PreprintPDF Available
Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

Buildings' energy consumption has dramatically increased over the last decade, accounting for more than 35% of global energy use. In this article, RESPOND's approach to improve building operation is shown for reducing energy demand peaks. Namely, RESPOND develops an AI system to dispatch optimal DR events maximizing the renewable energy generation and leveraging energy storage units. Furthermore, with views to ensuring dwellers' engagement with the project, the proposed DR events are aimed to generate the least disturbance in their everyday activities.
Content may be subject to copyright.
An Artificial Intelligent System
for Demand Response in Neighbourhoods
Iker Esnaola-Gonzalez1, Francisco Javier Diez1
, Dea Puji´
c2, 3, Marko Jeli´
c2, 3 and Nikola Tomasevi´
Abstract. Buildings’ energy consumption has dramatically in-
creased over the last decade, accounting for more than 35% of global
energy use. In this article, RESPOND’s approach to improve building
operation is shown for reducing energy demand peaks. Namely, RE-
SPOND develops an Artificial Intelligent system to dispatch optimal
Demand Response events maximizing the renewable energy gener-
ation and leveraging energy storage units. Furthermore, with views
to ensuring dwellers’ engagement with the project, the proposed DR
events are aimed to generate the least disturbance in their every-day
1 Introduction
Buildings’ energy consumption has dramatically increased over the
last decade due to different factors including the population growth,
the increase in time spent indoors or the increased demand for build-
ing functions and indoor quality [1]. As a matter of fact, according
to the UNEP (United Nations Environment Programme), buildings
account for more than 35% of global energy use and nearly 40% of
energy-related CO2 emissions [2]. However, significant energy sav-
ings can be achieved in buildings if they are properly operated.
In this domain, the residential sector is specially promising as it
is characterized by many end consumers with relatively low individ-
ual energy demand, but with very high demand when considered in
terms of home clusters, districts and residential communities. Evi-
dence of this is that in 2016, residential buildings represented the
25.4% of final energy consumption and 17.4% of gross inland en-
ergy consumption in the EU4. The major end-uses responsible for
this figures are the space and water heating, followed by appliances,
cooking and lighting [3].
Apart from the large energy consumption of buildings, peak en-
ergy demand certainly attracts lots of attention because of their nega-
tive impact on energy grid capital, operational cost and environmen-
tal pollution to name a few. This impact is a direct consequence of the
carbon-intense generation plants that grid operators deploy in order
to satisfy energy demand during peak periods [4].
Renewable Energy Sources (RES) are increasingly penetrating the
energy production side and could contribute to significantly reduce
peak demands [5]. However, due to their intermittent nature, their
1TEKNIKER, Basque Research and Technology Alliance (BRTA),
naki Goenaga 5, 20600 Eibar, Spain. Email: {iker.esnaola, fran-
2School of Electrical Engineering, University of Belgrade, Bulevar kralja
Aleksandra 73, 11120 Belgrade, Serbia
3Institute Mihajlo Pupin, University of Belgrade, Volgina 15, 11060 Bel-
grade, Serbia. Email: {dea.pujic, marko.jelic, nikola.tomasevic}
Energy consumption in households
availability commonly does not match the distribution of energy de-
mand in time, which may hinder their management and exploitation.
Energy storage units may contribute to the reduction of the impact
of these uncertainties by smoothing out fluctuations and reducing the
mismatch between renewable energy supply and demand [6, 7].
Demand Side Management (DSM) activities including load cur-
tailment (i.e. a reduction of electricity usage) or load reallocation (i.e.
a shift of energy usage to other off-peak periods) have a huge poten-
tial to match energy demand with energy supply side, thus avoiding
these undesirable peaks. As a matter of fact, Demand Response (DR)
programs are introduced into the smart grids so that reliable and eco-
nomical operation of power systems are ensured [8]. DR can be un-
derstood as the set of technologies or programs that concentrate on
shifting energy use to help balance energy supply and demand [9]. In
combination with energy generated from RES and an adequate use of
energy storage units, DR is envisioned as one of the crucial enablers
of curbing energy demand peaks [10].
