Energy & Buildings 256 (2022) 111708

Contents lists available at ScienceDirect

Energy & Buildings

journal homepage: www.elsevier.com/locate/enb

Cooling seasonal performance of inverter air conditioner using model

prediction control for demand response
Cuiling Wang, Baolong Wang ⇑, Mengdi Cui, Falin Wei
Beijing Key Laboratory of Indoor Air Quality Evaluation and Control, Department of Building Science, Tsinghua University, Beijing 100084, China

a r t i c l e i n f o a b s t r a c t

Article history: The supply-demand imbalance of electricity increases the operating burden on smart grids, decreases the
Received 4 August 2021 average efficiency of power generation equipment, and threatens the safe operation of power grids.
Revised 7 November 2021 Residential air conditioning is a flexible load and a major consumer of electricity. Therefore, demand
Accepted 19 November 2021
response control can be applied to air conditioners (ACs) to shift their peak energy consumption and save
Available online 25 November 2021
energy. Model predictive control (MPC) is an effective demand response control method. In this study, we
analyze the cooling seasonal performance of an inverter AC with MPC. A time-varying MPC was designed
Keywords:
and evaluated using a simulation testbed that was constructed using MATLAB. Subsequently, the energy,
Demand response
Time-varying model
cost, and temperature control performances of the MPC were analyzed in detail from electricity pricing
Predictive control model, weather conditions and fluctuation of real-time price. The results show that compared to the
Inverter air conditioners proportional–integral–derivative (PID) control method, MPC can shift the peak-hour energy consumption
Cooling seasonal performance by 6.34%–21.60% and reduce the total electricity costs by 13.44%–27.43%, while maintaining indoor ther-
mal comfort during the whole cooling season, and Demand response with MPC control is very suited to
hot weather conditions with highly fluctuating RTP. By applying MPC hybrid demand response under
real-time price, there are better performances on peak shifting and cost saving.
Ó 2021 Elsevier B.V. All rights reserved.

1. Introduction transmission and distribution capabilities. Energy storage systems,

including electricity storage, cooling storage, and heating storage,
Traditional electricity generation plants have poor rapid regula- should be properly designed and employed to reduce the discrep-
tion capabilities, which limits the transmission capacity of the grid ancy between the supply and demand of electricity [3,4]. More-
and leads to electrical imbalances owing to the fluctuating power over, the flexibility of electricity demand should be increased by
demands of end-users [1]. The share of renewable energy, such shifting demand from peak periods to valley periods [5].
as wind power and photovoltaic energy, is expected to increase Buildings are responsible for a significant proportion of global
significantly in the future owing to the push towards carbon neu- electricity consumption, with air conditioning systems being a
trality. Consequently, electricity supply will fluctuate and become major contributor to the total building energy demand [6]. On
more random, which could decrease the average efficiency of hot days or in hot regions, the electricity required for air-
power generation, decrease the utilization of renewable electricity, conditioning further increases the peak power demand. For exam-
and threaten the safe operation of the power grid [2,3]. To over- ple, on extremely hot days, space cooling can use more than 25% of
come these problems, all relevant aspects, including power gener- the peak electricity demand in the United States. In Singapore, air
ation, grids, storage, and use, should be considered synthetically. In conditioning can account for as much as 50% of the peak electricity
terms of electricity generation, the capacity modulation range and load, which is quite similar to that in Beijing [7]. Fortunately, air
speed of traditional electricity generation plants should be conditioning is a flexible load. By exploiting room thermal inertia,
increased, and the quality and modulation of renewable electricity the electricity demand for cooling can be reduced significantly,
generation should be improved. Considering power grids, smart without affecting temperature control [8].
grid technologies should be implemented to improve dynamic The energy consumption of residential buildings accounts for a
big share in the total buildings’ global final energy use [9], and the
⇑ Corresponding author. residential buildings have the highest demand response (DR)
E-mail addresses: [email protected] (C. Wang), wangbl@tsin- potential [10]. The IEA predicts that 5.6 billion ACs will be used
ghua.edu.cn (B. Wang), [email protected] (M. Cui), [email protected] in buildings by 2050, up from 1.6 billion today [7]. Therefore,
(F. Wei).

https://doi.org/10.1016/j.enbuild.2021.111708
0378-7788/Ó 2021 Elsevier B.V. All rights reserved.
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

Nomenclature

A geometric area (m2) max maximum

C capacitance (J/K) min minimum
b ost
C normalized electricity cost o outdoor air
Cost cost of electricity (RMB) set setpoint
d disturbance vector solar heat gain from solar radiation
E total energy consumption (J) t current time
f conversion coefficient of heat gain up upper limit value
Isolar solar radiation (W/m2) wall wall of room
Ncomp compressor speed (Hz) w.int internal wall layer
N prediction horizon wind window
P power of inverter AC (W) w.out external wall layer
Q heat gain (W)
R heat resistance (K/W) Abbreviations
t time (s) AC air conditioner
T temperature (°C) DR demand response
Tz solar–air temperature (°C) EER energy efficiency ratio
Tb normalized indoor air temperature ERCOT Electric Reliability Council of Texas
u manipulated variable vector HVAC heating, ventilation, and air conditioning
U constraint sets for manipulated variable MAPE mean absolute percentage error
w penalty weighting MhD model-predictive-control-hybrid demand response
x state vector MoD model-predictive-control-optimization demand
X constraint sets for state response
aout convective heat transfer coefficient of exterior sur- MPC model predictive control
faceW/(m2∙°C) PID proportional–integral–derivative
e slack variable PVFP peak-valley-flat electricity price
PVP peak-valley electricity price
Subscripts and Superscripts RC resistance–capacitance
a absorption factor of solar radiation RTP real-time price
comp compressor SSM state–space model
in indoor air STD standard deviation
inter heat gain from inter disturbance TOU time-of-use
k step k
low lower limit value
m internal mass

