2016 49th Hawaii International Conference on System Sciences

Decision Model for Sustainable Electricity Procurement

Using Nationwide Demand Response

Joscha Markle-Hu Stefan Feuerriegel Dirk Neumann

University of Freiburg University of Freiburg University of Freiburg
Freiburg, Germany Freiburg, Germany Freiburg, Germany
mjoscha@gmail.com stefan.feuerriegel@is.uni-freiburg.de dirk.neumann@is.uni-freiburg.de

AbstractAs part of the worldwide effort towards achieving

sustainable energy generation, an increasing share of renew-
80
able energy sources are connected to the electricity networks.
This intermittent electricity output from renewable energy
resources causes changes in prices and feed-in volumes. One
approach to compensate for the supply side uctuations is the
active management of the demand side via so-called Demand Average Price in EUR/MWh 60

Response (DR). Demand Response is dened as providing

incentives to electricity consumers to shift electricity consump-
tion to the economically most favorable time. Analyzing such
load shifts economically is challenging, as prices depend on
40
the quantity requested. We analyze this interdependency by
explicitly modeling the load dependence of electricity prices
and thus develop a decision model for nationwide electricity
expenses in the form of a quadratic optimization. Using real
market data, we show a decrease in average electricity prices 20

and price volatility.

1. Introduction 0 5 10 15 20
Time t in h

Electricity demand varies strongly throughout the day.

These variations are driven by consumers who usually start
using electricity in the morning with two peaks peak during
the day and then reduce demand again over night. This is
true for most private households, as well as many industrial
and commercial consumers. As electricity is traded on a
market, times of high demand especially during the day
often correlate with high prices. On the other hand, at
night, demand and prices are generally low. This is shown
in Figure 1 in which long-term averages of the electricity
prices and their 5 % and 95 % quantiles are plotted for each
hour throughout the day.
Because of the price differences during day and night,
electricity retailers who are purchasing energy on day-to-
day auctions on the spot market can achieve signicant cost
savings if volume and timing of their electricity purchases
were optimized to the economically most feasible solution
[1], [2]. Often a shift of a few hours is enough to realize
savings and for many applications does not impair the
customer [3], e. g., a freezers compressor is running at night
instead of during the day. Hence, such shifts usually occur in
Figure 1. Plot shows the uctuations in average electricity prices for each
hour of the day, including 5 % and 95 % quantiles (EPEX spot market
between 2009 and 2012).
times of high demand and high prices, where consumption
is shifted to hours of lower demand and also lower prices. It
is also possible that electricity supply may be high, e. g., due
to favorable weather for renewable generation, which could
make it benecial to consume more energy at such a time
even if overall demand is also high. Such shifts in consumer
electricity usage can be achieved, for example, by providing
incentives in programs or tariffs. In the following, we refer
to any lever which inuences the electricity consumption as
Demand Response (DR) [2], [4].
As a result of their central role in purchasing and dis-
tributing energy, electricity retailers are often considered as
innovators in establishing DR [5][7]. DR entails several
advantages, such as reduced procurement costs and improve-
ment of the grid stability; many of which benet electricity
retailers. In addition, consumers capable of shifting demand