However, the implementation of DR programs is not straightfor-
ward. The main barriers to adopt DR programs include regulatory,
economic, technological and social issues [11]. Furthermore, the en-
ergy demand optimization based on the collectively produced energy
is a complex multiobjective problem that requires from Artificial In-
telligent (AI) systems that allow generating optimal multiple demand
profiles. And this is where the RESPOND H2020 project5originates,
aiming to bring DR programs to neighborhoods across Europe. With
most contemporary DR implementation considering only the indus-
trial sector, this paper will propose a measure-optimize-control loop
that is suitable for residential consumers that would like to make
use of DR and incorporate it into their energy management systems.
By simultaneously utilizing multiple different AI-based technologies
such as machine learning and optimization, as well as novel concepts
such as Internet of Things (IoT) deployed using smart sensors and
actuators in smart homes, the proposed system makes use of pre-
vious behavior through the energy prediction subsystem, considers
estimated energy production from renewable sources and looks for
modifications that would mitigate potential instabilities in the energy
supply network by applying optimal energy utilization and load shift-
ing. Finally, the predicted and optimal behavior are analyzed and cor-
rective actions are suggested for end users, with the system even al-
lowing for the possibility of semi-automatic (remotely actuated with
individual user consent) and fully automatic (remotely actuated with
previous persisting user consent) actions to be taken in order to mini-
mize the additional effort that needs to be undertaken and that would
disrupt daily habits.
The rest of the article is structured as follows. Section 2 introduces
the previous work related to presented problem. Section 3 introduces
the RESPOND project. Section 4 presents RESPOND’s AI system
for dispatching the optimal DR events. Finally, the conclusions of
this work are presented in Section 6.
2 Related Work
Recent research work has been focused on removing the assump-
tion that the demand is given and fixed, and on investigating feasible
DSM approaches that dynamically adjust the demand, within given
bounds, in order to fulfill or improve a specified performance require-
ment. These methods yield additional flexibility and introduce new
degrees of freedom as part of novel energy management approaches
and commercially available products (e.g. Siemens DRMS or Akua-
com DRMS by Honeywell), since the demand had otherwise been
assumed to be passive and static in classic formulations. In this re-
gard, a number of approaches investigating the demand side opti-
mization have been analyzed. For example, a multi-objective genetic
algorithm approach to implementing the DSM in an automated ware-
house has been presented [12]. Furthermore, a modified genetic al-
gorithm has been used to optimize the scheduling of direct demand
control strategies [13]. An autonomous and distributed demand side
energy management system based on game theory has also been pro-
posed [14]. Additionally, an autonomous DR system that tries to
achieve both optimally and fairness with respect to the involved par-
ticipants was designed [15]. A fuzzy logic approach utilizing WSN
and smart grid incentives for load reduction in residential HVAC sys-
tems has also been presented [16]. Finally, an integration of RES and
electric vehicles with proper home DSM was evaluated through dif-
ferent scenarios [17].
An integrated approach to both supply and DSM has in gen-
eral received somewhat less attention in the literature. An integrated
DSM program for multiple entities (represented by designated En-
ergy Hubs) was proposed as a non-cooperative game within a cloud-
based infrastructure [18]. This DSM program was demonstrated only
for entities with critical loads hence optimizing only their supply
side. Each entity was incentivized to participate the program which
required the exploitation of different supply energy carriers, thus af-
fecting the overall energy supply price value. This approach didn’t
consider the possibility of influencing the non-critical demand, which
may be key to exploit the full capabilities of DSM approaches. Fi-
nally, the Energy Hub concept has been applied for optimization of
energy flows in simulated interconnected networks [19], but without
taking into account DSM actions.
Another existing problem in the scenario presented by this article
is the interoperability of smart home products. More often than not,
products from different manufacturers are incompatible with each
other, and even products of the same brand are unlikely to really
interoperate. True interoperability can be achieved when heteroge-
neous smart components can utilize data and suggest actions from
and to each other. Unfortunately, while the vision of smart build-
ings is nothing new, agreeing on the standards between larger groups
of companies has been unsuccessful, except in entertainment elec-
tronics. There are many possible typologies of architectures to im-
prove interoperability with smart home devices [20]. Traditionally,
such system is built with central managed architecture where the
sub-system interconnections can be guaranteed by a dedicated gate-
way. Thanks to the gateway it is possible to have interoperability in-
side home, it means an integrated environment with a lot of devices
that cooperate each together as a single entities, exchanging data and
providing high quality services. In the recent years many solutions
are proposed, but they do not have a high degree of interoperabil-
ity and are usually strongly technology-oriented. As a matter of fact,
an extensive amount of research has focused on smart home inter-
operability [21, 22]. Many studies have concentrated on connecting
heterogeneous devices and subsystems together, and providing a uni-
fied interface on top [23, 24, 25, 26]. Recently, the use of semantic
technologies in smart environments has been suggested by many au-
thors [27, 28, 29], as it provides a way to represent data on a higher
semantically meaningful level, and share common understanding of
the concepts using ontologies. In addition, different application do-
mains can define their own ontologies. In turn, these ontologies can
link concepts and properties to common ontologies, for example, in
Linking Open Data (LOD) cloud. Various studies propose ontology-
based context models for representing smart home context informa-
tion semantically [30, 31].
