Applied Energy 155 (2015) 195–203

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

An automated residential demand response pilot experiment, based

on day-ahead dynamic pricing
Koen Vanthournout a,⇑, Benjamin Dupont b, Wim Foubert c, Catherine Stuckens c, Sven Claessens a
a
EnergyVille and VITO, 2400 Mol, Belgium
b
EnergyVille and KU Leuven, 3000 Leuven, Belgium
c
Laborelec, 1630 Linkebeek, Belgium

h i g h l i g h t s

 We present experimental demand response pilot results, based on dynamic prices.

 The pilot included smart whitegood appliances and smart domestic hot water buffers.
 The whitegood appliances, esp. the dishwashers, outperform the hot water buffer.
 The larger energy consumption of the buffer yields larger absolute savings.
 The large share of non-smart consumption is a financial risk for the end user.

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

Article history: Dynamic pricing is a popular method to realize demand response. Automated response from smart appli-
Received 22 December 2014 ances reduces the comfort impact for the users and hence reduces response fatigue concerns, while
Received in revised form 28 May 2015 improving the price response. However, real-life experience with smart appliances is typically limited
Accepted 29 May 2015
to heating and cooling appliances. The Linear pilot was a residential demand response pilot with 240
Available online 18 June 2015
Belgian families using smart dishwashers, washing machines, tumble dryers and domestic hot water buf-
fers in various experiments. Goal was to evaluate the performance of those smart appliances in real life
Keywords:
circumstances for various applications of demand response. The results for the day-ahead dynamic pric-
Residential demand response
Pilot
ing experiments, conducted from September 2013 till July 2014 at 58 families, are presented. These
Dynamic pricing demonstrate a significant shift of the flexible share of the electricity consumption to the lower price peri-
Smart appliances ods. The dishwashers outperform the other appliances. The domestic hot water buffer shows the lowest
performance in terms of relative cost savings, but its much larger energy consumption translates to larger
absolute savings. As the flexible share of the total consumption remains small, the non-smart share rep-
resents a financial risk for the consumer. The smart appliances were well received by the users and no
response fatigue was observed. However, there was a high variation in the group of pilot participants,
both in terms of energy consumption as in terms of flexibility offered.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction basic requirement to enable sustained participation of families in

Demand response (DR) is increasingly deployed in the European
industry to extend the traditional production-based balancing and
congestion management reserves of Balancing Responsible Parties
(BRP), Distribution System Operators (DSO) and Transmission
System Operators (TSO). The large DR potential in the European
residential sector, on the other hand, remains hitherto unused, as
other criteria than for the industry apply; comfort protection is a
⇑ Corresponding author.
E-mail address: koen.vanthournout@vito.be (K. Vanthournout).
DR schemes, the sources of flexibility are small in power and
energy, but their numbers are large, etc. As such, the technology
required to unlock the residential potential is fundamentally dif-
ferent from the DR technology used in the industry.
The Linear residential demand response pilot [1,2] aspired to
bridge this gap and for this purpose, tested various residential
DR technologies and control schemes in practice. One of the
schemes developed and tested in Linear aims at realizing auto-
mated residential DR based on day-ahead dynamic pricing (DP),
i.e., to have ‘smart’ appliances autonomously react to variable
prices, within the comfort settings of the users. The technical goal

