Está en la página 1de 8

52nd IEEE Conference on Decision and Control

December 10-13, 2013. Florence, Italy

Predictive Control for Real-Time Frequency Regulation and
Rotational Inertia Provision in Power Systems

Andreas Ulbig, Tobias Rinke, Spyros Chatzivasileiadis and Goran Andersson

Abstract-This paper presents real-time optimal The presented control schemes enable the provision of
control schemes for regulating grid frequency and two crucial ancillary services, frequency control and rota­
providing rotational inertia in power systems, two
tional inertia provision, using any generic power system
control problems with fast dynamics on the time-scale
of milliseconds. Both control problems have received
unit for which only the relevant operational constraints,
wide attention in recent years due to rising shares power ramp rate (p), power rating ( 1f ) and - in case of
of variable renewable energy sources ( RES ) and the storage units - also energy storage ( E ) , need to be known.
thereby arising challenges for power system operation. The paper is organized as follows: In Section II cur­
The presented control schemes are based on explicit
rent trends and challenges in power systems operation,
model predictive control (MPC), which allows to
directly incorporate operational constraints of power
which serve as the paper's motivation are presented.
system units, i.e. ramp-rate, power rating and ener­ Section III explains the dynamics of grid frequency and
gy constraints, and to achieve real-time tractability inertia and the corresponding control problems. Secti­
while keeping the online computation effort low. Si­ on IV introduces MPC. Section V presents formulation,
mulations, including a performance comparison with
implementation and simulation results of the proposed
traditional frequency control, are presented.
MPC-based control schemes for frequency control and
1. INTRODUCTION inertia provision along with a performance comparison to
traditional frequency control based on P IPI controllers.
In this paper we present real-time optimal control
schemes for regulating grid frequency and providing grid II. CHALLENGES IN POWER SYSTEM OPERATION
frequency inertia based on a explicit model predictive Traditionally, power system operation is based on
control (MPC). Both control problems receive wide at­ the assumption that electricity generation, in the form
tention due to the rapidly growing shares of variable of thermal power plants, reliably supplied with fossil
renewable energy sources (RES), i.e. wind turbines and or nuclear fuels, or hydro plants, is fully dispatchable,
PV units, and the thereby arising challenges for power i.e. controllable, and involves rotating generators. Via
system operation, notably faster frequency dynamics due their stored kinetic energy they add rotational inertia, an
to reduced rotational inertia. important property of frequency dynamics and stability.
The control objective of frequency regulation is to Large-scale deployment of RES generation, notably in
minimize the frequency deviation [:).1 from the nominal the form of wind turbines and photovoltaic (PV) units,
grid frequency 10 (50 or 60 Hz depending on region). has led to significant shares of variable, non-controlled
Rotational inertia provision, i.e. increasing the inertia power in-feed in power systems. State-of-the-art wind
constant H, minimizes the frequency rate of change [:).j turbine systems are often grid-connected via AC-DC­
in case of power deviations. This renders power system AC converters, thus canceling (most of) the electro­
frequency dynamics more benign, i.e slower, and thus mechanical coupling of the as such large rotating mass,
increases the available response time to react to fault i.e. turbine blades and generator. PV units produce
events such as line losses, power plant outages or large­ direct current, have no rotating mass and are grid­
scale set-point changes of either generation or load units. connected via DC-AC inverters.
The choice of MPC as a control approach is motivated The increasing share of inverter-based power genera­
by its capability to incorporate operational constraints of tion and the associated displacement of usually large­
power systems directly in the controller design. Explicit scale and fully controllable generation units and their
MPC allows the shifting of most of the computational rotational masses, has the following consequences:
effort ofHine, thus enabling real-time tractability of MPC 1) The pool of suitable conventional power plants
schemes for systems with very fast dynamics on the time­ for providing traditional control reserve power is
scale of milliseconds and keeping online computation significantly diminished.
requirements low.
2) The rotational inertia of power systems becomes
A. Ulbig, s. Chatzivasileiadis and G. Andersson are with the
thus markedly time-variant and is reduced, often
Power Systems Laboratory, Department of Electrical Engineering, non-uniformly within the grid topology. Potential­
ETH Zurich, 8092 Zurich, Switzerland.
ly, this leads to new frequency instability pheno­
Email: ulbig I spyros I andersson@eeh.
T. Rinke completed his master thesis at the ETH Zurich Power
mena in interconnected power systems. Frequency
Systems Laboratory in 2010-11. Email: and power system stability may be at risk.
