A Cooperative Games Approach For Demand Side Management in Smart Grids

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/330798581

A Cooperative Games Approach for Demand Side Management in Smart Grids

Conference Paper · February 2019

CITATIONS READS
0 77

2 authors:

Luciana Marques Wadaed Uturbey


Federal University of Minas Gerais Federal University of Minas Gerais
1 PUBLICATION 0 CITATIONS 36 PUBLICATIONS 178 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Management of Distributed Energy Resources in the context of Smart Grids: a game theoretic approach. View project

Integration of Distributed Energy Resources View project

All content following this page was uploaded by Luciana Marques on 01 February 2019.

The user has requested enhancement of the downloaded file.


THE 12th LATIN-AMERICAN CONGRESS ON ELECTRICITY GENERATION AND TRANSMISSION - CLAGTEE 2017 1

A Cooperative Games Approach for Demand Side


Management in Smart Grids
L. S. M. Arnoux, and W. Uturbey

Abstract—Demand-Side Management (DSM) programs con- y Auxiliary variable for peak load (peak pricing
sider the energy consumers as intelligent agents and as key model).
players for system management, gaining importance on the Smart zt Auxiliary variable for total cost in inclining block
Grid era. In this context, we propose a cooperative approach rates model.
for the decision-making process of an aggregator that represents
a group of residential consumers in a DSM program. The
aggregator participates in a day-ahead market and has to decide Constants:
the best load curves for their consumers in terms of energy costs. αn,a Earlier starting time of controllable appliance a of
Our framework is divided in 3 steps: data submission, scheduling consumer n.
model and cost sharing. In the first one, consumers send to the βn,a Later finishing time of controllable appliance a of
aggregator their consumption preferences. With that information, consumer n.
the aggregator schedules consumers’ appliances, respecting their δ Size of time slot (in hours).
preferences and minimizing the total DSM cost, using a mixed-
integer scheduling model. In the last step, the aggregator attempts
ηt (Pt ) Inclining block rates pricing model as a function
to allocate total program cost to the participants in a fair manner, of power curve, in $/kW h.
using the Shapley Value. We test our framework in two random λ Peak load charge in $/kW (peak pricing model).
instances, with different appliances, preferences and number of πt Time-of-use tariff at time slot t in $/kW h.
consumers and verify that more flexible participants contribute ρt Price multiplier for when energy consumption ex-
more to the coalition value, and so forth pay less. Moreover, the ceed the threshold (inclining block rates model).
peak pricing and inclining block rates models have the capacity dn,a Duration of appliance a of consumer n (in number
to reduce peak-to-average ratio of the system. of time slots).
Keywords—Cooperative Games, Demand-Side Management, Ln,a Power rate of appliance a of consumer n.
Load Scheduling, Residential Power Management, Shapley Value. Rt Energy consumption threshold, in kW h (inclining
block rates model).
T Last slot of time horizon.
I. N OMENCLATURE wn,t Base load of consumer n at time slot t.
Sets and indices:
II. I NTRODUCTION
a Index for appliances.
n
t
Index for consumers.
Index for time slots. T HE introduction of the smart grid concept brought new
solutions to electric grid planning, management and con-
trol. It employs advanced communication systems and modern
An Set of all shiftable appliances of consumer n.
C Set of consumers in the coalition. computational technologies to increase grid’s efficiency and
In Set of interruptible appliances of consumer n. autonomy.
N Set of consumers. One of its aspects is demand side management (DSM),
T Set of time slots. defined as everything made at consumer side. DSM programs
Un Set of uninterruptible appliances of consumer n. consider the energy consumers as intelligent agents and as
key players for system management. Including consumers in
the operation and planning decision taken process of electrical
Variables and functions: systems can increase demand elasticity and add another degree
φn Payoff of consumer n. of freedom to grid operation and maintenance. Therefore, the
Pt Power curve of the consumers in time slot t. development of DSM models can be a competitive solution
v(C) Value of a coalition C. to consumption growth compared to building new generation
v(N ) Value of the grand coalition N . plants, installing storage systems or expanding transmission
xn,a,t Binary scheduling variable: is 1 if appliance a of and distribution network.
consumer n is on at time t and 0 otherwise. However, designing, implementing and operating DSM pro-
grams is not a trivial task and there is an extensive literature
This work has been supported by the Brazilian agency CAPES about this challenge, as can be seen in [1]. Pricing methods,
The authors are with the Graduate Program in Electrical Engineering -
Federal University of Minas Gerais - Av. Antônio Carlos 6627, 31270- as time-of-use tariffs or real-time pricing, are the most widely
901, Belo Horizonte, MG, Brazil (e-mail: lusantamarques@gmail.com and used mechanisms for implementing demand participation.
wadaed@cpdee.ufmg.br). They penalize certain peak periods of time with higher prices,
THE 12th LATIN-AMERICAN CONGRESS ON ELECTRICITY GENERATION AND TRANSMISSION - CLAGTEE 2017 2