3 The RESPOND Project
The RESPOND (integrated demand REsponse Solution towards en-
ergy POsitive NeighbourhooDs) project aims to deploy and demon-
strate an interoperable, cost effective, user centered solution, entail-
ing energy automation, control and monitoring tools, for a seam-
less integration of cooperative DR programs into the legacy energy
management systems. In this endeavor, RESPOND leverages an in-
tegrated approach for real-time optimal energy dispatching, taking
into account both supply and demand side, while exploiting all en-
ergy assets available at the site.
Figure 1. The RESPOND IoT platform for acquiring, processing and ex-
ploiting data.
Towards that goal, a central IoT platform has been developed for
the acquisition, processing and exploitation of relevant data collected
in neighborhoods. This platform is an addition of different compo-
nents and its architecture is depicted in Figure 1. It is worth mention-
ing that ensuring consumer’s data privacy has been a prime requisite
of this platform, since consumers are often concerned with sharing
their energy consumption data [32].
With the purpose of demonstrating the RESPOND solution, it is
being implemented in different types of residential buildings (i.e.
apartments, single-family and multi-family houses), situated in dif-
ferent climate zones (i.e. Mediterranean, oceanic and humid conti-
nental climate), having different forms of ownership (i.e. rental and
home-ownership), population densities and underlying energy sys-
tems. Namely, the three RESPOND pilot sites are located in Aarhus
(Denmark), the Aran Islands (Ireland) and Madrid (Spain).
Having such a heterogeneous group of end-users hinders the dif-
fusion and impact of DR solutions, and it makes more difficult to
ensure sustained user engagement with DR programs. This is why,
interaction with end-users is recognized as a key point in the RE-
SPOND project. Consequently, a set of tools and services are planned
to deliver measurement driven suggestions to end-users for energy
demand reduction and influence their behavior making them an ac-
tive indispensable part of DR loop. One of these tools is a multi-
lingual and cross-platform mobile app which contributes in the user-
engagement matter. The app is available for download both in Google
Play and App Store and gives the end user direct and detailed insight
into all relevant monitoring data6. The added value of the RESPOND
mobile app lies in its ability to suggest dwellers energy conservation
opportunities, which are a direct result of the proposed AI system.
4 An Artificial Intelligence System for Optimal DR
Artificial Intelligent systems are software (and possibly hardware)
systems that, given a complex goal, act in the physical or digital
dimension by perceiving the environment through data acquisition,
interpreting the collected structured or unstructured data, reasoning
on the knowledge or processing the information derived from this
data and deciding the best action(s) to take to achieve the given
goal. Although the AI is not something new, currently it is experi-
encing an upsurge that can be attributed to advances in computing
and the increasing availability of data7. Different definitions for AI
can be found in literature and according to the EC’s High-Level Ex-
pert Group on Artificial Intelligence8, AI systems are software (and
possibly hardware) systems that, given a complex goal, act in the
physical or digital dimension by perceiving the environment through
data acquisition, interpreting the collected structured or unstructured
data, reasoning on the knowledge or processing the information de-
rived from this data and deciding the best action(s) to take to achieve
the given goal.
RESPOND aims to allocate the most suitable demand profiles both
at a dwelling and neighborhood levels as a driver for reducing the
energy demand in specific time periods, as well as for maximizing
the exploitation of renewable energy. To do so, RESPOND has pro-
posed the AI system depicted in Figure 2 which has four main blocks:
Measurement, Forecasting, Optimization and Control blocks. Each
of these four main blocks is composed of different services, and the
adequate interaction between these services ensures the dispatching
of optimal DR actions.
