynamics when assumes the pro?le of Fig. 8(a) yields the result of Fig. 8(b). As predicted by the frequency response of Fig. 1, the wind power variations are smoothed by units’ inertia and primary controls. There is little AGC action in this case, as 50-mHz dead-band of does not allow AGC to detect the frequency error. The system , as units alter their rotor frequency error largely follows speeds, releasing/storing their rotational energies, in order to keep system generation and demand in balance.

BANAKAR et al.: IMPACTS OF WIND POWER MINUTE-TO-MINUTE VARIATIONS ON POWER SYSTEM OPERATION

155

Fig. 9. Responses to a wind rise and a wind die out (

= = 0;  = 4 s).

Fig. 10. Response to a wind lull and a wind gust (

= = 0; 

= 4 s).

The frequency response can be summarized here by an average error of 2.27 mHz, standard deviation of 10.60 mHz, and a variation range. These values are relatively small, noting that governor dead-bands are often set dead-band can be 50 mHz to 10 40 mHz and the or larger, depending on the system size. Based on these results, henceforth, the fast wind power ?ucis astuations are ignored, and, consistent with Fig. 3(b), sumed to have no variations during the simulations. VIII. IMPACTS OF WIND EVENTS This section presents system responses to the wind events of Section VI. To allow comparisons with other works, load dynamics is initially ignored to generate the worst-case results [14]. Then the simulations are repeated in the presence of load dynamics. All simulations start from the steady-state condition with no frequency error, while WFs and generators take up the values listed in Tables I and II. A. Worst-Case Analysis 1) System Responses to Wind Rises and Wind Die Outs: The chances that a wind rise or die out at a WF can pose risks to the power system operation are slim since: 1) power systems are equipped to handle their daily ramps, which are usually steeper than a WF output ramps; and 2) the resulting ramps typically last long enough for system operators to invoke system secondary reserves, if required. Yet, they are simulated here as there are exceptional situations where they can lead to operational complications (see Section X). Moreover, knowing the system response to these basic events allows responses to more complicated events to be anticipated. Wind rise and die out events are separately simulated by letvary as in Fig. 4(a) and (b), using the ramp data ting of Table II. For both events, as shown in Fig. 9, the system frequency largely follows the impelling ramp, engendering responses that appear to mirror each other. However, a closer examination reveals that for the wind die out, the frequency error tumbles to 515 mHz while for the wind rise, it ascends to 480 mHz. The difference is attributable to higher units’ ramp rates in the “Lower” direction, which leads one to conclude that, operationally, wind rises are less serious than wind die

outs. Also note that WFs with pitch-controlled WTGs can safely “spill” the wind rise excess power. When the wind event is in progress, due to the time delays, unit set-point adjustments do not correspond to the frequency error being measured at the time. By the time the adjustments turn into actual unit output changes, the system frequency has moved further away. As a result, unit output movements follow the event’s ramp with a time delay. The delay in this case is small, as the generators can supply their small shares of largely with their fast-responding stored boiler energies. At the conclusion of a wind rise/die out event, unit output changes are still driven by the earlier AGC commands, causing the demand-generation gap to rapidly narrow. Nevertheless, as the momentum of past control commands is lost, corrections occur at a slower rate. Since the system is stable, AGC eventually moves the units to their correct set points, diminishing the frequency error. The extreme frequencies in Fig. 9 are not large enough to be operationally problematic, as the lowest relay setting for frequency in many systems is 600 mHz ( 1% error), and it has to last 10 min or longer to trigger a relay [2]. 2) System Responses to Wind Lulls and Wind Gusts: A wind lull is a wind die out, followed by a pause in the wind speed change and then by a wind rise. A wind gust is made up of the same events, but in the reverse order. As their basic ramps (wind rise and die out) do not individually cause operational problems, wind gusts and wind lulls are not expected to do so either. Such a view, however, ignores the “slingshot effect” arising from the two opposing ramps occurring in succession. At the end of the ?rst ramp, long time delays of AGC and thermal units’ fuel supply systems create momentum for the units’ outputs. When the pause period is short, the units will be still responding to the ?rst ramp as the second ramp starts, which cause the unit outputs to gather much larger momentums at the end of the second ramp. As a result, well after the WF output returns to its original level, the system frequency is undergoing changes. Fig. 10 shows the system responses to a wind lull and a wind gust, which seem to be formed from the two basic responses in Fig. 9, again creating mirror images. As indicated in Table II, for both events, the WF regains its initial generation at the end of the second ramp. In the pause period, frequency errors behave as they did at the end of the wind rise/die out events in Fig. 9. However, starting with the second ramp, the frequency errors’

156

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

Fig. 11. Response to a wind die out (

= 0 02;

:



= 0 005;

:



= 4 s).

