Sunteți pe pagina 1din 12

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1

Equivalence of Multi-Time Scale Optimization


for Home Energy Management Considering User
Discomfort Preference
Yi Liu, Yun Zhang, Kairui Chen, Si Zhe Chen, and Bin Tang
, g, c

AbstractThe problem of home energy management (HEM)


optimization can be transformed from single-time scale into
multi-time scale to decrease the computational complexity,
however the equivalence of transformation should be guaranteed.
In this paper, we investigate the equivalence of multi-time scale
HEM optimization, which includes electric vehicles (EV), thermal
appliances, and uncontrollable devices. The total electricity cost
and user discomfort of temperature are considered. We propose a
thermal model in multi-time scale to reduce the error of
transformation. We show and prove the equivalent conditions of
transformation. Based on the conditions, we present an improved
optimization algorithm for the problem of multi-time scale HEM
optimization. Numerical results show that the proposed model is
more suitable for the transformation, the conditions are obeyed,
and the proposed algorithm achieves better performance while
solving the problem.
Index TermsHome energy management, multi-time scale,
optimization algorithm, thermal model, equivalence of
transformation, user discomfort.

Continuous time indoor temperature (C).

in

Fast time scale indoor temperature (C).

in

Slow time scale indoor temperature (C).

T (t )
T (k )
T {k s }
T

out

(t )

Continuous time outdoor temperature (C).

out

(k )

Fast time scale outdoor temperature (C).

Fast time scale desired temperature (C).

Slow time scale desired temperature (C).

T (k )
T {k s }
P

HVAC

(t )

Continuous time input power of HVAC (kW).

HVAC

(k )

Fast time scale input power of HVAC (kW).

HVAC

{k s } Slow time scale input power of HVAC (kW).

HVAC ,max

s(k )

t , ts

Fast and slow time scale time interval (hours).

K , Ks

Time slot number of fast and slow time scale.

k , ks

Time slot index of fast and slow time scale,

Kf

Time slot number of fast time scale in each time

M, N
C
S
R

interval of slow time scale.


Number of HVAC and EV.
Thermal capacitance (kWh/C).
Thermal resistance (C/kW).
Power efficiency of HVAC.

k 1, 2, , K , k s 1, 2, , K s .

This work was supported by the National Natural Science Foundation of


China under Grants U1501251 and 51307025, and by the PhD Fund of Ministry
of Education of China under Grant 20124420130001. (Corresponding Author:
Y. Zhang.)
Y. Liu is with the School of of Automation, Guangdong University of
Technology, Guangzhou 510006, China, and also with the College of
Information Science and Technology, Zhongkai University of Agriculture and
Engineering, Guangzhou 510225, China (e-mail: baddielao@126.com).
Y. Zhang, K. Chan, S. Z. Chen, and B. Tang are with the School of
Automation, Guangdong University of Technology, Guangzhou 510006, China
(e-mail:
yz@gdut.edu.cn;
cszscut@126.com;
tangbin316@163.com;
kraychen@139.com).

Maximum input power of HVAC (kW).


Fast time scale switch of HVAC, 1 for turned on,

k d , k sd

0 otherwise.
Fast time scale power of EV, positive when EV
charges and negative when EV discharges (kW).
Slow time scale power of EV, positive when EV
charges and negative when EV discharges (kW).
time slot of Fast and slow time scale when EV
arrives.
time slot of Fast and slow time scale when EV

E (k )

departs.
Fast time scale battery energy of EV (kWh).

P EV (k )
NOMENCLATURE

Coefficient of fast time scale thermal model.

in

P EV {k s }
k a , k sa

E{k s }

Slow time scale battery energy of EV (kWh).


Batterys efficiency of EV.

E min , E max Minimum and maximum battery energy of EV


(kWh).
EV ,min
P
Minimum power of EV (kW).
P EV ,max

Maximum power of EV (kW).

desired
j

Desired battery energy of EV when departs (kWh).

SH

Fast time scale total power consumption of home


(kW).
Slow time scale total power consumption of home
(kW).

P (k )
P SH {k}

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1

control method considering the electricity cost, the charging


target of EV, and the user discomfort of temperature. Reference
[17] studies a HEM system with HVACs, EVs and DGs. It
P u {k s }
builds Markov models for the wind turbines and photovoltaic
panels, regards the HVACs as hard loads, and schedules the
Bd
EVs through dynamic programming. Reference [18] uses an
D(k )
auto regressive moving average model of wind speed to
forecast the output power of wind turbine, and solves the
Slow time scale dynamic price (cent/kWh).
D{k s }
problem by genetic algorithm. Reference [19] builds models for
i , i
Fast and slow time scale weight of discomfort.
photovoltaic panel and hard load, solves the problem by
adaptive control method, and adjusts the charging power of
I. INTRODUCTION
batteries in real time to eliminate the disturbance of uncertainty.
n order to improve performance, reduce waste of energy and Reference [20] predicts the dynamic price in the future by a
emission of greenhouse gas, smart grid influences users statistical method and schedules the EVs by thresholds of price.
Reference [8], [17]-[20] describe the problem of HEM
energy consumption by dynamic pricing [1]-[3]. HEM is an
important part in demand side of smart grid. A HEM system optimization in single-time scale. The dimension of solution is
schedules its controllable loads according to the dynamic price, high because of the great amount of controllable loads and time
setting of user, power demand of hard loads, and output power slots in the period of optimization. The combinatorial
explosion problem is encountered [21]. Furthermore, with
of distributed generators (DG) in the HEM system.
The energy consumption of heating, ventilation, and uncertainties, the solution may become invalid in practice [22].
air-conditioning (HVAC) exceeds half of the total energy The HEM systems must refresh their predictions and
consumption in residential buildings [4]. On the other hand, re-optimize online. Its hard to get the solution with high
EVs are not only travel tools but also storage devices. EV will dimensions within a limited time.
Reference [23] and [24] describe the problem of HEM
be popular in the future [5]. In the literature, HEM optimization
with HVACs and EVs is a hot and extensive research topic. He optimization in multi-time scale. They divide the period of
optimization in slow time scale (STS) firstly, and describe a
we focus on modeling and optimization.
Bargiotas and Birdwell proposed a continuous thermal problem of STS HEM optimization. Every STS time interval
model for residential buildings in 1988 [6]. They considered the becomes a new period of optimization. Secondly, they divide
input power of air conditioner, outdoor temperature, and every STS time interval in fast time scale (FTS), and describe a
Wiener process as the factors that influenced the indoor problem of FTS HEM optimization for each STS time interval.
temperature. This model is a simple first-order dynamic. The The FTS time interval and the single-time scale time interval
models of hot water temperature and building temperature in [7] are the same. Thus, they transform the problem of single-time
are similar with the model in [6]. Reference [8], [9], and [10] scale HEM optimization into two problems with different time
consider more influencing factors while modeling, but their scales. The dimensions of the two problems are decreased.
continuous thermal models are still described in first-order Reference [11] considers user discomfort of temperature in
differential equations. Instead of calculations based on physical multi-time scale HEM optimization. But its STS thermal model
principles, estimations according to experimental data are is not suitable for multi-time scale, because it doesnt consider
widely applied to gain the parameters of thermal model [8]-[9], the equivalence while transforming the thermal model from
single-time scale to multi-time scale.
[11].
We continue the work of [11]. We study the multi-time scale
Reference [12]-[16] optimize HEMs by multiple agents. In
[12], the batteries, DGs, and controllable loads are controlled HEM optimization that includes controllable loads (EVs and
by the producer and consumer agents, the dynamic price and HVACs) and uncontrollable devices (hard loads and DGs). The
grid states are observed by the observer agents. The producer total electricity cost and user discomfort of temperature are
and consumer agents make decisions by operational cost and considered in the optimization. Unlike previous work, we
shared performances. Reference [13] and [14] are based on the concentrate on the equivalence of transformation. The main
game theory. The agents try to optimal the total cost. The HEM contributions of this paper are summarized as follows:
We propose a STS thermal model. The single-time scale
of [15] runs in centralized control mode generally. But when it
thermal model should be transformed into a STS one
is autonomous, decentralized control mode will be activated to
while the problem of HEM optimization is transformed
stabilize the voltage in real time. Reference [16] tries to reduce
from single-time scale into multi-time scale. Both the
the computational complexity and get satisfying performance
changing outdoor temperature and input power of HVAC
by a two-layer optimization, which is based on both centralized
influence the equivalence of thermal models
and distributed agents.
transformation. The proposed STS thermal model, which
Decentralized optimization reduces the computational
considers the changing outdoor temperature and input
complexity, but it is hard to get the optimal solution. Reference
power of HVAC, eliminates the influence of outdoor
[8], [17]-[20] schedule HEMs based on centralized
temperature and reduces the one of input power of HVAC.
optimization. Reference [8] builds models of thermal dynamic