978-1-4673-5717 -3/13/$31.00 ©2013 IEEE 2946

and hence stable operation. 1: ENTSO-E frequency control categories [3]. contribution of inertia is an inherent feature of rotating Deviations from its nominal value.e. Small local disturbances can evolve rescheduling is manually activated according to the ex­ into consequences influencing the whole power system. i.Eventually. Grid frequency 1 is directly coupled to the sudden imposed changes to the grid frequency. i. drawing on an internal or ex­ control categories etc. range is thus a necessary requirement for the stable 2) Grid rotational inertia: The decrease of rotatio­ operation of power systems. as damaging vibrations in synchronous machines ling. [2]. Challenges are the real-time code specifications of the ENTSO-E. for the provision of control reserve power. The responsible units Power systems are dynamic systems with a high de­ in the control zone of the imbalance start to take over gree of complexity along several dimensions. one of the world's traints that need to be respected by any control largest synchronous power systems.g. nal frequency control. many of these units ex­ • whole grid affected • control zone in which • control zone in which fault electro-mechanical fault occured occured hibit time-varying availability and stringent. of variable renewable power units on power systems It provides power output proportional to the deviation operation as well as on mitigation options [1]. Fig. 75 min. wind turbines. Tertiary frequency control manually adapts the stability. the nominal scheme aiming to provide rotational inertia.e. is provided to the grid in case of a frequency deviation. stabilizing the system frequency but not restoring it to 10' Generators of all grid control zones are III. likewise.[5]. control schemes is increasing with every new inverter­ Stable power system operation is provided by traditio­ based power unit that is commissioned and. tractability and the stringent operational cons­ for the European continental grid. providing secondary frequency space and grid hierarchy. part (PI control). the key system state This traditional categorization is missing an important to observe and control is the grid frequency 1 (in power contributor to frequency stability: the inertial response. Here. Large frequency devia­ by increasing the inertia contribution from non­ tions can be caused by errors in load demand or RES conventional units via so-called inertia mimicking. bat­ tery systems and demand-side management. time. after approximately 30 s. Please note. Power system processes extend control. PV units. depends on the ac­ power generation set-points and coordinates the power tive power balance. it restores both the grid frequency spatial distribution is high (100s-1000s km) and multiple from its residual deviation and the corresponding tie-line grid hierarchies. forecasts and the loss of load units. 10. by cheaper sources at a later stage (i. as defined in [3]. regarding frequency conventional units. exist. (P control). which in [3] has three categories every conventional generator that is decommissioned. in pected residual fault in order to relieve tertiary control the worst case ending in cascading faults and black-outs. generators or lines. should be kept synchronous generators. Due to electro-mechanical coup­ small. This is achieved through the proper control of non­ In the remainder of the paper. that The motivation to adapt the current power system classifications of other power systems usually differ. Regarding active power balance. Unlike con­ Inertia Response: Secondary Control (AGC): Generator Rescheduling • response: <Is • response: 30s . different voltage levels. One power exchanges with other control zones to the set-point main characteristic of power systems is that frequency values. 2947 .e. meaning that the total power in-feed production operation beyond the initial 15 min.15rnin • response: 75min . 1): Primary frequency control is provided within a Recent studies show the increasing interest on the impact few seconds after the occurrence of a frequency deviation.2h ventional power plants. kinetic energy stored in a generator's rotating mass and load shedding occur in case of larger deviations. the changes to these two important power Primary Control: Tertiary Control • response: 5-30s • response: > 15min system properties need to be reflected in the design of • whole grid affected • control zone in which fault occured • proportional control via L'lf • manual dispatch according to AP power system control schemes: 1) Control reserve power: The shrinking pool of sui­ table conventional power plants for control reser­ ve provision and frequency regulation increases the motivation to consider non-conventional power system units. grid frequency 10 is set to be 50 Hz. generator is kept close to zero. operational constraints such as depending on availability time-variant power and energy storage availability. Traditional P /PI-based frequency control cannot handle operational constraints explicitly.[5]. GRID FREQUENCY AND FREQUENCY INERTIA participating in primary control. we will refer exclusively to the grid ternal energy storage. This mo­ Maintaining the grid frequency within an acceptable tivates the need for novel grid control schemes. The rotational speed of a synchronous generator ( w = 27r j).) [3]. systems with several grid control zones also tie-line power a stabilizing quality of a power system to counteract exchanges). thus • coup Iing via ilf • PI control via L'lPAGe • re-dispaLCh via intra-day auction flexibility-reducing. Additionally. In normal operation small nal inertia in a power system can be mitigated variations occur spontaneously. e. (Fig. time­ minus the total consumption (including systems losses) frame after a fault has occurred. As secondary control has an integral control along several time-scales (milliseconds to yearly seasons).