Fig. 12. Response to a wind lull (

= 0 02;

:



= 0 005;

:



= 4 s).

rates of change are noticeably higher, as the power imbalances rapidly decrease. As anticipated, larger frequency swings occur at the end of the second ramp. Note that, for the wind lull, the , while frequency error varies in the range this range is for the wind gust. Yet again, the differences can be explained by the discrepancies in units’ ramp rates in “Raise” and “Lower” directions. Once more, the resulting frequency errors are small enough to rule out operational problems. This is largely because the system ramp rate is still much larger than the rate of WF output changes. In the future, as inland WFs form clusters and offshore WFs’ capacities reach 1000 MW and higher, larger ramp rates due to wind event can be expected. B. System Response Including Load Dynamic Model To include load dynamics in the simulations, appropriate values are needed for and in (6). These parameters are dif?cult to measure or estimate, since they vary with the composition of the system demand which changes during the day and seasonally. In a given system, rotational loads whose speeds are coupled to the system frequency, either by running synchronously or by having speed control mechanisms, set the . Since such loads are relatively few, is typically value of is set to 0.005, which adds 0.7 s of normalized small. Here . For the system of inertia to the system is set to 0.02, representing 15 MW of load relief Fig. 7, for each 1.0-Hz drop in the frequency. When in (6) is and , which agrees normalized, one has with the values reported in [10]. Figs. 11 and 12 show the simulation results for the same wind events generating Figs. 9 and 10 but this time for and . These responses differentiate themselves from their counterparts by: 1) faster initial response; 2) lower peak frequency errors; 3) almost constant frequency error over 75% of the ramp periods; and 4) at the end, faster returns to smaller frequency errors. These differences stress the signi?cance of including load dynamics in evaluating system dynamics. Attaching too much weight to the above listed differences and are may overstate the case since, here, values of rather arbitrarily chosen. Nonetheless, it can be safely concluded that excluding load dynamics from the simulations: 1) produces larger frequency deviations for the same events; and

Fig. 13. Response to a wind gust for different values of  (

=



= 0).

2) leads to underestimating the wind penetration limit, when it is set by the frequency dynamics. C. In?uence of AGC Time Delay The simulation results presented in Figs. 8–12 were obtained for an overall AGC time delay of 4 s. Thus, one may wonder if larger time delays could signi?cantly alter any of those results. Fig. 13 shows the system response to the wind gust of Fig. 10, and the AGC time delay is set to , 2, when 4, 6, 8 s. The graphs indicate that the selected time delays have relatively small impacts on the frequency dynamics. To establish that even in the presence of system demand dynamics, the above conclusion stays valid, system response to the and , while wind lull in Fig. 12 is obtained for and . Again, as shown in Fig. 14, the load dynamics and AGC time delay do alter the shape of the system response but only to some extent. IX. SYSTEM RESPONSE TO LOSS OF A WIND FARM Behavior of a power system following the loss of a WF is important to the system’s operation and generation expansion. Here, the system response to a WF loss is examined and its operational and planning implications are discussed. A. System Response to Loss of a WF System dynamics following the loss of are presented in Fig. 15 for two cases: 1) AGC online; and 2) AGC of?ine. For comparison, responses to the outage of

BANAKAR et al.: IMPACTS OF WIND POWER MINUTE-TO-MINUTE VARIATIONS ON POWER SYSTEM OPERATION

157

TABLE III UNDER-FREQUENCY RELAY RESPONSE TO LOSING P

OR

P

Fig. 14. Response to a wind gust for = 0:005).