http://dx.doi.org/10.1016/j.apenergy.2015.05.100
0306-2619/Ó 2015 Elsevier Ltd. All rights reserved.
196 K. Vanthournout et al. / Applied Energy 155 (2015) 195–203
of the Linear tariff structure is to help compensate for day-ahead
predicted variations in wind and solar energy, while at the same
time ameliorate distribution grid congestion.
After situating the Linear residential automated DR/DP
approach in the field of residential DR pilots (Section 2), we elabo-
rate on the tariff structures (Section 3). Section 4 describes the
Linear pilot setup, followed by the dynamic pricing control algo-
rithm specification (Section 5). In Section 6 the measurement
results from the Linear residential demand response pilot are pre-
sented and analyzed and Section 7 summarizes the feedback from
the users.
2. Related work
Due to its technical simplicity, dynamic pricing (DP) is a popular
method in attempts to influence the electrical energy consumption
of residential end-users. The most straightforward application of
dynamic tariffs is via manual demand response, i.e., to inform the
end-user of the variable prices and to rely on that same user to man-
ually shift electrical consumption from the expensive periods to the
cheaper ones. However, a point of debate is the concern regarding
the decreasing efficiency of such manual schemes due to ‘response
fatigue’; the end-user tires of continuously checking prices and the
resulting comfort impact, resulting in decreased involvement or a
switch to non-dynamic pricing schemes. The literature is indecisive
on this topic [3–6]. An alternative dynamic pricing DR scheme that
circumvents the response fatigue concern, is automated DR. Here,
smart appliances respond to the prices, where the impact on the
user’s comfort is limited to the configuration of those smart appli-
ances (see Section 4). As a complex tariff scheme was rolled out in
the Linear pilot (see Section 3), concerns regarding response fatigue
arose and it was decided to deploy both manual and automated DR
for the dynamic pricing experiments, the latter of which is dis-
cussed here. The choice for automated demand response was fur-
ther supported by the evidence that this improves the price
response [7], up to 200% compared to manual response [8–10].
Many examples of simulations [11–15] and lab tests [16–19]
can be found in the literature, proposing and evaluating a wide
range of pricing schemes, smart appliances and control strategies.
On the other hand, although a lot of residential demand response
pilots have been conducted and even a fair number of commercial
initiatives exist, especially in the U.S., the range of smart appli-
ances covered by these initiatives is limited [7–10,20–26]. Most
do not include smart appliances and are based on manual response
solely, and those that do include smart appliances limit automated
control to heating or cooling appliances, and – rarely – pool pumps.
For the European commercial variable pricing schemes, e.g., Option
tempo in France, PVPC in Spain or Vattenfall’s Nordic power
exchange coupled variable pricing in Sweden [27], no public data
is available if smart appliances are used within these schemes,
nor with what success.
Two noteworthy exceptions are the pilot results discussed in
[28,29]. In [28], the results for a German pilot are discussed, in
which smart washing machines and tumble dryers were tested
by 41 participants, using a pricing scheme composed of hourly
prices set day-ahead to one of three price levels. However, the per-
formance results are based on user interviews and dairies, and not
on measurements. Ref. [29] describes the results of a Dutch pilot,
more specifically of 50 smart washing machines used by partici-
pants to a dynamic tariff scheme. The tariffs are based on day
ahead energy market prices and on the local transformer load,
which contains a strong photovoltaic production component.
Note that the results presented in [29] do not separate the impact
of manual behavioral changes from the impact of the automated
demand response actions. In Sections 6 and 7, the results from
[29] are discussed in light of our own findings.
This paper presents the Linear pilot measurement results on the
performance of smart Domestic Hot Water (DHW) buffers, smart
washing machines, smart dishwashers and smart tumble dryers
during day-ahead dynamic pricing experiments. We fill the gap
of empiric data on the performance of all types of smart whitegood
appliances under dynamic pricing in real life conditions, and com-
pare this performance to that of the more commonly used DHW
buffer.
3. Time of use tariff structure
The day-ahead dynamic tariff scheme used within Linear, con-
sisted of several components: energy, distribution, transmission,
and levies. Various ways exist to construct dynamic pricing struc-
tures from these components [30,31]. The Linear project opted for
a dynamic energy and distribution component, and a constant
transmission and levies component. The following paragraphs
summarize how the dynamic components were constructed.
More elaborate descriptions of the methodology can be found in
[13,32]. The constant transmission and levies components are
based on the Belgian averages and amount 10 €/MW h and
70 €/MW h, respectively.
The dynamic energy component is based on the historic Belgian
day-ahead wholesale market prices of 2011 [33]. These historic
prices were corrected to reflect a future larger share of renewable
generation, according to the 2020 scenario defined in [34]. The [34]
data was combined with public data on the annual generation
profiles of solar and wind plants [35,36], to calculate the hourly
power deviation from the historic power generation due to the
increased share of renewable generation. This was converted into
an hourly financial correction on the historic wholesale prices,
using the wholesale price sensitivity for increases in offer or
demand, as presented in the market resiliency analysis in [37].
Result is the adjusted hourly wholesale price (WPadjusted;p ).
As the wholesale price only accounts for part of the energy com-
ponent, a second correction is needed in order to attain revenue
neutrality [38]. A rescaling factor (rfe) was applied to ensure that
the same revenues are accrued using the dynamic prices or the
previously flat energy tariff, if the averaged residential user1 does
not change his consumption pattern (SLPp ). Or:
X
8760
  