Typical values for H are in the range of 2-10 s Note. A. i.... Frequency stabilization system with different inertia (20 . by formulating •... becomes thus more difficult. the frequency inertia of power systems is reduced.. This highlights the importance of the rator and Pe the electric power demand. ( 4 ) = o 1� o 70 80 90 Here f o is the reference frequency and Do J ad denotes the --. in order to emulate the dynamic response of frequency dynamics slower and. conventional. Figure 2 illustrates the dynamic response of the un­ constant H for a synchronous machine is defined by controlled IEEE 14-bus test system to power balance Ekin J(27rfm) 2 deviations for several inertia setups.. The control scheme providing virtual 2948 . . ) of the synchronous generator following = sturbance. fo fo o fm . The rotational energy is given as For this purpose a control loop is added to the generator (1) bus ( (Pm . With an increasing penetration 5 of non-conventional inverter-connected power units. 2: Dynamic response of uncontrolled IEEE 14-bus ty not contributing any inertia. the smaller are frequency de­ 5 viations f m and its derivative 6.. inertial response for power systems as frequency fault Noting that frequency excursions are usually small dynamics may become too fast for existing frequency deviations around the reference value. The maximum frequency devia­ . e.. the slower are grid frequency dynamics. . 50% and 20% of Ho) . (3) duration to reach this post-fault steady-state frequency with Pm as the mechanical power supplied by the gene­ differs significantly. Padd Ll f80% . which mimicks the characteristics of the conventional generator response as given by its swing with J as the moment of inertia of the machine and equation ( Eq. descri­ contribute to the inertial response.... f and also 6. The high share of conventional generators is trans­ 5 lated into a large rotational inertia of the here pre­ 90 sented interconnected power system..•.. Hence the value of rotational inertia for P\21 shaping frequency dynamics in power systems with high­ f o. Appropriate adaptations of grid codes are needed. and complete the classical swing equation by adding frequency-dependent load damping. representing diffe­ H = = ' (2) rent shares of conventional generation units. 3).g.2HSBDJ ad fm+ 2HSB (Pm.100% Ho) to faults.:. frequency dynamics are faster for low­ as the change in rotational speed Wm ( or rotational fre­ inertia power dispatch setups.e. The higher the inertia constant H of the system. Here. : 1 ... with a high share of generation capaci­ Fig. Inertial Response of a Power System B. Following a frequency enables inverter-connected generation units to rapidly deviation.. rendering power system sources. nominally scheduled mechanical generator power. . rotating generators to frequency variations. bes the inertial response of the synchronous generator In = ---y. The inertia kinetic energy. in which with SB as the rated power of the generator and H only conventional generators are dispatched (100% Ho) . hence. a self-stabilizing pro­ shares of inverter-connected units is high. perty of power systems. The classical swing equation. f 100% graph shows the behavior of a system. easier to regulate.. the SB 2SB 6.. O .. a well-known to the share of inverter-connected units. denoting the time duration during which the machine The other graphs illustrate the system dynamics for can supply its rated power solely with its stored kinetic decreasing inertia levels (80%. 2]. 0 is the 5 . 6. 4 ) by timely providing otherwise missing fm the rotating frequency of the machine. that the rotational inertia H is inversely related [7.. energy. via internal or external power the generator system is released. which do not model representation for synchronous generators..Pe) = . kinetic energy stored in the rotating masses of change their power output. Table 3...Pe+Padd) .Ll f50% denotes additional power in-feed via a generic control o loop. ' 2HSB ' tion for the step change is the same for all setups but the Ekin = J(27r) 2 fm . This is no­ 5 ��������--� tably a concern for small power networks. we replace f m by control schemes. providing frequency regulation or inertia provision.•.f100% frequency-dependent load damping constant. .j are larger for grid setups with a power imbalance as lower rotational during identical o power imbalance faults. Pm. island grids o 10 20 30 40 50 60 70 80 90 time Is] such as Ireland. For the triangle-wave di­ quency f m �. .. Concep t of Inertia Mimicking The derivation of the inertial response follows the Inertia Mimicking ( 1M ) is a control concept that line of thought described in [6].