1530-1605/16 $31.00 2016 IEEE 1010

DOI 10.1109/HICSS.2016.130
can only be rewarded for their exibility if retailers provide
appropriate contracts. However, governments and policy
makers also have stakes in DR and should seek opportunities
to drive a nationwide adoption [8].
For governments and policy makers, DR is especially
important in its function as an enabler for sustainable energy
generation from intermittent renewable sources. Renewable
energy sources are promoted by many countries [9], [10].
However, due to specic properties of renewable generation,
such as their intermittency, this also poses a challenge to
electricity networks from the supply side [1], [11]. Active
management of the demand side can help to compensate [1],
[8], [12] for an increase in electricity prices and volatility
of power generation. Furthermore, this allows to shut down
environmentally unfriendly generation plants (because of the
decrease in peak demand due to DR) and one can also reduce
transmission infrastructure.
Previous literature has covered multiple individual as-
sessments of the benets of DR on household level [13],
[14]. However, we are not aware of any nationwide analytics
or decision models to evaluate the optimal re-distribution of
demand. Furthermore, no holistic evaluation of the impact
of DR on cost savings, prices and price volatility has been
undertaken. Evaluation of these parameters is particularly
challenging since prices are affected by load shifts.
In order to analyze the nationwide cost savings of De-
mand Response, we propose a novel decision model for elec-
tricity expense optimization. Using a quadratic optimization
problem, we minimize the electricity expenses and identify
the optimal timing for load shifts.
The remainder of this paper is structured as follows.
We dene the required terminology in Section 2 before pre-
senting related work on the benets of Demand Response,
as well as decision models in Section 3. Next, Section 4
presents the quadratic optimization problem, which we use
to identify the optimal amount of load shifting throughout a
single day. This is followed by an evaluation of our decision
model based on real-world data in Section 5. Finally, we
present policy implications in Section 6 and Section 7
summarizes the most important ndings.
2. Terminology
A general term describing any activities within the de-
mand side of the electricity network is given by Demand
Side Management (DSM). Three types of Demand Side
Management can be identied, which differ in their objec-
tives [1]:
1) Economic/market-driven DSM aims to reduce the
overall cost of electricity by various means. For
example, this may be achieved through a shift of
demand to times with lower prices.
2) Environmental-driven DSM focuses on improv-
ing the environmental aspects of energy production
and consumption, such as the general reduction of
demand and avoidance of generation technologies
that are particularly polluting.
1011
3) Network-driven DSM focuses on enhancing net-
work stability and reducing the need for additional
capacities in generation and transmission.
In addition to the above terminology, market-driven (and
partly network-driven) Demand Side Management ties to so-
called Demand Response (DR). Demand Response is a term
which has been used traditionally to describe peak clipping
a measure to actively cut down demand in times of very
high electricity usage. Recently, however, institutions such
as the U. S. Department of Energy and the Federal Energy
Regulatory Commission have used the term more generally
as a name for a tariff or a program established to motivate
changes in electric use by end-use customers in response
to changes in the price of electricity over time, or to give
incentive payments designed to induce lower electricity use
at times of high market prices or when grid reliability is
jeopardized [4]. This includes the use of incentive- or
price-based programs to steer demand [2]. In this paper,
we use the term Demand Response to describe the shifting
of load by customers in response to appropriate incentives,
whereas DSM could include any measure that inuences
the demand side, such as an awareness campaign for energy
efcient lighting.
3. Related Work
In this section, we present literature which evaluates the
benets of Demand Response and decision models for DR.
3.1. Benets of Demand Response
It is generally agreed that Demand Response entails a
large variety of benets for almost all stakeholders [2], [15]
[18]. Multiple studies point out how sustainable electricity
generation based on renewable energy resources can be sup-
ported by the active management of the demand side through
DR [1], [8], [12], [19]. For example, one study specically
mentions the intermittent properties of renewable energy
generation which prohibit their use as dispatchable re-
sources [1]. Hence, Demand Response is regarded as a very
suitable instrument [20], [21] for intermittency mitigation,
in addition to other measures such as energy storage systems
[22]. Furthermore, Demand Response and environmental
goals are called an optimal complementary approach [8].
In addition, cost savings are expected for customers
capable of shifting their load; as shifts atten the overall
load curve and hence reduce the general cost of electricity
production. Customers without Demand Response capability
benet from lower overall electricity prices [18]. This comes
at the cost of reduced prots for electricity retailers offering
expensive peak reserve generation capacities [23], [24].
While many studies evaluate the available load potential
(e. g. [8], [12], [19], [25], [26]), the results strongly depend
on the underlying assumptions. Klobasa [19], [25] provides
a very detailed bottom-up analysis for Germany calculating
a potential of 3 GW in the industrial sector, 8 GW in the
commercial sector and above 20 GW from private house-
holds.
In terms of nancial savings, it is estimated that es-
tablishing smart meters in Europe could provide savings
of up to EUR 67 bn if optimal adoption is achieved and
only EUR 14 bn for the less optimistic scenario [27]. Also,
the cost of smart meters would have to be covered rst.
Average household savings in the UK are estimated at
around GBP 1,800 present value over a 20 year period if at
least 30 % of the population participate and also accept 1 h
of reduced consumption per day [28].
Altogether, knowledge of the nationwide economic po-
tential of Demand Response in liberalized markets is still
limited [1], [13] and we are not aware of any analysis that
includes a simulation of load shifting whilst using actual
real-world data.
3.2. Decision Models for Demand Response
Using models and methods from Operations Re-
sarch (OR) to solve questions regarding the energy market
is widely established [29][32]. In this context, decision
models are used to provide recommendations on how to
achieve the optimal outcome for a given problem [33], [34].
Multiple studies related to Demand Response aim to
create decision models for households [35], [36], consumer
groups [13], [37], [38] or even individual appliances [39].
However, decision models on an aggregate level are scarce.
Faruqui et al. [27] use direct calculation of savings based
on a xed reduction of peak cost by a given percentage.
While such estimates are possible, they do not take into
account the electricity market dynamics and also provide no
insight into the required load shifts and how to use different
appliances with different maximum shift durations.
Nguyen and Nguyen [40] optimize electricity expenses
with a dynamic programming backward algorithm. Using
stochastic optimization uncertainties are taken into account.
Yet, the problem is solved only for individual appliances
such as ventilation and electric vehicles. As prices can
be assumed xed for such solitary problems, this is not
very accurate. Furthermore, the DP problem has exponential
dimensionality and is computationally inefcient.
Another approach [28] simulates the behavior of three
types of stakeholders: the utility, an aggregator and house-
holds. However, here the usage of Demand Response in form
of net curtailment and load shifting is presented as a binary
on-or-off decision depending on static thresholds.
In summary, none of the above decision models provides
full insights in how to optimally shift loads depending on
actual market data and demand curves for a given time
period.
4. Methodology
In this section, we describe how nationwide usage of
Demand Response can be modeled by methods from Oper-
ations Research. In a rst step, we describe the underlying
1012
model that aggregates expenses from electricity auctions.
In a second step, we extend the decision model to include
Demand Response usage. Here, we formulate a quadratic
optimization problem which allows us to simulate the price
dependency on the traded volumes of the (multi-)national
electricity exchange.
Electricity can be purchased via different products on
the markets. Usually, there is a baseload product, which is
traded in options and futures to ensure the long-term supply
of electricity. To further differentiate these long-term needs,
the baseload package which usually covers all 24 h of a
day, may be complemented by various peak and off-peak
products, which allow a differentiation of the time of the
day. The remaining energy is traded on a spot market, which
is particularly important to balance day-to-day variations in
electricity demand. In this paper, we focus on the expenses
for electricity on the spot market.
4.1. Optimization of Volume-Dependent Electricity
Expenses
Let us calculate the nationwide costs Cnation (in
EUR/MWh) spent on daily electricity auctions on the spot
market. Furthermore, let us introduce the hourly prices for
electricity auction contracts pA (t, qA (t)) and their corre-
sponding traded volumes qA (t) at hour t. Here, the expres-
sion pA (t, qA (t)) denotes that prices at the exchange depend
on the corresponding hours, as well as the nationwide quan-
tities. Then, the expenses for electricity traded as auctions
on a day-to-day basis may be written as