SLP p  WPadjusted;p  rfe ¼ FlatEnergy ð1Þ
p¼1
with SLPp the synthetic load profile during hour p (% of yearly con-
sumption) [39], WP adjusted;p the adjusted wholesale price during hour
p (€/MW h), rfe the rescaling factor for the energy component and
FlatEnergy the (old) flat energy tariff component (€/MW h). FlatEnergy
was set to 80 €/MW h.
The dynamic distribution component varies with the level of
the electricity usage of residential users according to the following
formula:
SLPp
RTPDistr;p ¼ P8760 2
 FlatDistr ð2Þ
p¼1 SLP p
with RTPDistr;p the dynamic distribution tariff component (€/MW h)
and FlatDistr the average distribution tariff component over the year
(70 €/MW h).
The hourly distribution price is determined based on the ratio
between the hourly usage and its weighted average over the year.
This results in a higher distribution tariff when the hourly electric-
ity usage of residential users is above the weighted average.
1
Based on the official Synthetic Load Profiles (SLP), used in the Belgian energy
market [39].
 K. Vanthournout et al. / Applied Energy 155 (2015) 195–203
After both the dynamic energy and distribution component are
attained and summed with the flat transmission and levies compo-
nents, the dynamic tariff design is applied. Within Linear, the daily
price pattern consisted of 6 price blocks. The beginning and end of
these blocks were fixed for every day of the year and were selected
based on a analysis of the similarity between the price levels in
adjacent hours. The prices for each block varied on a daily basis.
The tariff for one random day, split into the 4 tariff components,
is visualized in Fig. 1. Prices for the next day were published
each day at 16:00, i.e., after the clearing of the day-ahead
market. Minimum price was 47.82 €/MW h, maximum price
441.694 €/MW h. The minimum, average and maximum daily peak
to off-peak ratio was 1.19, 1.78 and 3.79, respectively.
As can also be seen in Fig. 1, patterns emerge in the tariff struc-
ture, which are useful to keep in mind when interpreting results.
Barring a few exceptions, the 0–7 h block and the 13–17 h block
are the cheapest periods of the day. During winter time, the highest
day price can typically be found in the 17–24 h blocks, coinciding
with the Belgian winter evening consumption peaks. This evening
price peak lowers towards summer time, following the Belgian sea-
sonal electricity consumption pattern. In summer, it depends
mainly on the solar and wind production whether the 7–13 h
blocks or the 17–24 h blocks contain the highest day price.
Summarized, for a good performance, the demand respond actions
should shift consumption towards the night or afternoon.
4. The Linear residential demand response pilot
The Linear residential demand response pilot [40] consisted of
240 participating families, 186 of which were equipped with smart
appliances, while the other 54 families participated in a manual
dynamic pricing scheme. The 186 families with smart appliances
participated to four demand response control schemes, i.e., the
use of automated DR to reduce temperatures in transformers and
hence increase their life length, automated DR to support the line
voltage in low voltage distribution grids, automated DR to reduce
the intra-day imbalance caused, e.g., by intermittent wind produc-
tion, and the use of automated demand response to optimize the
BRP’s day-ahead portfolio through the use of dynamic prices,
which is the focus of this work [1,40].
As comfort protection is an essential requirement for residential
demand response, Linear has selected and deployed two types of
smart appliances that offer significant flexibility and that can be
automated to minimize comfort impact (see Fig. 2).
First type are the postponable appliances, such as dishwashers,
washing machines and tumble dryers, which were supplied by two
different manufacturers, and 445 of which were deployed in the
Linear pilot. When configuring these appliances, users were
requested to set a deadline for the end of the appliance’s program
as far as possible in the future, with a maximum of 24 h. This gives
the Linear system a time window in between configuration time
and start deadline,2 during which it can freely choose the optimal
start time for the selected program. Once started, the cycle cannot
be interrupted. Note that users were not obliged to configure their
appliances ’smart’ and that they could always choose to start it
immediately.
Second type of smart appliances were the buffered appliances,
such as smart domestic hot water (DHW) buffers [16] (15
deployed) and electric vehicles (7 deployed). Once the comfort set-
tings of the DHW buffers are set, no further user interaction is
required. And since the default comfort settings of the DHW
2
Users always configured an end deadline; the Linear system then immediately
converted this to the start deadline for the cycle, which was the settings used by the
control.
197
Fig. 1. An example of the Linear dynamic tariffs and its composition for one day.
buffers proved satisfactory in all cases, no user interactions were
required at all. The electric vehicles are configured analogue to
the postponable appliances, i.e., the user is requested to set a
departure deadline as far as possible in the future. The electric
vehicles did not participate to the dynamic pricing experiments,
all other appliances did.
The flexibility made available by the participants was used for
various experiments and control schemes. To hide this complexity
from the users, they received a capacity fee as an incentive to
always offer as much flexibility as possible. This capacity fee was
based on the hours of flexibility offered, which equals the number
of hours between configuration and start deadline. As the DHW
buffers required no user interaction, and as the sole objective of
the capacity fee was to stimulate the amount of flexibility offered,
the buffers were not included in the capacity fee mechanism. Note
that this implies that the users where not able to consult the
dynamic tariffs.
Additionally, 110 families were equipped with smart meters,
extra measurement equipment was installed at the residences
without smart meter and close to 2000 submetering points were
rolled out. All smart appliances were equipped with submetering,
the data of which is used for the analyses presented in Section 6.
More details on the Linear ICT architecture can be found in [41],
the overview of the Linear project and a summary of the results
can be found in [40].
5. Control strategy
The next sections elaborate on the control algorithm, the prime
aim of which is to minimize the energy cost of the smart appli-
ances. As the consumption and hence the cost of each appliance
is mutually independent, each appliance can be controlled sepa-
rately based on the prices published and the settings of the user.
For the postponable appliances, a secondary peak avoidance con-
trol criterion is imposed to avoid concurrent consumption, if this
has no cost impact and if the user settings permit so.
The focus of the work presented here was to implement and roll
out automated DR technology in the field, where it was part of the
participant’s daily routine. Main target was to evaluate how suited
residential flexibility is for dynamic pricing demand response. As
such, the prime design criteria for the control logic were simplicity
and robustness to minimize technical issues during the pilot,
rather than optimality. More advanced methods can be found in
the literature, e.g. as summarized in [42].
5.1. Postponable appliances
Linear residences were equipped with a maximum of three
postponable appliances, i.e., one dishwasher, one washing machine
198 K. Vanthournout et al. / Applied Energy 155 (2015) 195–203