guiding consumers’ load curves. Even though they are easy devices of multiple consumers and to bring the load consump-
to implement, they depend on consumers’ disposal to change tion curve as close as to the objective load consumption curve.
their household routines, which is a challenging task [2]. [9] proposes a controlling strategy for a PHEV aggregator,
Moreover, it is difficult to consumers to respond manually to which schedules their batteries charging times considering the
the time-varying prices [3]. Finally, they can create new peaks local marginal prices (LMP) of the grid. Even though they
in cheaper hours [1]. can provide grid and market advantages, these models are
Direct load control (DLC) is another extensively studied more complex and difficult to solve, because they have more
method. It adds technology to the demand management pro- variables to be coordinated. Generally, they are solved using
cess, helping consumers to benefit from DSM advantages with- heuristic methods [8], [9].
out having to continuously analyze prices and consumption to When the control technology is centralized, guided by an
decide whether to turn on the television or prepare dinner. aggregator or the DSO, the controller challenge is to esti-
The control technology can be either real-time or solved for mate correctly their clients demand and contract the predicted
predefined time horizons, mainly the day-ahead. amount or bid ancillary services at the market. Moreover, the
In the first case, [2] develops an online and real-time control controller must charge their clients correctly. As the consumers
algorithm that optimizes the loads of a smart home as soon are rational and independent, with preferences and priorities,
as they are started. It analyses all the equipment status and game theory offers an interesting analytical and conceptual
modifies background electrical loads consumption, which do framework to deal with the study of their complex interactions
not impact users perceived comfort and have some scheduling [10].
flexibility (e.g. fridge and heaters). The objective is to flatten In this context, some researches use game theory in the
demand and results are interesting from the point of view of development of demand side management programs. For ex-
the overall grid. ample, [11] proposes a compensation model to consumers of
On the other hand, some control techniques work in an a DSM program using Shapley Value. The aggregator bids
offline basis, programming the loads in a future time horizon, spinning reserve offers in the market and consumers cut load
more frequently the day-ahead. In this category, there are when the reserve is triggered. They are fairly payed for their
algorithms for the optimal control of one load of a smart response. [12] and [13] also present scheduling load models for
home (e.g. water heater) [4], [5]. They can model the load in consumption optimization. The first one provides a model for
detail, including comfort constraints (e.g. minimum accepted an aggregator who needs to optimize the energy consumption
temperature of water) and non-linear behavior (e.g. thermal of a group of smart homes and to decide how much power
dynamics). However, these models do not consider global must be purchased from the market the next day. It shares
information about consumers’ consumption in general, being the cost between participants using a collaborative approach.
sub-optimal in the total energy cost minimization of a smart On the other hand, the second considers a competitive game
home. theory environment and Nash Equilibrium. This approach is
Another group of applications focus on scheduling all loads interesting because it decentralizes the controlling algorithms,
of a smart home or an industry. [6] proposes an optimal and and each consumer takes his decision considering information
automatic load control model for a residential consumer, which from the others until the equilibrium is reached. Finally,
schedules all the responsive appliances to minimize the total [14] focuses on Shapley Value aspects and implement this
cost of energy and the waiting time for the operation of each methodology to demand response program and to distributed
device. The pricing model guiding the optimization is the day- generation compensation.
ahed real-time prices (RTP) with inclining block rates (IBR) Given this background, we propose a cooperative approach
in order to result in a more balanced load curve. [3] includes for the decision-making process of an aggregator that repre-
plug-in electric vehicles (PHEVs) and storage systems to the sents a group of residential consumers in a DSM program.
model. The pricing environment uses the time-of-use tariff and The aggregator participates in a day-ahead market and has to
to cap the peak consumption, the model limits the energy decide the best load curves for their consumers in terms of
transfer. [7] design an scheduling model for the industry, energy costs. We propose a mixed-integer scheduling model
aiming to maximize profit, and consider many peculiarities of which respects consumers preferences. The total cost is divided
the sector. These three models are solvable with linear/integer between the participants using Shapley Value.
programming solvers given that they deal with a reduced Our contributions are as follows: we propose a new schedul-
number of loads, in other words, variables. Even though they ing model based on network optimization concepts to reduce
try to reduce the peak-to-average ratio (PAR) by means of the the total consumption cost of a consumer’s group respecting
pricing model, they do not consider the optimization of a group their preferences. We use a new model to divide the total cost
of consumers or the communication between them. between the participants, using cooperative game theory logic.
Some DLC programs assemble consumers and manage their We test our framework in two random instances, based on real
loads using a centralized approach. An aggregator or the data, which enables an extensive analysis.
distribution system operator (DSO) is responsible of a group The proposed approach has two advantages compared to
of consumers and control their consumption pattern in order the majority of previous research: it allows small consumers to
to reach some objective, like reduce their bills, reduce the participate in the wholesale market via the representation of an
network PAR, provide ancillary services to the grid etc. In aggregator, which is responsible for the optimal management
this framework, [8] develops a model to schedule the shiftable of their loads; and it reduces the total PAR of the group,
THE 12th LATIN-AMERICAN CONGRESS ON ELECTRICITY GENERATION AND TRANSMISSION - CLAGTEE 2017 3