24
Cnation = pA (t, qA (t)) qA (t) (1)
t=1
where we sum over all hours t = 1, . . . , 24 of a single day.
Within a reasonable neighborhood, we can assume a
linear relationship between volume and day-to-day prices
[41]. This is justied when looking into the bid-ask curves
of the electricity exchange. Thus, we model prices with the
linear model
pA (t, qA (t)) = (t) + (t) qA (t) (2)
with suitable coefcients (t) and (t). These coefcients
can, for example, be estimated via ordinary least squares or
other regression techniques. We show later how we derive
this linear model from actual demand curves (Section 5.2).
Combining the previous ideas from Equations (1)
and (2), we yield a formulation in which prices are depen-
dent on volumes via

24 
24
Cnation = (t) qA (t) + 2
(t) qA (t). (3)
t=1 t=1
While the above equation for Cnation states how to calculate
nationwide costs, the actual computation of qA (t) is not yet
specied; this is addressed in the following section where
we formulate a corresponding optimization problem.
4.2. Electricity Expense Optimization Here, G is a diagonal matrix G R2424 , where the
diagonal entries are 2(t). Then, our constraints need to
The previous section has outlined how to model nation- be rearranged in the form
wide spending on electricity, while the actual calculation
subject to C T x = b. (10)
of qA (t) in Equation (3) is left unclear. As a remedy, we
pursue the following approach: we rst rewrite the above As electricity cannot be purchased in arbitrary amounts
model as an optimization problem and then transform it into but must meet demand, we dene the following equality
matrix notation. The matrix formulation simplies notation constraint in matrix representation using a diagonal matrix
and serves as a default input for most solvers. of ones such that
The optimization problem in order to determine the
1 0 ... 0
quantities qA (t) is given by
0 1 . . . 0
CT =
.. .. . . . .. R
2424
. (11)
min Cnation . . .
qA (1),...,qA (24)
0 0 ... 1

24 
24 (4)
= min (t) qA (t) + 2
(t) qA (t) The demand on the right-hand side is written as a vector
qA (1),...,qA (24)
t=1 t=1
D(1)
with an additional constraint as follows. Let us introduce a D(2)
new variable D(t) which denotes the nationwide electricity b=
.. R .
24
(12)
demand in hour t. Then, the quantities bought at the elec- .
tricity exchange need to equal the nationwide demand for D(24)
electricity from auctions, e. g., excluding any demand that Using general information about the electricity market
is fullled with other products, such as options and futures. we have shown how the total expenses can be derived and
Mathematically speaking, we obtain the following constraint represented in quadratic form. We can easily add further
constraints to Equation (10) by appending more rows to C .
qA (t) = D(t) for all t = 1, . . . , 24. (5)
In the next section, we show how Demand Response can be
This states that the purchased electricity power from auc- included in our model.
tions must meet total demand D(t) in every hour t = Now, it becomes obvious that our optimization problem
1, . . . , N . is quadratic with linear equality constraints. This can be
To make notation easier, we now rewrite the above easily solved with a large variety of solvers from quadratic
equations in matrix form. Hence, we dene x R24 as programming [42], [43] in polynomial time [44] if the
the quantitites of electricity purchased at every hour t. Using matrix G is positive-denite.
this, we write the nationwide electricity expense in quadratic
form as 4.3. Modeling Demand Response
1
min aT x + xT Gx. (6)
x 2 In order to extend the above quadratic optimization to
include DR, we need to introduce the following notation
Transforming the minimization problem into quadratic form
(see Figure 2). It is important to note that not all energy can
can be achieved by dening the vectors and matrices of
be shifted by the same maximum duration j . For example,
Equation (6) and Equation (10) as
heating the at may be delayed by an hour or two, whilst
washing the laundry could be delayed by j = 12 h or more.
(1)
(2) Hence, we dene LSj (t, d) as the load shift from t by d
a=
.. R ,
24
(7) hours with a maximum shift by j hours according to the
. given appliance. In other words, we denote a shift from t
(24) to t via LSj (t, t t). For example, let LS2 (t, 1) be the
potential (i. e. all appliances with maximum shift of 2 h)
qA (1) which is shifted by 1 h from t to t 1. By denition, we
qA (2) def
set all LSj (t, d) = 0 where t + d < 1 or t + d > 24 (out
x=
.. R
24
(8)
. of range). In addition, a shift LSj (t, 0) by 0 hours is not
qA (24) dened.
This notation is now used, to integrate Demand Re-
and using sponse usage into the above model. We append further
constraints and include additional components in the pa-
2(1) 0 ... 0 rameter x. Then, we obtain a new optimization problem in
0 2(2) ... 0
G=
.. .. .. .. R2424 .
(9) the following high-level formulation
. . . . min Cnation Demand Response usage. (13)
0 0 ... 2(N ) qA (t),LSj (t,d)

1013
 t2 t1 t t+1 t+2

j=1
LS1(t,1) LS1(t,+1)

j=2 LS2(t,2) LS2(t,1) LS2(t,+1) LS2(t,+2)

Figure 2. Illustration of load shifted away from time slot t to adjacent time slots for appliances with maximum shifts of j = 1 h and j = 2 h.