studying the DR control of ACs is extremely important for future optimization-based control, wherein optimal control methods are
electricity distribution systems. used to minimize electricity costs over a given time span under
Several studies have focused on the DR management of residen- variable prices. Thomas et al. [12] designed an intelligent residen-
tial ACs. Based on the type of response signal, DR can be classified tial AC controller formulated as a stochastic dynamic program to
into two categories: incentive-based DR and price-based DR. In respond to both energy prices and randomly varying environmen-
incentive-based DR, consumers are directly incentivized by the tal conditions, to realize the trade-offs between cost and optimal
electric power company to change their electricity consumption comfort by scheduling the switching on or off of an AC. Hu et al.
limits, according to a pre-agreed protocol. In price-based DR, con- [13] investigated a model-based genetic optimization control
sumers voluntarily change their consumption patterns in accor- method that is an open-loop control method for inverter ACs to
dance with dynamic electricity prices. Incentive-based DR is realize electricity cost savings and peak power reductions in a
primarily used for large electricity consumers, such as industries, dynamic electricity price environment. Comparison of the two
as the agreements need to be signed in advance. Price-based DR types of control methods [14,15] has revealed that the
is more suitable for large-scale small-capacity users, such as resi- optimization-based approach can significantly decrease the total
dential users with ACs. Currently, the electricity pricing models energy consumption and cost, whereas the rule-based approach
used in price-based DR primarily include time-of-use (TOU) pricing provides a large reduction in the peak electricity demand. Conse-
model and real-time price (RTP) model. Compared to TOU price, quently, a combination of optimization-based control and rule-
wherein electricity prices can jump between two or three pro- based control, which can be called hybrid control, may further
jected prices over an entire day with an interval of several hours, improve the performance of DR.
RTP modulates electricity prices in minutes to respond to fluctua- The aforementioned cases of price-based DR are primarily real-
tions in the electricity market. Consequently, RTP has significant ized through real-time feedback control or model-based open-loop
potential in real-time DR. optimal control. It is difficult to achieve maximum cost saving over
In general, two types of control methods are used to realize a given time span with real-time feedback control as it only uses
price-based DR. The first is rule-based control, the operation of the currently available information and does not include future
the electricity is directly controlled based on the real-time price information. Owing to the errors of thermal load prediction and
of electricity. Yoon et al. [11] developed a price-responsive con- the system models, the operation of open-loop optimal control also
troller for residential heating, ventilation, and air conditioning deviates from the optimal condition. Model predictive control
(HVAC) systems, wherein temperature set-points are assigned if (MPC) can shift peak electricity in response to dynamic electricity
the retail price exceeds a preset price. The second is price using an optimization algorithm, realize multiple distur-

2
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

bances, multiple objections, and big-time lag system optimal con-

trol, and decrease prediction errors by using a model-based feed-
back method. Therefore, MPC can be implemented in residential
ACs for better DR control [16].
Several studies have reported significant improvements in
energy saving, cost saving, and peak shifting by adopting
optimization-based DR using MPC, which can be shortened to
MPC-optimization DR (MoD), or hybrid DR using MPC, which can
be shortened to MPC-hybrid DR (MhD), in residential ACs. In
MhD, the indoor temperature setting or compressor frequency is
directly adjusted in response to the dynamic price to further
strengthen the peak-cutting performance. Mesut et al. [17] pro-
posed a practical cost and energy-efficient MPC strategy for
single-speed ACs under RTP and developed an algorithm that
assigns temperature set-points in response to RTP. The results Fig. 1. The basic structure of an MPC controller [20]
demonstrated that compared to on/off control, MPC achieved
An MPC framework can be defined by the following finite-
reductions of 44.2% and 43.3% in the total energy consumption
horizon optimization problem [21]:
and energy costs, respectively, during peak hours. Compared to
fixed-speed ACs, inverter ACs have higher efficiency under part- X
N

load conditions and a higher thermal sensation, and are widely J ðuk Þ ¼ min lk ðxk ; uk Þ Cost function ð1Þ
u0 ;::::uN
k¼0
used in household ACs [18]. Therefore, Hu et al. [19] designed an
MhD to optimally control an inverter AC under a dynamic RTP Subject to:
and limited the operating frequency of the inverter AC in response
xkþ1 ¼ f ðxk ; uk ; dk Þ Prediction model ð2Þ
to the electricity price. The simulation results showed that com-
pared to a PID controller, the MhD control reduced the average
ðxk ; uk Þ 2 X k  U k ð3Þ
power consumption during peak demand periods by 17.31%–
38.86%. where N is the prediction horizon, Xk and Uk are the constraint
However, it should be noted that compulsively constricting the sets for states xk and manipulated variables uk, respectively, at the
working frequency of the compressor can threaten the safety of the time step k, and dk is the disturbance at time step k.
compressor and dramatically increase the indoor air temperature The structure of the optimal control system for an MPC con-
beyond the comfortable range. Moreover, most existing studies troller of an inverter AC is shown in Fig. 2. There are two aspects
using MPC in ACs were conducted over a few days, during which that must be considered while designing the MPC controller: (1)
disturbances such as the outdoor air temperature, solar radiation, a room model that can predict the indoor air temperature or load
and dynamic price remained relatively stable. Consequently, the and an inverter AC model that can predict the required cooling
cooling seasonal performance of ACs with MPC is still unclear. capacity; and (2) an optimizer that considers the cost function
To bridge these research gaps, the cooling seasonal performance and constraints to achieve energy saving, cost saving, and temper-
of an inverter air conditioner with MPC for DR is investigated ature control. Furthermore, before applying time-manipulated
herein. The effects of disturbances, such as the outdoor air temper- variables to the inverter AC, a signal converter is required to con-
ature, solar radiation, dynamic price fluctuations, and the electric- vert the manipulated variable signals to directly implementable
ity pricing models are investigated in detail. Based on this signals.
investigation, the long-term suitability of MPC for inverter ACs is
comprehensively evaluated. 2.2. Prediction models
The rest of this paper is organized as follows. Section 2 presents
the proposed model predictive controller, including the time- MPC requires appropriate models of the room and inverter AC.
varying prediction model, cost function, constraints, and control The mathematical models of the room and the AC must be able to
signal converter. The details of the MPC process—including discus- accurately predict the room thermodynamics and provide compu-
sions on the temperature set-point assignment in the tational efficiency for real-time control and optimization.
optimization-based and hybrid conditions, disturbances, and eval-
uation index, are presented in Section 3. Section 4 compares the 2.2.1. Room model
performance of inverter ACs with the proposed MPC. Various building energy simulation tools such as EnergyPlus
and TRNSYS can be used to construct an accurate physical model
of a room, but these models are too complicated for real-time con-
trol and optimization [22]. Therefore, in this study, we construct a
2. Model predictive control of inverter ACs data-driven gray resistance-capacitance (RC) network model based
on the state-space formulation given by Eqs. (4)–(7) [13].
2.1. Principle of model predictive control dT w:out T o  T w:out T w:int  T w:out
C wall ¼ þ þ f solar;wall Awall Isolar ð4Þ
dt Rw:out Rwall
MPC is an optimal control method. A digital model is used to
predict the future outputs of the system based on the proposed dT w:int T w:out  T w:int T in  T w:int
optimal future control actions called manipulated variables, past C wall ¼ þ ð5Þ
dt Rwall Rw:in
inputs, and past outputs. These optimal actions are calculated by
the optimizer, considering the cost function and constraints. How- dT in T m  T in T w:in  T in T o  T in
ever, only the first optimal action is applied to the real system. The C in ¼ þ þ þ f inter:in Q inter
dt Rin:m Rw:in Rwind
entire process is repeated during the next sampling time. Subse-
quently, the measured states are fed back to compensate for pre- þ Q HVAC ð6Þ
diction errors, as shown in Fig. 1.
3
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