includes loads that can be shifted in time to reduce consumers’


energy bills. They are further classified in loads that can be
interrupted (set In ), e.g. electric vehicles and air conditioning,
and those that can not (set Un ), e.g. washing machine and
dishwasher. We define the set An = In ∪Un of all controllable
loads of consumer n. The groups of appliances are indexed in
n because homes can have different sets of equipment.
The binary decision variables xn,a,t of the optimization
model define whether the controllable appliance a ∈ An of
consumer n ∈ N is on at time t ∈ T = {1, 2, . . . , T }, where
T is the day-ahead scheduling horizon set (24 hours divided
in time slots of size δ).
The scheduling process of the smart home loads must
respect consumers preferences. According to the next day
needs, each consumer n defines the earlier starting time αn,a
and the later finishing time βn,a of each controllable appliance
a ∈ An . He also specifies the appliances’ required duration
(dn,a ). Equations (1) and (2) represent these preferences.
βn,a
X
xn,a,t = dn,a ∀n ∈ N ∀a ∈ An (1)
t=αn,a
Fig. 1. A simplified illustration of the market context considered in this paper X
xn,a,t = 0 ∀n ∈ N ∀a ∈ An (2)
t<αn,a
minimizing some operational costs of the DSO. t>βn,a

The rest of this paper is organized as follows. We present Some loads, as discussed previously, are uninterruptable,
the market context and optimization model in Section III. which means once they started, they can not be stopped. We
We explain the Shapley Value concepts applied to the cost define equation (3) to guarantee this behavior.
sharing between participants in Section IV. We introduce a
simple example to understand our approach in Section V. The
simulation results are presented in Section VI. The paper is xn,a,t − xn,a,t+1 + xn,a,t+k ≤ 1 ∀n ∈ N
concluded in VII. ∀a ∈ An
(3)
∀k = 2, . . . , T − 1
III. S CHEDULING M ODEL ∀t = αn,a , . . . , βn,a − k
In this section, we present the mathematical formulation of If we consider Ln,a as the rated power of consumer’s
the optimal residential power scheduling problem. We consider appliances, the power curve of the group of participants can
the market environment shown in Fig. 1, where an aggregator be defined as:
serves a number of consumers interested in the demand side
management program. It receives the day-ahead time-of-use
!
X X
tariffs with peak prices or inclining block rates from the Pt = wn,t + Ln,a xn,a,t ∀t ∈ T (4)
wholesale market and the consumption preferences from the n∈N a∈An
consumers via a communication infrastructure. With that in-
formation, it performs a mixed-integer scheduling optimization
B. Peak Pricing Model
and sends back control messages to the smart loads. The
objective is to schedule the controllable loads in order to Generally, this pricing model is composed by two rates:
minimize participants energy bills, subjected to consumers usage charge and peak load charge [7]. We adopt time-of-use
preferences. tariffs for the usage charge to represent some electricity cost
variation in a day. They are energy prices and are denoted by
πt $/kW h. The peak load charge depends on the consumers’
A. Residential Consumers daily highest load. For this reason, it flattens the load curve and
Consider a set of consumers N represented by the ag- reduces the peak-to-average ratio. The peak price is a capacity
gregator in the wholesale market. Each consumer n ∈ N cost and is represented by λ $/kW . The total electricity cost to
has some household appliances classified into two categories: be minimized when the wholesale market adopts this pricing
uncontrollable or controllable. The first one represents non- police is:
shiftable loads, e.g. refrigerator, freezer and TV. Their power X
consumption are summed in one base load curve for each min πt δPt + λ max Pt (5)
consumer, denoted by the power vector wn . The latter group x t∈T
t∈T
THE 12th LATIN-AMERICAN CONGRESS ON ELECTRICITY GENERATION AND TRANSMISSION - CLAGTEE 2017 4