110ad the total The internal system model has a decisive role for the system load of the grid. assuming that fl. the chosen model must be capable of and uniform. For before it is applied to the power system as Padd. ( Pm 11oad) . The power system parameters used in all capturing the process dynamics so as to precisely predict simulations are H = Ho 8s. The triangle­ . with a 80%-inertia setup (i. the same grid bus.02 � and fo 50 Hz. The complexity of the losses of the l lines making up the grid topology and controlled system and. 3. e. the system state x E jRn. Considering the system a finite horizon open-loop optimal control problem..J. 3: General simulation setup for inertia 1 r-- I 2HlowSB __ I _ wave signal represents a slow variability in the power 5+ + 10 Dlood !1!jillIit 0 balance.u. Pm the dynamics of the controlled system [9]. k · k · . thus. the case of frequency regulation. total mechanical power of the generators.. H 2:7= = �:i SB'i . whereas the step signals can be seen as a 1 sudden disturbance. Fault ASE (high inertia) leads to the Aggregated Swing Equation (ASE) generator I introduced to the power system (Fig. e. 20% of generation capa­ rv fm = . its modeling representation f o 50 Hz. B E jRnx9. O . = = implementable.2HSBD ad fm+ 2HSB (Pm. which is assumed here to be constant In consequence. representing the center of inertia of the given grid.Pload . 6) is valid for a highly meshed grid. the employed ASE power system models for frequency f y E jR the system output. 4) for � 2N. a system for each time step.. Pload L Pload. compared with classical PID control schemes has been a in which all units can be assumed to be connected to barrier for MPC implementation in the past [9]. i..g.SB -- 10 5+ - 1 D/oad . which is internally integrated block) with 100% of conventional generation units.2HSBD ad fl. the behavior of The control input for correcting the response of the a "high-inertia" ASE power system model (middle model system state jm is Fadd. Since load-frequency disturbances are A.Pe+Padd) . SB 1. as it predicts the future H the aggregated inertia constant of the n generators. Scope . Ploss the total transmission performance of MPC schemes. An additional "low-inertia" ASE model (lower model block) An aggregated swing equation model of the well­ is employed as a reference of the dynamic behavior. The change (fl. Reformulating the classical swing equation (Eq.= + M"dd I Additional I pow" p. For A E jR nxn. The sampling time Ts 20 ms.g. i' = = = problems where its capability to handle system state and i=l i=l i=l control input constraints explicitly is of value. the additional power C. The term Do = l ad is the frequency damping of is directly affecting the necessary computational effort. is = .8p. system behavior based on an internal model of the plant SB the total rated power of the generators. The relatively high computational effort The ASE (Eq. (6) 10 D10ad 2HSBDol ad 2HSB ASE (low inertia) with Fig..e. is used for all presented simulation cases. . the linear system matrices regulation and grid inertia provision is given in Fig. ideally exactly. Do = = l ad = the future outputs as well as being simple enough to be 0.rotational inertia has to be based on the time-derivative the case of grid inertia provision.f .f + -. An MPC scheme determines the optimal open-loop response to Here f is the center of inertia (COl) grid frequency. r+I a power system with n generators. C E jRfxn and D E jRfX9 at 2949 . Ploss L Ploss. (7) 2HS-B fl.fl. IV. a variation in power in-feed or load demand ASE (low inertia) forecasts.Ploss) . The disturbance signal - a combination of triangle-wave and step signals . MODEL PREDICTIVE CONTROL i=l n j Model Predictive Control (MPC) is often used for Pm L Pm. the inertial response of of the swing equation a "low-inertia' ASE model (upper model block) initialized . 2H1owSB I fo fo s+_- f - = f+ (Pm . MPC Setup Definition normally relatively small. the relative problem is to stabilize the linear discrete-time system formulation of the ASE. the system load [5].. .) before and after a disturbance. known IEEE 14-bus test system. f= 2:7=1 Hil i 2:7=1 Hi ' SB = tSB. y ( k) Cx(k) + DU(k) = o l An illustration of the general simulation setup with with the control input U E jR9. i.f o can be used. j loads and l lines. Model Setup of Simulation Cases path is used for providing frequency regulation. which represents a small interconnected power system [8]. (5) city is inverter-connected) is shaped via another power l o path in order to mimic. linearized swing equations with MPC is formulated as a repeated online solution of fl. 2) .f i = f i ..11oss 0. i. load or generator disconnection.. x(k +1) Ax(k) +Bu(k) (8) fo fo fl.

