Automatic under-frequency load shedding mal-operation in power systems with high wind power penetration

Countries with a limited interconnection capacity suffer substantial frequency variations after large incidents so they use automatic under-frequency load shedding schemes to arrest the frequency decay. Some of these countries such as Portugal, Spain and Ireland also have very high wind penetrations. This can cause additional frequency excursions due to generation time variability but also to the fact that variable speed wind turbines do not add directly their inertia to the power system. Thus several transmission system operators have announced new grid codes requiring wind turbines to provide frequency response. In some scenarios, however, wind energy support may be detrimental to frequency control because it generates an extra energy that reduces decay and derivative but that cannot be maintained over time. These lower values of frequency decay and derivative are currently expected after a reduced incident or when conventional generation, which can maintain the extra generation, provides frequency support, so lead to low or no load shedding. This paper has studied, in particular, the effect of wind generation emulating inertia. A re∗Corresponding author Email addresses: aparicio@uji.es (Néstor Aparicio), sanyo@die.upv.es (Salvador Añó-Villalba), efbeleng@uji.es (Enrique Belenguer), r.blasco@ieee.org (Ramon Blasco-Gimenez) Preprint submitted to Elsevier January 12, 2017

cause additional frequency excursions due to generation time variability but also to the fact that variable speed wind turbines do not add directly their inertia to the power system. Thus several transmission system operators have announced new grid codes requiring wind turbines to provide frequency response.
In some scenarios, however, wind energy support may be detrimental to frequency control because it generates an extra energy that reduces decay and derivative but that cannot be maintained over time. These lower values of frequency decay and derivative are currently expected after a reduced incident or when conventional generation, which can maintain the extra generation, provides frequency support, so lead to low or no load shedding. This paper has studied, in particular, the effect of wind generation emulating inertia. A re-duction of frequency derivative is achieved, which looks positive at first, but in some cases leads to initial smaller load shedding than the incident requires.
A reduced frequency derivative triggers less under-frequency relays as if there were a significant amount of conventional generation that is online. However, this generation has been substituted by wind generation emulating inertia, and as it can maintain extra generation over time, the frequency continues to decay until the shedding of the next load step. As a result there is an excessive frequency deviation and an incorrect load shedding for the magnitude of the initial disturbance. In order to prevent this problem, automatic under frequency load shedding settings may need readjustment when a large amount of wind generation provides frequency support.
Keywords: wind energy, under-frequency load shedding, frequency control, power system simulation

Introduction
Wind energy has reached penetrations over 10% in several European countries, particularly Denmark, Portugal, Spain, Ireland and Germany [12]. However, the way these countries deal with wind integration is significantly different.
On the one hand, Denmark, whose penetration is by far the highest, has an 5 extraordinary exchange capacity with its neighbouring power systems, around 100% of its peak load, that permits to deal with wind variability. In the case of high wind energy production, Denmark is able to export its overproduction to Portugal and Spain are relatively well interconnected to each other and belong to a common electricity market. Although in the case of transmission restrictions each country may have different prices (market splitting), both countries can be considered as a single system, which will be referred to as Iberian 25 system in the following. The Iberian system is synchronously interconnected with France and Morocco. The thermal capacity of the interconnection is only 8.7% -the net transfer capacity (NTC) is around 5%-of the 2011 peak load, being 6% with the Continental Synchronous Area and the remaining 2.7% with Morocco. More detailed data about the exchange capacity of the aforementioned 30 countries are shown in Table 1, which collects data from [11].
With a limited interconnection capacity below 10%, the Iberian system can be considered as an electricity "island". Quasi-islanded systems require static sources of reserve, such as interruptible load, pumped storage hydroelectricity (PSH) and load shedding to reduce frequency variations, being the latter the 35 only way to prevent frequency collapse following a large incident [19]. However, as installed wind capacity has reached considerable figures in those systems, the almost negligible capability to adapt energy exchanges to wind production has forced their system operators to order wind generation curtailment at certain times of high wind penetration [1] and even the installation of new wind capacity  Variable-speed wind turbines have no inertial response so the power system inertia is reduced when they displace conventional generators. After an incident, that could affect the frequency relays, including those of the automatic underfrequency load shedding (AUFLS) scheme [2].

50
Some already announced grid codes try to prevent large frequency variations by requiring wind turbines to provide some form of frequency response [8,9,14,22,27]. In the Iberian system, a proper recovery is critical to avoid the risk of losing the interconnection with France, which would make the initial incident worse by leaving the system completely isolated.

