IET Renewable Power Generation

Research Article

Model predictive control-based load ISSN 1752-1416

Received on 30th November 2018
Revised 18th June 2019
frequency control for power systems with Accepted on 20th August 2019
E-First on 24th October 2019
wind-turbine generators doi: 10.1049/iet-rpg.2018.6179
www.ietdl.org

Jianwei Yang1, Xiaofei Sun2, Kai Liao1,2,3 , Zhengyou He1, Liangcheng Cai1
1School of Electrical Engineering, Southwest Jiaotong University, Chengdu, People's Republic of China
2National Rail Transit Electrification and Automation Engineering Technique Research Center, Chengdu, People's Republic of China
3Chengdu Guojia Electrical Engineering Co., Chengdu, People's Republic of China

E-mail: [email protected]

Abstract: In order to smooth the system frequency disturbances caused by wind power fluctuation, this study proposes a novel
load frequency control (LFC) strategy based on model predictive control (MPC) to regulate the system frequency. The proposed
strategy mainly aims to improve the system frequency responses by considering the dynamics of wind turbine (WT) and
forecasting the output power of WT. First, the frequency response model (FRM) of system including the power system LFC and
WT is established. Second, through discretisation of the FRM, a predictive model of system is obtained. Then, system state and
response in the next predictive period are forecasted by the predictive model. After obtaining the prediction information, the
control signal of system is optimised for improving system frequency response. On this basis, a novel MPC-based LFC strategy
is proposed based on the predictive model and the optimisation of control sequence. With the information of wind speed, this
control strategy is able to forecast the system states and responses easily and adjust the control signal according to wind power
fluctuation. Finally, numerical simulations are demonstrated to visualise the effectiveness and feasibility of the proposed MPC-
based LFC strategy and its advantages over existing methods.

1 Introduction response and better control effect compared with the conventional
PID LFC. The main advantage of optimisation-based methods is
Load frequency control (LFC) plays a key role in the operation of their ability for achieving multi-objective programming. However,
power systems of which major function is to match the power these methods cannot respond wind power in advance, which lead
generation with the load demand during load and source variations to the constrained response speed and robustness against wind
[1]. A traditional LFC is generally based on area control error power fluctuation. Besides, the participation of WTs in LFC leads
(ACE) to control the unit in the area. This input signal consists of to the decline in the utilisation of wind energy.
local area frequency and tie-line power exchanges. Among these control schemes, MPC performs a fast response
Wind energy is a clean and renewable energy with the most and robustness against wind power and load changes. It shows a
large-scale development prospects. Due to the randomness and straightforward optimisation procedure to obtain optimal control
intermittency of wind power, wind power will inevitably introduce actions in each control time domain. Meanwhile, another
new challenges to the power system stability and safety, especially advantage of MPC controllers is that it can forecast the system
to the frequency stability [2]. Maintaining frequency in an accepted state by the model of a plant in advance. Reference [16] proposes a
region is based on active power balance. Therefore, the intermittent linear MPC based on fuzzy logic techniques, in which the
wind power integration has a significant influence on the dynamic predictive model is designed according to the system LFC
characteristics of system frequency. Meanwhile, large-scale wind structure. In [17], the MPC based on offline non-linear model is
power integration will pose a greater challenge to LFC of presented. The simulation results show this controller can
interconnected power systems [3]. effectively reduce frequency deviation. These MPC strategies in
To meet the system frequency standards, the classical these literatures are able to forecast the system state in advance and
proportional–integral–derivative (PID) controllers are widely used change the controller output instruction according to the system
to control the ACE in a LFC [4]. Besides this, there are many new responds; so they are able to improve system performance.
control schemes, such as optimal control [5], robust control [6], However, for the interconnected power system with WTs, it is not
event-triggering LFC [7], fuzzy control [8], hybrid control realistic to get perfect wind speed prediction. The MPC modelling
algorithms [9], and model predictive control (MPC) [10], have in these literatures is based on the system LFC structure without
been presented. Some intelligent optimisation algorithms are the dynamic model of WTs, which causes that the output power
utilised to tune PID parameters to obtain better frequency command of controllers cannot adapt to wind power changes well.
performance and controller dynamics in the wind power grid- Therefore, it is necessary to reconstruct the MPC model
connected power systems [11]. Reference [12] proposes a considering the dynamic model of WTs so that this controller is
hierarchical optimal robust controller to deal with LFC parameter, able to deal with the randomness and intermittency of wind power.
load variations, errors in modelling. This controller guarantees In this paper, a novel LFC strategy is proposed based on MPC
stability and performance of systems. Reference [13] presents a algorithm and the LFC structure incorporating the dynamic model
decentralised fuzzy logic-based LFC scheme in wind power grid- of WTs. The proposed strategy aims to manage the system control
connected power systems. Based on this scheme, a new fuzzy- signal to improve the frequency performances. First, the frequency
based speed controller, with the fuzzy logic scheme and the genetic response model (FRM) of a power system with WTs is established.
algorithm, is designed in [14]. In [15], a LFC strategy by the The FRM consists of the multi-area interconnected power system
participation of wind turbines (WTs) based on the rotor kinetic LFC and WT generators. Then, the state-space model of FRM is
energy control is proposed to provide effective frequency support built by combining the mathematical models of power system LFC
for a power system. The simulation results in the above literature and the WT, which will be used in MPC as the system model.
show that the proposed LFC strategies present a faster dynamic Through discretisation of this state-space model, the system