Fig. 2. The Linear smart appliances lined up in one of the development and testing sites for the pilot; the VITO HomeLab.

and one tumble dryer. Each configured appliance is defined by the

user configuration time tconf , the start deadline set by the user tdl
and the load curve plc , i.e., the estimate of the power consumption
in function of the time, defined by the appliance and based on the
program the user selected. The start time t on which the appliance
is activated, is then the result of following optimization problem:
For the dishwasher:
 
t dw ¼ minðplc;dw  c þ t dw  tdw;conf aÞ
ð3Þ
s:t: t dw;conf 6 t dw 6 tdw;dl

with a very small, so price takes absolute precedence over the dish-
washer starting as early as possible. c is the array with the dynamic
prices in function of the time.
For the washing machine: Fig. 3. Example scheduling of a dishwasher from the pilot, captured on 21/09/2013.
  At 17:30 the user configured the dishwasher to start the latest at 03:45. The
t wm ¼ minðplc;wm  c þ twm;dl  twm bÞ dishwasher is scheduled and started at 0:00, when the tariff block with the lowest
ð4Þ cost begins. P dw;s shows the consumption of the dishwasher, as measured by a
s:t: t wm;conf 6 t wm 6 t wm;dl
submetering plug, and P res;s the total consumption of the residence, as measured by
the smart meter. To illustrate the effect of the control, P dw;c and P res;c illustrate what
with b very small, so price takes absolute precedence over the
the dishwasher and residence consumption would have been, if the appliance was
washing machine starting as late as possible. started at the configuration time. All power values are average power per quarter
For the tumble dryer: hour in Watt.

t td ¼ min plc;td  c
   
þ size ½t td ; ttd þ lengthðplc;td Þ \ ½t i ; t i þ lengthðplc;i Þ  c SoC > SoC min [16]. The latter is important to handle delayed control
   ð5Þ responses, e.g., caused by the quarter hour control loop. Following
þttd  t td;conf a
control sequence is repeated every quarter hour, for each DHW
s:t: t td;conf 6 ttd 6 t tw;dl buffer:
with i all currently configured whitegood appliances, and with c
very small, so price takes absolute precedence over the tumble if ccurrent 6 ci ; i 2 today½0h : 24h then
dryer not coinciding with the other appliances and a  c, so the if ðP ¼ 0Þ & ðSoC < 80%Þ then
overlap avoidance in turn takes precedence over starting the earli- P¼1
est as possible. end if
The time resolution used for t is a quarter hour, which limits the if ðP ¼ 1Þ & ðSoC P 88%Þ then
search space. An exhaustive search is used to solve the optimiza- P¼0
tion problem. See Fig. 3 for an example of this scheduling end if
procedure for a dishwasher. else
if ðP ¼ 0jjccurrent – cprev ious Þ & ðSoC < 80%Þ then
5.2. Thermally buffered appliances if ccurrent 6 ci ; i 2 ½tnow : tnow þ t SoC min  then
P¼1
Every quarter hour, the control system collected the DHW buf- else
fer status updates, which consists – amongst others – of the state P¼0
of charge (SoC, in %) and the minimum state of charge (SoC min , in end if
%), i.e., the SoC that must always be maintained, so that any imme- end if
diate hot water peak demands can be covered. SoC min was set to if ðP ¼ 1Þ & ðSoC P 80%Þ then
30% for all users. The control signal is P 2 f0; 1g, i.e., the buffer is P¼0
charging or not charging. Note that the buffer has comfort and end if
safety overrules; if the SoC P 100%, the buffer automatically stops end if
charging, if SoC 6 SoC min the buffer automatically recharges until
 K. Vanthournout et al. / Applied Energy 155 (2015) 195–203 199

Fig. 4. Example of the DHW buffer behavior during one day, measured in the Linear
pilot on 22/09/2013. Illustrated are the tariffs, the buffer’s average power per
quarter hour, as measured by a submetering plug and the SoC of the buffer. Note the
SoC control limit overshoot, due to the fact that an update of the buffer state is
available only once per quarter hour.
Fig. 6. The share per hour of the energy consumption of the smart DHW buffers –
when controlled based on the dynamic tariffs –, throughout the day. A significant
share of the consumption is shifted to the lower cost night and afternoon blocks.
The consumption in the more expensive blocks is caused by the comfort protection
logic. It must be noted, however, that this comfort protection logic was set very
conservative to minimize the risks for the Linear participants. Additional perfor-
mance gains are possible by more correctly dimensioning the comfort margins.