The objective function on equation (5) is subjected to IV. C OST S HARING


constraints (1)-(4). To solve this problem in a linear fashion
(see [12]), we can rewrite (5) as:
After the load scheduling optimization, the total cost of the
DSM program must be shared between the participants. We
X
min πt δPt + λy (6)
x
t∈T
use Shapley Value to do this task in a fair manner.
Consider the set C ⊆ N as a coalition of consumers inside
Where δ is the size of the time slot and y is a positive
the group of participants represented by the aggregator. We
variable representing the peak load, and is subjected to:
can define a characteristic function v(C) as the result of the
scheduling problem presented in section III for the group of
y ≥ Pt ∀t ∈ T (7)
consumers C.
Because we consider the group of consumers as a single The aggregator want to split the total DSM program cost
market agent represented by the aggregator, who purchases the v(N ) between their clients. This must be done in a fair manner,
energy needed in the day-ahead market, the peak load charge according to each participant contribution to the program. We
depends on the total load curve of all participants. This idea define φn the payoff of consumer n. Some conditions must be
is interesting to the grid control because it reduces the overall respected for what the calculation of φn is fair and interesting
PAR of the group. However, it can be noted that equation to the participants:
(7) couples all participants of the group, precluding the use
of a decentralized method of resolution. Without it, we could • Efficiency: the total payoff must be divided between the
separate the centralized aggregator problem in n smaller and participants, respecting the equation below:
easier to solve models, as in [3]–[7]. Nevertheless, as said
before, without a central controller to guide the process, the X
grid would not benefit from PAR reduction. v(N ) = φn (13)
n∈N

C. Inclining Block Rates Model


Inclining block rates (IBR) is another pricing method that • Symmetry: two participants which contribute equally to
promotes more balanced load curves [7]. In IBR, the energy the overall value must perceive equivalent payoffs;
price increases when total load exceeds a threshold. Con- • Null player: participants which do not contribute to the
sidering a context with TOU tariffs denoted by πt $/kW h total payoff of the group must have φn = 0;
(as in PP) and a price multiplier ρt for when the group • Linearity: the total value generated by two participants
consumption surpasses the limit Rt , we can write this pricing must be equal to the sum of their payoffs:
model mathematically as:

πt , if 0 ≤ Pt ≤ Rt v(a + b) = φa + φb ∀a, b ∈ N (14)
ηt (Pt ) = (8)
πt ρt , if Pt ≥ Rt
The total electricity cost to be minimized when the whole- Shapley proposed a method to share the payoff between
sale market adopts this pricing police is: the participants that respects all the conditions above with the
X additional property of being unique [15]. His method is based
min ηt (Pt )δPt (9) on the contribution of each player to all possible coalitions
x
t∈T C ⊆ N and can be calculated by equation:
Although the optimization function is not linear, it is pos-
sible to reformulate it and solve the resulting model using
commercial solvers. We can decompose the IBR pricing func- X |C|! (|N | − |C| − 1)!
tion in two intersecting lines and use auxiliary variables zt to φn = [v(C ∪ {n}) − v(C)]
|N |!
represent the total cost. For more details, see [6]. The problem C⊆N \{n}
(9) can be rewriten as: (15)
X It is worth mentioning that, when the payoff is divided using
min zt (10) the Shapley Value, no participant has the incentive to quit the
x
t∈T grand coalition N . Therefore, this equation is used to identify
Where zt are positive variables representing the total cost, the contribution of each participant to the DSM program and
and are subjected to: to help the aggregator to calculate how much each consumer
must pay. Since the calculation of the coalition value v(C)
zt ≥ πt Pt ∀t ∈ T (11) is computationally complex, because it is the result of the
combinatorial optimization problem proposed in section III,
we use the algorithm 1 from [14] to calculate each coalition
zt ≥ πt ρt Pt + πt (1 − ρt ) Rt ∀t ∈ T (12) value only once.
THE 12th LATIN-AMERICAN CONGRESS ON ELECTRICITY GENERATION AND TRANSMISSION - CLAGTEE 2017 5

