Model Predictive Control: A Survey of Dynamic Energy Management
Nsilulu T. Mbungu
1,2,3 a
, Raj M. Naidoo
1 b
, Ramesh C. Bansal
2 c
and Mukwanga W. Siti
3 d
1
Department of Electrical, Electronic and Computer Engineering, University of Pretoria, Pretoria, South Africa
2
Department of Electrical and Computer Engineering, University of Sharjah, Sharjah, U.A.E.
3
Department of Electrical Engineering, Tshwane University of Technology, Pretoria, South Africa
Keywords:
Energy Efficiency, Model Predictive Control, Optimal Control, Power System, Demand Response.
Abstract:
This paper presents the structure of the model predictive control (MPC), its development and application
through optimal energy system. The MPC is one of the algorithms that are used in a computer controlled
environment to predict the future behaviour of an explicit process model. It is devised by computing and
adjusting the next sequence of the input variables at each control interval. The MPC is an algorithm in which
the challenge is to optimize the behaviour of a future plant. The optimization sequence starts by sending the
first input into the plant and then at each subsequent control interval the entire computation is repeated to reach
the performance index function to follow. MPC offers a variety of applications in a wide range of industries.
This is due to its robustness in the optimal control design of a process. MPC is also widely used in aerospace,
automotive, chemical and food processing applications. This study describes the implementation of the energy
management scheme through the use of MPC design.
1 INTRODUCTION
Managing the energy system has transformed the con-
figuration of a conventional power grid in terms of
coordinating the optimal power flow, minimising the
system power losses and voltage stability on the elec-
trical network (Abdi et al., 2017a; Siti et al., 2019;
Abdi et al., 2017b; Mbungu et al., 2019a; Adefarati
et al., 2019; Foley et al., 2020). Currently, several re-
search works try to look for the alternative approaches
of dealing with different grid’s challenges, which con-
sists of analysing the system protection of the electri-
cal system and optimal integration and coordination
of energy mix system. The objective of these ap-
proaches aims to improve the efficiency of the power
system, which must be based on dynamic modelling
strategy.
MPC is frequently used in the industry as an op-
timal control strategy due to its ability to handle hard
system constraints. MPC system design offers the
possibility of controlling the input system and the out-
put constraints (Mesbah, 2016; Mbungu et al., 2018;
Kale and Chipperfield, 2005) as well as the incremen-
a
https://orcid.org/0000-0003-0498-5065
b
https://orcid.org/0000-0002-2439-4505
c
https://orcid.org/0000-0002-1725-2648
d
https://orcid.org/0000-0002-8085-4967
tal constraints of the control signal (Mbungu et al.,
2020). Manufacturing process has been widely in-
fluenced by developing the implementation of MPC
due to its approach to resolve and manage the uncer-
tainty of a processing system. It predicts the future
behaviour and keeps the input and output signal in the
acceptable optimal operation level.
Through the MPC strategy, a controller of a given
system can handle multiple inputs, multiple outputs
plant model that are subject to diverse constraints.
The MPC algorithm is also valued for its robustness in
handling unexpected process and system behaviour.
Therefore, this research works contributes on imple-
mentation strategy of a system behavior based on
MPC design. The approach aims to analyse the dy-
namic performance of the energy management under
the smart grid environment.
2 SYSTEM MODELLING
2.1 State-space Model
Consider a given function f (x, u) with its internal sys-
tem, in which the vector space is known as the state-
space. If this structure is also lumped together and
has finite state-space, then, the state-space equations
Mbungu, N., Naidoo, R., Bansal, R. and Siti, M.
Model Predictive Control: A Survey of Dynamic Energy Management.