minimizing operation costs or of grid frequency need [10]. oo}. the online compu­ tory does not result in the ideal system reaction described tation merely consists of search operations in a look­ by the system model at the end of the prediction horizon.1). R as the cost term for the control input u(k) and MPC setup parameters. from any generic power system unit. Further cost terms. i. restricting the solution space to permissible regi­ lers. or As frequency control and especially frequency inertia both. . hence significantly reducing The actual system behavior will be different from the the computation and storage requirements of controller predicted behavior. x(k) :S. [7].the discrete time instant kEN. up table of control laws.. result of this offline calculation of an explicit linear MPC control law has the form where u* := [u(k) T. To react to disturbances and mismat­ hardware and significantly speeding up sampling time. Using the initial system fault.u(k . The optimal control trajectory may not frequency. The The term I in II· III specifies the norm of the cost employed prediction horizon N further has an exponen­ function as I E {I. MPC schemes of the mated with satisfactory accuracy into a single polynomial form shown by Eq. and energy storage tractability of the employed control scheme is high.e. First. over all feasible states and stores them in a look-up X mm :s. MPC-BASED GRID FREQUENCY CONTROL AND B. Umax .I. consisting of the optimal control trajectory of the where xE]Rn represents the system state and Uj E]Rm control inputs for each step of the prediction horizon is the optimal control input for the region Pj defined k = 0. frequency control should be provided with a is confined to the flexibility volume spanned by these sampling rate that corresponds to one cycle of the grid three metrics. i. .e. which is well { t+(N-I) applicable for linear. ons. the L2-norm is used tial influence on the necessary offline computational cost. Explicit MPC ROTATIONAL INERTIA PROVISION In the past. bumax . Ts. The x(t) .g. and on the slew rate of the control input wing the development of very fast online MPC control­ bu(k) . se- 2950 . Nevertheless. u(k) :S. In an rating ( E ) . within milliseconds. play an important role with respect to the control input. immediate dynamic response of a power system after a for performing "conventional" frequency control and. power ramp rate (p). constraints and weights on consequently R8 the cost term for the slew rate of the the states. generators. can provide frequency control or rotational inertia. e. need to be known [14]. 9. a lead to first swing instability in a power system [5].g. model plant defined state-space is guaranteed as all control laws have mismatches or disturbances. explicit MPC retains key advantages for IIQ.. see overview and performance comparison in [12]. Rotational inertia should be battery degradation. on the control design have drastically reduced computation times. This is only be shaped by the control objective itself but also by also about the sampling rate that robust measurements other considerations. additional terms.e.. e. provision should ideally be accomplished very fast and storages or loads. bu(k) = u(k) . 2. table. control input u is determined over the prediction horizon Such fast control response times can be ensured via Tp = N . feasibility within a pre­ Due to uncertainties in the system. thus effectively shifting most of the computational s. [15]. allo­ signal u(k) .ermx(N) lll. i.. 20 ms for a grid frequency of 50 Hz.. Second. the optimal control trajec­ been computed beforehand. X m ax . constraints can Recent advances in MPC implementation and solver be imposed on the state vector x(k) . e. bu(k) :s. For a given regulation task. For linear time invariant (LTI) systems the bUmin :s. e.e. effort offline. which can be decisive as insufficient inertia may state measurement x(O) = [Xl (0). 9 possess no naturally guaranteed control law.. ches. first.. Xo = x(O) . provided on the same time-scale in order to shape the Two MPC-based control schemes are presented. low­ provides the methods for designing control schemes that dimensional problems or systems with slow dynamics. The explicit k=t optimization problem pre-computes optimal control laws X(k + 1) = Ax(k) + Bu(k) .g. Control operation ideal world. N . in the setup. i. real-time applicability of standard on­ We show that model-based optimal control theory line MPC optimizations was limited to simple. allowing even faster implementation times closed-loop stability. stability constraints.e. . feedback on the actual development is incorporated The resulting look-up table can furthermore be approxi­ by the receding horizon scheme. Umm :s. the motivation for real-time rating ( 1f ) .1) TV E R9XN is the resulting optimization vector of the above pro­ blem. i. complexity of the explicit MPC controller solution. Here. u(k + N . a terminal state-constraint. of which only the relevant operational since modern battery systems are in fact able to react constraints. hybrid systems of a manageable min u(·) L IIQxx(k) lll + IIRu(k) lll + IIR8bu(k) lll (9) size.t. categorized by the flexibility metrics power fast. practical implementation... can be added to Eq. throughout. systems that are sufficiently small. xn(O) V = Xo. Stability can be guaranteed via for practical control problems (<< 1 ms) [13].g. with Qx weighing the state vector within the sub-space partition Sli of the state-space. by solving the optimal control problem the offline computation of all possible online control moves using explicit MPC techniques [11]. V.