= 0s



= 8 s (

= 0:02,

Fig. 15. System response to separately losing P and P puts, with and without AGC ( = = 0;  = 4 s).

at 120-MW out-

for the same two cases are also shown. The relevant under-frequency relay settings (limits and durations) are also added to Fig. 15 (dotted lines) for reference. or the outage of , stored enFollowing the loss of ergies in online units’ inertias are initially tapped to replace the lost power. As Fig. 15 indicates, for the test system, this pe). By that time, riod last almost 3 s (simulation starts at thermal units’ primary controls “kick in,” supplying power ?rst by drawing on their “responsive reserves” (i.e., stored energies in their boilers) and, then, by increasing fuel ?ows at their supply systems. These two power production mechanisms, as shown in Fig. 15, lead to two distinctly different dynamic behaviors in the 15 25 s and 30 70 s periods, providing thermal units with “instantaneous” and “long-term” ramp rates. AGC is expected to have no in?uence on the initial system redead-band, sponse, as AGC commands are delayed by the the AGC feedback loop, and units’ slow response. Yet, there is a noticeably smaller frequency dip when AGC is online. This is mainly due to the way the simulations were conducted here. For each case, ?rst the loss of the WF or unit was simulated with “no AGC.” Then the lost WF or unit was put back online and AGC was activated. Once the steady-state condition was fully reestablished, the WF or the unit was tripped again to obtain

the system response with AGC online. This approach allows the thermal units to build up their responsive reserves by the time of the second outage, and it is the release of these fast reserves, following the outage, which results in smaller frequency dips. The small steady-state frequency errors are due to enforcing the dead-band. Table III summarizes frequency relays’ reactions to the frequency responses of Fig. 15, should they were simulated. For the 16 different cases considered (two events, two AGC status, four relay settings), there are ?ve instances where no relay limit is breached. Then, for six of the remaining 11 cases, the violations last long enough to trigger the relays (shown in grey). For both cases, the frequency dips subsequent to the loss of are roughly 20% smaller than the frequency dips for the outage. Since, at the time of their outage, the two facilities had identical outputs, one can reasonably conclude that the loss of a WF is operationally “softer” on the system than outage of a comparable thermal unit. This notion is supported by the following facts: 1) due to the asynchronous coupling of WTGs to the power system, loss of a WF does not reduce the system’s rotational inertia, as opposed to the outage of a thermal unit; 2) the outage of the thermal unit comes with the loss of that unit’s controls, including capabilities of its governor and power system stabilizer. In contrast, WFs have limited real power control capability, in the ?rst place; and 3) for a system with N conventional units and a WF, loss of one unit leaves the reserves of only N-1 units to absorb the ensuing dynamics, as WFs normally carry little reserves. Conversely, losing the WF leaves the fast reserves of N units behind to replace the lost generation. The frequency response in Fig. 15 is obtained for . As demonstrated before, nonzero values for and would reduce the size of the frequency dip, implying that a WF loss is in fact less severe than the one shown in Fig. 15. Although larger AGC time delays do alter the system responses, as shown in Fig. 13, the changes are generally inconsequential. B. Wind Penetration Limit For the expedited penetration approach, adding WFs to the system leads to the retirement of conventional units, reducing the system’s rotational inertia as well as its ability to control and provide reserves. Since it is not possible to operate reliably a power system without these capabilities, it follows that this approach is limited by the size of rotational inertias, controls, and reserves available from the remaining conventional units.

158

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

The conclusion that, for a power system, the loss of a WF is “softer” than outage of a comparable generator, can be combined with the coordinated approach to devise a new planning and operation scheme: add WFs to the system in step with the load growth while keeping their locations diverse and their outputs below the output of the most heavily loaded unit. Since the coordinated approach retains the original levels of system inertia and controls, it seems that this strategy allows constantly adding WFs to the system. The above scheme can be rejected on two grounds. Firstly, it presumes that, while wind power is being added to the system, “acceptable system dynamics” would remain the main criterion for increasing penetration. This is certainly questionable. At some point, another consideration, such as transmission capacity, required reserves, or short-circuits currents, could become the dominant factor in setting the penetration limit. Secondly, this conclusion is based on a deterministic view of the system reliability, centered on the outage of the largest grid in-feed. A probabilistic view of the system reliability obliges the system’s loss of load expectation (LOLE) to stay below a set limit [21], [23]. By increasing the number of WFs, this limit will be inevitably encroached, as the probability of a WF producing no or little power is much larger than the probability of a conventional unit outage. X. DISCUSSIONS A. Wind Events and System Demand Ramps On weekdays, power systems routinely deal with their morning up-ramps and afternoon down-ramps. For a system with peak and base demands of 15 000 and 7000 MW, respectively, on average the morning up ramp has a rate of 0.74 MW/s, when it lasts for 3 h. Thus, in principle, this power system should have no dif?culty handling wind events with ramp rates below 0.74 MW/s. The above statement comes with caveats. First, wind events have to be predictable with some accuracy to ensure that, when they actually occur, suf?cient ramping and reserve capabilities are already online. Secondly, considering that wind events statistically occur twice a day [5] and daily demand up ramps typically last 3 4 h, chances for the two ramps to coincide are high. When a system ramping capability is just enough to meet its morning up-ramp, such coincidences can lead to operational dif?culties. Thirdly, as reported in [11], WFs that are clustered in small geographical areas respond to wind events in unison. In that case, the resulting ramp rate will be an aggregate of the clustered WFs’ ramp rates, which could exceed the demand ramp rate. B. Control Area (CA) Performance and WF Location The new NERC performance metrics, CPS1 and CPS2, are variability for a CA over intended to measure the extent of speci?c reporting periods. In these periods, values of CPS1 and CPS2 are, respectively, calculated using 1- and 10-min averages and [12]. of The effects of adding WFs to a CA can be studied in terms components [see (2)]. The impact of of the basic on CPS1 and CP2 is generally small as it varies slowly and is