DOI: 10.5220/0010522201230129
In Proceedings of the 18th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2021), pages 123-129
ISBN: 978-989-758-522-7
Copyright
c
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
123
can describe the system. It is essential to note that
modelling of this type of structure in a state-space
model must satisfy the three superposition of the state
and the output equations and the linearity propri-
ety of state-space (Holkar and Waghmare, 2010; Se-
borg et al., 2010; Bishop, 2007; Lee, 2009; Mayne
et al., 2000; Dahleh et al., 2004; Rossiter, 2003; Ma-
ciejowski, 2002; Rawlings and Mayne, 2009). The
state of the system is the basis of the state-space rep-
resentation. It also considers the value of updating
internal elements of the system. This procedure can
change independently from the system output. The
function f (x, u) in the state-space model is constituted
of three components, namely input variables (u) or
manipulated variables (MVs), output variables (y) or
controlled variables (CVs), and the state variables (x)
(Mbungu et al., 2016; Mbungu et al., 2017b; Wang,
2009; Holkar and Waghmare, 2010; Seborg et al.,
2010; Bishop, 2007; Lee, 2009; Mayne et al., 2000;
Dahleh et al., 2004; Rossiter, 2003; Maciejowski,
2002; Rawlings and Mayne, 2009). The vectors be-
low describe these components as:
u(t) =
u
1
u
2
.
.
.
u
m
, y(t) =
y
1
y
2
.
.
.
y
q
, x(t) =
x
1
x
2
.
.
.
x
n
(1)
where m the number of input into the system, q the
number of output of the plant mode, and n the number
of state that defines the system or order of state-space.
2.1.1 Continuous-time System
If the function f (x, u, t) is considered as the state
evolution equation of a given system, and the out-
put vector y(t) can be described by the function of
the state variable and MVs over a given time t as
g(x, u, t), which is the instantaneous output equation
(Dahleh et al., 2004; Rossiter, 2003; Maciejowski,
2002; Rawlings and Mayne, 2009). Therefore, the re-
lation below can describe the system in a continuous
time model as:
˙x(t) = f (x(t), u(t), t) (2a)
y(t) = g(x(t), u(t), t) (2b)
where t R or R
+
, and ˙x(t) is the rate of change
of the state variables. Equation 2 can be simplified
in compact linear and time-invariant relations, which
describe the continuous state space model of a given
system as:
˙x(t) = Ax(t)+ Bu(t) (3a)
y(t) = Cx(t) + Du(t) (3b)
where A, B,C, and D are respectively, the state ma-
trix of dimension, input matrix, output matrix, and
feed forward matrix. For t 0, x(t) R
n
, u(t)
R
m
, and y R
p
these involve that dim[A] = n × n,
dim[B] = n × m, dim[C] = q × n, and dim[D] = q × m.
Due to the absence of direct feed through on the sys-
tem model developed in (Mbungu et al., 2016; Wang,
2009; Mbungu et al., 2017b), it is assumed that the
feed forward matrix is zero.
2.1.2 Discrete-time System
The discrete-time system model is considered a stan-
dard state-space model. Nowadays, all modern ap-
plications in control systems are discretised or dig-
italised for more accurate evaluation and robustness
of the design. The discrete time system offers the
possibility of determining the output of the system
in real time by using past information of the input.
The technical approach should be followed regard-
ing the amount of input information which defines
the present output. From Eq. 3a, the discrete time
system in a state-space model is developed. This con-
sists of using Euler’s forward approximation method
(Mbungu et al., 2016; Mbungu et al., 2017b; Dahleh
et al., 2004). Suppose that the system is operated at a
given period T , the discretization of state evolution in
that time for (x, u, t) can be determined as (Mbungu
et al., 2016; Mbungu et al., 2017b; Dahleh et al.,
2004; Rossiter, 2003; Maciejowski, 2002; Rawlings
and Mayne, 2009).