Fig. 2. MPC scheme for a room with an inverter AC.

dT m T in  T m of Tw.out, Tw.int, Tin, and Tm. The SSM is a first-order time-varying

Cm ¼ þ f solar:m Awind Isolar þ f inter:m Q inter ð7Þ
dt Rin:m equation wherein matrix B varies with the price of electricity
(Eqs. (11) and (12)).
where R and C are the heat resistance and capacitance of the
room, respectively; T is the temperature; the subscripts wall, w. T kþ1 ¼ AT k þ Buk þ Edk ð11Þ
out, w.int, o, in, m, and wind denote the wall, external wall layer,
internal wall layer, outdoor air, indoor air, internal mass, and win- T k ¼ CT k þ Ddk ð12Þ
dow, respectively; Qinter and QHVAC are the internal heat gain and T
where the system states T = [Tw.out Tw.int Tin Tm] , the distur-
heat removed by the inverter ACs, respectively; Isolar is the solar
bance inputs d = [To Isolar Qinter 1]T, and the manipulated variable
radiation; A is the geometric area of the wall; and f is the conver-
u = Cost.
sion coefficient of the heat gain.
The room considered herein is a single residential room in a hot
climate, located in Guangzhou in China. The room (L = 4.8 m,
2.2.2. Inverter AC model W = 3.7 m, H = 3 m) has one east-facing exterior wall (4.8 m  3
There are two types of inverter AC models: the transient model m) and one south-facing exterior wall (3.6 m  3 m). The overall
and the steady-state model. As the thermal dynamic variation of heat transfer coefficients of the single glazed window and the exte-
the refrigerant is much faster than that of the room, the steady- rior wall are 5.69 and 2.57 W/(m2K) and the window-wall-ratio of
state model is sufficient for the MPC designed herein [23]. The each wall is 0.2. The parameters in Eqs(4)-(7) can be identified
steady-state performances of the inverter AC (cooling capacity, based on actual test data or simulated building energy data. In this
energy efficiency ratio (EER), and power) are determined by the paper, the identification results from [19], which is gotten through
indoor air temperature (Tin), outdoor air temperature (To), and dif- a TRNSYS simulation, are used. The parameters are as follows:
ferent compressor rotational speeds (Ncomp) (Eqs. (8)–(10)). A 3- Rwall = 0.006 K/W, Rwind = 0.0051 K/W, Rw.in = 0.0041 K/W, Rin.-
dimensional linear model that can predict performance under
m = 0.0023 K/W, Rw.out = 0.001 K/W, fsolar.m = 0.0012, finter.in = 3.0367,
time-varying indoor air and outdoor air temperatures was devel- fsolar.w = 0.4834, finter.m = 0.9976, Cwall = 8381 kJ/K, Cm = 13904 kJ/K,
oped herein. and Cin = 871 kJ/K.
  A total of 80 sets of experimental data of a 2.5 kW inverter AC
EER ¼ f T in ; T o ; Ncomp ð8Þ
from Gayeski [24] were used to identify and validate the inverter
  AC model, of which 60% were used for identification and the rest
Q ¼ f T in ; T o ; Ncomp ð9Þ were used for validation. The linear model is given by Eq. (13).
Fig. 3 compares the identified and validated data. The deviations
Q ¼ f ðT in ; T o ; PÞ ð10Þ were primarily in the range of ± 15%. The mean absolute percent-
where T is the temperature, Ncomp is the compressor speed, and age error (MAPE) was 7.1%.
EER, Q, and P are the energy efficiency ratio, cooling capacity, and Q HVAC ¼ 54:048T in - 41:356T o þ 1:644P þ 1170:28 ð13Þ
power, respectively; the subscripts in, o, and comp denote the
indoor air, outdoor air, and compressor, respectively. The initial states are [28.09, 27.52, 26.48, 27.27], and the matri-
ces A, B, E, C, and D in the discrete-time SSM, which are derived
from Eqs. (4)–(10), are as follows:
2.2.3. System model 2 3
The combination of the room and the inverter AC is the target 959:12 5:8003 0:0230 0:0012
system. The system model is an extremely important part of the 6 5:8003 985:76 7:4167 0:5848 7
6 7
MPC controller that predicts the comfort of a room considering A ¼ 103 6 7
4 0:2214 71:298 727:03 127:77 5
several future manipulated variables on the prediction horizon. A
0:0004 0:1763 4:0058 995:64
multiple-input single-output state–space model (SSM) was con-
structed herein, with disturbance inputs of To, Isolar, and Qinter, elec-
tricity cost as the manipulated variable, an output of Tin, and states
B ¼ 103 ½ 0:0537 26:590 5813:8=EP 14:335 

4
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

the indoor air temperature. The prediction horizon and sampling

time have a significant influence on the controlled results. The pre-
diction horizon is determined by thermal inertia of the room, the
variation frequency of RTP, computation time cost and so on. For
rooms with large thermal inertia, the prediction horizon should
be long enough, approximately 5–48 h [26–28]. When MPC is
applied to small thermal inertia rooms, the prediction horizon
should be short, as 0.5–4 h [19,22,29]. The investigated room is
built with light thermal mass and has small thermal inertia. Conse-
quently, the prediction horizon herein is selected as 3 h. To
respond to the dynamics of the room and reduce the mechanical
wear of the compressor, the sampling time was set to 5 min
(N = 36).