5) is used. At the beginning of the step fault (t = 22.. 4) 7r: -0. the controlled low-inertia system (solid) state X2 is created.:••• :. .:...50). has an improved available system output 6. ""ow . 5-7. The initial setting XSoc(O) = o allows control action in both directions.Padd (p) and Eadd ( E ) . A.u.g. The second FC control scheme. Padd ( 7r) . The structure of the control problem is similar for both approaches: the swing equation in its normal form (Eq.: •• :. The controller performs well while obeying rate of change (min 6. frequency deviation (same color coding). The third state is similar to the control for fre­ dot).5 . MPC-based Inertia Mimicking (1M) The MPC cost function weights are defined as The setup for inertia mimicking (1M) is more complex Qpc = [ 100 0 ] o term = . The MPC-based 1M control scheme reacts to 6. as well as its derived form (Eq.u..cond. the additional system state Eadd is introduced as 0. f ± 0. and -1 (empty). The second direct output u will be the derivative of the additional plot shows the frequency deviation of the high inertia power (Fadd) .u. which is the desired control input and the uncontrolled low-inertia reference system (dash­ Padd. for mimicking frequency inertia of conventional continues to reduce the frequency deviation until the generators. Please note. 6. (12) E: -0. _6.25 p. By internally integrating u.!�>""� o 20 units. . .. Increasing inertia. QPC [ 00 than for frequency control (FC).___----" '--v--" A x � B with the control input u = u = Padd. as expected.fhi9h . The MPC FC control problem is defined as follows x = [ 2HIOw�� Dload o 0 0 6.50 Hz. f Eadd ].8 s) and no further imba­ frequency deviations (min 6. 6. (11) _____ .00 p. the input YI = X l to the 1M control scheme.10 p." . f therefore has to be time-differentiated. The MPC 1M control problem is defined as follows this p lot structure will also be maintained for Fig . respectively. .. < Fadd < 1.. the lower three subplots (4-6)." 6. and 6. e.�.u.j is then Figure 4 shows the system response of the FC strategy. ___"". a significant reduction of the frequency deviation. the defined operational constraints u also leads to a reduction of frequency deviations. -I Eadd = J Padddt = Cbat J i:socdt = CbatXSOC. B. load.fFe-SOCO. input is the regulation power Padd that is supplied to the power system through an additional power source. f ] [ ] [ XSOC + l�: 2H c- SB 1 u ---_. whereas the In the top plot the fault signal is depicted.. using twice The frequency control (Fe) approach uses the directly the storage capacity (FC-SOC 0. indirectly. a generator. f as control signal.5 s). the second system (dash). 6..u.00 p. are shown in somewhat increasing the controller complexity.50 with Cbat as the capacity of the additional power source and the limits of the state-of-charge (SOC) XSOC being +1 (full). The third plot shows the time derivative of the quency control: it formulates the energy constraint Eadd..1) .. . inputs.j.:.5 s) the controller with the control input u = u = Fadd.u. battery or other energy storage unit. The control The model structure is thus augmented by one state. < Padd < 0. the frequency deviation shall not exceed 6.. MPC-based Frequency Control (FC) Furthermore.25) shows. For the triangle-wave signal the frequency deviation is reduced by � 80% compared to the non-controlled high inertia reference system. < Eadd < 0. while inertia provision lance power can be absorbed until the reverse step event u increases the inertial response by reducing the frequency (t = 37.25 p. The frequency-controlled aggregated power system (FC-SOC 0.j) . p: -1. Frequency control aims at the reduction of Eadd limit is reached (t = 26. which is integrated 2951 . The control performance during the step faults. . The power system output and Rpc = 1e-3 and RIi.10 p. To account for energy constraints in the case of storage :�P\21·· · ·rr •••••:••••• :• • : •••:. PC = 2 with N = 5.

::�:E:. the two control inputs Padd of FC MPCFC+IM and Padd of 1M MPCFC+IM counteract each other..j only. The plot further shows that for the time-frame from t = 45 s till t = 60 s.:J:J:: : :J:J:. For this durati­ additio�al power to the power system.. the frequency deviation is higher compared to the cing [:). and .IM = 0 with N = 5 and a sampling frequency control. -...50 1 reacts to local measurements of [:). The FC MPC scheme than has to regulate o �..6.!-ring the initial triangle-wave signal." 6. The step change causes the for t = 36 .. v C x The MPC cost function weights are defined as All ASE use the "low-inertia" setup. which is similar to existing fast o r----4-. .e. and (c) the combination of both the MPC-based frequency and inertia mimicking control Fig... 5). and ( MPCFC+IM ) ' For the case (c) (Fig.