IET Renew. Power Gener. 1

© The Institution of Engineering and Technology 2019
Fig. 1 Structure of FRM of a power system with WTs

2.1 Wind turbine model

In order to improve the power generation efficiency of WTs, the
WT is operating at the MPPT mode, with its structure shown in
Fig. 2.
The WT model is consisted of a drive train system and a blade
rotor aerodynamics section. The WT converts wind energy into
mechanical energy, which drives the rotor of the generator. The
output mechanical power Pm can be calculated as follows [18]:

Fig. 2 Structure of wind turbine Pm = 0.5πρR2C p(λ, β)vw3 (1)

predictive model is obtained. By the predictive model, the
where ρ, vw, and R are air density, wind speed, and blade length,
frequency responses and states of system are able to be forecasted
with wind speed information in advance. With this predictive respectively. Cp is the aerodynamic power coefficient. It can be
information, system control signal can be optimised for better represented with a non-linear function of the tip speed ratio λ and
system frequency responses. The proposed strategy can fully utilise the blade pitch angle β given as:
the advantages of MPC, which significantly improves dynamic
performance of power system frequency. Furthermore, the WT C p(λ, β) = c1(c2 /λi − c3 β − c4)ec5 / λi + c6λ (2)
dynamic is considered in the proposed strategy to deal with wind
power fluctuation. Besides, the inertia of the drive train in WT where λ = Rwm/vw, 1/λi = 1/(λ + 0.08β) − 0.035/(0.08β3 + 1), wm
results in a delay of its output power relative to real-time wind is the mechanical rotation speed, and ci (i = 1,2,3) is the constant
speed input. By using this delay, output power of WT can be coefficients.
obtained in advance of real time. Therefore, the proposed strategy When the WT operates at MPPT mode to harvest maximum
is able to effectively control frequency without wind speed wind power, β is set as zero. On this basis, Cp can be represented as
forecast. This guarantees the feasibility of the proposed strategy.
Thus, the controller can fast response according to the wind power a second-order polynomial of λ [19]:
fluctuation, which significantly improves dynamic performance of
power system frequency. At last, simulation experiments on a C p(λ) = k2λ2 + k1λ + k0 (3)
single-area power system and a four-area power system are carried
out to verify the effects of proposed MPC-based LFC strategy. where k1, k2, and k0 are constant coefficients. Under an invariant
The remainder of this paper is organised as follows: the original wind speed, the maximum power can be obtained when
dynamic models of power system LFC, the dynamic models of dC p(λ)/dλ = 2k2λopt + k1 = 0, which gives the optimal tip speed
WTs and the FRM of a power system with WTs are established in ratio:
Section 2. In Section 3, a predictive model is introduced in the
MPC. Then, the MPC-based LFC strategy is proposed based on the k1
λopt = − (4)
MPC formulation. Simulation results carried on a single-area and a 2k2
four-area power system are shown in Section 4. Section 6 gives the
conclusion of this paper. Since the rotor of the generator is connected to the WT though a
gearbox, the optimal generator rotor speed is:
2 Frequency response model of a power system
with WTs p p λoptvw pk1kg
ωropt = kw = k = − v (5)
2 g mopt 2 g R 4k2R w
FRM of a power system with WTs consists of multi-area
interconnected power system LFC and WT model, which is where p is the number of pole pairs, kg is the gearbox ratio, ωropt
depicted in Fig. 1. The WT is operating at maximum power point
and ωmopt are the optimal rotor speeds of generator and WT,
tracking (MPPT) mode. In this section, the models of WT and
multi-area interconnected power system LFC are described, and the respectively. Substituting (4) into (1) and (3), the maximum
FRM employed in the predictive model of MPC is derived. mechanical power can be obtained:

2 IET Renew. Power Gener.

© The Institution of Engineering and Technology 2019
 respectively; ΔPLi denotes the load variations; Hi and Di are the
inertia of the synchronous machine and machine damping
coefficient, respectively; ΔPtie, i is the difference of the tie-line
power between the actual and the scheduled power flows.
The dynamics of the turbine can be represented as:

1 1
ΔṖmi = ΔP − ΔP (14)
T ti gi T ti mi

Fig. 3 Structure of the ith area LFC model in an n-area power system where ΔPgi is the deviations of valve position; T ti denotes the
constant time of turbine.
k12 k2 The dynamic equation of the governor can be expressed as:
Pmax = 0.5πρR2C pmaxvw3 = 0.5πρR2vw3 − 1 + k0 (6)
4k2 2k2
1 1 1
ΔṖgi = ΔP − Δf − ΔP (15)
By manipulating (5) and (6), the mathematical description between T gi ci RiT gi i T gi gi
ωropt and Pmax can be computed as:
where T gi and Ri are the time constant of the governor and speed
3
Pmax(ωropt) = koptωropt (7) drop.
The total tie-line power change between areas i and j and other
areas can be calculated as:
where the optimal coefficient is expressed as
kopt = 8(k12 − 4k2k0)πρR5k22 / p3k13kg3 . N
In order to derive the following FRM, this section transforms ΔṖtie, i = 2π ∑ T i j(Δ f i − Δ f j) (16)
the above power form of WTs into mechanical torque form of WTs. j = 1, j ≠ i

3
1 2 vωC p(λ, β) where T i j is the tie-line synchronisation coefficient between area i
Tm = ρR π (8) and area j, and N is the number of area.
2 ωt
The ACE of control area i, which is the power mismatch
where Tm is the mechanical torque of WTs. between the area load and generation, can be expressed as:
In the maximum power tracking mode of WTs, there are the
following mathematical relationships among electromagnetic ACEi = βiΔ f i + ΔPtie, i (17)
torque, electromagnetic power and mechanical speed.
where βi is the frequency bias factor.
Pe
Te = (9)
ωt 2.3 Frequency response model of a power system with WTs
Based on the mathematical models of the WT and frequency
T e = kaωt2 (10) control systems in the previous two sections, a FRM considering
the WT model is established in this section. First, (8), (9) and (11)
Pe = kaωt3 (11) are expanded according to the Taylor series to obtain the following
mathematical model:
where Te is the electromagnetic torque of WTs; Pe is the
electromagnetic power; ka is a constant coefficient. ΔT m = kwtΔωt + kvwΔv
The electromagnetic torque of WTs is related to the mechanical ΔT e = kwgΔωt2 (18)
torque. The mechanical torque is determined by the drive shaft
3
model from the prime mover to the generator. The mathematical ΔPe = kwgΔωt
description of the single mass drive shaft model is as follows.
where the linear coefficient is defined around the first-order Taylor
∂ωt series expansion of the operating point as:
T m − T e − klossωt = 2Hω (12)
∂t
∂T m ∂T m ∂T e ∂Pe
kwt = kvw = kwg = kwp = (19)
where Kloss is loss torque coefficient; Hω is the inertia constant. ∂ωt (0) ∂vw (0) ∂ωt (0) ∂ωt (0)

2.2 LFC model of multi-area interconnected power system Combined with formula (18), the mathematical model (12) of
single-mass drive model of WTs in the previous section is
A multi-area interconnected power system consists of several LFC expressed in incremental form.
units, connected by tie-lines. The load fluctuation in any area can
result in transient frequency turbulence of system. The function of kwt − kwg − kloss kvw
LFC is to balance the output of generators and load demand in each ω̇t = Δωt + Δv (20)
2Hw 2Hw w
area and to achieve the stability of frequency. A typical block
diagram of a non-reheated power system LFC model is illustrated
Considering the dynamic relationship of generator and load in the
in Fig. 3, which is widely used in related literatures, such as [5, 11,
frequency control model (13), according to the structure of FRM of
12]. The dynamic relationship of generator and load in ith area
a power system with WTs, the following relationship between
between the incremental mismatch power and the frequency
frequency and power can be derived:
deviation Δ f i can be expressed as:
1
1 1 Di 1 Δ f i = (ΔPmi − ΔPLi − ΔPtiei − ΔPei) ⋅ (21)
Δ f˙ i = ΔP − ΔP − Δf − ΔP (13) 2His + Di
2Hi mi 2Hi Li 2Hi i 2Hi tie, i
Substituting (18) and (19) into the above equation, the following
where i is the area number; Δ f i and ΔPmi are the deviations of relationship can be obtained:
system frequency and synchronous machine mechanical output,