Next, each of these four blocks is detailed.
4.1 Measurement
The sensor technology embedded in IoT devices is continuously be-
coming cheaper, more advanced and more widely available, thus
moving beyond disruption to become a mainstay of daily life. The
residential sector is no exception to this expansion, and according to
a report conducted by Navigant Research9, the global annual revenue
from residential IoT device sales will reach $167.2 billion in 2027.
In the context of the RESPOND project, the monitoring of real-
world qualities and events within dwellings and neighborhoods is
approach-excellence-and-trust en
performed with Smart Home Monitoring solutions provided by En-
ergomonitor10 and Develco11. These solutions comprise the equip-
ment necessary for the acquisition of observed ambient (e.g. temper-
ature or humidity) and energetic (e.g. electric demand or gas con-
sumption) data. Monitoring is complemented with an OpenMUC12
based gateway which is able to acquire data from other monitoring
and control applications that are not from Energomonitor or Develco.
Regarding other relevant data such as energy price or weather infor-
mation, which is not monitored with installed physical devices, it is
collected from external sources. All these data are sent to the MQTT
broker middleware, which allows the integration of information com-
ing from different sources, as well as the communication between
different components. The communication with the MQTT broker is
done via the publish/subscription method, which decouples the client
that sends the message (the publisher) from the client or clients that
receive the messages (the subscribers). Since there are differences in
hardware and software implementations of devices produced by dif-
ferent vendors, a Canonical Data Model (CDM) is designed to work
along with the MQTT message exchange protocol and to ensure in-
teroperability among different system components.
Complimentary to the data acquisition equipment and services, the
RESPOND Measurement block counts on data repositories. Due to
the diverse types of data collected, three different database systems
are considered: Time Series Databases (TSDB), triplestores or Se-
mantic Repositories, and Relational Databases.
IoT data, which is characterized by its abundance, is recommended
to be stored in TSDBs. These databases are optimized for time series
data and designed to handle high write and query loads as well as
down-sampling and deletion of old data, thus being able to manage
an amount of data while ensuring a high performance. This is why,
one of RESPOND’s data storage systems is InfluxDB, an open source
In the built environment, the integration of static building informa-
tion and IoT data has become one of the main challenges [33]. Fur-
thermore, easy and intuitive ways to rapidly browse, query and use
building information combined with IoT data are not usually avail-
able [34]. Semantic Technologies can aid to solving these issues, as
they allow a more dynamic manipulation of the building information
in RDF graphs by means of query and rule languages. Therefore,
RESPOND platform uses a Semantic Repository to store the static
building information. Semantic Repositories are optimized for host-
ing this type of data and usually support a SPARQL endpoint where
data can be queried using SPARQL queries. Namely, a Openlink Vir-
tuoso repository is used. Both practice and research suggest the use
of a graph-based format to capture building data, nevertheless keep-
ing numeric data explicitly out of the semantic graph for computa-
tional performance reasons [35]. And this is the approach followed
by RESPOND [36].
Last but not least, structured data that is massively instantiated
but is not time based is stored in a relational database. Namely, RE-
SPOND’s relational database system is MySQL.
4.2 Forecasting
Once the collected data is stored in the adequate data repositories,
it remains accessible to be exploited for different purposes. Some
of this data such as the monitored electric consumption of appliances
and the dwelling as a whole are used for visualization purposes in the
Figure 2. The RESPOND AI system.
RESPOND mobile app. This data can also be exploited by analytic
services, which are the core of this Forecasting block.
Being able to accurately predict the amount of energy to be pro-
duced over a period of time and knowing in advance when demand
peaks will occur, can definitely contribute to a better management
of their disparity, thus allowing the suggestion of the most suitable
actions to end-users. Therefore, in this block, two main services are
considered: the RESPOND Energy Production Forecasting Service
and the RESPOND Energy Demand Forecasting Service.