When using Demand Response, we actively modify the rst present our data sources and then outline how we
demand at certain hours. Furthermore, according to our model the relationship between traded electricity volume
denition any load shift must be positive, such that and corresponding price. Finally, we present our simulation
results which are based on real-world data.
LSj (t + i, i) 0 and
(14)
LSj (t i, +i) 0 for all t, j, i.
5.1. Data Sources
In addition, the load shifted away is limited by a maxi-
mum available capacity j (t). For example, the sum of all Several publications aim at quantifying the available load
LS2 (t, d) is thus bounded, i. e. LS2 (t, t 2) + LS2 (t, t shifting potential in Germany (e. g. [12], [25], [26]). In this
1) + LS2 (t, t + 1) + LS2 (t, t + 2) must be smaller or equal paper, we use the estimated DR potential from Klobasa [25]
than 2 (t). We formalize this in a second constraint as these values feature the highest granularity. In addition,
1 this DR potential ranges in the middle of the spectrum
 
j
LSj (t, i) + LSj (t, i) j (t) for all j, t. (15) when compared to other publications. These potentials are
based on a bottom up assessment of available capacities
i=j i=1
in households, commercial and industrial applications. We
Finally, our model requires a new constraint that vali- calculate the average daily shift potential in Germany and
dates if electricity demand is satised by traded quantities include only appliances which are applicable to a winter
and Demand Response usage. This is given by day. As such, we exclude, for instance, air conditioning but
qA (t) = D(t) load shifted away include heating applications. In addition, we scale the data
to also include Austria.1,2
+ load shifted here (16)
The previous description addresses the theoretically pos-
for t = 1, . . . , 24 sible Demand Response potential, based on which we derive
This expression can then be extended to the following three scenarios. These scenarios vary in terms
1 of the penetration of Demand Response that is assumed:
  
j
accordingly, we set the ratio of Demand Response utilization
qA (t) =D(t) LSj (t, i) LSj (t, i) to 1 %, 10 % and 25 %. The corresponding volumes available
j i=j i=1 for load shifting are given in Table 3.
1

(17)
 
j
In order to achieve realistic estimates of the savings pos-
+ LSj (t i, i) + LSj (t i, i) sible through Demand Response, we use actual market data
i=j i=1 from EPEX spot3 for the combined German and Austrian
for t = 1, . . . , 24 and with the original demand D(t). Then, electricity market. Here, we use hourly auction prices and
the variable qA (t) gives the quantity of electricity that needs volumes for December 19, 2012 as an exemplary simulation
to be purchased at a time t to fulll the original demand after horizon. In our calculations, we determine the electricity
load shifting for the given time slot. demand based on data from EEX transparency4 .
The previously proposed additions require multiple Finally, we need to specify how we determine electricity
changes to the quadratic implementation of this minimiza- prices when simulating traded electricity volumes that differ
tion problem. The solution vector x denoting the demand from the historic data. For this, we employ actual demand
is now extended to comprise of one value for each possible 1
Here, we choose a scaling factor of 1.11. This factor is based on
demand shift in addition to the 24 values for each hour, Eurostat data of gross inland consumption of energy for 2012 in Germany
while the cost function remains the same. and Austria.
2
Eurostat. URL: http://ec.europa.eu/eurostat/statistics-explained/index.
php/Consumption of energy. Retrieved on June 6, 2015.
5. Evaluation 3
EPEX Spot. URL: http://www.epexspot.com/. Retrieved on May 15,
2015.
This section describe how we solve the quadratic opti- 4
EEX transparency. URL: http://www.eex-transparency.com/. Re-
mization problem to determine optimal load shifting. We trieved on May 14, 2015.

1014
 TABLE 3. AVAILABLE ELECTRICITY VOLUME AVAILABLE FOR LOAD
SHIFTING WITH A BREAK DOWN BY THE MAXIMUM SHIFT DURATION
FOR THREE SCENARIOS WITH 100 %, 25 %, 10 % AND 1 % DR
UTILIZATION ( DATA BASED ON [25]; SCALED FOR THE COMBINED
MARKET OF G ERMANY AND AUSTRIA ; APPLIANCES FILTERED FOR A
WINTER DAY ).

Maximum Shift Maximum Load Shift j in MW

Duration j in h
100% 25% 10% 1%
1 2,117 529 212 21
2 2,819 705 282 28
3 199 50 20 2
4 197 49 20 2
12 4,391 1,098 439 44
16 567 142 57 6
24 400 100 40 4

curves from EPEX spot. These curves feature all bids for an
electricity auction in a given hour. On average, an auction
consists of 691 bids.
Figure 4. Demand curve shows individual electricity offers (blue dots); the
red dot indicates the nal market price. The shade area highlights values
5.2. Dependency of Electricity Prices on Trading selected for the actual regression to calculate the slope (Data from 4:00 am
Volumes on December 19, 2012).