IET Renew. Power Gener. 3

© The Institution of Engineering and Technology 2019
 1 3.1 Predictive model based on FRM for MPC
Δ f˙ i = (ΔPmi − ΔPLi − ΔPtiei) ⋅ − kwp
2Hi
(22) At the design stage, the linearisation model should be discretised
1 Di
⋅ ⋅ Δωt − ⋅ Δfi first. The discrete linear state-space model of (23) can be expressed
2Hi 2Hi as:
According to the above equation, FRM is established for multi-area xi(k + 1) = Ai, d xi(k) + Bi, dui(k) + Fi, dwi(k)
interconnected power system with WTs. By combining the above (24)
WT model (1)–(7) and multi-area power system LFC model (13)– yi(k) = Ci, d xi(k)
(17), the state-space model for multi-area power system LFC with
WTs can be uniformly described as the following equation: where Ai, d, Bi, d, Ci, d, and Fi, d are the state matrix, control matrix,
observation matrix and disturbance matrix corresponding to the
ẋi(t) = Ai xi(t) + Biui(t) + Fiwi(t) discretisation of the continuous state-space model of (23).
(23) According to the principle of the model control algorithm, the
yi(t) = Ci xi(t)
system response yi(k + N p k) is predicted in the prediction time
domain N p first. The prediction is based on the state information
where the state variable matrix can be expressed as
xi(t) = [Δ f i ΔPtie, i ΔPmi ΔPgi Δwti], the control variable is the xi(k + 1) and the state (24) of the current system time (time k), as
input power command of generators ui = ΔPci, the disturbance shown in (25).
vector wi(t) = ∑Nj = 1, j ≠ i Ti jΔ f j Δvi , yi is ACEi in area i which is yi(k + 1 k) = Ci, d Ai, d xi(k) + Ci, d Bi, dui(k k)
taken as the output of the system, the coefficient matrices in the
+Ci, dFi, dwi(k k)
state-space model
yi(k + 2 k) = Ci, d Ai, d xi(k + 1) + Ci, d Bi, dui(k + 1 k)
− Di /2Hi −1/2Hi 1/2Hi 0 −kw pi /2Hi +Ci, dFi, dwi(k + 1 k)
N
= Ci, d Ai2, d xi(k) + Ci, d Bi, d Ai, dui(k k)
2π ∑ Ti j 0 0 0 0
j = 1, j ≠ i +Ci, dFi, d Ai, dw(k k) + Ci, d Bi, dui(k + 1 k)
Ai = 0 0 −1/Tti 1/Tti 0 +Ci, dFi, dwi(k + 1 k) (25)
−1/T gi Ri 0 0 −1/T gi 0 ⋮
κ N N −1
0 0 0 0 yi(k + N p k) = Ci, d Ai, dp xi(k) + Ci, d Bi, d Ai, dp ui(k k)
2Hwi
N − Nu
0 0 +⋯ + Ci, d Bi, d(Ai, dp + ⋯ + I)ui(k + Nu − 1 k)
0 βi
−2π 0 N −1
+Ci, dFi, d Ai, dp wi(k k) +
0 1
0 0
Bi = 0 , Fi = , Ci = 0 . +⋯ + Ci, dFi, d(Ai, dp
N − Nu
+ ⋯ + I)wi(k + Nu − 1 k)
0 0
1/T gi 0
kvω
0 0 − 0 where Nu is control time domain, Nu ≤ N p; yi(k + j k) is system
2Hω
output response at time k + j predicted at time k, j = 1, 2, …, N p;
where the parameter κ in matrix Ai can be expressed as ui(k + j k) is the calculated value of control variable at time k,
j = 1, 2, …, Nu − 1; I is a unit matrix; when Nu ≤ j ≤ N p,
κ = kwt − kwg − kloss . ui(k + j k) = ui(k + Nu − 1 k).
The matrix form of (25) can be expressed as:
In the FRM presented in (13), the WT dynamic model has been
considered. Therefore, the output power of WT can be obtained by Y i = Gi xi(k) + LiUi + MiW i (26)
the input of wind speed, as well as the frequency fluctuation and
state variable of system. Meanwhile, this FRM shows the linear where
relationship between wind speed fluctuation and frequency
T
variation, i.e. it is able to obtain the system output frequency and Y i = yi(k + 1 k), yi(k + 2 k), …, yi(k + N p k) ;
other variables by the input of wind speed according to WT and
system LFC characteristics. Therefore, this FRM will be used as a Ui = ui(k k), ui(k + 1 k), …, ui(k + Nu − 1 k) T;
predictive model after transformation, which is explained in
Section 3. T
W i = wi(k k), wi(k + 1 k), …, wi(k + N p − 1 k) ;
3 Design of MPC-based LFC strategy N T
The general MPC model consists of prediction unit and controller Gi = Ci, d Ai, d, Ci, d Ai2, d, …, Ci, d Ai, dp ;
unit [20]. The prediction unit forecasts future control action of the
system based on its current output, disturbance, and control signal. Ci, d Bi, d 0 ⋯ 0
The control unit is to minimise the objective function according to Ci, d Ai, d Bi, d Ci, d Bi, d ⋯ 0
system constraints by using the predicted output. MPC Li = ,
compensated the measured disturbance by the feedforward control ⋮ ⋮ ⋯ ⋮
method. Feedforward control compensates most of the measured N −1 N −2
Ci, d Ai, dp Bi, d Ci, d Ai, dp Bi, d ⋯ ξ
disturbance before effect on the system, which is different to the
traditional feedback control. The feedforward control cooperates N − Nu
the feedback control to reject most of the measured disturbance in which ξ = Ci, d(Ai, dp + ⋯ + I)Bi, d;
effect. Considering this advantage of MPC, the proposed strategy is
designed based on the FRM of system with WTs.