2.4. Control signal converter

The cooling capacity of an inverter AC can be modulated by the

compressor frequency, indoor fan speed, outdoor fan speed, and
Fig. 3. Comparison of the simulated data and experimental data of the inverter AC.
superheating degree. However, the compressor frequency is the
primary impact factor and is always used for capacity control. In
this study, the optimization result follows the trajectory of the
2 3
350:56 3:4163 0:0001 0:0319 electricity cost (Costðt þ 1jtÞ Cost ðt þ 2jtÞ:::Costðt þ NjtÞ). There-
6 4:2467 0:0102 0:0409 15:767 7 fore, a control signal converter must be built to convert the cost
6 7
E ¼ 104 6 7 into frequency. First, the power (P) of the inverter AC can be
4 699:48 0:0003 8:9530 3447:5 5
obtained by dividing the electricity cost by the real-time electricity
1:7246 0:0007 0:1294 8:5004
price and sampling time Dt. Subsequently, a 3-dimensional nonlin-
ear lumped parameter model is used to convert the power (P) to
C ¼ ½ 0 0 1 0 D ¼ ½ 0 0 0 0  the compressor frequency (Ncomp). A model with 20 coefficients
where EP denotes the electricity price. was used herein (Eq. (18)) and the resulting MAPE was 1.4%.

Ncomp ¼ f ðT in ; T out ; P Þ ð18Þ

2.3. Optimization
where the frequency range of the inverter AC is 20–100 Hz.
When Ncomp is less than 20 Hz, the inverter AC is switched off.
The proposed MPC optimizes the cooling supply of inverter ACs
to realize cost savings and reduce the deviation of the inside tem-
perature from the desired value in accordance with the cost func- 3. Implementation of MPC
tion (Eq. (14)) in the prediction horizon, N [25]. The constraint is
given by Eq. (15). The cost function is a quadratic cost that is solved A simulation testbed was constructed to test the performance of
by quadratic programming. the proposed MPC considering different operating conditions,
" # weather conditions, and electricity prices, in comparison with
XN X
N X
N
J ¼ min ^ ðt þ kjtÞ þ
wC Cost wT T^ ðt þ kjtÞ  T set ðt þ kjtÞÞ þ
2
we e2 ðt þ kjtÞ PID and MhD. Computer-based dynamic simulation is effective
k¼0 k¼0 k¼0
and cost-efficient under various conditions. MATLAB/Simulink
ð14Þ
was used as the simulation environment herein owing to its visu-
alization, flexibility, and calculation efficiency. The room energy
T low 6 T 6 T up ð15Þ
system testbed, as illustrated in Fig. 4, consisted of a controller, vir-
where Cb ost , T
b , w, and e are the normalized electricity cost, nor- tual testbed including the thermal room and inverter AC, and tem-
malized indoor air temperature, penalty weighting factor, and perature setting module to assign the indoor air temperature at
slack variable, respectively; N refers to the number of prediction different times and electricity prices, in which the model of the
horizons; the subscripts t and k denote the current time and time thermal room is continuous state-space model that is different
step, respectively; and the subscripts low and up denote the lower from that in prediction model of MPC controller and the model
and upper indoor air temperature limits, respectively. On the right- of inverter AC is a polynomial model that is more complex and
hand side of Eq. (14), the first term is the electricity cost, the sec- accurate than that of MPC controller.
ond term is the temperature deviation cost, and the last term is
the constraint violation cost. Different weighting factors for each 3.1. Preparation of input variables
term yield different results.
As the weighting factor ratio (wT : wC) is high, the cost function 3.1.1. Weather conditions
is a thermal comfort-dominated setting. The normalized electricity During the simulation process, the weather data must be
cost and indoor air temperature are defined by Eqs. (16) and (17). obtained in advance from a local observatory or a prediction
model. In this study, the weather forecast was obtained from the
^
Cost ¼ Cost=Costmax ð16Þ website of the local observatory. The weather conditions during
the cooling season in Guangzhou were chosen to study the MPC
_   performance, assuming the weather data was accurately predicted.
T ¼ ðT  T set Þ= T up - T low ð17Þ
Fig. 5 shows the daily average weather between June 1 and
where Costmax is the maximum electricity cost, which is equal to September 1. To analyze the relationship between the weather
the sum of the rated power of the inverter AC and the maximum and the performance of the controller, the solar–air temperature
electricity price and sampling time, and T set is the set-point of was calculated to represent the combined influence of the weather
5
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

3.1.3. Electricity price

Compared to a TOU price, a high-resolution RTP is more benefi-
cial to power systems in terms of flattening the load profile and
reducing peak demand [31]. Therefore, RTP was selected as the
main DR object herein [32]. However, currently, RTP is not
employed in the power market in China, and a TOU price is used
in most cities to shift the peak-hour energy, which only reflects
the long-term or mid-term power demand–supply balance. There-
fore, a TOU price was also used herein to investigate the DR perfor-
mance and evaluate the influence of different electricity pricing
models on the performance of the proposed MPC.
The RTP determines the matrix B in the SSM. The time-varying
coefficient matrix is assigned a prediction horizon in advance. Con-
sequently, the RTP must be predicted or obtained from an electric-
ity retailer in advance. Although RTP has not yet been employed in
China, it is an important trend to make the price reflect the real-
time balance of supply and demand as much as possible [33]. In
our simulation, typical RTP at 15-min intervals was obtained from
a mature real-time electricity price market—the Electric Reliability
Council of Texas (ERCOT) for June, July, and August [34]. To
increase the computation speed, the electricity price was con-
verted from USD to RMB, thereby ensuring that a similar nominal
value could be used for both RTP and TOU prices in China. Fig. 6
Fig. 4. Testbed for room air conditioner system with MPC.
shows the typical daily RTP profiles during the cooling season.
Two types of TOU prices that are used in China, the residential
on the thermal environment (Eq. (19)) [30] The average solar–air peak-valley electricity price (PVP) and the residential peak-
temperatures in June, July, and August were 31.93 °C, 34.64 °C, valley-flat electricity price (PVFP), were employed herein (Fig. 7).
and 33.80 °C, respectively.
3.2. Set-point of indoor air temperature