In economics, relationships between variables are fre-

quently assumed to be linear for a given neighborhood of
observed data [45]. Accordingly, we adopt this approach
and also assume a linear relationship between electricity
price and traded volume [41]. We validate this procedure
visually by depicted all bids for an exemplary electricity
auction in Figure 4. Here, the red dot indicates the actual
market price and we see that a linear relationship is valid
in its neighborhood.
In order to calculate the gradient, we rst lter the data
and then perform a linear regression. For the former, we
choose the 0.5 % and 99.5 % quantiles of all electricity
prices between 2009 and 2012. These represent a set of
feasible prices which we can expect during normal op-
erations on the electricity exchange. Hence, our regres-
sion is thus based on prices between 0.04 EUR/MWh and
93.81 EUR/MWh in order to compute the gradient. This
range is highlighted by an shaded area in Figure 4. Interest-
ingly, the intercept from the linear model is not needed for
further evaluations, instead we only use the slope from the
regression and then calculate the simulated price based on
gradient and the historic price.
5.3. Results
We now evaluate the effects from Demand Response
usage on electricity prices and their volatility. Thereby, we
compare three different scenarios that vary in the utilization
(1 %, 10 % and 25 %) of the theoretically available DR po-
tential. In our analysis of shifted volumes, we predominantly
focus on the 25 % simulation to provide insights on how
the shifts work. In addition, we compare and discuss the
economic effects across all three scenarios, nding a similar
pattern for the expected savings.
Figure 5 illustrates how electricity demand is shifted
to minimize costs. Here, the x-axis shows the hour of the
day and the y-axis the volumes shifted. Bars in positive
direction indicate that electricity was shifted from some time
to this hour, whereas negative bars reveal that electricity
was shifted to other hours. The colors of the bar denote the
duration of the shift. Overall, we see the following pattern:
electricity is mostly shifted to night hours from 0 to 5 am
and from 9 pm to midnight. During the day, electricity
is mostly shifted away, except for a peak at 3 pm when
relatively low electricity prices promote higher usage. Fur-
thermore, these shift mainly originate from appliances that
are highly exible in terms of scheduling. More precisely, all
shift durations j are used to relocate the electricity demand.
One and two hour shift potentials are activated to shift peak
demand to times of lower electricity usage at the boundaries
and the potentials with longer shifts, such as 12 h, 16 h and
24 h, are used in the morning and evening to shift electricity
towards midnight.
The previous ndings are conrmed by Table 6 in more
detail: the net effect from Demand Response remains posi-
tive for a given hour if this hour receives more energy than
is shifted away. Night hours, especially in the early morn-
ing, receive up to 4.1 GW electricity, whereas our decision
model shifts electricity most away during day time. The
maximum Demand Response potential for each individual
hour accounts for 2.7 GW, which is equivalent to using the
full 25 % potential. This happens at 7 pm, which is also one
of the hours with highest electricity prices (cf. the original

1015
 4000

Max. Shift (j) in h

2000 01
Volume in MWh

02
03
04
12
0
16
24

2000

0 5 10 15 20
Time t in h

Figure 5. Bar chart shows load shifts by hour. In short, load is shifted from the morning and afternoon to the night times. When electricity volume is
received, this is shown in the positive range of the y -axis; load shifted away appears as a negative bar. The color indicates the maximum possible shift
duration in hours (scenario with DR utilization of 25 % of the theoretically possible potential).

electricity prices in Table 6). In contrast, receiving electricity

is not limited by this volume, since electricity can be shifted DR potential 0% 1% 10% 25%

from multiple hours to another.

Figure 7 compares visually the simulated electricity
prices across all three Demand Response scenarios. The
solid line represents the benchmark without Demand Re- 70
sponse. These prices are equivalent to the originally ob-
served ones. Based on the simulated prices, we observe
two peaks in the late morning and early evening, which
Price in EUR/MWh

both mirror the previous load shifting. In the scenarios 60

with Demand Response potential, load is shifted, resulting

into prices that are more balanced. Interestingly, we nd
evidence of considerably more stable and less volatile prices
50
in scenarios with higher Demand Response utilization. For
instance, the simulated prices in the 25 % scenario (dashed
line) is above the original prices during night hours and
below during daytime. 40

Furthermore, we observe one interesting pattern in sim-

ulation using 25 % of the theoretically available shift po-
tentials. In this case, some prices during the night are
higher than during the day. These effects are driven by the
large volumes of shifted load. As a consequence, it is not
realistic to study DR utilization of above 25 % with the
given decision model as this might violate the assumption
of a linear relationship between electricity price and traded
volume. For such evaluations, one needs to either develop
a different decision model or validate the outcomes with
results from practical experiments.
Finally, we compare the effects on electricity prices
across all Demand Response scenarios. As such, we pro-
0 5 10 15 20
Time t in h
Figure 7. Plot shows simulated electricity prices for each hour across
different Demand Response scenarios. A higher share of DR utilization
attens the price curve and thus reduces volatility.
vide descriptive statistics of simulated electricity prices
from Demand Response usage in Table 8. When using
Demand Response, the mean electricity price decreases for

1016
 TABLE 6. OVERVIEW OF HOURLY NET EFFECT INDUCED BY D EMAND R ESPONSE USAGE ACROSS EACH HOUR OF THE DAY FROM VOLUME SHIFTS
( SCENARIO WITH 25 % DR PENETRATION ).

Time Slot t (in h) 0 1 2 3 4 5 6 7 8 9 10 11

Original Volume (in GW) 23.8 24.4 23.8 23.5 23.0 23.3 25.0 24.3 27.8 28.4 28.8 29.4
Volume After DR (in GW) 27.8 28.2 27.8 26.5 25.2 25.1 24.3 23.0 25.5 25.9 26.5 26.9
Net Effect (in GW) 4.0 3.7 4.1 2.9 2.3 1.8 0.7 1.3 2.3 2.5 2.3 2.5
Original Price (in EUR/MWh) 36.07 35.32 35.21 34.22 34.20 36.99 44.28 60.45 66.35 66.22 65.10 63.91
Price After DR (in EUR/MWh) 55.60 53.50 55.09 49.97 47.28 47.38 40.16 50.77 42.67 40.65 41.71 39.10
Price Change (in EUR/MWh) 19.53 18.18 19.88 15.75 13.08 10.39 4.12 9.68 23.68 25.57 23.39 24.81

Time Slot t (in h) 12 13 14 15 16 17 18 19 20 21 22 23

Original Volume (in GW) 29.7 28.9 27.6 25.4 23.7 25.4 27.1 26.9 24.4 23.6 23.6 24.6
Volume After DR (in GW) 28.2 27.5 27.1 26.0 23.0 23.3 24.5 24.3 23.1 25.2 25.1 26.6
Net Effect (in GW) 1.4 1.4 0.4 0.6 0.8 2.2 2.6 2.7 1.3 1.6 1.5 2.0
Original Price (in EUR/MWh) 60.02 59.90 58.27 57.95 63.92 77.40 69.94 62.16 55.08 48.48 42.13 37.65
Price After DR (in EUR/MWh) 47.54 47.23 54.64 62.62 56.58 54.98 43.31 33.47 46.32 56.95 50.63 48.60
Price Change (in EUR/MWh) 12.48 12.67 3.63 4.67 7.34 22.42 26.63 28.69 8.76 8.47 8.50 10.95