We
show and prove the equivalent conditions of
and battery of EV. It schedules the energy by a stochastic

P u (k )

Fast time scale total power consumption of


uncontrollable device (kW).
Slow time scale total power consumption of
uncontrollable device (kW).
Daily budget (cents).
Fast time scale dynamic price (cent/kWh).

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1

problems transformation considering total electricity cost


and user discomfort of temperature. The objective
function that considers total electricity cost and user
discomfort of temperature should satisfy the conditions to
keep the equivalence of problems transformation.
We propose an optimization algorithm for the problem of
FTS HEM optimization. The optimal solution must satisfy
the equivalent conditions. We utilize the conditions to
construct the proposed algorithm that gives the solution
directly according to the conditions. Unlike heuristic
algorithm, the algorithm has no iteration; it can give an
approximate optimal solution speedily.
The remaining of this paper is organized as follows. The STS
thermal model is presented in Section II. The equivalent
conditions of problems transformation are proved in Section
III. The optimization algorithm for the problem of FTS HEM
optimization is proposed in Section IV. The numerical results
are provided in Section V. Conclusions are presented in Section
VI.

When t is short enough, (2) is approximately equivalent to


a discrete-time thermal model as
T in ((k 1)t ) T in (k t )
(3)
(T out (k t ) T in (k t )) gP HVAC (k t ) c
where , g , and c are coefficients. Thermal model (3) is used
in the single-time scale HEM optimization [7]-[8], [11], and the
FTS HEM optimization of multi-time scale as well.
In the problem of multi-time scale HEM optimization, a
discrete-time STS thermal model is needed. Based on (2),
Reference [11] gives the STS thermal model as
T in ((k s 1)ts ) T in (k s ts )
(4)
HVAC (k s ts ) c
(T out (k s ts ) T in (k s ts )) gP
where , g and c are coefficients.
According to (3), we get:
in
T ((k K f )t ) T in (k t )

((1 )

II. STS THERMAL MODEL

Kf

1)T in (k t )

k K f 1

(1 )

k K f 1 r

T out (r t ) . (5)

r k

In this section, we describe the relationship between FTS and


STS firstly, and discuss the defect of STS thermal model in [11].
Finally, we propose a new STS thermal model.
A. Relationship between FTS and STS
The time scale of single-time scale HEM optimization is
called FTS [11]. The time intervals of FTS and single-time
scale are the same. t , ts are defined as the FTS and STS
time intervals. K and K s are defined as the time slot numbers
of FTS and STS. Let k and k s be the time slot indexes of FTS
and STS, k 1, 2, , K , k s 1, 2, , K s . There are K f FTS
time slots in each STS time interval:
K f ts / t .

(1)

K FTS time slots are separated into K s groups. FTS time


slot k ( k [ K f (k s 1) 1, K f k s ] ) belongs to STS time slot k s .
B. Defect of STS Thermal Model in [11]
The indoor temperature of a home is influenced by the
thermal resistance, thermal capacitance, cooking, solar
radiation, wind speed, humidity, outdoor temperature, HVAC
power, etc. For simplicity, we assume the outdoor temperature
and HVAC power are the major factors of thermal model. This
assumption is made commonly in the literature [6]-[8], [10].
The continuous thermal model can be described as follow
1
(2)
CdT in (t ) (T out (t ) T in (t ))dt RP HVAC (t )dt dt
S
where T in (t ) is the indoor temperature, T out (t ) is the outdoor
temperature, C is the thermal capacitance, S is the thermal
resistance, P HVAC (t ) is the input power of HVAC and R is the
power efficiency of HVAC. is the total impact of other
influencing factors. When the other influencing factors (such as
solar radiation, cooking, etc) are non-ignorable, they can be
modeled (such as by Wiener process [6], [25]) and added into

k K f 1

(1 )

k K f 1 r

r k

K f 1

P HVAC (r t ) c (1 ) r
r 0

When k k s K f , the equals signs left parts of (4) and (5)


will be the same. When :
T out (r t ) T out , P HVAC (r t ) P HVAC , r [k , k K f 1] , (6)
(4) and (5) will be equal, and their coefficients will have
relations:
K
[1 (1 ) ] , g g
f

1 (1 )