aIsolar 3.2.1. MPC-optimization DR

T z ¼ T in þ ð19Þ
aout In optimization-based DR, MPC is directly applied to inverter
where Tz is the solar–air temperature, a is the absorption factor ACs by optimizing the cost. The indoor temperature set-points
of solar radiation and is equal to 0.65, and is the convective heat are determined based on only the comfort preference of the occu-
transfer coefficient of the exterior surface of the envelope and is pants and the occupancy fluctuation pattern. When the occupants
equal to 18 W/(m2∙°C) [30] are active, i.e., between 8:00 and 24:00, the indoor air temperature
set-point was 24 °C, and the fluctuation range was 22–26 °C. When
the occupants are at rest, i.e., between 24:00 and 8:00, the indoor
air temperature set-point was 26 °C, and the fluctuation range was
3.1.2. Internal thermal load 24–28 °C.
The internal thermal load in residential rooms is primarily gen-
erated by equipment such as televisions, computers, lighting, etc. 3.2.2. MPC-hybrid DR
and the occupants themselves, and is determined by occupant To further realize peak shifting and cost saving, the indoor air
behavior. Occupant behavior has inherent uncertainty owing to temperature set-points of the MPC were varied in accordance with
its stochastic nature and modeling errors. This inherent uncer- the daily fluctuation of RTP in one day, which is hybrid DR method.
tainty of the internal thermal load affects the performance of the Lower temperature set-points were assigned to lower prices, and
MPC. To better exploit the strengths of MPC in practical applica- higher temperature set-points were assigned to higher prices.
tions, a further study of human behavior is required to build a The algorithm first determines the minimum and maximum
more accurate model. For simplicity, the internal load when occu- electricity prices over the next 24-h period and divides the price
pants are in the room was assumed to be 100 W herein. range averagely (Eq. (20)). Simultaneously, the same number of

Fig. 5. Daily average weather conditions during the cooling season in Guangzhou.

6
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

Fig. 6. Daily RTP profiles on typical days during the cooling season.

Fig. 7. Two types of diurnal TOU pricing models.

temperature set points were assigned to the comfortable tempera- price). The energy saving and electricity cost savings during peak
ture range. Finally, each price range was coupled to a temperature hours are direct indices of the DR. The percentage of energy savings
set-point at an interval of 15 min. The assigned temperature set- and cost savings were calculated using Eqs. (21) and (22).
points are listed in Table.1.  
EBasic  EStudy
RTPmax  RTPmin Energy savingðDtÞ ¼  100% ð21Þ
RTPi ¼ RTP min þ i ð20Þ EBasic
4
 
where RTPmin and RTPmax are the minimum and maximum elec- Cost Basic  CostStudy
Cost savingðDtÞ ¼  100% ð22Þ
tricity prices, respectively, and i is an integer between 0 and 5. Cost Basic
where E and Cost are the total energy consumption and electric-
3.3. Evaluation index of different control methods ity cost for a given duration; 4t represents the duration of energy
consumption and electricity cost, such as the entire cooling season,
The energy saving and electricity cost savings during the entire an entire day, or the peak hours; and the subscripts Basic and
cooling season or an entire day were calculated to evaluate the Study denote the basic and study control cases, respectively.
integrated performance of the different control methods consider-
ing different disturbances (such as the weather and electricity
4. Results and discussions

Table 1 4.1. Performance differences between MPC and PID

Indoor air temperature set-points in response to RTP.

RTP range Indoor air temperature set-points (°C) The system performances of the PID and MPC controllers with a
(RMB/kWh)
Occupant active (8:00– Occupant at rest (24:00–
weight factor ratio wT : wC = 1 : 1 were tested and compared. To
24:00) 8:00) analyze the different control methods in detail, three typical days
RTP0  RTP  RTP1 24 26
between August 3 and August 5 were selected herein. Fig. 8 shows
RTP1 < RTP  RTP2 24.5 26.5 the weather conditions and RTP during the three days. The weather
RTP2 < RTP  RTP3 25 27 was hot on days two and three, with solar–air temperatures of
RTP3 < RTP  RTP4 25.5 27.5 36.1 °C and 34.9 °C, respectively. The electricity prices on day three
RTP4 < RTP  RTP5 26 28
fluctuated more than those on the previous two days.
7
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

Fig. 8. (a) Weather conditions and indoor-heat gain, and (b) RTP on three typical days between August 3 and August 5.

Fig. 9. Temporal system control results with PID and MPC over three typical days: (a) indoor air temperature, (b) power, and (c) electricity costs.

8
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

Fig. 10. System performance with PID and MPC: (a) all-day energy saving and cost saving on three typical days, (b) peak-hour energy saving and cost saving on three typical
days, and (c) energy saving and cost saving during cooling season and peak hours of cooling season.

Fig. 11. Statistical distributions of indoor air temperature between 8:00 and 24:00 over different time periods: (a) three typical days and (b) entire cooling season.

9
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

Fig. 12. (a) Weather condition and indoor-heat gain, and (b) RTP on three typical days between June 13 and June 15.