4 IET Renew. Power Gener.

© The Institution of Engineering and Technology 2019
 Ci, d Fi, d 0 ⋯ 0 1, based on this prediction information, the optimisation object (27)
Ci, d Ai, d Fi, d Ci, d Fi, d ⋯ 0 is solved in each controller. The MPC-based LFC strategy for
Mi = , multi-area interconnected power system with WTs can be
⋮ ⋮ ⋯ ⋮ summarised by the following algorithm.
N −1 N −2
Ci, d Ai, dp Fi, d Ci, d Ai, dp Fi, d ⋯ ζ
Step 1 (Initialisation): Set t = k, control time domain Nu, and
N −N
prediction time domain Np. Obtain control action ui(k) and
in which ζ = Ci, d(Ai, dp u + ⋯ + I)Fi, d. previous predictive variables xi(k).
Equation (26) is the predictive model of LFC in area i, which is Step 2 (Optimisation): ui(k + 1) is obtained by solving the optimal
the prediction unit of MPC model. With this predictive model,
object (18).
system frequency and state information can be obtained in advance
based on its current output, disturbance, and control signal. This Step 3 (Assignment): If the optimal control action ui(k + 1) is
predictive information will be used in controller unit for the feasible, apply it, otherwise, return to Step 2.
optimisation of control signal. Step 4 (Prediction): Predict system states of next prediction time
domain.
3.2 Proposed MPC-based LFC strategy Step 5 (Implementation): Apply the control action ui(k). Set t = k +
1, and return to step 2 at the next control time domain.
LFC is to regulate the output power of each generator to reach a
balance between the generation and the total load demand. The Through these steps, the control sequence ui(τ), τ ∈ [k, k + N p] can
function of MPC-based LFC is to forecast system output and adjust be obtained from the proposed MPC. Then, this control signal will
control signal for better frequency response. Therefore, it is modulate system generators' output power according to wind
important to define a suitable control objective and subject for the power fluctuations, which can smoothen system frequency.
MPC that takes into account the special dynamics of the WTs [21].
Fig. 4 shows a simple structure of MPC method.
The MPC algorithm uses the predictive model to predict the 4 Simulation and analysis
future output of the system. Then, it optimises a series of control To verify the effectiveness of the proposed LFC strategy in
actions from the current time to the prediction time domain by improving system frequency response, this section performs
minimising the controller objective function. Only the actions of simulations on a single-area and a four-area power systems to test
control time domain in the optimised control sequence are carried the proposed MPC-based LFC strategy. The results of the proposed
out in each prediction cycle. Then, the previous step is repeated in strategy considering the WT dynamic model, the MPC strategy
the next prediction cycle to recalculate the control action sequence without considering the WT dynamic model, and the conventional
[22]. PID control are compared and discussed in this section.
For a multi-area power system with WTs, the objectives of
MPC are to optimise the reference to the thermal generator based 4.1 Case 1: single-area power system
on the system frequency standard and actual frequency of the
system. Therefore, for each control area, the control objective In this case, the proposed MPC-based LFC strategy for a single-
function of the MPC design takes both the dynamic performance of area power system with WTs is simulated. The parameters of
the thermal power plant and the minimisation of the frequency generator, governor, and non-reheated turbine are given in Table 1,
deviation of the power system. In order to take into account the which are widely used in [23]. This single-area includes the
dynamic performance of generators and reduce frequency deviation aggregated WT, of which parameters is shown in Table 2. The
at the same time, this paper chooses the optimisation objective as simulations for the presented scenarios are performed with a
shown in (27). This objective function is widely used in frequency control operation cycle of T = 0.1 s, prediction horizon of Tp = 0.3
control research, such as [23]. By the optimisation target, the s, and the control horizon of Tc = 0.1 s.
proposed strategy is able to achieve the maximum safe and
economical profit simultaneously. The objective function of area i
is expressed as:
Np
Ji(k) = ∑ [yiT(k + n k)Qiyi(k + n k)
n=1 (27)
+ui(k + n k)Ti Rui(k + n k)]

where Qi and Ri are the symmetric weighting and positive definite

matrices. Fig. 4 Block diagram of MPC method
The optimal control problem at instant k is formulated as:

min Ji(k) Table 1 Parameters of the single-area power system with

ui(k + n k)
(28) non-reheated turbines
Tt Tg H D
subject to 0.4 0.08 0.08335 0.015
xi(k + 1) = Ai, d xi(k) + Bi, dui(k) + Fi, dwi(k)
(29)
yi(k) = Ci, d xi(k) Table 2 Parameters of the aggregated WT
Parameters Symbol Value and units
Equation (28) and constraints (29) form a quadratic programming radius of rotor R 38.5 m
problem. By using the rolling optimisation algorithm, the optimal
air density ρ 1.901 × 10−3(m/s)−3
control sequence in first control domain is taken as the actual
control action of the system. At instant k + 1, the system status number of pole pairs p 2
information is re-measured, the optimisation program is started generator inertia constant HDFIG 5.28 s
again, and the above process is repeated. By the predictive model nominal frequency fnom 50 Hz
(26), controllers of all areas predict the future state at time k with
rated turbine power Pbase 1.5 MW
the optimal control sequence over the control horizon. At time k +