Kf

K f 1

, c c (1 ) r . (7)
r 0

Condition (6) shows that the outdoor temperature and input


power of HVAC must be static in every ts , otherwise (3)
and (4) are contradictory.
As (3) is used in the problem of single-time scale HEM
optimization, when the problem is transformed into
multi-time scale with STS thermal model (4), Condition (6)
must be satisfied. Otherwise the problems of single-time scale
and multi-time scale will be unequal because (3) can not be
transformed into (4).
The main purpose of problems transformation is to reduce
the dimension of solution. ts should be much longer than

t in applications. The outdoor temperature is unlikely to


remain constant over a long time interval, and it is
unreasonable to hold the input power of HVAC in ts
(because the goal is to optimize the input power of HVAC in
FTS). It is hard to satisfy (6), so it is hard to guarantee the
equivalence of problems transformation.
C. Modeling of STS Thermal Dynamic
For simplification, we omit the time interval t and ts in
the following discrete-time equations, and use Symbol () and

{}
to indicate FTS and STS.
Unlike previous modeling work, the modeling goal of this

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1

paper is to construct a model that works in STS and is


transformed from the single-time scale thermal model.
According to [11], each kind of STS power (including the
input power of HVAC, the power of EV, and the power of
uncontrollable device such as hard load and DG) at time slot k s ,
is the average of its FTS power within k s :
K f ks

1
P{k s }
Kf

P(k ) .

(8)

k K f ( ks 1) 1

The relationship between FTS temperature and STS


temperature is the same as (8). As the FTS input power of
HVAC at time slot k will decide the indoor temperature at time
slot k 1 [8], the STS temperature (including the indoor,
outdoor, and desired temperature) at time slot k s is defined as

T {k s 1}

1
Kf

K f ks

s (k )T (k 1)

(9)

k K f ( ks 1) 1

K f 1

T in {k s 1} (1 ) f T in {k s } c (1 ) r
K

r 0

Kf

Kf

g
Kf

(1 )

K f r

r 1

Kf

(1 )

K f r

r 1

K f ( ks 1) r

T out (k )

.(12)

k K f ( ks 2) 1 r

(K

r ) P HVAC {k s 1} rP HVAC {k s }

We calculate FTS indoor temperature T in (k ) from (3) and


get STS indoor temperature T in {k s } based on (9). This T in {k s }
is got directly from (3); it is the standard of following
comparison. We compare (4) (the STS thermal model in [11])
and (12) (the STS thermal model of this paper) with the
standard in the remaining of this section. Let , g , and c
equal to 0.25, -0.5, and 0.125, the values of , g , and c are
calculated by (7). Let K f 6 and the init indoor temperature

where s (k ) is the switch of HVAC. s (k ) 1 when user turns

T in (1) T in {1} 24.5o C . The known FTS input power of

on HVAC, otherwise s (k ) 0 .
According to (3) and (9), we get a thermal equation:

HVAC P HVAC (k )

temperature T in (k ) and the known FTS outdoor temperature

K f 1

T in {k s 1} (1 ) f T in {k s } c (1 ) r
K

r 0

Kf

Kf

g
Kf

(1 )
r 1

Kf

(1 )

K f ( ks 1) r

K f r

T out (k ) .

(10)

K f r

r 1

P HVAC (k )

k K f ( ks 2) 1 r

When K f 1 , STS and FTS are the same, (10) becomes (3).
When T out (k ) T out and P HVAC (k ) P HVAC , (10) becomes (4)
with coefficients (7). The STS thermal model in [11] is a
special case of (10). (10) is deduced from (3) directly. It wont
influence the equivalence of problems transformation.
The second term of the equals signs right part in (10) is a
known constant. T out (k ) is predicted and known before
optimization, therefore the third term of equals signs right part
in (10) is a constant as well. But P HVAC (k ) is a FTS variable,
therefore (10) can not be used as a STS thermal model directly.
We define an approximate conversion as

g
Kf

Kf

(1 )

K f r

r 1

g
Kf

Kf

(1 )
r 1

K f ( ks 1) r

P HVAC (k )

k K f ( ks 2) 1 r
K f r

(K

r)P

HVAC

{k s 1} rP

T out (k ) are shown in Fig. 2. The STS indoor temperatures of


the standard, (4), and (12) are shown in Fig. 3. The relative
errors of (4) and (12) are shown in Fig. 4. When P HVAC (k ) or
T out (k ) changes, (4) doesnt match the standard because
condition (6) is not satisfied, (3) and (4) are contradictory.
When P HVAC (k ) changes, (12) doesnt match the standard

k K f ( ks 2) 1 r
K f ( ks 1) r

is shown in Fig.1. The FTS indoor

HVAC

{k s }

(11)

to solve the defect.


According to (10) and (11), we propose a new STS thermal
model:

because P HVAC {k s } of (12) is approximately converted from

P HVAC (k ) according to (11).


We change the coefficients values of FTS thermal model (3)
and repeat the comparison. The maximum relative errors (MRE)
of (4) and (12) with different coefficients values are shown in
Table I. Model (3) shows that , g , and c are the weights of
T out (k ) , P HVAC (k ) , and the sum of other influencing factors
respectively. As , | g | ( g is negative when the HVAC is
working in cooling mode), and c increase, the effects of their
corresponding factors increase, and the effects of others
decrease. The relative error of (4) is affected by T out (k ) and

P HVAC (k ) , it increases as or | g | increases but decreases as


c increases. The relative error of (12) is only affected by
P HVAC (k ) , it increases as | g | increases but decreases as or
c increases. When K f is equal to 1, STS and FTS are the same,

and the relative errors reduce to 0 as both (4) and (12) become
(3).
The comparison results show that the STS indoor
temperature got by the STS thermal model (12) is more close to
the one got by the FTS thermal model than the one got by (4).
(12) is more suitable to be the STS thermal model of multi-time
scale optimization than (4).

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1

5
III. EQUIVALENCE OF PROBLEMS TRANSFORMATION
As the problem of HEM optimization is transformed from
single-time scale to multi-time scale, its necessary to analyze
the equivalence of problems transformation. In this section,
the problem formulations in single-time scale and multi-time
scale are described firstly, and then the equivalent conditions of
problems transformation are proved secondly. The dynamic
price is assumed constant over every STS time interval [11].
The assumption is practical as the power supplier wont adjust
the dynamic price frequently.

Fig. 1. FTS input power of HVAC.

A. Problem Formulation of HEM Optimization


A HEM system may include HVACs, EVs, DGs, and hard
loads [8], [11], [17]-[20]. Hard loads and DGs are all regarded
as uncontrollable devices in the paper. The HEM system
controls the powers of HVACs and EVs according to the
optimal solution.
Let M and N be the numbers of HVAC and EV. The
problem of single-time scale HEM optimization is generally
described as follow
K M
K

J min D(k ) PSH (k )t i si (k ) Ti in (k 1) Ti d (k 1)


k 1 i 1
k 1

Subject to: i 1, 2, , M , j 1, 2, , N

Fig. 2. FTS outdoor temperature and indoor temperature.