Fig. 9 shows the system time-series control results of the two have good control performances and the temperature fluctuation
control systems. Fig. 9(a) demonstrates that both control methods ranges are mostly within 1℃, which is a proven thermal comfort
could satisfactorily maintain the indoor air temperature within the range for human beings. As shown in Fig. 11(a), the mean value
temperature limits. Compared to the PID control method, MPC and fluctuation range of the indoor air temperature were a little
makes indoor air temperature cooler than the temperature set- large on days two and three. And on day three, the fluctuation
points at 8:00, when the weather was cool and RTP was low, to range of the indoor air temperature was larger as the RTP was very
realize cooling storage through thermal inertia; during the peak high during the peak hours. As shown in Fig. 11(b), the mean
hours between 15:00 and 17:00, when the RTP was high and the indoor temperature with the MPC and PID control systems was
cooling load was large, the indoor air temperature with MPC was 24.2 °C and 24.0 °C, respectively, during the entire cooling season.
higher than the temperature set-points to save costs. This over- The Temperature control performances of PID and MPC are quite
cooling strategy increases energy consumption during off-peak close.
periods and overheating strategy results in less electricity cost To conclude, the reasons of MPC to realize energy shifting, cost-
energy consumption and electricity cost during peak hours, as saving, and thermal comfort keeping are mainly to utilize room
shown in Fig. 9(b) and 9(c). Especially on days two and three, this thermal inertia and the high performance of inverter air condi-
resulted in an obvious peak shift. tioner under different outdoor to respond to RTP. Firstly, by utiliz-
As shown in Fig. 10, compared to the PID control method, the ing room thermal inertia to realize cold storage ahead when RTP
reduction in the all-day energy consumption of the inverter AC and outdoor air temperature are low. when RTP and outdoor air
with MPC was  3.58% on day one, 19.37% on day two, and temperature are high, the room thermal mass could release cold
12.78% on day three, whereas the reduction in the peak-hour to keep thermal comfort and reduce energy consumption. Mean-
energy consumption was  1.77% on day one, 41.45% on day while, an increase in the indoor temperature may decrease the
two, and 19.11% on day three. The energy saving and peak shifting cooling load and enhance the energy efficiency of the AC to affect
were influenced by the RTP profiles and the weather conditions. As the energy-saving rate and cost-saving rate.
the MPC method can shift energy consumption from peak hours to
off-peak hours, it provided better electricity cost savings over an 4.2. Performance differences between MPC-optimization DR and MPC-
entire day and during peak hours even though on day one, the hybrid DR
all-day electricity cost saving was 16.36% and the peak-hour elec-
tricity cost saving was 30.29%. The cost savings are also influenced In this section, the indoor air temperature set-points are fixed in
by the weather and fluctuation of RTP. Over the entire cooling sea- the MoD and varied in the MhD. Three consecutive typical days
son, although more energy was consumed on some days, the MPC with different weather patterns and RTP curves were used to study
method realized an energy saving of 13.64%, a peak-hour energy the performances of the MoD and MhD, as shown in Fig. 12.
saving of 21.60%, an electricity cost saving of 27.43%, and a peak- Fig. 13 shows the temporal system control results of the MoD
hour cost saving of 43.11%. and MhD. As shown in Fig. 13(a), the indoor air temperature set-
The statistical distributions of the measured indoor air temper- points of the MhD were higher than those of the MoD during peak
ature between 8:00 and 24:00 over different periods are shown in hours. Owing to the good temperature control capability of the
Fig. 11. It can be found that both PID and MPC control methods MPC method, the indoor air temperature was maintained within
10
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

Fig. 13. Temporal system control results of MoD and MhD over three typical days: (a) indoor air temperature, (b) power, and (c) electricity costs.

the temperature limits even when the temperature set-points were 4.3. Effects of electricity pricing model on MPC performance
high. The higher set-points reduced the peak-hour cooling load,
resulting in an obvious reduction in the energy consumption and The performance of the proposed MPC was compared with that
electricity cost during peak hours, as shown in Fig. 13(b) and 13 of the PID control method under different pricing models—RTP and
(c). Fig. 14 compares the system performance of the MoD and TOU price including PVFP and PVP—as shown in Fig. 15(a). Com-
MhD. Compared to the MoD, the reduction in peak-hour energy pared to the PID, the total energy saving was  2.65%, the cost sav-
with MhD was 3.92% on day one, 72.15% on day two, and 40.83% ing was 0.25%, the peak-hour energy saving was 3.93%, and the
on day three. As the MhD shifts energy consumption from the peak-hour cost saving was 4.63% over the entire cooling season
high-RTP period to the low-RTP period, the all-day cost reduction with MPCPVFP. With MPCPVP, the total energy saving was  4.78%,
was 46.62% on day two and 28.37% on day three. On day one, as the cost saving was  4.86%, the peak-hour energy saving was
the weather was cooler, the fluctuation in RTP was significant, 0.83%, and the peak-hour cost saving was 0.83% over the entire
making indoor air temperature set-points of the MoD and MhD cooling season. Under the TOU pricing model, the proposed MPC
were similar for most of the day, except during the peak hours. could realize peak shifting, but its ability to shift the peak was
However, owing to the sharp dynamic set-points, the inverter AC weaker than that of the MPC under the RTP model. Furthermore,
was less stable, negatively impacting its performance, and increas- under the TOU pricing model, the MPC saved lesser energy and
ing its all-day energy consumption and electricity cost. During the costs compared to the PID. Compared to MPCPVP, MPCPVFP has bet-
entire cooling season, although the energy costs and consumption ter performance, as PVFP can better reflect the power response and
were high on some days, the MhD realized an energy saving of demand. For example, the highest price of electricity between
14.18%, a peak-hour energy saving of 33.83%, an electricity cost 14:00 and 17:00 was almost the same as the duration of the peak
saving of 17.48%, and a peak-hour cost saving of 33.62%. Therefore, load of the room during the day. Figure 15(b) illustrates the statis-
higher energy and electricity cost savings and more effective peak tical distributions of the indoor air temperature between 8:00 and
shifting were achieved by assigning temperature set-points. 24:00 over the entire cooling season. Compared to MPCRTP, the

11
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

Fig. 14. System performance with MoD and MhD: (a) all-day energy saving and cost saving on three typical days, (b) peak-hour energy saving and cost saving on three typical
days, and (c) energy saving and cost saving during the cooling season and the peak hours of the cooling season.

Fig. 15. The control performances:（a）comparison of system performance of MPC and PID under the RTP, PVFP, and PVP models. (b) statistical distributions of indoor air
temperature between 8:00 and 24:00 under the RTP, PVFP, and PVP models

12
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

Fig. 16. Daily mean values and STDs of solar–air temperature and RTP during the entire cooling season

Fig. 17. Daily system performance of MPC (1:1) compared to the PID control method: (a) energy saving and peak-hour energy saving, and (b) electricity cost saving and peak-
hour electricity saving