IET Renew. Power Gener. 5

© The Institution of Engineering and Technology 2019
 Table 3 Maximum frequency deviation and ITAE of the
single-area power system
PID control MPC without Proposed control
strategy considering WTs strategy
|Δfmax| 0.173 Hz 0.144 Hz 0.014 Hz
ITAE 2.114 2.060 0.180

Table 4 Parameters of the single-area power system with

non-reheated turbines
Tt Tg H D
area 1 0.4 0.08 0.08335 0.015
area 2 0.33 0.072 0.111 0.04
area 3 0.35 0.07 0.08 0.05
area 4 0.375 0.085 0.065 0.0667

Fig. 5 System frequency simulation results under various wind speed

(a) The 50 s wind speed sequence,
(b) The wind power output from WT,
(c) The frequency deviation,
(d) The system input power command

Fig. 6 Structure of the studied four-area power system

The 50 s wind speed sequence shown in Fig. 5a from the

standard wind speed data is added to the system. Fig. 7 Wind speed variation and wind power output of WT in four-area
Fig. 5b is the wind power output from WT. In Figs. 5c and d, system
they show the frequency deviation dynamic responses and the (a) The 50 s wind speed sequence in area 1,
system control variable. (b) The 50 s wind speed sequence in area 4,
The red solid line corresponds to the proposed MPC-based LFC (c) The wind power outputs of the WT in area 1,
strategy, the blue-dashed line corresponds to the PID controller (d) The wind power outputs of the WT in area 4
with the parameters reported in [24], and the green dash-dotted line
corresponds to the MPC strategy without considering WT dynamic can obviously reduce the frequency deviation and improve the
model. The maximum frequency deviation and the ITAE of frequency dynamic response.
frequency deviation in the simulation results are shown in Table 3.
It can be seen from the table that the frequency deviation of the 4.2 Case 2: four-area power system
MPC strategy considering the dynamic model of WTs is the
smallest, which is 0.014 Hz. Meanwhile, the ITAE value of the In this case, the proposed MPC-based LFC strategy is tested on a
error integral is 0.180, which indicates that the strategy proposed in widely used four-area power system [5, 25]. The system structure
this paper can effectively suppress the frequency fluctuation. As is given in Fig. 6. A four-area power system with non-reheated
the blue line shown in Fig. 5c, with the PID controller, the system turbines has the nominal parameters for simulation listed in
frequency is significantly effected due to wind power fluctuation. Table 4. To verify the effectiveness of the proposed strategy in
Comparing with it, the frequency dynamic response becomes better improving system frequency performance, the aggregated WT
when adopting the MPC strategy without considering the WT models are both added to area 1 and area 4, which are printed in
dynamic model shown in the green line. However, because it does red. The 50 s wind speed sequences shown in Figs. 7a and b are
not consider the WT dynamic model in MPC controller, the control input in the WT of area 1 and area 4, respectively. The wind power
effect of the MPC controller is not significant. When the proposed outputs of the WT in area 1 and area 4 are shown in Figs. 7c and d,
MPC controller considering the WT dynamic model is adopted, it respectively. Meanwhile, the simulations for the presented