has zero readily tracked by the load-following units. Since mean and is largely ?ltered by the units’ inertias and governor systems, it can only marginally in?uence the CA 1- and 10-min and and, hence, CPS1 and CP2. Additionaverage has practically no in?uence on CPS1 and CP2 of the ally, heavily in?uences CPS2 neighboring CAs. In contrast, as it lasts for many minutes, producing non-zero 1- and 10-min averages. Furthermore, it negatively affects the performances of the neighboring CAs, as their tie-line ?ows (and, therefore, their frequencies) have to change to assist the affected CA. The above discussion suggests that the location of a WF within the interconnection is important to CPS2, but largely inconsequential to CPS1. This, in part, explains why some favor the use of CPS1 as the main performance metric. C. Loss of System Level Controls Governor system and the power system stabilizer (PSS) are the generating units’ components that partake in system level controls. By retiring a conventional unit, the system can no longer rely on its governor system to perform regulation, load tracking, or provide damping. Yet, operationally, the loss of a governor system may not be as serious as the loss of a PSS, when it happens to have a key role in damping out critical inter-area oscillations. In large interconnections, power transfer levels heavily depend on successful damping of inter-area oscillations. Since this is accomplished primarily by the PSS, displacing PSS-equipped units with WFs may run contrary to the goal of increasing wind penetration. This follows the facts that WFs are mostly located at remote sites and, to deliver their productions to the load centers, substantial increases in power grids’ transfer capabilities are often necessary. D. Wind Events and Protection System As mentioned earlier, the chances for the WFs’ ramps to coincide with system demand ramps are high. In that case, even when the power system has suf?cient ramping capability to handle its daily up-ramps, an ill-timed WF down-ramp could lead to activation of its under frequency relays. Thus, for high wind penetration, not only current and voltage relays need re-coordination [13], but frequency relay settings are to be revisited. XI. CONCLUSIONS The investigations reported here show that the operational impacts of the wind power fast ?uctuations are largely absorbed by the thermal units’ large mechanical and thermal time constants as well as control dead-bands. In addition, the effects of wind events can be operationally contained, as WF output ramps are typically less severe than the system demand ramps. Yet, one cannot rule out operating dif?culties due to steeper ramps resulting from a coincidence of system demand and WF ramps or formation of WF clusters. While the results indicate that the system dynamics varies with the size of the AGC time delay, they also establish that the resulting changes are not signi?cant enough to alter any of the stated conclusions. Conversely, adding load dynamics reduces