3
slot of the next day. Furthermore, he uses his AC at morning.
Consumer 2 has less flexibility and uses the washing machine
2.5
at morning, the dryer and AC at afternoon and the dishwasher
during the day. Consumer 3 has the least flexibility and prefers
2 to turn on all the appliances at evening.
base load (kW)

We use hypothetical time-of-use tariffs and peak charge.


1.5
We divide the 24 hours in six time intervals—see table II.
Moreover, the peak charge is 0.070 $/kW.
The results are shown in table III. The Shapley Value com-
1
putation for each participant, from equation (15), is presented
on (17)-(19). As it was expected, agent 1 pays less than the
0.5 others, because his appliances can be scheduled at almost any
time of the day-ahead, giving flexibility to the aggregator. On
0 the contrary, consumer 3 has to use them at evening, the most
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 expensive time interval, what makes him pay more.
time slot

Fig. 2. Base load of each consumer 1 × (5.47 − 0) 1 × (11.08 − 5.62)


φ1 = +
3 6
Algorithm 1 Algorithm for Shapley Value calculation 1 × (11.96 − 6.56) 1 × (17.56 − 12.10) (17)
+ +
Input: Set of consumers N and characteristic function v() 6 3
= 5.45 $
Output: Shapley Value vector SV
1: SV (n) ← 0, ∀n ∈ N ;
2: for all C ⊆ N do 1 × (5.62 − 0) 1 × (11.08 − 5.47)
3: coef W henAgentIn ← ω (max (|C| − 1, 0)); φ2 = +
3 6
4: coef W henAgentOut ← −ω (min (|C|, |N − 1|)); 1 × (12.10 − 6.56) 1 × (17.56 − 11.96) (18)
5: for all n ∈ N do + +
6 3
6: if n ∈ C then = 5.60 $
7: SV (n) ← SV (n) + coef W henAgentIn × v(C);
8: else
9: SV (n) ← SV (n) + coef W henAgentOut × v(C); 1 × (6.56 − 0) 1 × (11.96 − 5.47)
φ3 = +
10: end if 3 6
11: end for 1 × (12.10 − 5.62) 1 × (17.56 − 11.08) (19)
+ +
12: end for 6 3
13: return SV = 6.51 $
Figure 3 presents the schedule of the controllable loads
Where the function ω(x) is given by equation: of each consumers and the TOU tariffs on the right axis.
x!(|N | − x − 1)! As discussed previously, consumer 3 has strict preferences,
ω(x) = (16) represented by shorter and more expensive time intervals.
|N |! On the other hand, consumer 1 has more flexibility and the
aggregator can place his loads on cheaper time slots.
V. E XAMPLE
We apply our approach in a simple example to illustrate
VI. S IMULATION R ESULTS
how it works. Consider the case of an aggregator managing the
energy consumption of a group of 3 agents (N = {1, 2, 3}) in For a more extensive analysis, we test our approach on a
a wholesale market which adopts peak pricing method. They set with seven consumers, where the parameters are generated
have the same base load curve, as can be seen in figure 2. randomly from [12]. This results in the calculation of 128
The uncontrollable loads are: water heater, lighting, electric coalition values for each pricing method (2|N | subgroups). We
kitchen, fridge, freezer, electric oven, microwave, TV, desktop use the terms payoff, cost and payment as synonyms, because
computer and laptop. Their consumption patterns, including the Shapley Value is equal to the payment allocated to each
rated power and time of use, were generated from [12]. consumer in our context.
Each consumer has 3 uninterruptible loads (washing ma- The peak prices considered in this scenario are the same
chine, dryer and dishwasher) and one interruptible load (air presented on the table II. For the inclining block rates, we also
conditioning). Their preferences for each equipment as well use the time-of-use tariffs aforementioned for when the total
as the rated powers and the durations are presented on table consumption is smaller than the threshold (πt ), and a price
I. We consider a scenario where consumer 1 is more flexible multiplier ρt = 1.2 for all time horizon (see equation (8)). At
and can have his uninterruptible loads programmed at any time first, we use a threshold Rt equal to the peak load generated
THE 12th LATIN-AMERICAN CONGRESS ON ELECTRICITY GENERATION AND TRANSMISSION - CLAGTEE 2017 6