x((t + 1)T ) x(tT )
T
= Ax(tT ) + Bu(tT ) (4)
If it is supposed that the parameters (tT ) can be
replaced by (k), where k Z denotes the time sample,
Eq. 4 can therefore be rewritten as follows
x(k + 1) = (I + TA)x(k) + T Bu(k) = A
d
x(k) + B
d
u(k)
(5)
By combining Eq. 5 with the output function of
continuous state space model Eq. 3b which is sup-
posed to be discretised by Euler’s forward approxi-
mation like Eq. 43a, the linear discrete state space
model can be described as
x(k + 1) = A
d
x(k) + T B
d
u(k) (6a)
y(k) = C
d
x(k) + D
d
u(k) (6b)
where A
d
is the discrete state matrix, B
d
is discrete
input matrix, C
d
is discrete output matrix, and D
d
is the discrete feedforward matrix. It is also impor-
tant to notice that all discrete time matrices of Eq. 6
have the same dimension with the continuous matri-
ces that are defined in Eq 3. However, in most cases,
ICINCO 2021 - 18th International Conference on Informatics in Control, Automation and Robotics
124
for example, in predictive control, it can sometimes
be assumed that the feed-forward matrix D
d
equals
to zero (Mbungu et al., 2016; Mbungu et al., 2017b;
Zhang et al., 2014; Wang, 2009; Rossiter, 2003; Ma-
ciejowski, 2002; Rawlings and Mayne, 2009; Ma
et al., 2011; Xie et al., 2007). This is dependent on the
system design due to the principle of receding horizon
(Wang, 2009). In this study, the linear discrete state
space model is used to design the system.
2.2 Augmented State Space Model
Suppose that at a given MV with sample k, finding the
relationship between the sample k and the instant be-
fore it, i.e., k 1, this relation could describe the aug-
mented state space model. Equation 6a could, there-
fore, be rewritten as (Mbungu et al., 2017b; Wang,
2009)
x(k +1)x (k) = A
d
(x(k) x(k 1))+B
d
(u(k) u(11))
(7)
Through Eq. 7, the increment of state variables
and the MV can be rewritten respectively by x(k +
1) = x(k + 1) x(k), x(k) = x(k) x(k 1) and
u(k) = u(k) u(k 1). By considering these in-
cremental function and Eq. 7, the increment of the
state-space equation is therefore expressed as
x(k + 1) = A
d
x(k) + B
d
u(k) (8)
As the system input is changed to the increment
of MV, it is, therefore, a question of connecting the
increment of state vector to the CV. This method in-
troduces a new state vector of the system that is de-
veloped as follows.
x
a
(k) =
x(k)
y(k)
(9)
where x
a
denotes the augmented state vector.
From Eqs. 8 and 9, the output vector at the sample
(k + 1) needs to be determined for the stability of the
system. If, at a given sample k of a CV, the designed
model can predict a future CV at the sample (k + 1).
By using the developed strategy of Eq. 7, which can
be identified with the effect of CV in Eq. 9, thus, Eq.
6a as a function of current and future CV with D
d
= 0
is expressed as follows.
y(k + 1) y(k) = C
d
(x(k + 1) x(k)) = C
d
x(k + 1)
(10)
By substituting Eq. 8 in Eq. 10, the relation-
ship between the current and predicted CV can be ex-
pressed as follows.
y(k + 1) y(k) = C
d
A
d
x(k) +C
d
B
d
u(k) (11)
Equations 12a and 12b define the compact format
of the augmented state-space model, which derives
from Eqs. 8, 9, and 11 as follows.
x
a
(k + 1) = A
a
x
a
(k) + B
a
u(k) (12a)
y(k) = C
a
x
a
(12b)
where A
a
=
A
d
0
T
d
C
d
d
1
, B
a
=
B
d
C
d
B
d
, C
a
=
0
d
C
d
1
,
x
a
(k + 1) =
x
d
y(k)
, and 0
d
=
0 0 ... 0
with 0
T
d
as a zero column vector of n dimension. The aug-
mented state space system is also used in MPC design
(Wang, 2009).
3 PREDICTIVE CONTROL
The MPC approach is the composition of three base
components of the predictive controller. These are
the prediction, optimisation and receding horizon im-
plementation. The advantages of predictive control
strategy are its stability of driver for constrained sys-
tems. Moreover, due to the opportunity of real-time
computation and the improvement of the predictive
controller efficiency, the application predictive con-
trol has extended many controller structures that in-
clude high-speed sampling systems (Cannon, 2015).