55
Although many studies have proved the benefits of frequency response by wind turbines [17,4,21,25,28], only [17] considers load shedding. Interaction between wind energy and load shedding has been recently studied but only for small systems. In this kind of systems, energy storage is a cost-effective option to reduce load shedding [6] or a simple PC with very short calculation time can 60 provide adaptive adjustments or limitations in normal operation to frequency relays, spinning reserve, load reconnection, or wind generation [7]. In [15], the interaction of wind and load shedding is studied in detail but only from the intermittency point of view.
This paper studies the impact of increasing wind penetration on larger power      Figure  disconnection ∆P ls in case of frequency decay. All the variables are considered as small deviations from normal operation.
Models of these characteristics, which only consider a uniform frequency across the system, have been proved valid for island systems over many years [23,85 18].
A deviation in generation or load will lead to a deviation in frequency following the power system dynamics. A first order transfer function includes the power system inertia constant H eq and the self-regulation of load (also known as load-damping constant) D. The parameters of all the power system elements 90 are provided in the Appendix.

Conventional generation
Commonly accepted models based in [16] have been used for the steam and hydraulic turbines. The CCG turbine model is based on [24]. Fig. 2 shows the schematic of the a) steam, b) hydraulic and c) CCG turbine models. In a 95 CCG turbine, only the gas turbine provides frequency support, in contrast to the assembled steam one. Therefore, the gas turbine output is multiplied by 0.65, which is its share of the total CCG turbine output. All turbines include a governor with the same characteristics, including dead band DB, droop −1/R, time constant τ G and maximum power increase ∆P max .

Wind energy
The total wind power plants are considered as an aggregate model based on a variable speed wind turbine. With additional controls, this kind of wind turbines can provide frequency control [5], including inertia emulation and even primary control when there is energy storage [26]. Spain experienced significant also includes the characteristics of both frequency controls for commercial use 110 offered by General Electric, inertia emulation and active power control. A supplementary control strategy for DFIGs that is specific for mitigating the impact of limited inertia can be found in [13]. Only inertia emulation has been considered in this work because in the event of under frequency, wind turbines can provide this control with relatively ease and, contrary to active power control, 115 do not need to operate continuously deloaded. Several grid codes consider deloaded operation of wind turbines, although in practice they are not required to continuously spill wind just in case their upward regulation is required at some point due to under frequency. In the absence of grid restrictions, wind turbines are allowed to produce their maximum power and may be required to provide

Exchange capacity
According to the figures shown in section 1, the Iberian system has a very limited interconnection capacity. In the event of a contingency in the tie-lines connecting Spain to France, the interconnection capacity becomes almost negli-135 gible. Moreover, primary regulation from the rest of countries cannot contribute to the Iberian system stability as much as it should because it would trip the few interconnection tie-lines. As a result, no exchange capacity has been taken into account in the simulations, which consider contingency scenarios. Thus power system has been considered as completely isolated. RoCoF is expected, as is the case of our system. As a result, the AUFLS settings used in the simulations are shown in Table 2, which include two steps of PSH. Each one represents 0.05 p.u. as the installed capacity of PSH in Spain is around 10% of its off-peak demand. 165 The trip time of the AUFLS has been considered 220 ms, which corresponds to the sum of 6 cycles (120 ms) for detection plus 100 ms for the time from the relay signal sent to circuit breaker operation.

System inertia constant and self-regulation of the load
The inertia constant of a power system depends on the characteristics of   [17].
A study of the Spanish system [20] obtained the minimum values of the inertia constant that guarantee stability after a three-phase fault during peak and off-peak load. The results were 3.6 s for peak demand and 1.3 s for offpeak demand. The latter value is so low that is only possible with massive amounts of non-synchronous generation substituting conventional generation, 180 which can only happen during off-peak hours. The former supposes a higher amount of conventional generation that is online, which at off-peak hours implies operation at minimum technical load offering spinning reserve. This scenario is more realistic today so it will be used to simulate the behaviour of the power system. If the steam generation is considered to have a penetration of 40% and 185 inertia of 6 s, hydro generation to have a penetration of 15% and inertia of 3 s and CCG generation to have inertia of 4 s, then the wind penetration is around 25%, which is considered a significant value.
The self-regulation of the load has been considered to be 0.5 p.u. since the European Network of Transmission System Operators for Electricity  E) establishes that in the Continental Synchronous Area it is assumed to be 1%/Hz [10].

Studied scenarios
Two scenarios have been studied. One represents peak hours whereas the other represents off-peak hours. In each scenario, three different wind pen-195 etrations have been considered, low (LW), medium (MW), and high (HW), corresponding to 10, 30, and 50%, although 50% is only achievable during offpeak load. While production of steam and hydraulic turbines remains constant throughout the day, being 35 and 15% respectively, the CCG turbines adapt their production depending on the wind penetration. • Peak load (S1). PSH units are not operating and the reference incident assumes a power deviation of 7% so considered incidents will be of this magnitude or lower.

215
• Off-peak load (S2). PSH units are operating and the reference incident supposes a power deviation of 15%.