1. Ti in ( k 1) Ti in (k ) i (Ti out (k ) Ti in (k )) gi Pi HVAC (k ) ci


2. 0 Pi HVAC (k ) Pi HVAC ,max
3. Pi HVAC ,max 0 , if si (k ) 0
4. E min
E j (k aj )
j

Fig. 3. STS indoor temperatures of the standard, (4), and (12).

k k aj

, k [k aj , k dj 1]
PjEV (k)t E max
j

EV ,min
PjEV (k ) PjEV ,max , k [k aj , k dj 1]
P
5. j
PjEV (k ) 0, otherwise

6. E

desired
j

E j (k )
a
j

k dj 1

PjEV (k )t

k k aj
K

D(k ) P
7.

(k )t

. (13)
N
M HVAC

EV
u
d
D(k ) Pi
(k ) Pj (k ) P (k ) t B

k 1
j 1
i 1

Constraint 1 is FTS thermal model (3). Constraint 2 and 3


limit the range of Pi HVAC (k ) . When HVAC i is turned off,

Fig. 4. Relative errors of (4) and (12).

Pi HVAC (k ) must be zero. Constraint 4 limits the range of battery

TABLE I
MAXIMUM RELATIVE ERRORS OF (4) AND (12)
Coefficients

SH

k 1

energy of EV E j (k ) . j is the batterys efficiency of EV j (it

MRE of (4)

MRE of (12)

0.25 , g 0.5 , c 0.125 , K f 6

8.16%

0.85%

0.5 , g 0.5 , c 0.125 , K f 6

10.40%

0.31%

0.25 , g 2 , c 0.125 , K f 6

63.67 %

19.27%

is a constant as we assume that the EV has a secondary battery


such as ultracapacitor or electromechanical flywheel to
stabilize the efficiency when the power of EV changes).
PjEV (k ) is the power of EV j , it is positive when EV j

0.25 , g 0.5 , c 0.25 , K f 6

7.97%

0.84%

charges and negative when discharges. k aj and k dj are the time

0.25 , g 0.5 , c 0.125 , K f 1

0%

0%

slots when EV j arrives and departs. Constraint 5 limits the


range of PjEV (k ) . When EV j is not at home, PjEV (k ) must be
zero. Constraint 6 is the charging target of EV. E desired
is the
j

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1

desired batterys energy of EV j when it departs. Constraint 7


d

SH

limits the total electricity cost by daily budget B . P (k ) is


the total power consumption of home. P u (k ) is the total power
consumption of uncontrollable device. The power of hard load
and DG can be predicted by Markov chain [17], [26]-[27]. The
first part of objective function is the total electricity cost. D(k )
is the dynamic price at k . The second part of objective function
is the total user discomfort of temperature, which is measured
only by the temperature deviation from the user desire [8], [11].
Ti d (k ) is the desired temperature of HVAC i . When users are
more concerned about comfort of temperature, they can
increase FTS weight of discomfort i . The user discomfort of
temperature is calculated at time slot k 1 as the input power
of HVAC at time slot k decides the indoor temperature at time
slot k 1 [8].
There are two steps in the problem of multi-time scale HEM
optimization [11]. The first step is the problem of STS HEM
optimization. We replace the STS thermal model of STS HEM
optimization in [11] with the proposed thermal model, and
rewrite the problem of STS HEM optimization as follow
Ks M
Ks

J min D{ks }P SH {ks }ts i Ti in {ks 1} Ti d {ks 1}


ks 1 i 1
ks 1

Subject to: i 1, 2, , M , j 1, 2, , N

Ti {k s 1} (1 i ) Ti {k s } ci
Kf

in

1.

i
Kf

gi
Kf
2. 0 Pi

Kf

(1 i )

(1 )

(1 )
r 0

k K f ( ks 2) 1 r

K f r

r 1

K f 1

K f ( ks 1) r

K f r

r 1

Kf

HVAC

in

(K

Ti out (k )

{k s } Pi HVAC ,max

3. Emin
E j {ksa, j }
j

ks

ks ksa, j

EV
j

, ks [ksa, j , ksd, j 1]
{ks }ts Emax
j

EV ,min
PjEV {k s } PjEV ,max , k s [k sa, j , k sd, j 1]
P
4. j
PjEV {k s } 0, otherwise

5. E

desired
j

E j {k }
a
s, j

ksd, j 1

j PjEV {k s }ts

ks ksa, j
Ks

D{k }P

6.

ks 1

SH

{ks }ts

(14)
N
M HVAC

EV
u
d
Pi
{ks } Pj {ks } P {ks } ts B
D{ks }
ks 1
j 1
i 1

where i is STS weight of discomfort. For equivalence of


Ks

problems transformation, it should be set as K f

times the

value of i . In Section III B, we will explain the setting in


detail. When EV j arrives or departs in the middle of a STS
EV ,min
j

time interval, P

EV ,max
j

and P

according to Pi HVAC {k s } and PjEV {k s } , which are the solutions


of (14). Pi HVAC (k ) and PjEV (k ) are constrained by (8). Since
HVACs and EVs are decoupled in the first step, Pi HVAC (k ) and

PjEV (k ) are optimized in FTS separately [11].


As the dynamic price is unchanged in each STS time interval,
there is no object to optimize PjEV (k ) . In order to reduce the
varieties of power of EV, (15) is a good choice
PjEV (k ) PjEV {k s } , k [ K f (k s 1) 1, K f k s ] .

should be adjusted. When

(15)

Pi HVAC (k ) is optimized in FTS to decrease user discomfort of


temperature [11]:
K f ks

Jiks min si (k ) Ti in (k 1) Ti d (k 1)
k K ( k 1) 1

f s

Subject to
1. Ti in ( k 1) Ti in (k ) i (Ti out (k ) Ti in (k )) gi Pi HVAC (k ) ci
2. 0 Pi HVAC (k ) Pi HVAC ,max
3. Pi HVAC ,max 0 , if si (k ) 0
4. Pi HVAC {k s }

r ) Pi HVAC {k s 1} rPi HVAC {k s }

HVAC i is turned on or turned off in the middle of a STS time


interval, Pi HVAC ,max should be adjusted as well [11].
The second step is the problem of FTS HEM optimization.
Pi HVAC (k ) and PjEV (k ) are scheduled in the second step

1
Kf

K f ks

k K f ( ks 1) 1

Pi HVAC (k ) .