3.1 Prediction
By considering the dynamic model of discrete-time
linear state-space model Eq. 6a and the MPC sam-
ple time strategy, to generate the predicted behaviour
of a given plant system consists of assuming that at
sampling instant kth, the future state vector is in a re-
lationship with the next input. This strategy is exe-
cuted for N sampling intervals. If only at each given
predicted MV sequence when the model is simulated
forward over a given prediction horizon, the corre-
sponding sequence of predicted state is, therefore,
generated to describe the future sequence behaviour
of the system. The vector state and below input define
the predicted sequence of the discrete-time dynamic
model.
x(k) =
x(k + 1|K)
x(k + 2|K)
x(k + 3|K)
.
.
.
x(k + N|K)
, u(k) =
u(k|K)
u(k + 1|K)
u(k + 2|K)
.
.
.
x(k + N 1|K)
(13)
where x(k + i|K) and u(k + i|K) are respectively the
state vector and MV at a time (k + i) and the parame-
ter k of each variable denotes the predicted sampling
Model Predictive Control: A Survey of Dynamic Energy Management
125
instant. Through Eq. 13, the demonic model of linear
discrete state-space is rewritten in predictive environ-
ment as:
x(k + i + 1|k) = A
d
x(k + 1|k) + B
d
u(k + 1|k) (14)
with i = 0, 1, ..., N, and for the initial condition i.e.
i = 0 the state vector is x(k|k) = x(k ).
3.2 Optimisation
References (Wang, 2009; Holkar and Waghmare,
2010; Seborg et al., 2010) describe the predicted con-
trol law computation framework. The optimisation
strategy of MPC is based on minimising anticipated
performance cost. The predicted sequence of state
and MV play an influential role in the optimisation
strategy. Equation 15 defines the optimal approach to
predictive control as:
J(K) =
N
i=0
[x
T
(k + i|k)Qx(k + i|k) + u
T
(k + i|k)Ru(k + i|k)]
(15)
where J(k) denotes the performance index of pre-
dicted sequences, and Q and R are the positive def-
inite weighting matrices. But Q or R can also be a
positive semi-definite matrix. It is also important to
notice that Q and R are diagonal matrices that con-
tain only the positive elements. It has been noted that
the performance index is a function of state and MV
at each instant k. During the optimization strategy of
predicted input sequence which consists of minimiz-
ing Eq. 15 to find the minimum argument of the input
sequence, at each instant the optimal MV can be cal-
culated as:
u
(k) = argmin
u
J(k) (16)
It is also important to note that finding the mini-
mum MV can be subject to the input, state, and output
constraints. Therefore, these can be included in the
optimization strategy to determine the optimal solu-
tion. The structure of the restrictions in MPC design
will be established further in section 4.
3.3 Receding Horizon Implementations
Once the initial value is computed, i.e. at i = 0 by
using Eqs. 13 and 16 under a finite horizon, the op-
timal predicted MV sequence that is introduced into
the plant through the MPC control law is determined
as follows.
u(k) = u
(k|k) (17)
For k = 0, 1, ..., N, at each sampling instant as de-
scribed in (Wang, 2009; Holkar and Waghmare, 2010;
Seborg et al., 2010) the same process that is computed
for the first element is then repeated at each sampling
instant. The effect of repeating the optimisation of
future time instants describes the online optimisation
strategy of predictive control. This strategy defines
the receding horizon approach that keeps the predic-
tion horizon at the same length. The concept of feed-
back as described in (Holkar and Waghmare, 2010;
Seborg et al., 2010) determines the degree of robust-
ness of the system.
4 MPC DESIGN
If the dynamic of state-space design for a given dig-
ital mode Eq. 12 can verify either the controllability
or the observable laws, this dynamic model can also
be implanted in MPC controller. The MPC design
entails controlling the optimum approach that the ob-
servation sets at each predicted sequence as defined in
Eq. 15, which is defined as a quadratic equation. This
performance index can develop in terms of MPC gain
for the robustness of the designed controller (Wang,
2009). Thus, the performance index can be rewritten
in function of CVs and the targets of the system as
follows (Mbungu et al., 2016; Mbungu et al., 2017b;
Wang, 2009).
J(k) = (Y (k) r
w
)
T
(Y (k) r
w
R(k)) (18)
where J(k), R(k) and r
w
are the output system,
target to follow, turning parameter respectively. Af-
ter computation of a given sample k with a given
predicted horizon N
p
and a given control horizon N
c
through an MPC design, the optimum output of sys-
tem is descried in (Mbungu et al., 2018; Tungadio
et al., 2018; Mbungu et al., 2017a).