BANAKAR et al.: IMPACTS OF WIND POWER MINUTE-TO-MINUTE VARIATIONS ON POWER SYSTEM OPERATION

159

TABLE IV UNITS’ GOVERNOR DROOP SETTINGS AND PARTICIPATION FACTORS

REFERENCES
[1] C. Luo and B. T. Ooi, “Frequency deviation of thermal power plants due to wind farms,” IEEE Trans. Energy Convers., vol. 21, no. 3, pp. 708–716, Sep. 2006. [2] C. Luo, H. Golestani Far, H. Banakar, P. K. Keung, and B. T. Ooi, “Estimation of wind penetration as limited by frequency deviation,” in Proc. IEEE Power Eng. Soc. General Meeting, Jun. 18–22, 2006, 8 pp. [3] H. Banakar, C. Luo, B. Shen, and B. T. Ooi, “Attenuation of wind power ?uctuation by wind turbine generator,” in Proc. IEEE T&D Meeting, New Orleans, LA, Nov. 2005. [4] P. Li, H. Banakar, P.-K. Keung, H. Golestani Far, and B. T. Ooi, “Macro-model of spatial smoothing in wind farms,” IEEE Trans. Energy Convers., vol. 22, no. 1, pp. 119–128, Mar. 2007. [5] B. Ernst, Y. Wan, and B. Kirby, “Short-term ?uctuation of wind-turbines looking at data from the German 250 MW measurement program from the ancillary point services viewpoint,” in Proc. AWEA Wind Power ’99 Conf., Washington, DC, Jun. 1999. [6] IEEE Committee Report, “Dynamic models for steam and hydro turbines in power system studies,” IEEE Trans. Power App. Syst., vol. PAS-92, no. 6, pp. 1904–1915, 1973. [7] E. Cheres, “Small and medium size drum boiler models suitable for long term dynamic response,” IEEE Trans. Energy Convers., vol. 5, no. 4, pp. 686–692, Dec. 1990. [8] “HYPERSIM 8.9 Reference Guide Manual,” TransEnergie, June 2002. [9] D. Flynn, Thermal Power Plant: Simulation & Control. London, U.K.: IEE, 2003. [10] IEEE Task Force on Load Representation for Dynamic Performance, “Load representation for dynamic performance analysis,” IEEE Trans. Power Syst., vol. 8, no. 2, pp. 472–482, May 1993. [11] H. Banakar and B. T. Ooi, “Clustering of wind farms and its sizing impact,” IEEE Trans. Energy Convers., submitted for publication. [12] [Online]. Available: http://www.ercot.com/mktrules/compliance/nerc/ bal/BAL-0010.pdf. [13] T. Ackermann, Wind Power in Power Systems. New York: Wiley, 2005. [14] G. Lalor, A. Mullane, and M. O’Malley, “Frequency control and wind turbine technologies,” IEEE Trans. Power Syst., vol. 20, no. 4, pp. 1905–1913, Nov. 2005. [15] P. B. Erikson et al., “System operation with high wind penetration,” IEEE Power Energy, vol. 3, no. 6, pp. 65–74, Nov./Dec. 2005. [16] C. Strauss, Practical Electrical Network Automation and Communication Systems. Oxford, U.K.: Newnes, 2003. [17] I. Erlich, W. Winter, and A. Dittrich, “Advanced grid requirements for the integration of wind turbines into the german transmission system,” in Proc. IEEE Power Eng. Soc. General Meeting, Montreal, QC, Canada, Jun. 18–22, 2006. [18] Communications with Mr. Y. Wan, National Wind Technology Center (NWTC), National Renewable Energy Laboratory, Colorado. [19] J. Pease, “Challenges with the integration of large scale wind by a regional utility,” in Proc. AWEA Wind Power ’06 Conf., Pittsburgh, PA, Jun. 2006. [20] M. R. Milligan, Modeling Utility-Scale Wind Power Plants–Part 2: Capacity Credit, NREL/TP-500-29701, Mar. 2002. [21] R. Billinton and W. Li, Reliability Assessment of Electric Power Systems Using Monte Carlo Method. New York: Plenum, 1994. [22] Y.-L. Chou, Statistical Analysis, With Business and Economic Applications. New York: Holt, Rinehart and Winston, 1975. [23] R. Doherty and M. O’Malley, “A new approach to quantify reserve demand in systems with signi?cant installed wind capacity,” IEEE Trans. Power Syst., vol. 20, no. 2, pp. 587–595, May 2005.