(16)

B. Equivalent conditions of Problems transformation


The problems transformation has two parts: transformations
of constraint and objective function. In Section II, we had
discussed the relationship between problem (13) constraint 1
and problem (14) constraint 1. Problem (13) constraint 2 and 3
are equal to problem (14) constraint 2 because of problem (16)
constraint 2, 3, and 4. With (15), problem (13) constraint 4, 5,
and 6 are equal to problem (14) constraint 3, 4, and 5. Based on
the definition of (8) and the assumption of dynamic price,
problem (13) constraint 7 and problem (14) constraint 6 are
equal (theorem 1 proves it in detail). Except the thermal model,
the other constraints remain equal after transformation.
In the following of this section, we discuss the equivalent
conditions of objective functions transformation.
Theorem 1: when the dynamic price is constant over every
STS time interval, the total electricity costs in problem (13) and
(14) are equal.
Proof 1: the condition of theorem 1 is:
D{k s } D(k ) , k [ K f (k s 1) 1, K f k s ] .
(17)
From the first term of objective function in problem (13), we
get:
K f ks
Ks
K

K f t
SH
D
(
k
)
P
(
k
)

D(k ) P SH (k )

k 1
ks 1 K f
k K f ( ks 1) 1
. (18)
Ks

D{k s }P SH {k s }ts
ks 1

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1

Equation (18) shows that the first terms of objective function


in problem (13) and (14) are equal. Theorem 1 is proved.
Theorem 2: when i is K f times the value of i , and when
the indoor temperature keeps not lower than (or not higher than)
the user desire in each STS time interval (except the case when
HVAC is turned off), the total user discomforts of temperature
in problem (13) and (14) are equal.
Proof 2: the conditions of theorem 2 are:
(19)
i K f i

Ti in (k 1) Ti d (k 1)

k [ K f (k s 1) 1, K f k s ]

si (k ) 0

and
(20a)

Ti in (k 1) Ti d (k 1)

k [ K f (k s 1) 1, K f k s ]

and

(20b)
si (k ) 0 .
From the second terms of objective function in problem (13)
and (14), we get:
K

s (k ) T
i i

k 1 i 1

in

(k 1) Ti d (k 1)

K f ks

Ks

i 1 ks 1 k K f ( ks 1) 1
Ks

i
i 1 ks 1

Ks

ks 1 i 1
M

Ks

i 1 ks 1

k K f ( ks 1) 1

si (k )Ti in (k 1)

K f ks

k K f ( ks 1) 1

(21)

i
Kf

k K f ( ks 1) 1

si (k )Ti in (k 1)

K f ks

k K f ( ks 1) 1

the
the
let

Step 2: if si (k ) 0 , then Ti in (k 1) Ti d (k 1) , and


calculate Pi HVAC (k ) by the FTS thermal model. Otherwise, let

Pi HVAC (k ) 0 and k M .
Step 3: if Pi HVAC (k ) breaks maximum or minimum limit,
then set it as the limit.
Step 4: k k 1 , repeat step 2 and 3 until k K f k s .

K f Pi HVAC {k s }

K f ks

k K f ( ks 1) 1

K f M

Pi HVAC (k )
, renew

Pi HVAC (k ) by: Pi HVAC (k ) Pi HVAC (k ) P HVAC , k M and


k [ K f (k s 1) 1, K f k s ] .
Firstly, we assume that the HEM system will completely
eliminate user discomfort of temperature, that is, the indoor
temperature will match the user desire strictly. Step 1 to 4
sequentially process all the FTS time slots in STS time slot k s ,

si (k )Ti d (k 1)

Ti in {ks 1} Ti d {ks 1}
K f ks

for

k K f (k s 1) 1 .

Step 5: P HVAC

i si (k ) Ti in (k 1) Ti d (k 1)

K f ks

propose an optimization algorithm without iteration


problem (16).
The solution of HEM optimization satisfies (20), so do
solution of problem (16). We utilize this feature to design
optimization algorithm:
Step 1: define an empty set M , and

.(22)

si (k )Ti d (k 1)

When (19) and (20) ((20a) or (20b)) are satisfied, the equals
signs left parts of (21) and (22) are equal. That is, the second
terms of objective function in problem (13) and (14) are the
same. Theorem 2 is proved. We do not consider the case when
HVAC is turned off because the FTS indoor temperature cant
influence the objective functions as si (k ) 0 .
Optimization will try to decrease the input power of HVAC
because of electricity cost. When the desired temperature is
lower than the outdoor temperature, the indoor temperature will
be not lower than the user desire; (20a) is satisfied certainly.
Correspondingly, when the desired temperature is higher than
the outdoor temperature, (20b) is satisfied certainly. Therefore,
(17) and (19) are the equivalent conditions of objective
functions transformation.
If we set the STS time interval according to the dynamic
price update cycle, and set i K f i , we will satisfy (17) and
(19) easily. The equivalence of thermal models transformation,
which is discussed in section II, is the only factor that
influences the equivalence of problems transformation indeed.
IV. OPTIMIZATION ALGORITHM FOR PROBLEM (16)
The transformation from single-time scale to multi-time
scale decreases the dimension of solution, but the multi-time
scale optimization will still be time-consuming if heuristic
algorithm is used to solve the problem. In this section, we

and try to find out how much Pi HVAC (k ) should be to achieve


the assumption. When the HVAC is turned off or when
Pi HVAC (k ) breaks its limits, Pi HVAC (k ) should be adjusted. Step
2 and 3 satisfy problem (16) constraint 1, 2, and 3. Secondly,
we adjust Pi HVAC (k ) in step 5 to meet the demand of problem
(16) constraint 4. P HVAC is the adjusting value of Pi HVAC (k )
and keeps unchanged in k s . Therefore, Pi HVAC (k ) is lower than
(or higher than) its desire in k s , so do the indoor temperature.
Condition (20) is realized. When the HVAC is turned off, it
cant be adjusted in step 5.
The proposed algorithm hasnt any iteration. Its
computational speed is fast as it is not influenced by the
dimension of solution. Notice that although the proposed
algorithm satisfies all the constraints of problem (16) and
condition (20), its solution is an approximate optimal one as the
objective function of problem (16) is not calculated in the
algorithm. That is, the solution only satisfies the necessary
conditions of optimal solution.