TABLE 8. D ESCRIPTIVE STATISTICS OF SIMULATED PRICES ACROSS Response can also compensate for uctuations in the power
DIFFERENT DR SCENARIOS . generation of renewable energies [1], [11]. In addition, this
paper provides evidence that load shifting reduces average
DR Scenario Mean Median Stand. Dev. Min. Max. Skew. Kurt.
electricity prices and their volatility. Hence, it is desirable
0% 52.97 58.11 13.58 34.2 77.4 0.13 1.49
1% 52.78 58.36 12.95 34.18 76.52 0.13 1.46 to advance smart metering and generally any technology
10 % 51.16 52.65 8.16 37.69 68.64 0.02 0.86 which allows load shifting [27], [46], [47]. Consequently,
25 % 48.61 48.07 6.94 33.47 62.62 0.14 0.68
governments should consider supporting this trend with
appropriate policies [48].
all scenarios; for example, from 52.97 EUR/MWh down to Challenges in the nationwide adoption of Demand Re-
48.61 EUR/MWh in the 25 % scenario. Similarly, Demand sponse have been clustered into four groups in a recent
Response usage also reduces the volatility of electricity study for the Federal Energy Regulatory Commission [49]:
prices, as well as the minimum and maximum price. regulatory, technical, economic and other. These are as fol-
Overall, nationwide costs for electricity procurement lows: (1) regulatory challenges refer to any barriers created
from day-ahead auctions accounts for EUR 33.12 mn. De- through policies. (2) Technical obstacles denote the need
mand Response can diminish these costs. In the scenarios for new types of appliances, communication protocols and
with 1 %, 10 % and 25 % DR potential, total costs account advances in metering technology. (3) Economic challenges
for EUR 32.95 mn, EUR 31.65 mn and EUR 30.00 mn include all cases in which the nancial benets are not clear.
respectively. This totals to savings of EUR 0.17 mn, (4) Finally, the category other comprises the remaining
EUR 1.47 mn and EUR 3.12 mn for each of the scenarios difculties. Examples include customer perception of DR
one the given day. and their willingness to enroll. Clearly, governments should
These ndings fully conrm the positive impact of De- tackle these challenges in the pursuit of increasing Demand
mand Response on electricity prices. Even in the very con- Response adoption [27], [48], [50]. Also, any initiative in
servative scenario, which uses only 1 % of the theoretically support of DR should be carefully designed to provide an
available DR potential, signicant savings are possible. If effective measure [51].
we assume that this single day represents average electricity In future, additional Demand Response potentials can be
consumption, annual savings of approximately EUR 62 mn unlocked as technology improves, for instance, from smart
could be achieved. In case of higher DR potentials such as cars [16] and home electricity storage systems [15], [36].
25 %, these could potentially increase to about EUR 1.1 bn These technologies entail the advantage of providing highly
savings from simple load shifting. We also note that further exible load shifting, which fosters their use in stabilizing
savings can originate from second order effects, such as the electricity grid. Hence, it should be a key goal of policy
the reduction of transmission infrastructure and reduced makers to bolster the introduction of such technologies from
expenses for load balancing. an early point onwards.

6. Policy Implications 7. Conclusion and Outlook

The manifold benets of Demand Response make it an Demand Response serves as an effective instrument to
intriguing lever for any sustainability strategy. Individual mitigate the intermittency of renewable energy resources,
advantages are as follows: Demand Response usage can whilst also providing very promising nancial savings to
result in cost and infrastructure savings [16][18]. Demand electricity retailers and consumers. While previous research