TABLE V UNITS’ BOILER PARAMETERS (VALUES DIFFERENT FROM [6])

the range of frequency deviations, suggesting that, for some systems, absence of load dynamics would cause their penetration limits to be underestimated. System response subsequent to the loss of a wind farm has been compared with those obtained for the outage of a thermal unit with the same output. The results indicate that the frequency dips due to the WF loss are roughly 20% less than that of the thermal unit outage. It is cautioned not to devise a long-term penetration strategy based on this result, as it is rooted in a deterministic view of the system reliability. When higher wind penetration entails displacing some thermal units, the results point out that, for acceptable system dynamics, the remaining conventional units must provide not only suf?cient rotational energy and reserves but also adequate controls. Above all, displacing units with a critical role in damping inter-area oscillations should be avoided. An analysis of WFs’ impacts on the NERC performance measures indicates that WF location can markedly in?uence the CPS2 measures across the interconnection. Conversely, the CPS1 measures remain largely unaffected when a new WF is introduced in the interconnection. To improve their CPS2, control areas with new WFs need to have suf?cient ramp rates to handle concurrent demand and WF output up-ramps. APPENDIX A. Simulation Parameters Table IV shows the units’ governor droop settings and participation factors, and Table V shows the units’ boiler parameters (values different from [6]).

ACKNOWLEDGMENT The authors would like to thank Dr. L. Dessaint, ?cole de technologie supérieure, for providing access to TransEnergy Hypersim, and Mr. Y. Wan, National Wind Technology Center, NREL, for the supply of wind farms data.

Hadi Banakar (S’79–M’81) received the M.Eng. and Ph.D. degrees, both in electrical engineering, from McGill University, Montreal, QC, Canada, in 1977 and 1981, respectively. Since 1981, he has held key positions at CAE Electronics and ALSTOM ESCA developing EMS and electricity market applications. Presently, he is a power utility consultant and a research associate at McGill University. His research interests are power system operation, operations planning, electricity markets, and wind power integration into power grids.

160

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

Changling Luo (S’01–M’06) was born in Beijing, China. He received the B.Eng. (Distinction) degree from Concordia University, Montreal, QC, Canada, and the M.Eng. degree from McGill University, Montreal. He is presently with SIMSMART Technologies, Inc., Brossard, QC, Canada. His research interests include power electronics, electric machinery, wind energy, and DSP applications.

Boon Teck Ooi (S’69–M’71–SM’85–F’02–LF’05) was born in Malaysia. He received the B. Eng. (Honors) degree from the University of Adelaide, Adelaide, Australia, the S.M. degree from the Massachusetts Institute of Technology, Cambridge, and the Ph.D. degree from McGill University, Montreal, QC, Canada. He is presently a Professor in the Department of Electrical and Computer Engineering, McGill University.


相关文章:
风力发电的实际困难和应对策略
Impacts of Wind Power Minute-to-Minute Variations on Power System Operation. IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008. (3)...
风力发电的研究
Hadi Banakar.ChangLing.Boon Teck Ooi.Impacts of Wind Power Minute-to-Minute Variations on Power System Operation[J].Ieee Transactions on Power Systems,...
130-minute与130 minutes的区别
130-minute与130 minutes的区别_设计/艺术_人文社科_专业资料。1. The _________ action film has attracted millions of young people to the cinema. A...
数字信号处理作业题
system specifications (in the case of a power ...Random or repetitive variations in the RMS voltage... rises above 110% for more than 1 minute. ...
english in a minute 一分钟英语
english in a minute 一分钟英语_英语学习_外语学习_教育专区。Welcome to ...The human nervous system is a collection of the nerves in our bodies that...
up-to-the-minute
force D. power 12. They ______ their business by giving good service ... The One Minute Manager 23页 1下载券 The 30-Minute Guide to... 暂无...
2 minute presentation[1]
2 minute presentation[1]_英语考试_外语学习_教育专区 暂无评价|0人阅读|0次下载|举报文档 2 minute presentation[1]_英语考试_外语学习_教育专区。Good ...
4minute
4minute 4minute 成员 : 南智贤 , 许嘉允 , 全智允 , 金泫雅 , 权昭贤 全智允, 1, 2, 3, 4minute GIRL 出道经历 2009 年 5 月 14 日 CUBE 娱乐...
Value every minute
Value every minute_英语学习_外语学习_教育专区。珍惜每一分钟英语作文,600字左右。Value every minute Good morning,ladies and gentlemen. Today I am very ...
【造福人类】One-minute Talks 朱昱佳
【造福人类】One-minute Talks 朱昱佳_英语学习_外语学习_教育专区。One-minute...One Minute Manager 20页 免费 One Minute Fly 2页 2下载券 ...
更多相关标签:
operation system | ieee power system | power system | simpowersystem | power system toolbox | simpower system在哪 | solar power system | simulink powersystem |