indoor air temperature control deviation was lesser under the TOU 82%, the daily total cost saving was  27.03%–80.79%, and the daily
pricing model with MPC. As the curve of the electricity price is flat peak-hour total cost saving was  22.46%–99.36%. Owing to the
and the difference between the peak-hour cost and valley-hour fluctuations in the weather conditions and RTP, the performance
cost is small, the optimal control sequences of the calculated fre- of the MPC varied significantly during the cooling season in
quency have a negligible effect on the energy and cost savings Guangzhou. The energy and electricity consumptions with the
and peak shifting, and a strong effect on temperature control. MPC were the highest on June 11, and were 34.51% and 27.03%
higher, respectively, than those with the PID control method. The
daily average solar–air temperature on June 11 was 27.9 °C, result-
4.4. Effects of weather and RTP on MPC performance ing in a small cooling load that caused the inverter AC to start and
stop frequently, which led to large temperature fluctuations. As the
Previous studies have shown that weather conditions and RTP control objective of the MPC is to maintain a balance between cost
fluctuations affect the performance of an MPC. In this section, the savings and temperature control, more electricity is consumed to
relationship between the MPC performance and weather condi- maintain the temperature. In addition, owing to the flat
tions and fluctuation of RTP are analyzed by comparing the daily electricity-price fluctuation, the shifting peak capability of the
average solar–air temperature, the standard deviation (STD) of MPC decreases. The best MPC performance, including energy sav-
RTP, and the daily total performance. Figure 16 shows the weather ing, cost saving, peak-hour energy, and peak-hour cost saving, dur-
conditions and RTP between June 1 and September 1. The daily ing the entire cooling season, was achieved on June 19. The daily
average solar–air temperature was 24.8–41.7 °C, and the STD of average solar–air temperature was 36.4 °C, whereas the daily
RTP was 0.1–8.1. STD of RTP was 7.28, which was the highest during the entire cool-
As shown in Figure 17(a) and 17(b), compared to the PID control ing season. The peak shifting performance of the MPC was the
method, the daily total energy saving with the MPC was  34.51%– worst on August 2, when the daily average solar–air temperature
34.17%, the daily peak-hour total energy saving was  22.25%–98.
13
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

Fig. 18. Correlation between system performance parameters and disturbances including daily mean values of solar–air temperature and STDs of RTP: (a) energy saving, (b)
peak-hour energy saving, (c) cost saving, and (d) peak-hour cost saving

Fig. 19. Comparison of system performances of MPC and PID under different weighting factor ratios.

was 28.4 °C, and the daily STD of RTP was only 0.31. Therefore, the r of the energy saving and the daily STD of RTP. This implies
based on this analysis, as the weather becomes hotter and the that the energy saving has a stronger correlation with the daily
RTP fluctuation increases, the performance of the MPC becomes average solar-air temperature than the daily STD of RTP. Cost sav-
better than that of the PID control method. ing, which is important to residents, has a stronger correlation
Figure 18 shows the values of the correlation coefficient (r) with the daily STD of RTP than the daily average solar–air tem-
between the four coefficients of performance and the daily aver- perature, as shown in Figure 18(b). The shifting peak performance
age solar–air temperature and STD of RTP. As the p-values were has a strong correlation with the daily STD of RTP, as shown in
less than 0.05, the correlation coefficients can be used for analy- Figure 18(c) and 18(d). In summary, the proposed MPC offers
sis. Figure 18(a) shows that the r of the energy saving and the good performance even in terrible weather conditions and power
daily average solar–air temperature is 0.84, which is larger than imbalances.

14
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

response performances of MPC on the inverter air conditioner; to

classify and evaluate the performance of different demand
response MPC methods. This may help many utilities know the
operating characteristic of MPC and adopt MPC in real projects
with better economic benefit. The main conclusions are as follows:
1) DR using MPC can reduce the peak-hour energy consumption
by 21.60% and the total energy cost by 27.43% over the entire cool-
ing season, while maintaining indoor comfort;
2) Compared to Mo-DR, Mh-DR reduces peak energy use by
33.83% and the cost by 33.62% by adjusting the indoor air temper-
ature according to the dynamic tariff;
3) Compared to MPC under a TOU price, MPC under an RTP
offers better performance as it can reflect power supply and
demand in a timely and effective manner. Therefore, to improve
the performance of MPC, a reasonable electricity pricing model
such as RTP should be employed in power distribution grids;
Fig. 20. Statistical distributions of indoor air temperature between 8:00 and 24:00
4) DR with MPC control is suitable for hot weather with widely-
under different weighting factor ratios.
fluctuated RTP.

Declaration of Competing Interest

The authors declare that they have no known competing finan-

cial interests or personal relationships that could have appeared
to influence the work reported in this paper.

Acknowledgment

This research is funded by the National Key Research Program

of China, Chinese-Norwegian collaboration project on Energy
(Grant No. 2019YFE0104900-04). The support is gratefully
acknowledged.