Results
Initially, all scenarios are in steady state with constant 50 Hz frequency until t = 5 s, when an incident occurs causing a frequency drop. The frequency 220 variation has been measured with the following indicators: • maximum dynamic frequency deviation, ∆f max (mHz). According to ENTSO-E it must be lower than 800 mHz from the nominal frequency in response to the reference incident.
• quasi-steady state frequency deviation, ∆f qss (mHz). According to ENTSO-225 E it must be lower than 180 mHz from the nominal frequency in response to the reference incident.
• RoCoF at certain frequency X,ḟ X (mHz/s). Part of the shedding relays trip depending on this value.
• load shed, ∆P ls (%). It is the amount of load or PSH that is shed.

No wind energy contribution to frequency control
First, both scenarios 1 (S1 & S2) have been simulated when wind energy does not contribute to frequency control. These simulations are representative of the current situation of the Spanish power system. Wind energy has displaced some conventional generation but there is still sufficient online to maintain system 235 inertia equal to 3.6 s. The obtained frequency deviations after a 3% incident are shown in Fig. 4 and all indicators are given in Table 3.
With low wind penetration (LW), the frequency drop is identical in both scenarios because the frequency deviation is below 500 mHz and hence no load is shed. However, with medium wind penetration (MW), the frequency drop 240 clearly differs. In S1 no load is shed either since the frequency drops 650 mHz, i.e. less than 1 Hz, whereas in S2 a part of the PSH (25%) is shed and the frequency drops less, 512 mHz. With high wind penetration (HW), which is highlighted with two black dots in Fig. 4, is below 0.4 Hz/s 1 .
With a 7% incident, the obtained frequency deviations increase, as Table 3 shows. In S1 the frequency reaches 49 Hz and 4% of the load is shed whereas in S2 frequency drop is lower because PSH is available at 49.5 Hz, and the whole of this step (50%) is shed. All quasi-steady state frequencies are below 180 mHz.

250
Finally, a 15% incident has been simulated in S2 and the results show that shedding of the whole PSH (100%) and 4% of load is required to achieve the frequency recovery with all penetrations. The quasi-steady state deviation is only 51 mHz.
Previously obtained results prove that the AUFLS scheme is correctly de-255

Wind energy contribution to frequency control
The effect of wind turbines providing inertia emulation has been simulated.

260
For a 3% incident, Table 4 shows the frequency variation in S2.
Simulation of S1 is considered not necessary because although wind energy support certainly reduces frequency deviation, in this scenario there are no discernible benefits. A lower frequency drop is only clearly beneficial if load shedding is avoided and Table II shows that no load is shed in S1 without wind 265 energy support.
However, PSH shedding is avoided in S2 when wind turbines emulate 7.5 s inertia and the wind penetration is medium. As a counterpart, the quasi-steady state frequency deviation increases from 38 to 134 mHz, which is acceptable because remains below 180 mHz. In the rest of cases wind turbines reduce both 270 maximum deviation and RoCoF but the same amount of load is shed.
With a 7% incident, Table 5 shows that not even the emulation of the largest inertia is able to avoid some kind of load shedding. In fact, the amount of shed load does not change but the order does. In S2 the RoCoF at 49.5 Hz is below 0.4 Hz/s so only half the PSH of this step is shed. As a consequence 275 the frequency continues falling until half the PSH of the next step is shed.
The same amount of shed load implies the same quasi-steady state frequency deviation. However, the wrong step order implies larger maximum dynamic frequency deviations.
Finally, Table 5  Load shedding mal-operation is largely due to current schemes having been designed according to the conventional generation mix where a small RoCoF is only possible in two situations, namely when the incident is small or when there is a considerable amount of conventional generation increasing the system 300 inertia and providing primary control to arrest frequency. Wind generation, however, reduces RoCoF while not providing primary control. Frequency is not arrested and continues to decay until more load is shed.

Conclusion
Power systems with limited interconnection capacity are equipped with AU-305 FLS to prevent excessive frequency decays. When high wind penetrations are reached in these systems, the wind turbines are forced to provide frequency response.
Wind energy inertia emulation is beneficial since the frequency deviation is reduced. However, this fact can only be considered indeed a benefit if some load 310 shedding is avoided. When the wind turbines emulate a large inertia, there is no load shedding in the smaller incidents considered in the simulations.
However, with higher emulated inertias the RoCoF is excessively reduced and some frequency trend relays of the AUFLS do not trigger. As a result, the system frequency continues decaying until the next step, where part of its 315 load is shed. This leads to both larger frequency deviations and the shedding of loads that do not correspond to the initial disturbance magnitude. In these cases, lower values of emulated inertia have been found to give better results.
The results obtained in this work suggest that AUFLS schemes that use derivative frequency may need revision in power systems with high amounts of 320 wind generation providing inertia emulation.