V. NUMERICAL RESULTS
In this section, we show a HEM example to test the
performances of our proposed model, equivalent conditions
and algorithm. We describe the HEM example in single-time
scale, multi-time scale with STS thermal model (4), and
multi-time scale with STS thermal model (12). , g , and c
are equal to 0.25, -0.5, and 0.125. The proposed algorithm is
applied to the optimizations, besides cooperative particle

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1
swarm optimization (CPSO) [28]. The population size of CPSO
is 500.
For simplicity, we assume that there are only one HVAC and
one EV in the HEM example. They are scheduled in a day
period. The STS time interval is set as one hour, and the FTS
time interval is set as 3 minutes. There are 24 STS time slots in
a day period and 20 FTS time slots in a STS time interval. We
set the daily budget as 9 dollars. We assume the EV has
following parameters: it arrives at 7:00 and departs at 18:00; its
maximum charge and discharge power are 3kW; its battery
capacity is 20kWh; its maximum and minimum states of charge
(SOC) are set as 80% and 20%; its batterys efficiency is 1; the
init SOC is 40%; the desired SOC when the EV departs is 80%.
We assume the HVAC runs in cooling mode; its maximum
input power is 3kW; it is turned off from 12:00 to 16:00; the
weight of discomfort is 1. We assume that the outdoor
temperature and the desired temperature are known, and the
initial indoor temperature is 28o C . We assume that the
dynamic price and the total power consumption of
uncontrollable device are predicted perfectly, as shown in Fig.
5.
Based on CPSO, we run the simulation twice (which are
marked as A and B) in single-time scale. Fig. 6 shows the two
simulative results of single-time scale optimization based on
CPSO (STSO-CPSO). The objective functions values
(performance indexes) of the simulations are the same, and the
total electricity costs both reach the daily budget. We discuss
the result of simulation A firstly. Fig. 6(a) and (d) show the
solution. The EV almost keeps discharging when it arrives at
7:00, as the dynamic price is very high from 6:00 to 10:00. The
EV charges with almost the maximum power when the
dynamic price is very low from 14:00 to 18:00. The EV also
charges from 10:00 to 14:00 because it should meet the desired
SOC when the EV departs. Before 7:00 and after 18:00, the EV
is not at home, the power of EV is zero. The input power of
HVAC varies according to the user discomfort of temperature
and the dynamic price. When the dynamic price is very high
from 6:00 to 10:00, the indoor temperature is far higher than the
desired temperature as the input power of HVAC is decreased
to reduce the total electricity cost. Except the time interval
when the HVAC is turned off, the indoor temperature is not
lower than the desired temperature. Condition (20) is satisfied.

8
Fig. 5. Dynamic price and total power consumption of uncontrollable device.
(a) Day-ahead dynamic price. (b) Total power consumption of uncontrollable
device.

Fig. 6. Results of STSO-CPSO. (a) Power of EV. (b) Average power of EV. (c)
SOC of EV. (d) Input power of HVAC. (e) Temperature.

The powers of EV of simulation A and B are different. But


the average ones from 7:00 to 10:00 are the same, besides the
ones from 10:00 to 14:00, and from 14:00 to 18:00, as shown in
Fig. 6(b). Therefore the SOCs of simulation A and B coincide at
10:00, 14:00, and 18:00 (the values are 20%, 21%, and 80%), as
shown in Fig. 6(c). As the dynamic price keeps constant over
the 3 time intervals, the contributions of the two powers of EV
to the total electricity cost are the same. The objective
functions values of A and B are the same; the optimal solution
of power of EV is not unique.
Fig. 7 shows the results of multi-time scale optimization with
model (12) based on CPSO (MTSO-M12-CPSO). We solve the

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1
problem of STS HEM optimization firstly. We search the
power of EV and input power of HVAC in STS based on CPSO.
The STS solution is shown in Fig. 7(a) and (b). According to
the STS solution, we solve the problem of FTS HEM
optimization secondly. We search the FTS input power of
HVAC of each hour based on CPSO. The search is repeated 24
times, the FTS input power of HVAC is shown in Fig. 7(a).
According to (15), we set the FTS power of EV, which is shown
in Fig. 7(b).

9
We replace model (12) by (4) and run the simulation of
multi-time scale again. Fig. 8 shows the results of multi-time
scale optimization with model (4) based on CPSO
(MTSO-M4-CPSO). The total electricity cost also reaches the
daily budget. MTSO-M4-CPSO also gets the optimal power of
EV, as the SOC of EV at 10:00, 14:00, and 18:00 also is 20%,
21%, and 80% respectively. Fig. 6, 7, and 8 show that the
difference of indoor temperature between STSO-CPSO and
MTSO-M4-CPSO is more obvious than the one between
STSO-CPSO and MTSO-M12-CPSO, because the relative
error of (4) is far greater than the one of (12). MTSO-M4-CPSO
schedules far more (or less) input power of HVAC than
STSO-CPSO and MTSO-M12-CPSO when the input power of
HVAC increases (or decreases) greatly. The users feeling on
temperature is worse than the ones of STSO-CPSO and
MTSO-M12-CPSO, although their total electricity costs are the
same.

Fig. 7. Results of MTSO-M12-CPSO. (a) Power of EV in FTS and STS. (b)


Input power of HVAC in FTS and STS. (c) SOC of EV in FTS. (d) Temperature
in FTS.

The total electricity cost reaches the daily budget. The SOC
of EV at 10:00, 14:00, and 18:00 is 20%, 21%, and 80%
respectively, this is the same as the one of STSO-CPSO, as
shown in Fig. 7(c). MTSO-M12-CPSO gets the optimal power
of EV. MTSO-M12-CPSO gets the approximate optimal input
power of HVAC as the indoor temperature is almost the same
as the one of STSO-CPSO. Fig. 3 shows that (12) gives a higher
STS indoor temperature when the input power of HVAC
increases. And it gives a lower one when the input power of
HVAC decreases. When the input power of HVAC increases
(or decreases) greatly, the HVAC is assigned more (or less)
power in STS than its actual demand. Thus the indoor
temperature is not completely the same as the one of
STSO-CPSO. As the relative error of (12) is small, the
difference of indoor temperature between STSO-CPSO and
MTSO-M12-CPSO is not obvious.

Fig. 8. Results of MTSO-M4-CPSO. (a) Power of EV in FTS and STS. (b)


Input power of HVAC in FTS and STS. (c) SOC of EV in FTS. (d) Temperature
in FTS.

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1

Fig. 9. Results of MTSO-M12-CPSO&PA. (a) Power of EV and input power


of HVAC in FTS. (b) SOC of EV in FTS. (c) Temperature in FTS.

Fig. 10. Impacts of iteration number on objective functions values.