RESPOND Energy Production Forecasting Service. As a part
of the RESPOND project, two different RES are identified: Photo-
voltaic (PV) panels in Aarhus and Aran Islands, and Solar Thermal
Collectors (STC) in Madrid. Therefore, three different day-ahead en-
ergy production forecasters with hourly time resolution were devel-
The crucial motivation for developing various models for RES
production forecasting corresponds to their stochastic nature, which
is most evident as there is a high correlation between the produced
energy and weather conditions. Therefore, day-ahead hourly weather
forecasts for various factors were obtained from Weatherbit13, with
the solar irradiance being the most important one. Namely, the cor-
relation between solar irradiance and production is expected to be
extremely high, which is why it was crucial to include it as an in-
put for the forecaster. Apart from the irradiance, UV, wind direction
and speed, outside temperature, cloud coverage and relative humid-
ity were considered as relevant inputs for the renewable production
Current state of the art solutions for PV production forecast model-
ing are mainly focused on various machine learning approaches [37,
38] as they achieve the highest performances when there is ample
available data. Unlike the Aarhus pilot site, where a couple of pilot
buildings are sharing a single PV plant and a historical data of a pe-
riod of 2 years was available, the Aran Island pilot is formed out of
different geographically separated houses with each of them having
its own PV production. Furthermore, for more than half of the partic-
ipant dwellings, no production measurements were available, which
is why the traditional physical model approach has been chosen for
the Aran pilot [39]. On the contrary, for STCs, even though physi-
cal approaches are the most frequent ones [40], due to the fact that
production measurements were stored as a part of the RESPOND
platform, a Machine Learning approach has been applied instead.
Various Machine Learning approaches were considered and tested
using Python for Aarhus and Madrid pilot sites. In all cases, optimal
hyper parameters were chosen using grid search and the Mean Abso-
lute Error (MAE) was used as an indicator of their performance. For
forecasting energy coming from PV panels, Random Forest models
were the ones with the best performance, whilst for the STC Neural
Networks were the chosen. For the Aran Islands, the physical model
presented in [41] was employed, as it required only parameters that
can most commonly be found in the PV cells data sheets. The main
concept of this methodology is to estimate the final production using
the cell temperature, which is estimated using two groups of input pa-
rameters: proprietary PV cells static parameters (longitude, latitude,
time zone offset, slop of the PV cell surface, rated capacity, of the
PV array, temperature coefficient, surface area of the PV cell, nomi-
nal operating cell temperature) and dynamic ones (global horizontal
radiation, ambient temperature, number of the day in the year, cloud
coverage, current time). Hence, for each Aran house PV plant, static
parameters were obtained, and models were developed to estimate
production of each households.
Finally, after training ML models and the development of the
physical one, they have been tested and the mean absolute errors
are as follows: 8.3% for Aarhus model, 21% for Aran and 6.2% for
Madrid. Additionally, Figure 3 shows the comparison between the
real and the forecasted STC energy generation in Madrid as an ex-
ample of model’s performance.
Figure 3. Madrid STC production forecaster performance.
RESPOND Energy Demand Forecasting Service. Alongside
with the forecasting of the energy produced from RES, the forecast-
ing of the energy demand is essential in RESPOND to allow the dis-
patching of optimal DR strategies. This service aims at forecasting
short-term energy demand, that is, the energy to be consumed dur-
ing the next 24 hours with an hourly frequency. Focus is placed in
the electric demand at a house level, therefore, a model is built for
predicting the short-term electric consumption of each dwelling par-
ticipating in the RESPOND project. Afterwards, these predictions
are aggregated to estimate the neighborhood demand prediction, as
it is considered to be an an effective way to mitigate the impact of
randomness in the behavior of different dwellers [42].
This service is based on data-driven predictive models and Ma-
chine Learning algorithms, and exploits the data previously collected
in the Measurement block. After testing different algorithms, hyper-
parameter tunings and sets of input data, best results were obtained
with a k-NN (k-Nearest Neighbours) algorithm with a kparameter
value 10 and a set of inputs comprised of historical electric demand
and temporal variables (e.g. hour, weekday or type of day in terms of
working or non-working).
All the developed predictive models within the Forecasting block
were integrated in the RESPOND platform and the predictions per-
formed were stored in a MySQL database where they remained avail-
able for their visualization in the RESPOND mobile app, as well as
for their further exploitation by the Optimization block.
4.3 Optimization
The Optimization block is aimed at converting inferred data and cus-
tom grid-related requests into an optimal demand curve for an entire
neighborhood. This curve can later be used to generate both non-
user-specific and user-specific DR event suggestions to modify the
demand with views to mitigating potential problems with grid stabil-
ity and respond to requests from a (virtual) DR aggregator. Namely,
it takes into consideration day-ahead energy prices (collected in the
Measurement block), the forecasted renewable production and the
predicted demands from individual users (both of them generated in
the Forecasting block) aggregated into a neighborhood profile. Using
the supposed demand flexibility, the optimizer shifts the demands
in intensity and in time to generate a profile that is the most cost-
effective for end users and most stable for the grid operator.