Afterwards, optimising the given system by us-
ing the MPC design is the effect of implanting a
quadratic equation as described in (Mbungu et al.,
2016; Mbungu et al., 2017b; Wang, 2009), whuch
can compute whether a constraining or unconstrain-
ing plant model. This consists of finding the argu-
ment of CV in relation with the minimum value of the
objective function of the MPC gain.
4.1 Quadratic Programming
A quadratic programming offers several advantages in
the industrial environment due to its opportunity of a
real-time application. This consists of safely writing a
jacket software implementation and the possibility to
update and change the code (Holkar and Waghmare,
2010; Seborg et al., 2010; Mbungu et al., 2020). If
the objective function of a given plant model in MPC
ICINCO 2021 - 18th International Conference on Informatics in Control, Automation and Robotics
126
computation is subject to some linear inequality con-
straints, finding the optimum control solution of the
MVs in receding horizon implementation consists of
resolving a quadratic programming equation below
J(k) = H(k)u(k) +
1
2
u(k)
T
G(k)u(k) (19a)
Mu(k) γ (19b)
where M and γ are constraints matrix and vector. Eq.
19b can be either or not combined with the equal-
ity constraints (Seborg et al., 2010). This system re-
striction mostly depends on what the controller has to
achieve on the performance of a given model.
time (h)
0 6 12 18 24
Energy (kWh)
-40
-30
-20
-10
0
10
20
30
40
50
60
Optimum e
target e
Figure 1: Daily TOU-MPC cost of electricity vs target of
electricity cost.
time(h)
0 6 12 18 24
Energy(kWh)
0
10
20
30
Optimal Energy Demand
Reference Energy Demand
Figure 2: Daily Prepaid-MPC cost of electricity vs target of
electricity cost.
5 IMPLEMENTATION ANALYSIS
This section consists of analysing the dynamic behav-
ior of MPC design. The simulation of the results are
based on the data developed in (Mbungu et al., 2016;
Holkar and Waghmare, 2010) of the energy manage-
ment system for a commercial load demand. Besides,
this research study deals on the possibility of finding
the dynamic energy system of the increment of the
control variable. The strategy implements the demand
response scheme based the energy management in the
consumers side (Mbungu et al., 2019b), namely real-
time electricity pricing.
5.1 Simulation Analysis
Table 1 provides different biased values that are used
to simulate the dynamic behavior of the given data.
The system implementation computed different time
of use (TOU) and prepaid electricity tariff as de-
scribed in Table 1. The simulation of the results is
presented in Figs. 1 and 2. Figure 1 gives the dy-
namic behavior of the optimal energy that flows on
the system. It is necessary to notice that this model
computes only the increment of the control signal as
described by the canonic form of state space model
in Eqs. 12a and 12b. Besides, the performance in-
dex of the MPC design as described in Eq. 18, with
its developed strategy which includes constraints and
simplified model of the objective function (Eqs. 19a
and 19b) are computed in the fashion of the increment
model. It is also important to notice that Fig. 2 is not
computed by the augmented model.
5.2 Discussion Analysis
Tables 2 and 3 present different values of the energy
cost and the saving energy. When it is about to com-
pare the results of the optimal input signal with target
input as described in Figs. 1 and 2, it is clearly shown
that this result is roughly running close to one another
for the increment signal during TOU computation,
and both signals (target and optimal energy demand)
are close in prepaid mode. However, at some time,
the optimal result does not follow the target energy.
This interpretation can be controversial in the context
of energy-saving and optimal computation. Neverthe-
less, the total cost of energy and the percentage of
the energy cost saving as described in Table 2 gives
another profile of the system performance. Besides,
Table 3 and Fig. 2 present perfect results which are
not often guaranteed during the computation process.
Based on this approach of interpreting the simulations
results, it can be seen that the system dynamics of the
proposed MPC scheme provides satisfactory results
to the consumer side. This is due to the important
value of the optimal energy cost and the significant
percentage rate of the cost-saving.