switching. See Fig. 4 for an example of the DHW buffer control dur-
ing one day in the Linear pilot.
Table 1
Overview automated DR/DP experiments in the Linear pilot.
6. Pilot results
Experiment start Experiment end # Participating families
01/09/2013 17/09/2013 36
Presented are the measurement results for the automated
17/09/2013 05/10/2013 48
05/10/2013 20/11/2013 52 DR/DP experiments in the Linear pilot, captured in the periods as
06/01/2014 23/01/2014 57 listed in Table 1. During these periods, data was gathered for in
03/03/2014 07/03/2014 58 total 37 unique dishwashers, 51 unique tumble dryers, 49 unique
17/03/2014 28/03/2014 58 washing machines and 6 unique DHW buffers. Of these 6 boilers,
31/03/2014 16/04/2014 58
the data of only 4 was valid, as the other 2 participants used a
30/04/2014 26/05/2014 58
16/06/2014 01/07/2014 58 timer controlled switch, set to coincide with the Belgian day/night
tariff, despite the fact that the Linear project compensated any
day/night shifts with a negative return. The power measurements
on the whitegood appliances are analyzed using the appliance pro-
file detection algorithms presented in [43]. All whitegood appli-
with c cost, ccurrent the cost when the control sequence is executed
ance cycles and all DHW buffer control periods, for which both
and cprev ious the cost during the previous execution of the control
the submetering measurements and the smart appliance status
sequence. t now is the time now, in quarter hours. tSoC min is an estimate
data was correctly registered, are included in the analysis.
of when the buffer will reach SoC ¼ SoC min , also in quarter hours:
Whitegood cycles configured by the user to immediately start,
 
SoC  SoC min i.e., ’non-smart’ configurations, are excluded from the analyzes.
t SoC min ¼ ð6Þ All results presented are based on in situ energy measurements.
k=4
As explained in Section 3, the 0–7 h night and the 13–17 h after-
with k the discharge rate, which was based on historic averages and noon blocks are typically the cheapest periods. The performance of
set to 10% per hour. Note that the control sequence is a trade-off the system, which is a combination of the flexibility offered by the
between cost optimization and avoidance of excessive buffer participants and the behavior of the control logic, can be measured

Fig. 5. The share of smart user configurations and smart starts by the DR logic, per hour and throughout the day, for each of the three whitegood appliance types. The cycles
are strongly shifted to the cheaper night and afternoon tariff blocks. Dishwashers outperform the other appliances, as they are typically configured in the evening and can
more easily be shifted into the lower cost night. Note also the effect of the control logic to start washing machines as late as possible. The washing machine results align with
the results presented in [29].
200 K. Vanthournout et al. / Applied Energy 155 (2015) 195–203

Table 2 DHW buffers would have been in the absence of DR control.

Savings smart start time vs. configuration time (per cycle). Therefore, Table 3 compares the cost of the appliance consumption
Device Average savings (%) Standard deviation (%) subjected to the dynamic tariffs to that of the same consumption
Dishwashers 18.26 15.29 subjected to the flat tariff of 200 €/MW h (see Section 3). The results
Tumble dryers 10.07 16.07 are consistent with the earlier findings: significant reductions, most
Washing machines 9.18 13.21 explicitly so for the dishwashers. The performance of the DHW buf-
fers is lower, which can be attributed to the comfort protection.
Also clearly shown in Tables 2 and 3, is the high standard devi-
ation for all appliances. This is confirmed in Figs. 7–9. There is a
Table 3
Savings dynamic tariff vs. flat tariff per appliance type. The savings for the whitegood
large spread of the pilot participant’s behavior, both in terms of
appliances are averages per cycle, and absolute for the DHW buffers. energy consumption per appliance, as in the amount of flexibility
made available per appliance. This translates into a large variation
Device Average savings (%) Standard deviation (%)
in savings per user (Fig. 7). The deciding factor for the whitegood
Dishwashers 18.33 18.76 appliance savings is the time of configuration, exemplified by the
Tumble dryers 8.92 20.16
Washing machines 10.97 18.51
link between the typical evening configuration and larger savings
DHW buffer 1 8.93 / for the dishwasher. As Fig. 7 shows, there is also a link with how
DHW buffer 2 5.05 / often a user offers flexibility with a specific appliance; the more
DHW buffer 3 1.97 / a user configures the appliances with a flexible window, the more
DHW buffer 4 1.71 /
the (higher probability) configurations with a higher return out-
weigh those with a low return, resulting in a better average profit.
However, with the spread on the prices deployed in the Linear pilot
by the energy shifts to those two cheaper blocks and is illustrated (average daily peak to off-peak ratio of 1.78), the savings remain
in Figs. 5 and 6. Note the large shifts realized for the whitegood modest.
appliances, esp. the dishwasher, mainly because the flexibility The whitegood appliances outperform the DHW buffers in
offered by the participants coincides with what is required to shift terms of relative savings. However, the energy consumed by
consumption to the cheaper periods. This is also confirmed in DHW buffers is much higher than that of whitegood appliances
Table 2, that lists the savings per cycle that can be realized when and not all energy consumed by the whitegood appliances is avail-
a whitegood appliance is started by the demand response system, able for the demand response system (Figs. 8 and 9). Hence the
compared to when it would start when the user configured the absolute savings for a DHW buffer are significantly larger than
appliance, both subject to the dynamic tariffs. Important here is for whitegood appliances.
to note that day-ahead dynamic tariffs is but one of the demand The profits from the flexible loads can be enlarged by increasing
response experiments the users participated to, and that they were the price spread. However, this also increases the financial risk of
remunerated using a capacity fee, as not all experiments were the non-flexible share for the consumer. Due to the revenue neu-
based on dynamic pricing (see Section 4). Users were requested trality, and if the consumers do not deviate strongly from the aver-
and stimulated to provide as much flexibility as possible, regard- age consumption profile, the non-flexible share should not incur
less of the time of day. They could not consult the prices and as significant financial losses or profits. On average this is so, but
such the user behavior was not influenced by the dynamic prices. there are large individual differences (see Table 4).
With the measurements Linear collected, it is not possible to Only a small share of the total consumption is flexible, espe-
deterministically define what the consumption profile of the cially for those families with solely smart whitegood appliances.