-.j: : :J:J:J:t:] o w � 00 00 timers] 100 120 140 hence does not necessitate communication or coordinati­ on.-:Jo"'C""-"""'''''' '' . is indepen- Fig. which IM_SOC0. . f (Fig. Y= 1 0 d ' d A fourth case is also plotted.. Therefore.j - .50 D. to flow MPC FC+IM scheme. i. energy constraint ( E ) is at ± 1.d6.fhi9h of a low-inertia system to look like that of a high-inertia 6.._ . The operational constraints are the i. Dl. This behavior indicates that it is of advantage to realize the combined MPCFC+IM via an integrated multi-objective 0.-. f causes the control scheme to provide and during t = 30 . f also shifts the local maxima in the The traditional frequency control with energy constraint plots of frequency and frequency derivative. showing the (trad FC limited) also causes the control to be inactive increase of rotational inertia. However. 6). control input M PC frequency control. o The two MPC controllers outperform the traditional and RIM = 2 and Ro. controlling two separate energy sources (Eadd) . 6. the two control­ lers FC MPCFC+IM and 1M MPCFC+IM are connected in [� ][ �. limited energy capacity Eadd. 5. the size of the individual control problems. the two MPC-FC's control The fault sign�l Pfault and the resulting frequency inputs are zero during t = 30 ... 6: Scheme of the combined MPC for 1M and FC. for the non-controlled low inertia setup (low). (IM-saeo. showing the frequency 0 Cbat X SOC � Eadd deviation.d6.����'-..e.and then applied as J u dt = Padd. -.I------'---� • . 2952 . unconstrained traditional frequency control (trad FC). I'll C. size of state x and control input vector u.. Details on traditional AGC schemes can be 1e-3 found in [16. Appendix B].25 and 1M MPC control schemes was a deliberate choice .fIM-SOCO. f and [:).. (IM-soeo. By changing the capacity of Eadd (Fig. ���������������������� . primary frequency control.con ro npu MPCIM_SOC0.50 here. as long as they are active (Fig. .5 ) . this reduction of [:). • " .----r'. The 1M MPC scheme shapes the dynamic response r---�. i.l1"1"001. 1 [ :: 1 0 0 parallel._.50) controller saturation can be avoided. 5: MPC Inertia Mimicking with constraints on power rating ( ]f ) . 5: IM­ consumption of MPCFC+IM (black graph) for which the SOC 0. while no saturation limits are reached._.. ramp rate (p) and energy rating ( E ) .e.: :::: : T::::: · :T: J t l i t -. Note that the energy constraint 1M controller to react until the energy limit is reached plot (lowest sub-figure) depicts the combined energy (t = 28s).45 s for FC MPCFC+IM rate of change [:). 80% conventional generators. Ch. d" .. i.39 s.-��c._.1.e. 12).0 (instead of ± 0.3.5 .f low system (Fig.. as well as the time derivative of frequency '� deviation.35 s for MPCFC. Controller Performance and Scalability ·: · : I:: ·:::·:T.. The setup for the MPC controllers is identical o to Section V. effectively redu­ on. 6). Due to their same as before (Eq. ( 1M-SOC0. Combined Frequency Control and Inertia Mimicking I'll versus Traditional PIPI-based Frequency Control This section builds upon the aforementioned scenarios and compares the performance of three different con­ System Output trol setups: (a) the traditional primary and secondary frequency control (droop control [P] and AGC [PI]) I'lf with and without limits (trad FC limited and trad FC ) .-". time of Ts = 20 ms. 7). generator ASE (low inertia) (b) the MPC-based frequency controller as described in virtual "high" inertia system Section V-A ( MPCFC ) ...d .25 Both the FC and 1M control schemes are implemented as local control loops.25 only the slower dynamics of the high-inertia system." . separating the FC MPC o _"""..�"""'_-____�� -..

." control schemes. the MPC setups furthermore depends ( exponentially ) on 2009. 2003. C.���� . Domahidi. 2010. 7: Comparison of conv. Realistic values York: McGraw-Hill. control schemes. Morren. . "On Operational Flexibility in Power Systems. Ma. S. and M.. 98 162. and M. Andersson. 02 188. Jun. have to run on embedded systems. N. A. Frequency 1M: 3 states. M. and hence complexity. ser. .. tems ( phd ) thesis . 1 input . [12] A.�� . CONCLUSION & OUTLOOK Power Systems. ee.. [13] / tx-ethpublications Rinke_2011. and C. Power system stability for all our simulation cases. Colle­ for Hand Dload. EPRI power system engineering series. We do not expect significant dif­ [9] E. Sweden.ftradFC FC+IM online 5.��"'-'-'-� .f1ow FC+IM explicit 27. 28 315 . frequency control ( conv. San Diego.. Morari. -0. 2000.... M. vol.. entsoe. and S..control input Irad Fe REFERENCES ��"*'��p����¥'=�==l=��*=i ' " pO " control input MPCFC [1] J.5 . time 151 "Dynamic frequency control with increasing wind generation.. of Industrial Electrical Engi­ neering and Automation.. Baotic.. "Operational Impacts of W ind Generation on California Power Systems...