6 IET Renew. Power Gener.

© The Institution of Engineering and Technology 2019
Fig. 8 Simulation results of area 1 in four-area system
(a) The input power command, Fig. 9 Simulation results of area 2 in four-area system
(b) The output power of the system generator, (a) The input power command,
(c) The frequency deviation, (b) The output power of the system generator,
(d) The tie-line power deviation (c) The frequency deviation,
(d) The tie-line power deviation
scenarios are performed with the prediction horizon of Tp = 0.3 s
and the control horizon of Tc = 0.1 s. in a delay between the mechanical power extracted by the
Similarly, the proposed MPC-based LFC strategy considering incoming wind speed and the real power generated by WTs.
the WT dynamic model, the MPC strategy without considering the Therefore, the power output of a WT can be predicted. By using
WT dynamic model and the conventional PID control are this delay, output real power of WTs can be obtained in advance, so
simulated. The results of area 1 to area 4 are shown in Figs. 8–11, the proposed strategy is able to effectively control frequency
respectively. without wind speed forecast. This guarantees the feasibility of the
The maximum frequency deviation of area 1 to 4 is shown in proposed strategy.
Table 5, and the ITAE of each area is given in Table 6. Through
Tables 5 and 6, it can be known that the frequency deviation of the 4.3 Sensitivity of MPC-based LFC strategy
proposed strategy is significantly smaller than the MPC strategy In order to investigate the sensitivity of the performance of MPC
without considering the WT dynamic and the PID frequency with regard to modelling uncertainty, assuming that the time
control strategy. The proposed strategy can effectively consider the constant of the governor Tg in the system is uncertain, let Tg be
WT dynamics to predict the system status and response. With these
0.06, 0.08, and 0.1, respectively. The strategy proposed in this
predictive information, the corresponding optimised control signals
paper is simulated. The simulation system adopts the single-area
is optimised to enhance the system's ability dealing with the
system and parameters are same with Case 1. The system
frequency deviation caused by the wind power fluctuation.
frequency deviation curve is obtained as shown in Fig. 12.
These results show that the proposed strategy can control the
According to the results shown in Fig. 12, it can be concluded
input power command so that the generator can output the
that the proposed strategy still has good control effect when the
appropriate power to stabilise the fluctuation of wind power.
system parameters are uncertain.
Therefore, the system frequency responses have very small
deviations compared to the PID controller and the MPC strategy
without considering the WT dynamic model. Meanwhile, the 5 Conclusions
frequency responses of the areas whether added the WT or not all In this paper, in order to smooth the frequency fluctuation caused
have a good dynamic performance. This indicates that the proposed by wind power integration, an effective MPC-based LFC strategy
MPC-based LFC strategy can offer a control command to the is proposed for power system with WTs. To forecast the output
system generator to help it track the wind power fluctuation well. power of WT, the WT dynamic model is considered in the
Therefore, the output power of generators can effectively balance derivation of the system FRM. This FRM of system with WTs is
the wind power fluctuation. The change in power output of each established by combining the WT model with the system LFC
generator unit settles at their expected values at steady state. Then, model. Based on this FRM, a MPC-based LFC strategy is designed
the local frequency changes of each area converge to zero. It via a rolling forecast and optimisation. By discretising the FRM, a
ensures that the frequency deviation is maintained within the safe system predictive model can be obtained to forecast the output
operation range. Furthermore, by using the proposed strategy, the power of WT and the system state. According to these predictive
power flow on the tie-line also tends to be smooth and stable. information, the power control signal can be optimised. Then, the
Actually, it is difficult to get an accurate sequence of wind speed output power of system generator is adjusted to smooth the
by forecasting. However, the inertia of the drive train in WT results frequency fluctuation. With this strategy, the output power of the

IET Renew. Power Gener. 7

© The Institution of Engineering and Technology 2019
Fig. 10 Simulation results of area 3 in four-area system
Fig. 11 Simulation results of area 4 in four-area system
(a) The input power command,
(a) The input power command,
(b) The output power of the system generator,
(b) The output power of the system generator,
(c) The frequency deviation,
(c) The frequency deviation,
(d) The tie-line power deviation
(d) The tie-line power deviation

Table 5 Maximum frequency deviation of the four-area

power system
PID control MPC without Proposed control
strategy, Hz considering WTs, strategy, Hz
Hz
area 1 0.216 0.095 0.018
area 2 0.186 0.086 0.009
area 3 0.186 0.086 0.009
area 4 0.303 0.106 0.042 Fig. 12 System frequency deviation curve when Tg = 0.06, 0.08 and 0.1

6 Acknowledgments
Table 6 ITAE of frequency deviation of the four-area power The authors acknowledge the support by Joint Funds of the
system National Natural Science Foundation of China under Grant
PID control MPC without Proposed control U1766208 and National Natural Science Foundation of China
strategy considering WTs strategy under Grant no. 61603311.
area 1 3.7 1.2 0.4
area 2 3.2 1.1 0.3 7 References
area 3 3.2 1.1 0.3 [1] Cintuglu, M.H., Mohammed, O.A.: ‘Behavior modeling and auction
area 4 4.5 1.3 0.4 architecture of networked microgrids for frequency support’, IEEE Trans. Ind.
Inf., 2017, 14, (4), pp. 1772–1782
[2] Zhang, W., Fang, K.: ‘Controlling active power of wind farms to participate in
load frequency control of power systems’, IET Gener. Transm. Distrib., 2017,
WT can be fully taken into account while the wind speed changes. 11, pp. 2194–2203
The frequency deviation is significantly reduced within the safe [3] Ono, T., Arai, J.: ‘Frequency control with dead band characteristic of battery
energy storage system for power system including large amount of wind
operation range, and the dynamic performance of the system power generation’, Electr. Eng. Jpn., 2013, 185, pp. 1–10
frequency can be easily tuned to the expected results. Furthermore, [4] Khodabakhshian, A., Edrisi, M.: ‘A new robust PID load frequency
because of the delay of the WT output, the proposed LFC strategy controller’, Control Eng. Pract., 2008, 16, (9), pp. 1069–1080
has the feasibility without wind speed forecast. Thus, the [5] Daneshfar, F., Bevrani, H.: ‘Multiobjective design of load frequency control
using genetic algorithms’, Int. J. Electr. Power Energy Syst., 2012, 42, (1), pp.
advantages of the MPC can be fully utilised. To verify the 257–263
effectiveness of the proposed strategy, the simulation results are [6] Chuang, N.: ‘Robust H∞ load-frequency control in interconnected power
carried on a single-area power system and a four-area power systems’, IET Control Theory Appl., 2016, 10, (1), pp. 67–75
system with comparison against the MPC without considering the [7] Wen, S.P., Yu, X.H., Zeng, Z.G., et al.: ‘Event-triggering load frequency
control for multiarea power systems with communication delays’, IEEE
WT dynamic model and the conventional PID control. Trans. Ind. Electron., 2016, 63, (2), pp. 1308–1317