References

[1] Graham, L. Robert, Francis, Julieta, Bogacz, J. Richard, Challenges and

Opportunities of Grid Modernization and Electric Transportation, in, 2017
[2] Z. Jinju, H. Lina, L. Canbing, C. Yijia, L. Xubin, G. Yinghui, What’s the difference
between traditional power grid and smart grid? - From dispatching
Fig. 21. Statistical distributions of indoor air temperature between 8:00 and 24:00 perspective, in:2013 IEEE PES Asia-Pacific Power and Energy, Engineering
under different weighting factor ratios.1 Conf. (2013).
[3] X. Fang, S. Misra, G.L. Xue, D.J. Yang, Smart Grid - The New and Improved
Power Grid: A Survey, IEEE Commun. Surv. Tutorials 14 (4) (2012) 944–980.
[4] X. Hanping, L. Yaoming, M. Shihong, L. Chunjian, X. qiushi, Optimization
4.5. Effects of weighting factor ratio on MPC performance dispatch strategy considering renewable energy consumptive benefits based
on ‘‘source-load-energy storage” coordination in power system, Power System
The weight factor ratio wT : wC in the cost function has a signif- Protection and Control, 45 (17) (2017) 18-25. [in Chinese]
[5] N. O‫׳‬Connell, P. Pinson, H. Madsen, M. O‫׳‬Malley, Benefits and challenges of
icant influence on performance. The performance of the MPC (1:1) electrical demand response: A critical review, Renew. Sustain. Energy Rev. 39
is much better than that of the PID, especially in terms of the peak- (2014) 686–699.
hour cost and energy shifting. A higher weighting factor ratio of 2:1 [6] V. Aggarwal, C.S. Meena, A. Kumar, T. Alam, A. Kumar, A. Ghosh, A. Ghosh,
Potential and Future Prospects of Geothermal Energy in Space Conditioning of
(MPC (2:1)) was used to analyze the effect of the weighting factor Buildings: India and Worldwide Review, Sustainability 12 (20) (2020).
ratio on MPC performance. Figure 19 depicts the cooling-season [7] The future of cooling: opportunities for energy-efficient air conditioning, in,
performance of the MPC under different weighting factor ratios. International Energy Agency, www.iea.org/t&c/, 2018.
[8] H.e. Hao, B.M. Sanandaji, K. Poolla, T.L. Vincent, Aggregate Flexibility of
Compared to the PID control method, the total energy saving with Thermostatically Controlled Loads, IEEE Trans. Power Syst. 30 (1) (2015) 189–
MPC (2:1) over the entire cooling season was 1.06%, the cost saving 198.
was 13.44%, the peak-hour energy saving was 6.34%, and the peak- [9] , International Energy Agency Paris (2019).
[10] X. Chen, J. Wang, J. Xie, S. Xu, K. Yu, L. Gan, Demand response potential
hour cost saving was 26.63%. Therefore, the energy and cost saving
evaluation for residential air conditioning loads, IEEE Generation,
and peak shifting ability of MPC (2:1) was weaker than that of MPC Transmission & Distribution 12 (19) (2018) 4260–4268.
(1:1). However, as shown in Figure 20, MPC (2:1) provided the [11] J.H. Yoon, R. Bladick, A. Novoselac, Demand response for residential buildings
least fluctuation in the indoor air temperature, which explains based on dynamic price of electricity, Energy Build. 80 (2014) 531–541.
[12] A.G. Thomas, P. Jahangiri, D.i. Wu, C. Cai, H. Zhao, D.C. Aliprantis, L. Tesfatsion,
why MPC (1:1) had a higher energy consumption. Intelligent Residential Air-Conditioning System With Smart-Grid
Functionality, IEEE Trans. Smart Grid 3 (4) (2012) 2240–2251.
[13] M.M. Hu, F. Xiao, Price-responsive model-based optimal demand response
5. Conclusion control of inverter air conditioners using genetic algorithm, Appl. Energy 219
(2018) 151–164.
[14] S. Li, D. Zhang, A.B. Roget, Z. O’Neill, Integrating Home Energy Simulation and
In this study, a time-varying MPC was designed to control an Dynamic Electricity Price for Demand Response Study, IEEE Trans. Smart Grid
inverter AC in a residential room to realize DR. A simulation 5 (2) (2014) 779–788.
testbed for a room energy system was built using MATLAB to ana- [15] U.K. Jha, N. Soren, A. Sharma, An efficient HEMS for demand response
considering TOU pricing scheme and incentives, in: 2018 2nd International
lyze the performance of the proposed MPC. This research bridges Conference on Power, Energy and Environment: Towards Smart Technology,
the knowledge gaps in two points: to evaluate seasonal demand 2018, pp. 6 pp.-6 pp.

15
C. Wang, B. Wang, M. Cui et al. Energy & Buildings 256 (2022) 111708

[16] A. Afram, F. Janabi-Sharifi, Theory and applications of HVAC control systems - [26] J. Rehrl, M. Horn, Temperature Control for HVAC Systems based on Exact
A review of model predictive control, MPC, Build. Environ. 72 (2014) 343–355. Linearization and Model Predictive Control, in, IEEE Int. Conf. Cont. Appl. 2011
[17] M. Avci, M. Erkoc, A. Rahmani, S. Asfour, Model predictive HVAC load control in (2011) 1119–1124.
buildings using real-time electricity pricing, Energy Build. 60 (2013) 199–209. [27] J. Široký, F. Oldewurtel, J. Cigler, S. Prívara, Experimental analysis of model
[18] T.Q. Qureshi, S.A. Tassou, Variable-speed capacity control in refrigeration predictive control for an energy efficient building heating system, Appl. Energy
systems, Appl. Therm. Eng. 16 (2) (1996) 103–113. 88 (9) (2011) 3079–3087.
[19] M.M. Hu, F. Xiao, J.B. Jorgensen, S.W. Wang, Frequency control of air [28] J.A. Candanedo, A.K. Athienitis, Predictive control of radiant floor heating and
conditioners in response to real-time dynamic electricity prices in smart solar-source heat pump operation in a solar house, Hvac&R Research 17 (3)
grids, Appl. Energy 242 (2019) 92–106. (2011) 235–256.
[20] E.F. Camacho, C. Bordons, Model predictive control, Advanced Textbooks in [29] S. Yang, M.P. Wan, B.F. Ng, S. Dubey, G.P. Henze, W. Chen, K. Baskaran,
Control and Signal Processing. Web, London, 2007 Experimental study of model predictive control for an air-conditioning system
[21] F. Oldewurtel, A. Parisio, C.N. Jones, D. Gyalistras, M. Gwerder, V. Stauch, B. with dedicated outdoor air system, Appl. Energy 257 (2020) 113920, https://
Lehmann, M. Morari, Use of model predictive control and weather forecasts for doi.org/10.1016/j.apenergy.2019.113920.
energy efficient building climate control, Energy Build. 45 (2012) 15–27. [30] Z. Yingxin, Built Environment, China Architecture & Building Press, Beijing,
[22] S.Y. Yang, M.P. Wan, B.F. Ng, T. Zhang, S. Babu, Z. Zhang, W.Y. Chen, S. Dubey, A 2010 [in Chinese].
state-space thermal model incorporating humidity and thermal comfort for [31] P. Damien, R. Fuentes-Garcia, R.H. Mena, J. Zarnikau, Impacts of day-ahead
model predictive control in buildings, Energy Build. 170 (2018) 25–39. versus real-time market prices on wholesale electricity demand in Texas,
[23] H. Xiangdong, L. Sheng, H. Asada, Modeling of vapor compression cycles for Energy Econ. 81 (2019) 259–272.
advanced controls in HVAC systems, in: Proceedings of the 1995 American [32] P. Faria, Z. Vale, Demand response in electrical energy supply: An optimal real
Control Conference, 1995, pp. 3664-3668 vol.3665. time pricing approach, Energy 36 (8) (2011) 5374–5384.
[24] N.T. Gayeski, P.R. Armstrong, L.K. Norford, Predictive pre-cooling of thermo- [33] H. Guo, M.R. Davidson, Q. Chen, D. Zhang, N. Jiang, Q. Xia, C. Kang, X. Zhang,
active building systems with low-lift chillers, Hvac&R Research 18 (5) (2012) Power market reform in China: Motivations, progress, and recommendations,
858–873. Energy Policy 145 (2020).
[25] A. Bemporad, Morari, M, Model Predictive Control Toolbox User’s Guide, The [34] E.R.C.O. Texas, Historical RTM Load Zone and Hub Prices, in, Electric Reliability
MathWorks, 2017. Council of Texas.

16