�u... Assessment: 'Simulation" is the run-time of a Matlab Simulink / simulation of 200 s length (Ts = ) 20 ms . more precisely the local grid parameters riable speed wind turbines. 2953 . The size. Jo­ mulation. TABLE 1: Overview of Computational Efforts. 1715 .. J. no. Calculation: All calculations were performed using the MPT tool­ box [17] with the solvers MP-QP ( explicit MPC ) and QuadProg ( online MPC) . Lalor. Dumur. S. Morari." in Proceedings on the performance was similar for online and explicit MPC si­ 39th IEEE Conference on Decision and Control. Traditional online MPC .Q. pp.--.L """"�.-dftradFC FC explicit 11. Table I gives an overview of the computational efforts [11] A. -. control input 1 M MPCFC+1M . Bordons. Zeilinger. Bemporad. Ulbig. http: //www. Makarov." in IEEE Conference on run-time of 200 s length is calculated in less than one Decision and Control.-dll. C.. [7] P. the time-constant of the frequency dynamics of a given [6] J. 12. Rourke.- !---.. while the full simulation Arising in Receding Horizon Control. 61 427 ... Olaru. Model predictive control. --! ." in IEEE PES General Meeting. de Haan.. The proposed MPC FC and MPC 1M [10] M." Electric Power Systems Research.with the tools Asian Journal of Control.. / Paris: ENTSO-E UCTE.. Advanced textbooks in control and signal processing. However.55 0. forced. Lund. vol. ETH ZUrich Automatic Control Laboratory. 11.03 " • • . 2006. Grieder. "Explicit second. no. Nov. Ritchie.. Control strategy Computation s [] Simulation s [] Regions . July 2012._. Control of LP-based Model Predictive Control. May 2010. 2004. Its choice depends indirectly on tems. FC ) vol. 96 1M online 2. .-.52 0. 1997.ethz. power systems. pdf. New result in a better control performance. were chosen for the Aggregated Swing / www. washington. S.. Simulations are described in detail in this section. 2. vol. "Contribution of DG units to primary frequency control. Compliance gy sources. at the controller bus. 2004. edu research pstca . and P. 2... with MPC-based FC and MPC-based FC+1M. P. Ulbig and G. 24.. eeh. . M. and inertia provision into one unified control strategy..Load-Frequency Control and Performance.�� .. We show that there exist meaningful applications for [16] T..05 ." in 2005 International Conference on Future Power Systems. a higher value for N did not and control. and F. 3.. 51 742 -.-. and M. _. Akke.� ft•ad FCljm�ed FC online 3.." '" dfMPC-FC 1M explicit 18. and C.. F. 1 input ) . CRC press. confirming the real-time capability of the explicit nonlinear predictive control for a magnetic levitation Lauby. The offiine computation takes less than 30 s for nes. Borrelli. of and Voltage Control in Electric Energy Systems. Dynamics and Control of Electric Power Sys­ the prediction horizon N. Ferreira. and the sampling time. "Computation" is time . notably for frequency micking for improved grid integration of renewable ener­ regulation and frequency inertia provision.. ETH ZUrich._. too slow and not applicable.-!I-. dent of the power system size ( FC: 2 states. "Explicit solution of the here employed FC and 1M control schemes. . Zgraggen.� . Rinke.." in IEEE Power Engineering Society General Meeting. D. [5] G. 2011. 2005. "Efficient Interior Point Methods for Multistage Problems all presented control schemes. de Mello. Kvasnica. D..fMPC_FC Hardware: Pentium Dual Core 3 GHz with 3GB RAM. especially since in practical application controllers will USA. Fig. F. [3] Union for the Coordination of Transmission of Electricity. www. 77 0. and J.L_� . Flynn. / / Equation ( ASE ) model. Loutan. 1994. Kundur. Future work investigates the addition of stability [17] M. IEEE Transactions on. Balu. vol." VI." LTH.1720. "Inertial response of va­ power system." Master's thesis. ch.68 374. Sydney. Pierik. Dept." dll. de Haan. ser. University of Washington. p. Christophersen. Operation Handbook Appendix 1 . "Some control applications in electrical power sys­ controllers should scale well for any power system size. and P. "MPC-based frequency regulation and inertia mi­ MPC in power systems control. and solvers used here . J. Cafiizares. [ 8] Power Systems Test Case Archive. ETH ZUrich Power Sys­ with operational constraints of power system units is en­ tems Laboratory. J. Andersson.r- . .. similar also for large interconnected ge of Engineering. Alvarez Bel.. [14] A. constraints and the incorporation of frequency control Multi-Parametric Toolbox. New ferences for the practical application of the here proposed York: Springer. May 2009. Boucher. Lecture Notes. O'Malley. control input Fe MPC + FC 1 M for offline computation of the explicit controller. [15] Y. Andersson. 76. Camacho and C. o 10 20 30 40 50 60 70 80 90 [2] G. 9.1-9. [4] G.