6 CONCLUSION
The MPC is an optimisation strategy because it of-
fers the opportunity of computing the control vari-
Model Predictive Control: A Survey of Dynamic Energy Management
127
Table 1: Turning parameter and Energy tariffs.
Tariff scheme Weighted coefficient Energy prices (R/kWh)
Off-peak (TOU) 1 0.6150
Standard (TOU) 1 and 5 1.073
Peak(TOU) 1.0526 and 1.143 4.115
Prepaid 2 1.2774
Table 2: TOU Cost of energy and percent of costs analysis.
Type of strategy Cost (Rand) Parameters Saving analysis (%)
Cost to pay 2254.3 Cost-target 48.1818
Target cost 1086.2 MPC cost 38.1196
Optimal cost 895.3235 Cost-saving 61.8804
Table 3: Prepaid Cost of energy and percent of costs analysis.
Type of strategy Cost (Rand) Parameters Saving analysis (%)
Cost to pay 1475.4 Cost-target 52.3810
Target cost 772.8270 MPC cost 52.3810
Optimal cost 772.8270 Cost-saving 47.6190
ables based on a given target. This optimal con-
trol method aimed to minimising the cost of electric-
ity for a given electrical system. It was found that
the model is robust in conjunction with consumers
0
actions. It involves creation of an optimal strategy
where the user can custom the amount of electricity
to use. The optimisation tactics of the MVs for an
MPC is an online algorithm that can compute any lin-
ear model in the real-time environment. The MPC ap-
proach is also considered as a suitable strategy algo-
rithm to be implemented in smart grid technology due
to its robustness in the optimal control solution. The
MPC can also perform a control problem as an opti-
misation problem that is made by an on-line optimi-
sation with receding horizon implementation. Thus,
the real-time optimisation through the quadratic pro-
gramming strategy in the framework of the MPC per-
forms a suitable scheme between data transfer and op-
timisation calculation. It also aims to resolve at each
sampling time the optimal controller within the given
set-point. Besides, the system provides satisfactory
performance in terms of energy-saving and cost opti-
misation. Therefore, future research work can look at
different implementation strategy of the MPC design
through a dynamic energy metering based on sensors
networking within the applications of the smart tech-
nologies.
REFERENCES
Abdi, H., Beigvand, S. D., and La Scala, M. (2017a). A
review of optimal power flow studies applied to smart
grids and microgrids. Renewable and Sustainable En-
ergy Reviews, 71:742–766.
Abdi, H., Beigvand, S. D., and La Scala, M. (2017b). A
review of optimal power flow studies applied to smart
grids and microgrids. Renewable and Sustainable En-
ergy Reviews, 71:742–766.
Adefarati, T., Bansal, R. C., Naidoo, R. M., and Mbungu,
N. T. (Aug 12-15, 2019). Techno-economic evalua-
tion of a grid connected microgrid-cogeneration sys-
tem using wind turbines, microturbine and battery sys-
tem. In International Conference on Applied Energy,
V
¨
aster
˚
as, Sweden.
Bishop, R. H. (2007). Mechatronic systems, sensors, and
actuators: fundamentals and modeling. CRC press,
Texas, USA, 2nd ed edition.
Cannon, M. (2015). C21 Model Predictive Control. Uni-
versity of Oxford, Oxford, Tech. Rep.
Dahleh, M., Dahleh, M. A., and Verghese, G. (2004).
Lectures on dynamic systems and control. A+ A,
4(100):1–100.
Foley, A. M., McIlwaine, N., Morrow, D. J., Hayes, B. P.,
Zehir, M. A., Mehigan, L., Papari, B., Edrington,
C. S., Baran, M., et al. (2020). A critical evaluation
of grid stability and codes, energy storage and smart
loads in power systems with wind generation. Energy,
205:117671.
Holkar, K. and Waghmare, L. (2010). An overview of
model predictive control. International Journal of
Control and Automation, 3(4):47–63.
ICINCO 2021 - 18th International Conference on Informatics in Control, Automation and Robotics
128
Kale, M. and Chipperfield, A. (2005). Stabilized mpc
formulations for robust reconfigurable flight control.