The obtained optimal demand curve, combined with the forecasted
profile and with the current measurements registered by smart sen-
sors and actuators (smart plugs and cables that can control on/off
statuses of appliances remotely, occupancy measurements, tempera-
ture measurements, etc.), can create mass recommendations (which
are not user-specific) with suggestions regarding how to modify the
demand. Likewise, specific users can also be targeted through the
analysis of their appliance activation status. Furthermore, if adequate
permission is given, the system can even be allowed to take automatic
This optimization model is developed upon the core constraints
that govern the way the Energy Hub is used to model energy trans-
mission and transformation, as described in [43]. Since the proposed
architecture of the system regards the demand as an aggregate value
equal to a sum of individual consumptions of different RESPOND
participants, the demand is also managed and optimized in an aggre-
gated form, without disaggregating into values that would correspond
to individual users. In order to do so, and considering that in energy
management solutions that employ DR, either the required demand
curve is known before hand or that the modifications that should be
made to the demand profile are given some time ahead, a key variable
defined as the demand deviation given by
L=Lrequired L
where Lis the real (measured) demand and Lrequired is the pro-
file that is desirable to achieve. The demand variable Lis limited to
values in the range of the forecasted value Lforecasted(k)±a prede-
fined flexibility margin which is (for this use case) set to be 20% of
the forecasted value at the considered time stamp as to mimic poten-
tial demand flexibility that can be achieved with user interaction and
suggestions [44]. As the main goal of the optimization is to provide
a demand curve that is as close as possible to the required one, these
demand deltas need to be penalized in the objective function. In or-
der to do so, its values must be split into positive instances L+(k)
and negative instances L(k)so that each one of these variables
can be penalized with positive and negative values, respectively. This
is achieved by redefining the demand delta as
L(k) = ∆L+(k) + ∆L(k) = Lrequired (k)L(k).
However, in order to force these newly introduced values to equal
positive and negative deviations, a set of constraints is introduced
L+(k)+I(∆L+(k)) ·L+
L(k)≥ −I(∆L(k)) ·L
that limits these variables to the maximum absolute positive de-
viation L+
max(k)and negative deviation L
max(k), but also in-
troduces the indicator variables I(∆L+(k)) and I(∆L(k)) that
should equal 1 only if the appropriate deviation exists in time step
k. Furthermore, the previously mentioned maximum deviations are
determined based on the maximum demand flexibility. The indicator
variables are forced to correct values by marking them as Boolean
variables in the Mixed-Integer Linear Programming (MILP) model
and also adding a constraint
I(∆L+(k)) + I(∆L(k)) 1.
Finally, in order not to allow the engine to reduce all demands to
the minimum possible value, an integral constraint is added with
L(k) =
where k1and k2represent the beginning and ending indexes of a
sliding window (usually 24 hours apart) for the optimization.
In order to implement DR events into the optimization engine, two
approaches can be utilized: implicit (price-based) DR and explicit
DR. In the former case, a price profile of energy acquisition is cre-
ated so that it discourages higher consumptions at times where the
demand should be decreased using higher prices, and motivates the
increase of loads with lower prices. In the latter case, DR events are
considered to be predefined and the criterion function is constructed
to estimate total costs for the end user plus a factor that allows for
penalization of previously mentioned demand deviations, but only
for those time steps when a DR event is defined. This approach also
allows for different DR events to be differently stressed. Through the
utilization of the required profile (which is not a variable but a pre-
defined sequence), both demand increasing and demand decreasing
DR events can be created with different modification values to the
forecasted profile that is considered as the baseline. With a set of
miscellaneous constraints, the aforementioned set of equations and
the appropriate objective function form a MILP optimization prob-
lem implemented by the RESPOND approach. The example of the
optimization utilization will be given in the Use-case Section.