We apply the proposed algorithm to solve the problem of


multi-time scale optimization. That is, according to the STS
solution of MTSO-M12-CPSO, we search the FTS input power
of HVAC of each hour based on the proposed algorithm. Fig. 9
shows the results of multi-time scale optimization with model
(12), its first step is based on CPSO, and second step is based on
the proposed algorithm (MTSO-M12-CPSO&PA). The indoor
temperature in fig. 9(c) is similar to the one of STSO-CPSO,
except the beginning minutes of each hour from 6:00 to 12:00
and from 16:00 to 24:00. Without iteration, the proposed
algorithm has not any comparison and adjustment; it can not
assign the input power of HVAC in FTS perfectly. As the STS
input power of HVAC is relatively large from 6:00 to 12:00 and

10
from 16:00 to 24:00, the error of proposed algorithm is
relatively obvious.
Fig. 10 shows the impacts of iteration number on objective
functions values. All the simulations are convergent with the
increase of iteration number. The convergent value of
MTSO-M12-CPSO is closer to the one of STSO-CPSO than the
one of MTSO-M4-CPSO, as model (12) is more precise than
model (4). As mention in Section IV, the proposed algorithm
gives an approximate solution, thus the convergent value of
MTSO-M12-CPSO&PA doesnt match the one of
STSO-CPSO.
We suppose the convergent value of STSO-CPSO is the
objective functions optimal value. The relative errors are
shown in Table II. STSO-CPSO directly searches the optimal
solution in a space that has 960 dimensions (there are one EV,
one HVAC, and 480 time slots in a day period). The relative
error of STSO-CPSO decreases slowly with the increase of
iteration number. In multi-time scale, there are 24 STS time
slots in a day period, so the solution of STS optimization has 48
dimensions. There are 20 FTS time slots in each hour, and the
FTS power of EV is given directly by (15). Therefore only the
FTS input power of HVAC should be searched in each hour,
and the solution of FTS optimization has 20 dimensions. The
FTS search should be repeated 24 times as there are 24 hours.
In multi-time scale optimization, there are 25 searches in total.
We assign 1/25 the iteration number to each search of
MTSO-M12-CPSO and MTSO-M4-CPSO. As the dimension
numbers of the 25 searches are all been reduced greatly, the
objective functions value of MTSO-M12-CPSO converges
faster than the one of STSO-CPSO, and the relative error of
MTSO-M12-CPSO decreases faster than the one of
STSO-CPSO in the early stage of iteration. MTSO-M4-CPSO
shows similar performance, but the relative error is greater than
the one of MTSO-M12-CPSO while their iteration numbers are
the same. As the 24 searches in FTS are all based on the
proposed
algorithm
that
contains
no
iteration,
MTSO-M12-CPSO&PA assigns all the iterations to the only
search of STS. After 2000 iterations, the relative error of
MTSO-M12-CPSO&PA
decreases
to
6.41%.
MTSO-M12-CPSO achieves similar performance after 50000
iterations, as each search is assigned 2000 iterations. The
proposed algorithm finishes all the 24 searches in FTS within
31 milliseconds in our simulation. When the time of
optimization is limited, the proposed algorithm gives a
relatively good solution.
TABLE II
RELATIVE ERRORS OF OBJECTIVE FUNCTIONS VALUES (A)
Iteration

STSOCPSO

MTSO-M1
2-CPSO

MTSO-M4
-CPSO

MTSO-M12CPSO&PA

2,000

88.26%

55.21%

68.45%

6.41%

50,000

45.34%

5.83%

9.62%

3.70%

1,000,000

12.23%

1.06%

6.31 %

3.66%

10,000,000

0.19%

0.44%

5.89%

3.65%

Coefficient g influences the relative errors of (4) and (12)


greatly, as shown in Table I. We increase | g | to test the

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1

11

relative errors of objective functions values again. Table III


shows the results when g is -2. The relative errors of MTSOM12-CPSO, MTSO-M4-CPSO, and MTSO-M12-CPSO&PA
are larger than the one of STSO-CPSO after convergence. And
the relative error of MTSO-M4-CPSO is the largest one as
model (12) is more precise than (4).
TABLE III
RELATIVE ERRORS OF OBJECTIVE FUNCTIONS VALUES (B)
Iteration

STSOCPSO

MTSO-M1
2-CPSO

MTSO-M4
-CPSO

MTSO-M12CPSO&PA

2,000

76.68%

62.52%

50.96%

14.51%

50,000

54.50%

12.74%

44.24%

10.18%

1,000,000

9.66%

7.25%

28.52 %

9.54%

10,000,000

0.33%

6.93%

27.77%

9.46%

VI. CONCLUSION
We propose a STS thermal model and analyze the
performance of the model in this paper. As the model considers
the changing outdoor temperature and input power of HVAC, it
is more suitable for the problems transformation from
single-time scale to multi-time scale. We show the conditions
that must be obeyed while transforming the problem into
multi-time scale, and test the conditions in simulation. We
propose an algorithm without iteration for the problem of FTS
HEM optimization. The solution of proposed algorithm that is
very close to the optimal one is obtained in a short time in
simulation.
In the paper, the coefficients of thermal model (3) and the
batterys efficiency of EV are constants, and the optimization
only considers the total electricity cost and user discomfort of
temperature. As future study, it is suggested to look into the
problem of transformation while the coefficients are variables,
and while the optimization has more considerations (such as the
stabilization of total power consumption of home, user
discomfort of ventilation).
REFERENCES
[1]C. Kang, Q. Chen and Q. Xia, Prospects of low-carbon electricity, Power
System Technology, vol. 33, no. 2, pp. 1-7, 2009.
[2]I. E. Agency. CO2 emission from fuel combustion Highlights. International
Energy Agency. OECD/IEA. Paris, France. [Online]. Available:
http://www.iea.org/publications/freepublications/publication
/CO2EmissionsFromFuelCombustionHighlights2013.pdf
[3]S. M. Amin and B. F. Wollenberg, Toward a smart grid: Power delivery for
the 21st century, IEEE Power Energy Mag., vol. 3, no. 5, pp. 34-41,
2005.
[4]U. S. Environment Protection Agency. [Online]. Available:
www.epa.gov/greenbuilding /pubs/gbstats.pdf.
[5]A. Y. Saber and G. K. Venayagamoorthy, One million plug-in electric
vehicles on the road by 2015, in Proc. IEEE Conf. Intell. Transp. Syst.,
2009.
[6]D. Bargiotas and J. Birdwell, Residential air conditioner dynamic model
for direct load control, IEEE Trans. Power Del., vol.3, pp. 2119-2126,
Oct, 1988.
[7]M. Tasdighi, H. Ghasemi, and A. Rahimi-kian, Residential microgrid
scheduling based on Smart meters Data and temperature dependent
thermal load modeling, IEEE Trans. Smart Grid, vol. 5, no. 1, pp.
349-357, 2014.
[8]D. T. Nguyen, L. B. Le. Joint optimization of electric vehicle and home
energy scheduling considering user comfort preference. IEEE Trans.
Smart Grid, vol. 5, no. 1, pp. 188-199, 2014.