TABLE I. P REFERRED TIME SLOTS FOR STARTING TIME αn,a , FINISHING TIME βn,a , DURATION dn,a AND POWER RATE Ln,a OF EACH APPLIANCE

Power Rate Consumer 1 Consumer 2 Consumer 3


Load Duration
(kW) Starting Time Finishing Time Starting Time Finishing Time Starting Time Finishing Time
WASHING 12 0.19 1 144 36 72 114 132
DRYER 9 1.24 1 144 78 108 114 132
DISHWASHER 12 0.66 1 144 36 108 114 132
AC 12 1.50 36 72 78 108 114 132

TABLE II. T IME - OF - USE TARIFFS USED TO TEST THE APPROACH TABLE IV. C OMPARISON BETWEEN THE CONSUMERS ’ PAYOFF WHEN
MARKET ADOPTS PEAK PRICING (PP) OR ICLINING BLOCK RATES (IBR)
Period Time Interval Price ($/kWh)
Dawn 00:00 to 06:00 0.160 Payoff φn (%)
Consumer n
Morning 06:00 to 11:00 0.176 PP IBR Flat Rate
Lunch time 11:00 to 13:00 0.208 1 15.72 15.75 15.50
Afternoon 13:00 to 17:00 0.192 2 13.52 13.57 12.86
Evening 17:00 to 22:00 0.320 3 12.44 12.47 12.66
Night 22:00 to 24:00 0.192 4 17.75 17.75 17.81
5 16.64 16.53 17.98
6 16.40 16.35 15.70
TABLE III. C OALITION VALUE v(C) OF ALL POSSIBLE COALITIONS
C⊆N 7 7.55 7.58 7.49
Total (%) 100.00 100.00 100.00
Coalition (C) Cost (v(C)) $ Total ($) 20.32 19.77 -
{} 0
{1} 5.47
{2} 5.62
{3} 6.56 notice the importance of consumption pattern and preferences
{1,2} 11.08 on the sharing. In a flat rate scenario, the consumers would
{1,3} 11.96 pay the share presented on the last column (calculated from
{2,3} 12.10 day-ahead total load). However, with the pricing models and
{1,2,3} 17.56 the optimization procedure, consumers’ share increases or
decreases depending on their controllable loads preferences
0.8 0.4 and on base load pattern. As an example, consumer 7, who do
Consumer 1
energy consumption of controllable loads (kWh)

Consumer 2
not have any controllable load, has to pay a larger share in the
0.7 0.35
Consumer 3 collaborative scenarios.
TOU tariff
0.6 0.3 In table V we present the consumers’ payoffs when they
optimize their consumption alone (decentralized case) and
TOU tariffs ($/kWh)

0.5 0.25 together in the grand coalition via the aggregator (centralized
case). It can be seen that all consumers reduce their electricity
0.4 0.2 payment when optimizing their loads together, for the peak
pricing model. Even consumer 7, who do not have controllable
0.3 0.15
loads, enjoys from the peak load reduction. However, for the
IBR pricing model, the outcomes are almost the same for
0.2 0.1
both cases and there is no advantage to consumers to optimize
0.1 0.05
consumption in the group. It is important to emphasize that the
threshold Rt for every coalition on the optimization procedure
0 0 is equal to the optimal peak load y of the peak pricing model
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 for the same coalition. This may justify the equal results in
time slot centralized/decentralized cases in IBR scenario, because the
Fig. 3. Energy consumption of consumers’ controllable loads in kWh with
value of the boundary is optimal.
price curve of TOU tariffs In terms of peak-to-average ratio—see eq. (20), Fig. 4 shows
that, in both models, the peak load is reduced when a central
controller optimizes consumers’ appliances (centralized curve),
from the peak pricing optimization (for each coalition) to compared to the case where each consumer optimizes his
compare the pricing methods efficiency on reducing PAR. consumption alone and the resulting power curves are summed
The consumers’ payoff depends on their total load on the (decentralized curve). In the peak pricing model, the PAR is
day and on their preferences. As can be seen on table IV, reduced from 2.29 to 1.95. In the second model, it is reduced
results of peak pricing and inclining block rates scenarios are from 2.34 to 1.95. The PAR is the same in centralized scenario
almost the same in terms of total cost split. It is interesting to for both cases.
THE 12th LATIN-AMERICAN CONGRESS ON ELECTRICITY GENERATION AND TRANSMISSION - CLAGTEE 2017 7