Fig. 7. The average savings per whitegood appliance cycle (smart start time vs. configuration time) for the 30 users that offered most flexibility during the dynamic pricing
experiments, together with a count per user and per whitegood appliance of the total number of smart cycles configured during the dynamic pricing experiments.
 K. Vanthournout et al. / Applied Energy 155 (2015) 195–203 201

Fig. 8. The average energy consumption per week of the whitegood appliances (kW h/week) for the 30 users that offered most flexibility during dynamic pricing experiments,
with the flexible (configured with smart delay) and non-flexible shares indicated. The same ordering as for Fig. 7 is maintained. The flexible share is an underestimate, as only
the cycles with valid registration and valid measurements are taken into consideration.

Fig. 9. The average energy consumption per week (kW h/week) of the DHW buffers during the dynamic pricing experiments. The same ordering as for Fig. 7 is maintained.

This implies that – although significant flexible volumes could be
aggregated on national levels – the physical impact on the local
distribution grid is limited, as is confirmed by the experiments
conducted within the Linear pilot that specifically target local con-
gestion management [40].
7. User acceptance
54 of the 240 families participating to the Linear pilot did not
receive smart appliances, and were provided access to a portal, dis-
playing the dynamic prices. They were requested to manually alter
their consumption according to the prices, and received a financial
bonus based on their shifting performance. Analysis of their
consumption profiles showed that there were very limited to no
behavioral changes and that response fatigue was widespread
[44]. A questionnaire post the pilot, confirmed that these users
found responding manually to the Linear dynamic tariff scheme
too invasive, too much effort, and too complex. On the other hand,
no user fatigue was observed for the users of the smart whitegood
appliances, although there was a large spread of the share of smart
delay configurations. The smart DHW buffer required no interven-
tions of the users, and no user reported reduced comfort, confirm-
ing the effectiveness of the comfort protection measures (see
Section 5.2).
Questionnaires were conducted before and after the pilot.
Comfort impact was measured using questions regarding the
202 K. Vanthournout et al. / Applied Energy 155 (2015) 195–203
Table 4
Statistics on the total consumption per household for 50 households and the non-
smart share of that consumption, together with the corresponding savings of the
dynamic tariff vs. flat tariff.
Average Std. dev. Minimum Maximum
Total 2385 kW h 1400 kW h 570 kW h 6215 kW h
Profit total 0.11% 1.84% 4.86% 5.97%
Non-smart 2323 kW h 1348 kW h 534 kW h 6207 kW h
Profit non-smart 0.15% 1.90% 6.26% 6.04%
expected impact on the daily schedule and regarding the loss of
comfort. Before the pilot this yielded a score of 4.03/5 (5 means
no impact). After the pilot the score was 4.02/5. The small per-
ceived comfort impact by the users before the pilot was hence con-
firmed during the pilot. The perceived ease of use before the pilot
was 3.96/5 (measured using 4 ease of use related questions, with a
high score meaning low complexity). After the pilot this dropped to
3.41/5. The lower score, however, was mainly caused by the tech-
nical issues with the prototype equipment used in the pilot. For
more detailed user acceptance results from the Linear pilot, see
[40].
This confirms that for complex pricing schemes that require fre-
quent price consultation in case of manual response, automated
response by means of smart appliances is the better method from
a user’s point of view.
Ref. [29] presents the results for 50 smart washing machines. A
functionally identical user interface to the one deployed in the
Linear pilot was used and the dynamic prices were comparable
in complexity. Also here, little demand response was observed
with regard to the use of the smart washing machines. However,
only 14% of the washing machine configurations were smart start
configurations, while smart configurations did result in superior
load shifting. The authors of [29] suggest to investigate methods
to stimulate users to use more automation. The capacity fee used
in the Linear pilot, which links user remuneration directly to the
use of smart starts, while hiding the complexity of the dynamic
pricing scheme, is such a method. 29% of the washing machine
configurations in the Linear pilot were smart start configurations.
8. Conclusions
The Linear pilot tested the performance of a wide range of res-