[9]J. Siroky, F. Oldewurtel, J. Cigler, and S. Privara, Experimental analysis of


model predictive control for an energy efficient building heating system,
Appl. Energy, vol. 88, no. 9, pp. 3079-3087,Sep. 2011.
[10] R. Halvgaard, N. K. Poulsen, H. Madsen, and J. B. Jorgensen, Economic
model predicitive control for building climate control in a smart grid, in
Proc. IEEE PES Innov. Smart Grid Technol. (ISGT), Jan. 2012, pp. 1-6.
[11] Z. Yu, L. Jia, M. C. Murphy-Hoye, A.Pratt, L. Tong. Modeling and
stochastic control for home energy management, IEEE Trans. Smart
Grid, vol. 4, no. 4, pp. 2244-2255, 2013.
[12] C. M. Colson and M. Hashem Nehrir, Comprehensive real-time
microgrid power management and control with distributed agents, IEEE
Trans. Smart Grid, vol. 5, no.1, pp. 617-627, 2013.
[13] P. H. Nguyen, W. L. Kling, and P. F. Ribeiro, A game theory strategy to
integrate distributed agent-based functions in smart grids, IEEE Trans.
Smart Grid, vol. 5, no.1, pp. 568-576, 2013.
[14] P. Chavali, P. Yang, and A. Nehorai, A distributed algorithm of
appliance scheduling for home energy management system, IEEE Trans.
Smart Grid, vol.5, no.1, pp. 282-290, 2014.
[15] C. Dou and B. Liu, Multi-agent based hierarchical hybrid control for
smart microgrid, IEEE Trans. Smart Grid, vol. 4, no. 2, pp.771-778,
2013.
[16] B. Jiang and Y. Fei, Smart home in smart microgrid: a cost-effective
energy ecosystem with intelligent hierarchical agents, IEEE Trans.
Smart Grid, vol. 6, no. 1, pp.3-13, 2015.
[17] M. H. K. Tushar, C. Assi, M. Maier, M. F. Uddin. Smart Microgrids:
Optimal Joint Scheduling for Electric Vehicles and Home Appliances,
IEEE Trans. Smart Grid, vol. 5, no. 1, pp. 239-250, 2014.
[18] M. Ghofrani, A. Arabali, M. Etezadi-Amoli, M. S. Fadali. Smart
scheduling and cost-benefit analysis of grid-enabled electric vehicles for
wind power integration, IEEE Trans. Smart Grid, vol. 5, no. 5, pp.
2306-2313, 2014.
[19] Y. Wang, X. Lin, M. Pedram. Adaptive control for energy storage
systems in households with photovoltaic modules, IEEE Trans. Smart
Grid, vol. 5, no. 2, pp. 992-1001, 2014.
[20] L. Sun, Z. Yang, Optimal charging schedule for PHEV taxies with
time-varying prices, Journal of University of Science and Technology of
China, vol. 42, no. 9, pp. 746-752, 2012.
[21] R. E. Bellman, Dynamic Programming. Princeton, NJ: Princeton
University Press, 1957.
[22] C. Wang, Z. Wu, and P. Li, Research on key technologies of microgrid,
Transactions of china electrotechnical society, vol. 29, no. 2, pp. 1-12,
2014.
[23] L.Jia, Z. Yu, M. C. Murphy-Hoye, A. Pratt, E. G. Piccioli, and L. Tong,
Multi-scale stochastic optimization for home energy management, in
Proc. 4th IEEE Int. Workshop Comput. Adv. Multi-Sensor Adaptive
Process., Dec. 2011, pp. 113-116.
[24] Z. Yu, L. Mclaughlin, L. Jia, M. C. Murphy-Hoye, A. Pratt, and L.Tong,
Modeling and stochastic control for home energy management, in Proc.
IEEE PES Gen. Meet., Jul. 2012.
[25] D. Revuz and M. Yor, Continuous martingales and Brownian motion.
Berlin, Germany: Springer, 1999, pp. 15-48.
[26] S. Roy, Market constrained optimal planning for wind energy
conversion systems over multiple installation sites, IEEE Trans. Energy
convers., vol. 17, no. 1, pp. 124-129, Mar. 2002.
[27] M. K. C. Marwali, S. M. Shahidehpour, and M. Daneshdoost,
Probabilistic production costing for photovaltaics-utility systems with
battery storage, IEEE Trans. Energy Convers., vol. 12, no. 2, pp.
175-180, Aug. 2002.
[28] F. V. D. Bergh and A. P. Engelbrecht, A cooperative approach to particle
swarm optimization, IEEE Trans. Evol. Computation, vol. 8, no. 3, pp.
225-239, 2004.

Yi Liu received the B.S. and M.S. degree


in automatic engineering from the South
University
of
Technology,
China
Guangzhou, China, in 2002 and 2005. He is
currently working toward the Ph.D. degree
in control theory and control engineering in
the Guangdong University of Technology,
Guangzhou, China.
He is currently an associate professor

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2015.2510222, IEEE
Transactions on Smart Grid

TSG-00706-2015.R1
with the College of Information Science and Technology,
Zhongkai University of Agriculture and Engineering,
Guangzhou. His research interests include intelligent control
and optimal control.
Yun Zhang received the B.S. and M.S.
degrees in automatic engineering from the
Hunan University, Changsha, China, in
1982 and 1986, respectively, and the Ph.D.
degree in automatic engineering from the
South China University of Technology,
Guangzhou, China, in 1998.
He is currently a Professor with the
School of Automation, Guangdong University of Technology,
Guangzhou. His research interests include intelligent control
systems, network systems, and signal processing.
Kairui Chen received the B.S. degree in
network engineering from Guangdong
Univerity of Technology, Guangzhou,
China, in 2012, and the M.S. degree in
control science and engineering from
Guangdong Univerity of Technology, in
2014. Currently, he is a Ph.D. candidate
with the School of Automation,
Guangdong Univerity of Technology, Guangzhou, China. His
research interests include multi-agent system control, adaptive
control and optimal control.

12

Sizhe Chen was born in Shantou,


Guangdong, China, in 1981. He received
the B.Sc. and Ph.D. degrees from the
South China University of Technology,
Guangzhou, China, in 2005 and 2010,
respectively. Dr. Chen currently is a
Associate Professor in the School of
Automation, Guangdong University of
Technology, Guangzhou, China. His
general research interests include the control and power
electronics technology in renewable energy.
Bin Tang received the B.S. degree in
measurement and control technology and
instrument from Xiangtan University,
Xiangtan, China, in 2001, the M.S. degree
in control theory and control engineering
and the Ph.D. degree in control science
and engineering from Central South
University, Changsha, China, in 2004 and
2008, respectively.
He is currently an Associate Professor with the School of
Automation, Guangdong University of Technology,
Guangzhou, China. His research interests include networked
control systems, sampled-data control systems, and delay
systems.

1949-3053 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

S-ar putea să vă placă și