TABLE V. C ONSUMERS ’ PAYOFF WHEN OPTIMIZING THEIR


CONSUMPTION ALONE VERSUS IN THE GRAND COALITION
1) larger than the peak load from the solution of the peak
pricing optimization;
PP Payoff ($) IBR Payoff ($) 2) equal to the mean consumption resulting from the peak
Consumer n
Alone In group Alone In group pricing solution;
1 3.25 3.19 3.12 3.11
2 2.80 2.75 2.68 2.68
In both cases the PAR increased to 2.09 and 2.18, respec-
3 2.60 2.53 2.47 2.47
tively. Therefore, the choice of the threshold value is very
4 3.67 3.61 3.51 3.51
important to IBR efficiency. If it is too large or too tiny, the
5 3.43 3.38 3.27 3.27
model does not have any incentive to optimize the peak-to-
6 3.37 3.33 3.23 3.23
average ratio.
7 1.61 1.53 1.50 1.50 We can conclude that peak pricing model focuses on PAR
minimization because it penalizes the peak consumption, what
Peak Pricing causes demand flattening. Thereby, in a centralized approach,
12
this pricing model has more degrees of freedom to allocate
centralized
descentralized
loads, which results in an always win solution, from con-
10 sumers’ point of view. However, the optimization is more
costly in terms of running time.
8
On the other hand, inclining block rates can reduce the peak,
but it depends largely on the choice of price multiplier ρt
load (kW)

and the threshold Rt . It has the advantage to generate good


6
results in a decentralized scenario and to be faster in terms
of running time, although the optimal PAR is always given
4 in the centralized optimization. Therefore, it can be a good
decentralized control option.
2

VII. C ONCLUSION
0
0 50 100 150 In this paper, we proposed an optimal, autonomous and
time slot centralized load scheduling model in order to minimize the
total electricity cost and also flatten consumption pattern.
Inclining Block Rates
12 Unlike most of the previous DSM strategies that focus on the
centralized optimization of each smart home independently, our approach
descentralized
10
considers a group of consumers represented by an aggregator
and the attempt to also minimize the system PAR, allowing
small consumers to participate in the wholesale market and
8 minimizing some operational costs of the DSO. Simulation
results confirm that the proposed approach can reduce con-
load (kW)

6 sumers’ energy cost and the PAR.


As future work, it is necessary to develop methods to reduce
4
the running time of the optimization model. In addition, the
Shapley Value calculation is a combinatorial problem and it
should be necessary to implement other calculation methods.
2
Moreover, it should be interesting to seek for more cooperative
approaches to decentralize the scheduling optimization, avoid-
0 ing computational complexity at the same time as reinforcing
0 50 100 150
time slot
information security. Finally, it would be interesting to test the
proposed approach in real data from Brazilian consumption
Fig. 4. Total load curves of consumers when they optimize their consumption patterns.
alone - and they are summed (blue) and in the coalition (red): peak pricing
model (top); inclining block rates model (bottom)
ACKNOWLEDGMENT
The authors would like to thank the Brazilian agency
CAPES for the financial support.
max(Pt )
P AR = 1
P (20)
T t∈T Pt
R EFERENCES
To analyze the impact of threshold choice on the efficiency [1] P. Palensky and D. Dietrich, “Demand side management: Demand re-
of inclining block rates model, we solve the optimization sponse, intelligent energy systems, and smart loads,” IEEE transactions
model with another two values for Rt : on industrial informatics, vol. 7, no. 3, pp. 381–388, 2011.
THE 12th LATIN-AMERICAN CONGRESS ON ELECTRICITY GENERATION AND TRANSMISSION - CLAGTEE 2017 8