idential smart appliances in real life conditions when used in a
day-ahead dynamic pricing demand response scheme. We pre-
sented how these dynamic tariffs were constructed and detailed
how the various smart appliances, installed at the residences of
the Linear participants, were controlled. Results for automated
DR/DP experiments, conducted from September 2013 till July
2014 with 58 families, are analyzed. Although there was a large
spread in the amount of flexibility made available through the
smart appliances by the users, smart appliances were well received
by the users and no response fatigue was observed.
The experiments demonstrate a significant shift of the flexible
share of the electricity consumption to the lower price periods.
The dishwashers outperform the other appliances, as the flexibility
pattern of the dishwashers (time of configuration and length of
delayed start window) coincides best with the price pattern.
Although the DHW buffer shows the lowest performance in terms
of relative savings, its much larger energy consumption translates
to larger absolute savings. With the price spread used in Linear, the
overall gains remain modest. However, increasing the spread to
increase smart appliance gains would also increase the financial
risk for the consumer, due to the large non-flexible share of the
consumption. Another finding is the high variation in the group
of pilot participants, both in terms of energy consumption as in
terms of flexibility offered.
For dynamic pricing to be risk free and financially more inter-
esting than flat tariffs, either the flexible share of the consumption
must be larger, e.g., with smart electric heating and/or smart
charging of electric vehicles, or the dynamic tariff should be
applied only to the separately measured flexible share of the
consumption.
Acknowledgment
This work was supported in part by the Flemish Ministry of
Science via the project Linear organized by the Institute for
Science and Technology (IWT).
References
[1] Linear (local intelligent networks and energy active regions). <http://
www.linear-smartgrid.be/?q=en>.
[2] Dupont B, Vingerhoets P, Tant P, Vanthournout K, Cardinaels W, De Rybel T,
et al. Linear breakthrough project: large-scale implementation of smart grid
technologies in distribution grids. In: 2012 3rd IEEE PES international
conference and exhibition on innovative smart grid technologies (ISGT
Europe); 2012. p. 1–8.
[3] Faruqui A, Hledik R, Palmer J. Time-varying and dynamic rate design, Tech.
rep., The Brattle Group (July 2012).
[4] Kim J-H, Shcherbakova A. Common failures of demand response. Energy
2011;36(2):873–80.
[5] Cappers P, Mills A, Goldman C, Wiser R, Eto JH. An assessment of the role mass
market demand response could play in contributing to the management of
variable generation integration issues. Energy Policy 2012;48(C):420–9.
[6] Bartusch C, Alvehag K. Further exploring the potential of residential demand
response programs in electricity distribution. Appl Energy 2014;125(0):39–59.
[7] Frontier Economics. Sustainability first, demand side response in the domestic
sector – a literature review of major trials, Tech. rep., UK Department of Energy
and Climate Change (August 2012).
[8] Stromback J, Dromacque C, Yassin MH. The potential of smart meter enabled
programs to increase energy and systems efficiency: a mass pilot comparison,
Tech. rep., VaasaETT; 2011.
[9] Woo C, Li R, Shiu A, Horowitz I. Residential winter kW h responsiveness under
optional time-varying pricing in British Columbia. Appl Energy
2013;108(C):288–97.
[10] Herter K, McAuliffe P, Rosenfeld A. An exploratory analysis of california
residential customer response to critical peak pricing of electricity. Energy
2007;32(1):25–34.
[11] Iacovella S, Geth F, Ruelens F, Leemput N, Vingerhoets P, Deconinck G, et al.
Double-layered control methodology combining price objective and grid
constraints. In: SmartGridComm. Vancouver, Canada: IEEE; 2013. p. 25–30.
[12] Gottwalt S, Ketter W, Block C, Collins J, Weinhardt C. Demand side
management – a simulation of household behavior under variable prices.
Energy Policy 2011;39(12):8163–74.
[13] Dupont B, Tant J, Belmans R. Automated residential demand response based on
dynamic pricing. In: 2012 3rd IEEE PES international conference and
exhibition on innovative smart grid technologies (ISGT Europe); 2012. p. 1–7.
[14] Gaiser K, Stroeve P. The impact of scheduling appliances and rate structure on
bill savings for net-zero energy communities: application to west village. Appl
Energy 2014;113(C):1586–95.
[15] Cole WJ, Rhodes JD, Gorman W, Perez KX, Webber ME, Edgar TF. Community-