1017
has predominantly addressed the nancial savings on house-
hold or retailer level, little is known on the economic
consequences on a nationwide dimension. Hence, this paper
proposes a decision model in order to simulate implications
of Demand Response usage for nancial savings and elec-
tricity prices. Since price and volume are dependent on each
other, we need to implement a corresponding price model
in form of a quadratic optimization problem.
Based on our decision model, we infer optimal load
shifts for a simulation scenario based on the German and
Austrian spot market. When assuming a 25 % penetration
of the theoretic Demand Response potential, our simula-
tion reveals nationwide savings for a single day of up to
EUR 3.12 mn. This result should make the adoption of
Demand Response an obvious choice for electricity retail-
ers wherever feasible. In addition, governments and policy
makers should evaluate the benets of Demand Response
especially with regards to enabling sustainable electricity
generation and cost savings.
While we have chosen a sample day for all evaluations in
this paper, future work will focus on making the results more
reliable by studying longer time horizons. Furthermore,
policy makers would also benet from detailed assessments
of the cost structure for all Demand Response scenarios in
order to evaluate the total net benet of Demand Response.
References
[1] J. Aghaei and M.-I. Alizadeh, Demand Response in Smart Elec-
tricity Grids Equipped with Renewable Energy Sources: A Review,
Renewable and Sustainable Energy Reviews, vol. 18, pp. 6472, 2013.
[2] M. Albadi and E. El-Saadany, A Summary of Demand Response
in Electricity Markets, Electric Power Systems Research, vol. 78,
no. 11, pp. 19891996, 2008.
[3] S. Gottwalt, W. Ketter, C. Block, J. Collins, and C. Weinhardt,
Demand Side Management: A Simulation of Household Behavior
under Variable Prices, Energy Policy, vol. 39, no. 12, pp. 81638174,
2011.
[4] U.S. Department of Energy, Benets of Demand Response in Elec-
tricity Markets and Recommendations for Achieving them: A Report
to the United States Congress Pursuant to Section 1252 of the Energy
Policy Act of 2005, 2006.
[5] EU-DEEP, The Birth of a European Distributed Energy Partnership
that will Help the Large-scale Implementation of Distributed Energy
Resources in Europe: FP6 - Sustainable Energy Systems: Project
reference 503516, 2009. [Online]. Available: http://cordis.europa.
eu/project/rcn/73969 en.html
[6] Smart Energy Demand Coalition, The Demand Response Snap Shot:
The Reality For Demand Response Providers Working in Europe To-
day, 2011. [Online]. Available: http://sedc-coalition.eu/wp-content/
uploads/2011/09/SEDC-DR-Snap-Shot.-FINAL-20112.pdf
[7] A. Faruqui and S. Sergici, Household Response to Dynamic Pricing
of Electricity: A Survey of 15 Experiments, Journal of Regulatory
Economics, vol. 38, no. 2, pp. 193225, 2010.
[8] C. Bergaentzle, C. Clastres, and H. Khalfallah, Demand-Side Man-
agement and European Environmental and Energy Goals: An Optimal
Complementary Approach, Energy Policy, vol. 67, pp. 858869,
2014.
[9] The White House, U.S.-China Joint Announcement on Climate
Change, Ofce of the Press Secretary, 2014. [Online].
Available: https://www.whitehouse.gov/the-press-ofce/2014/11/11/
us-china-joint-announcement-climate-change
1018
[10] European Union, Decision No 406/2009/EC of the European
Parliament and of the Council of 23 April 2009 on the Effort
of Member States to Reduce their Greenhouse Gas Emissions
to Meet the Communitys Greenhouse Gas Emission Reduction
Commitments up to 2020, Ofcial Journal of the European Union,
vol. L 140/136, no. Decision No 406/2009/EC, p. 136, 2009. [Online].
Available: http://eur-lex.europa.eu/legal-content/EN/TXT/?uri=OJ:L:
2009:140:TOC
[11] R. Walawalkar, S. Fernands, N. Thakur, and K. R. Chevva, Evolution
and Current Status of Demand Response (DR) in Electricity Markets:
Insights from PJM and NYISO, Energy, vol. 35, no. 4, pp. 1553
1560, 2010.
[12] M. E. Dyson, S. D. Borgeson, M. D. Tabone, and D. S. Callaway,
Using Smart Meter Data to Estimate Demand Response Potential,
with Application to Solar Energy Integration, Energy Policy, vol. 73,
pp. 607619, 2014.
[13] S. Feuerriegel and D. Neumann, Measuring the Financial Impact of
Demand Response for Electricity Retailers, Energy Policy, vol. 65,
pp. 359368, 2014.
[14] M. Vasirani and S. Ossowski, Smart Consumer Load Balancing:
State of the Art and an Empirical Evaluation in the Spanish Electricity
Market, Articial Intelligence Review, vol. 39, no. 1, pp. 8195,
2013.
[15] O. Erdinc, N. G. Paterakis, I. N. Pappi, A. G. Bakirtzis, and J. P.
Catalao, A New Perspective for Sizing of Distributed Generation
and Energy Storage for Smart Households under Demand Response,
Applied Energy, vol. 143, pp. 2637, 2015.
[16] P. Siano, Demand Response and Smart Grids: A Survey, Renewable
and Sustainable Energy Reviews, vol. 30, pp. 461478, 2014.
[17] A. Safdarian, M. Fotuhi-Firuzabad, and M. Lehtonen, Benets of
Demand Response on Operation of Distribution Networks: A Case
Study, IEEE Systems Journal, vol. PP, no. 99, pp. 19, 2014.
[18] C.-L. Su and D. Kirschen, Quantifying the Effect of Demand Re-
sponse on Electricity Markets, IEEE Transactions on Power Systems,
vol. 24, no. 3, pp. 11991207, 2009.
[19] M. Klobasa, Analysis of Demand Response and Wind Integration
in Germanys Electricity Market, IET Renewable Power Generation,
vol. 4, no. 1, pp. 5563, 2010.
[20] A. Papavasiliou and S. S. Oren, Large-Scale Integration of De-
ferrable Demand and Renewable Energy Sources, IEEE Transactions
on Power Systems, vol. 29, no. 1, pp. 489499, 2014.
[21] P. Finn and C. Fitzpatrick, Demand Side Management of Industrial
Electricity Consumption: Promoting the Use of Renewable Energy
hrough Real-time Pricing, Applied Energy, vol. 113, pp. 1121, 2014.