Control Engineering Practice, 13(6):771–788.
Lee, J. H. (2009). A Lecture on Model Predictive Control.
Pan American Advanced Studies Institute Program on
Process Systems Engineering.
Ma, J., Qin, S. J., Li, B., and Salsbury, T. (2011). Economic
model predictive control for building energy systems.
IEEE, Anaheim, USA.
Maciejowski, J. M. (2002). Predictive control: with con-
straints. Pearson education, London, UK.
Mayne, D. Q., Rawlings, J. B., Rao, C. V., and Scokaert,
P. O. (2000). Constrained model predictive control:
Stability and optimality. Automatica, 36(6):789–814.
Mbungu, N. T., Bansal, R. C., and Naidoo, R. (2019a).
Smart energy coordination of autonomous residential
home. IET Smart Grid, 2(3):336–346.
Mbungu, N. T., Bansal, R. C., Naidoo, R., Miranda, V., and
Bipath, M. (2018). An optimal energy management
system for a commercial building with renewable en-
ergy generation under real-time electricity prices. Sus-
tainable Cities and Society, 41:392–404.
Mbungu, N. T., Bansal, R. C., Naidoo, R. M., Bettayeb, M.,
Siti, M. W., and Bipath, M. (2020). A dynamic energy
management system using smart metering. Applied
Energy, 280:115990.
Mbungu, N. T., Naidoo, R., Bansal, R. C., and Bipath,
M. (2017a). Optimisation of grid connected hybrid
photovoltaic–wind–battery system using model pre-
dictive control design. IET Renewable Power Gen-
eration, 11(14):1760–1768.
Mbungu, N. T., Naidoo, R. M., and Bansal, R. C. (2017b).
Real-time electricity pricing: TOU-MPC based en-
ergy management for commercial buildings. Energy
Procedia, 105:3419–3424.
Mbungu, N. T., Naidoo, R. M., Bansal, R. C., and Vahid-
inasab, V. (2019b). Overview of the optimal smart
energy coordination for microgrid applications. IEEE
Access, 7:163063–163084.
Mbungu, T., Naidoo, R., Bansal, R., and Bipath, M. (2016).
Smart SISO-MPC based energy management system
for commercial buildings: Technology trends. In Fu-
ture Technologies Conference (FTC), pages 750–753,
San Francisco, USA.
Mesbah, A. (2016). Stochastic model predictive control: An
overview and perspectives for future research. IEEE
Control Systems Magazine, 36(6):30–44.
Rawlings, J. B. and Mayne, D. Q. (2009). Model predictive
control: Theory and design. Nob Hill Pub, New York,
USA.
Rossiter, J. A. (2003). Model-based predictive control: a
practical approach. CRC press, New York, USA.
Seborg, D. E., Mellichamp, D. A., Edgar, T. F., and
Doyle III, F. J. (2010). Process dynamics and control.
John Wiley & Sons, USA, 3rd ed edition.
Siti, M. W., Tungadio, D. H., Sun, Y., Mbungu, N. T., and
Tiako, R. (2019). Optimal frequency deviations con-
trol in microgrid interconnected systems. IET Renew-
able Power Generation, 13(13):2376–2382.
Tungadio, D. H., Bansal, R. C., Siti, M. W., and Mbungu,
N. T. (2018). Predictive active power control of
two interconnected microgrids. Technology and
Economics of Smart Grids and Sustainable Energy,
3(3):1–15.
Wang, L. (2009). Model predictive control system design
and implementation using MATLAB
R
. Springer Sci-
ence & Business Media, Girona, Spain.
Xie, L., Li, P., and Wozny, G. (2007). Chance constrained
nonlinear model predictive control. Assessment and
Future Directions of Nonlinear Model Predictive Con-
trol, pages 295–304.
Zhang, Y., Liu, B., Zhang, T., and Guo, B. (2014). An intel-
ligent control strategy of battery energy storage sys-
tem for microgrid energy management under forecast
uncertainties. International Journal of Electrochemi-
cal Science, 9(8):4190–4204.
Model Predictive Control: A Survey of Dynamic Energy Management
129