scale residential air conditioning control for effective grid management. Appl
Energy 2014;130(0):428–36.
[16] Vanthournout K, D’hulst R, Geysen D, Jacobs G. A smart domestic hot water
buffer. IEEE Trans Smart Grid 2012;3(4):2121–7.
[17] Widergren S, Subbarao K, Chassin D, Fuller J, Pratt R. Residential real-time
price response simulation. In: 2011 IEEE power and energy society general
meeting; 2011. p. 1–5.
[18] Dütschke E, Paetz A-G. Dynamic electricity pricingWhich programs do
consumers prefer? Energy Policy 2013;59(C):226–34.
[19] Finn P, O’Connell M, Fitzpatrick C. Demand side management of a domestic
dishwasher: wind energy gains, financial savings and peak-time load
reduction. Appl Energy 2013;101:678–85.
[20] Faruqui A, Sergici S. Household response to dynamic pricing of electricity: a
survey of 15 experiments. J Regul Econ 2010;38(2):193–225.
[21] Newsham GR, Bowker BG. The effect of utility time-varying pricing and load
control strategies on residential summer peak electricity use: a review. Energy
Policy 2010;38(7):3289–96.
[22] Saele H, Grande O. Demand response from household customers: experiences
from a pilot study in Norway. IEEE Trans Smart Grid 2011;2(1):102–9.
[23] Chassin DP, Kiesling L. Decentralized coordination through digital technology,
dynamic pricing, and customer-driven control: the GridWise testbed
demonstration project. Electr J 2008;21(8):51–9.
[24] Torriti J. Price-based demand side management: assessing the impacts of
time-of-use tariffs on residential electricity demand and peak shifting in
northern Italy. Energy 2012;44(1):576–83.
 K. Vanthournout et al. / Applied Energy 155 (2015) 195–203
[25] Herter K, Wayland S. Residential response to critical-peak pricing of
electricity: California evidence. Energy 2010;35(4):1561–7.
[26] Thorsnes P, Williams J, Lawson R. Consumer responses to time varying prices
for electricity. Energy Policy 2012;49(0):552–61. special Section: Fuel Poverty
Comes of Age: Commemorating 21 Years of Research and Policy.
[27] Torriti J, Hassan MG, Leach M. Demand response experience in Europe:
policies, programmes and implementation; 2010.
[28] Stamminger R, Anstett V. The effect of variable electricity tariffs in the
household on usage of household appliances. Smart Grid Renew Energy
2013;4(4):353–65.
[29] Kobus CB, Klaassen EA, Mugge R, Schoormans JP. A real-life assessment on the
effect of smart appliances for shifting households’ electricity demand. Appl
Energy 2015;147(0):335–43.
[30] Dupont B, Jonghe CD, Olmos L, Belmans R. Demand response with locational
dynamic pricing to support the integration of renewables. Energy Policy
2014;67(0):344–54.
[31] Borenstein S, Jaske M, Ros A. Dynamic pricing, advanced metering, and
demand response in electricity markets. J Am Chem Soc
2002;128(12):4136–45.
[32] Dupont B, De Jonghe C, Kessels K, Belmans R. Short-term consumer benefits of
dynamic pricing. In: 2011 8th international conference on the European
energy market (EEM); 2011. p. 216–21.
[33] Belpex <http://www.belpex.be/>.
[34] Federal-Regional Energy Consultation Goup, National renewable energy action
plan pursuant to directive 2009/28/ec, Brussels, Belgium (November 2010).
203
[35] Soda. Solar energy data for professionals. <http://www.soda-pro.com>.
[36] Koninklijk Nederlands Metereologisch Instituut (KNMI), Potential wind speeds
of the netherlands <www.knmi.be>.
[37] Belpex. Market resilience analysis; 2013. <https://www.belpex.be/wp-
content/uploads/marketdata/belpex/public/CWE_Market_Resilience_Analysis/
Market-resilience-analysis-2013.pdf>.
[38] Faruqui A, Hledik R, Sergici S. Piloting the smart grid. Electr J
2009;22(7):55–69.
[39] Synergrid, Synthetic load profiles (slp’s). <http://www.synergrid.be/index.
cfm?PageID=16896>.
[40] Cardinaels W, et al. Linear – demand response for families; 2014. <http://
www.linear-smartgrid.be/sites/default/files/boekje_linear_okt_2014_boekje_
web.pdf>.
[41] Strobbe M, Vanthournout K, Verschueren T, Cardinaels W, Develder C. Large-
scale residential demand response pilot ict architecture. In: Proc 5th IEEE PES
innovative smart grid technologies European conf. (ISGT-EU 2014), Istanbul,
Turkey; 2014. p. 1–6.
[42] De Craemer K. Event-driven demand response for electric vehicles in multi-
aggregator distribution grid settings, Ph.D. thesis, K.U.Leuven, Kasteelpark
Arenberg 10, 3001 Leuven (July 2014).
[43] Labeeuw W. Characterisation and modelling of residential electricity demand,
Ph.D. thesis, KU Leuven; 2013.
[44] Dupont B. Residential demand response based on dynamic electricity pricing:
theory and practice, Ph.D. thesis, K.U.Leuven, Leuven, Belgium (January 2015).