8 IET Renew. Power Gener.

© The Institution of Engineering and Technology 2019
[8] Yousef, H., Kharusi, K.A., Albadi, M., et al.: ‘Load frequency control of a [18] Arani, M.F.M., El-Saadany, E.F.: ‘Implementing virtual iner-tia in DFIG-
multi-area power system: an adaptive fuzzy logic approach’, IEEE Trans. based wind power generation’, IEEE Trans. Power Syst., 2013, 28, (2), pp.
Power Syst., 2014, 29, (4), pp. 1822–1830 1373–1384
[9] Prakash, S., Sinha, S.K.: ‘Load frequency control of multi-area power systems [19] Ye, H., Pei, W., Qi, Z.: ‘Analytical modeling of inertial and droop responses
using neuro-fuzzy hybrid intelligent controllers’, IETE J. Res., 2015, 61, (5), from a wind farm for short-term frequency reg-ulation in power systems’,
pp. 526–532 IEEE Trans. Power Syst., 2016, 31, (5), pp. 3414–3423
[10] Ersdal, A.M., Imsland, L., Uhlen, K.: ‘Model predictive load-frequency [20] Venkat, A.N., Hiskens, I.A., Rawlings, J.B., et al.: ‘Distributed MPC
control’, IEEE Trans. Power Syst., 2016, 31, (1), pp. 777–785 strategies with application to power system automatic generation control’,
[11] Liu, F., Li, Y., Cao, Y., et al.: ‘A two-layer active disturbance rejection IEEE Trans. Control Syst. Technol., 2008, 16, (6), pp. 1192–1206
controller design for load frequency control of interconnected power system’, [21] Kouro, S., Perez, M.A., Rodriguez, J., et al.: ‘Model predictive control:
IEEE Trans. Power Syst., 2017, 31, (4), pp. 3320–3321 MPC's role in the evolution of power electronics’, IEEE Ind. Electron. Mag.,
[12] Rahmani, M., Sadati, N.: ‘Hierarchical optimal robust load-frequency control 2015, 9, (4), pp. 8–21
for power systems’, IET Gener. Transm. Distrib., 2012, 6, (4), pp. 303–312 [22] Elsisi, M., Soliman, M., Aboelela, M.A.S., et al.: ‘Model predictive control of
[13] Bevrani, H., Daneshmand, P.R.: ‘Fuzzy logic-based load-frequency control plug-in hybrid electric vehicles for frequency regulation in a smart grid’, IET
concerning high penetration of wind turbines’, IEEE Syst. J., 2012, 6, (1), pp. Gener. Transm. Distrib., 2017, 11, (16), pp. 3974–3983
173–180 [23] Liao, K., Xu, Y.: ‘A robust load frequency control scheme for power systems
[14] Ashouri-Zadeh, A., Toulabi, M., Ranjbar, A.M.: ‘Coordinated design of based on second-order sliding mode and extended disturbance observer’,
fuzzy-based speed controller and auxiliary controllers in a variable speed IEEE Trans. Ind. Inf., 2018, 14, (7), pp. 3076–3086
wind turbine to enhance frequency control’, IET Renew. Power Gener., 2016, [24] Khousa, E.A., Ismail, A.: ‘Active disturbance rejection control-based load
10, (9), pp. 1298–1308 frequency controller of interconnected power systems involving wind power
[15] Zhang, X., Zha, X., Yue, S., et al.: ‘A frequency regulation strategy for wind penetration’. IEEE GCC Conf. and Exhibition, Doha, Qatar, November 2013
power based on limited over-speed de-loading curve partitioning’, IEEE [25] Cai, L., He, Z., Hu, H.: ‘A new load frequency control method of multi-area
Access, 2018, 6, pp. 22938–22951 power system via the viewpoints of port-Hamiltonian system and cascade
[16] Rao, C.S., Nagaraju, S.S., Raju, P.S.: ‘Automatic generation control of TCPS system’, IEEE Trans. Power Syst., 2017, 32, (3), pp. 1689–1690
based hydrothermal system under open market scenario: a fuzzy logic
approach’, Electr. Power Energy Syst., 2009, 31, (7), pp. 315–322
[17] Spencer, M.D., Stol, K.A., Unsworth, C.P., et al.: ‘Model predictive control of
a wind turbine using short-term wind field predictions’, Wind Energy, 2013,
16, (3), pp. 417–434

IET Renew. Power Gener. 9

© The Institution of Engineering and Technology 2019