[2] S. Barker, A. Mishra, D. Irwin, P. Shenoy, and J. Albrecht, “Smartcap: Wadaed Uturbey received her degree in Electri-
Flattening peak electricity demand in smart homes,” in Pervasive cal Engineering from Universidad de La República,
Computing and Communications (PerCom), 2012 IEEE International Uruguay, in 1991, the M.Sc. and Dr. Eng. degree
Conference on. IEEE, 2012, pp. 67–75. from Federal University of Santa Catarina, Brazil,
[3] M. Rastegar, M. Fotuhi-Firuzabad, and F. Aminifar, “Load commitment in 1995 and 2002, respectively. She is currently an
in a smart home,” Applied Energy, vol. 96, pp. 45–54, 2012. associate professor at the Department of Electrical
Engineering of the Federal University of Minas
[4] P. Du and N. Lu, “Appliance commitment for household load schedul- Gerais, Brazil. Her research interests include proba-
ing,” IEEE transactions on Smart Grid, vol. 2, no. 2, pp. 411–419, bilistic and optimization methods for power systems
2011. planning and energy markets.
[5] C. Wang, Y. Zhou, J. Wang, and P. Peng, “A novel traversal-and-pruning
algorithm for household load scheduling,” Applied Energy, vol. 102, pp.
1430–1438, 2013.
[6] A.-H. Mohsenian-Rad and A. Leon-Garcia, “Optimal residential load
control with price prediction in real-time electricity pricing environ-
ments,” IEEE transactions on Smart Grid, vol. 1, no. 2, pp. 120–133,
2010.
[7] A. Gholian, H. Mohsenian-Rad, and Y. Hua, “Optimal industrial load
control in smart grid,” IEEE Transactions on Smart Grid, vol. 7, no. 5,
pp. 2305–2316, 2016.
[8] T. Logenthiran, D. Srinivasan, and T. Z. Shun, “Demand side manage-
ment in smart grid using heuristic optimization,” IEEE transactions on
smart grid, vol. 3, no. 3, pp. 1244–1252, 2012.
[9] D. Wu, D. C. Aliprantis, and L. Ying, “Load scheduling and dispatch for
aggregators of plug-in electric vehicles,” IEEE Transactions on Smart
Grid, vol. 3, no. 1, pp. 368–376, 2012.
[10] W. Saad, Z. Han, H. V. Poor, and T. Basar, “Game-theoretic methods
for the smart grid: An overview of microgrid systems, demand-side
management, and smart grid communications,” IEEE Signal Processing
Magazine, vol. 29, no. 5, pp. 86–105, 2012.
[11] G. O’Brien, A. El Gamal, and R. Rajagopal, “Shapley value estimation
for compensation of participants in demand response programs,” IEEE
Transactions on Smart Grid, vol. 6, no. 6, pp. 2837–2844, 2015.
[12] M. Vasirani and S. Ossowski, “A collaborative model for participatory
load management in the smart grid,” in Workshop on AI Problems and
Approaches for Intelligent Environments, 2012, p. 21.
[13] A.-H. Mohsenian-Rad, V. W. Wong, J. Jatskevich, R. Schober, and
A. Leon-Garcia, “Autonomous demand-side management based on
game-theoretic energy consumption scheduling for the future smart
grid,” IEEE transactions on Smart Grid, vol. 1, no. 3, pp. 320–331,
2010.
[14] S. Maleki, “Addressing the computational issues of the shapley value
with applications in the smart grid,” Ph.D. dissertation, University of
Southampton, 2015.
[15] L. S. Shapley, “A value for n-person games,” Contributions to the
Theory of Games (AM-28), vol. 2, pp. 307–317, 1953.

Luciana S. M. Arnoux received the B.Sc. degree


in industrial engineering from Federal University of
Minas Gerais (UFMG), Brazil, in 2011. She also
received the M.Sc from the same department in
2015, with the emphasis on Stochastic Modelling
and Simulation applied to decentralized generation.
Currently, she is working toward the Ph.D. degree
in applications of optimization theory, intelligent
multiagent systems and game theory for control and
management of distributed power systems. She is
doing this research with the Graduate Program in
Electrical Engineering of UFMG.

View publication stats

You might also like