[22] J. P. Barton and D. G. Ineld, Energy Storage and Its Use with
Intermittent Renewable Energy, IEEE Transactions on Energy Con-
version, vol. 19, no. 2, pp. 441448, 2004.
[23] S. Stoft, Power system economics: Designing Markets for Electricity.
Piscataway, NJ, New York: IEEE Press, Wiley-Interscience, 2002.
[24] C.-L. Su, Optimal Demand-Side Participation in Day-Ahead
Electricity Markets, Ph.D. dissertation, The University of
Manchester, 2007. [Online]. Available: http://www.ee.washington.
edu/research/real/Library/Thesis/Chua-Liang SU.pdf
[25] M. Klobasa, Dynamische Simulation eines Lastmanagements und
Integration von Windenergie in ein Elektrizitatsnetz, ser. ISI-
Schriftenreihe Innovationspotenziale. Stuttgart: Fraunhofer-IRB-
Verl., 2009.
[26] H. C. Gils, Assessment of the Theoretical Demand Response Poten-
tial in Europe, Energy, vol. 67, pp. 118, 2014.
[27] A. Faruqui, D. Harris, and R. Hledik, Unlocking the e53 Billion
Savings from Smart Meters in the EU: How Increasing the Adoption
of Dynamic Tariffs Could Make or Break the EUs Smart Grid
Investment, Energy Policy, vol. 38, no. 10, pp. 62226231, 2010.
[28] S. Dave, M. Sooriyabandara, and M. Yearworth, System Behaviour
Modelling for Demand Response Provision in a Smart Grid, Energy
Policy, vol. 61, pp. 172181, 2013.
[29] C. H. Antunes and A. Gomes, Operational Research Models and
Methods in the Energy Sector, European Journal of Operational
Research, vol. 197, no. 3, pp. 997998, 2009.
[30] R. Banos, F. Manzano-Agugliaro, F. G. Montoya, C. Gil, A. Alcayde,
and J. Gomez, Optimization Methods Applied to Renewable and
Sustainable Energy: A Review, Renewable and Sustainable Energy
Reviews, vol. 15, no. 4, pp. 17531766, 2011.
[31] K. Kowalski, S. Stagl, R. Madlener, and I. Omann, Sustainable
Energy Futures: Methodological Challenges in Combining Scenar-
ios and Participatory Multi-criteria Analysis, European Journal of
Operational Research, vol. 197, no. 3, pp. 10631074, 2009.
[32] S. D. Pohekar and M. Ramachandran, Application of Multi-Criteria
Decision Making to Sustainable Energy Planning: A Review, Re-
newable and Sustainable Energy Reviews, vol. 8, no. 4, pp. 365381,
2004.
[33] S. Ahmad and R. M. Tahar, Selection of Renewable Energy Sources
for Sustainable Development of Electricity Generation System Using
Analytic Hierarchy Process: A Case of Malaysia, Renewable Energy,
vol. 63, pp. 458466, 2014.
[34] K. Smith Stegen and M. Palovic, Decision-Making for Supplying
Energy Projects: A Four-Dimensional Model, Energy Conversion
and Management, vol. 86, pp. 644652, 2014.
[35] J. Valenzuela, P. R. Thimmapuram, and J. Kim, Modeling and
Simulation of Consumer Response to Dynamic Pricing with Enabled
Technologies, Applied Energy, vol. 96, pp. 122132, 2012.
[36] O. Erdinc, Economic Impacts of Small-Scale Own Generating and
Storage Units, and Electric Vehicles under Different Demand Re-
sponse Strategies for Smart Households, Applied Energy, vol. 126,
pp. 142150, 2014.
[37] F. d. Ridder, M. Hommelberg, and E. Peeters, Four Potential Busi-
ness Cases for Demand Side Integration, in 2009 6th International
Conference on the European Energy Market. IEEE, 2009, pp. 16.
[38] N. Mahmoudi, T. K. Saha, and M. Eghbal, A New Demand Response
Scheme for Electricity Retailers, Electric Power Systems Research,
vol. 108, pp. 144152, 2014.
[39] J. H. Yoon, R. Baldick, and A. Novoselac, Dynamic Demand Re-
sponse Controller Based on Real-Time Retail Price for Residential
Buildings, IEEE Transactions on Smart Grid, vol. 5, no. 1, pp. 121
129, 2014.
1019
[40] M. Y. Nguyen and D. M. Nguyen, A Generalized Formulation
of Demand Response under Market Environments, International
Journal of Emerging Electric Power Systems, vol. 16, no. 3, 2015.
[41] W. Bonte, S. Nielen, N. Valitov, and T. Engelmeyer, Price elasticity
of demand in the EPEX spot market for electricityNew empirical
evidence, Economics Letters, vol. 135, pp. 58, 2015.
[42] D. Goldfarb and A. Idnani, Dual and Primal-Dual Methods for
Solving Strictly Convex Quadratic Programs, in Numerical Analysis,
ser. Lecture Notes in Mathematics, J. P. Hennart, Ed. Springer Berlin
Heidelberg, 1982, vol. 909, pp. 226239.
[43] , A Numerically Stable Dual Method for Solving Strictly Con-
vex Quadratic Programs, Mathematical Programming, vol. 27, no. 1,
pp. 133, 1983.
[44] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, The Polynomial
Solvability of Convex Quadratic Programming, USSR Computational
Mathematics and Mathematical Physics, vol. 20, no. 5, pp. 223228,
1980.
[45] P. A. Samuelson, Some Implications of Linearity, The Review of
Economic Studies, vol. 15, no. 2, pp. p 8890, 1947.
[46] J. Carroll, S. Lyons, and E. Denny, Reducing Household Electricity
Demand Through Smart Metering: The Role of Improved Information
About Energy Saving, Energy Economics, vol. 45, pp. 234243,
2014.
[47] J. Torriti, M. G. Hassan, and M. Leach, Demand Response Experi-
ence in Europe: Policies, Programmes and Implementation, Energy,
vol. 35, no. 4, pp. 15751583, 2010.
[48] P. Warren, A Review of Demand-Side Management Policy in the
UK, Renewable and Sustainable Energy Reviews, vol. 29, pp. 941
951, 2014.
[49] The Brattle Group, Freeman, Sullivan & Co, and Global Energy
Partners, LLC, A National Assessment of Demand Response
Potential: Staff Report, 2009. [Online]. Available: http://www.ferc.
gov/legal/staff-reports/06-09-demand-response.pdf
[50] Z. Ming, X. Song, M. Mingjuan, L. Lingyun, C. Min, and W. Yuejin,
Historical Review of Demand Side Management in China: Man-
agement Content, Operation Mode, Results Assessment and Relative
Incentives, Renewable and Sustainable Energy Reviews, vol. 25, pp.
470482, 2013.
[51] P. Grunewald and J. Torriti, Demand Response from the Non-
Domestic Sector: Early UK Experiences and Future Opportunities,
Energy Policy, vol. 61, pp. 423429, 2013.