4.4 Control
Once the optimal energy profiles are generated for each dwelling and
neighborhood, some specific control actions need to be performed
in order to achieve those profiles. It has been demonstrated that the
potential in the flexibility of appliances’ operation and time of use
allows them to be exploited for matching the needs of specific DR
programs [45]. Therefore, the main goal of the Control block is to
translate the optimal energy demand profiles into specific DR events
mainly related to the scheduling of appliances. Furthermore, the DR
events proposed are aimed to generate the least disturbance in the
every-day operations of building occupants. To do so, dwellers spec-
ify a set of preferences that the RESPOND AI system takes into ac-
The Control block has a heuristics service aimed at translating the
optimal profiles generated from the Optimization block into the ac-
tual DR control actions (e.g. turning the dishwasher on at a certain
time). These control actions include not only home devices control,
but also the management of energy assets related to the energy gener-
ation and dispatching both at a dwelling and district level. The actual
actions proposed by this heuristics service are guided by the prefer-
ences set for each dwelling.
The second service that conforms the Control block is the User
Adapter service. Balancing energy efficiency and user satisfaction is
another unresolved DR challenge [46], therefore, the User Adapter
service is of utmost importance towards the success of the overall
RESPOND system. This service aims at collecting the preferences
that dwellers may have with regards to the different events and situ-
ations involved in DR events. Namely, the collected preferences in-
The type of actions allowed by each dweller (e.g. switch on or
switch off).
The appliances upon which these actions may be performed (e.g.
a dishwasher of a washing machine).
The periods of time when these actions may be allowed (e.g. from
19:00 onwards).
The type of notifications preferred by dwellers (e.g. informative,
prescriptive or none of them).
The set of preferences for each dweller are specified in the RE-
SPOND mobile app and stored in the MySQL database. Dwellers
may easily modify their preferences as they wish. These preferences
are then considered by the system to propose specific DR events to-
wards the achievement of the obtained optimal demand profiles.
5 Use-case
This section aims at pointing out what are the main results and bene-
fits of the proposed AI system by presenting a real-world use-case, in
which inputs and outputs of all the blocks are accompanied with rele-
vant conclusions and interpretations. This real-world scenario shows
how RESPOND helps users adjusting their loads optimally in accor-
dance with the production from RES, DR events and their flexibility.
This use-case showcases the load modifications for a set of social
houses in Aarhus using explicitly defined DR events, more precisely,
using value-based DR events whereby a specific amount by which the
load should be corrected is defined. After data is collected within the
Monitoring block, PV production (Figure 5 - orange line on the left)
and demand (Figure 5 - blue dashed line on the right) have been pre-
dicted by the Forecasting Block. These predictions have been consid-
ered by the Optimization block, which produced the optimal profile
given as a full blue line in Figure 5 on the right.
This example considers separately the electric load and thermal
(district heating) by specifying two DR events per each carrier, as
shown in Figure 4 referenced previously. For the electric domain, a
load increase event is specified during the mid-day period to maxi-
mize the utilization of renewable generation while the reduction of
the afternoon peak is enforced by a load decrease event. Similarly,
the morning peak in thermal demand is reduced with the “missing”
energy moved to the afternoon period when the demand is generally
With the optimizer monitoring the total energy consumed on a
daily basis and keeping the total optimized energy equal to what the
predicted curve suggests, although load modifications do exist, user
comfort is kept in check. Making use of explicit DR events, makes
more specific and granular modifications necessary to achieve the
optimized profile, allowing for better control of this key aspect in
maintaining grid stability. Finally, in accordance with the optimized
profile, the Control block sends suggesting to the end user to achieve
the desired profile. With regards to the electric curve, a notification to
turn the washing machine on during the mid-day is sent and to turn
the dishwasher off during the afternoon is sent via the mobile app.
As for the thermal curve, the suggestion to reduce the thermostat set-
point is sent.
Figure 4. Explicit DR events
Having in mind use-case presented above, it can be concluded that
AI-based analytical services are beneficial for encouraging users to
improve their habits and helping them in increasing energy savings.
Additionally, what is even more crucial is the system as hole, which
utilises benefits of all particular blocks, providing understandable
suggestion for the final user.
6 Conclusions
Buildings account for more than 35% of global energy use, although
it is estimated that significant energy savings can be achieved if
proper operation strategies are implemented. The residential sector
is specially promising, as it is responsible for around the 25% of
energy usage. RESPOND aims to dispatch optimal DR events to
dwellers towards the reduction of energy peak demands by maxi-
mizing the renewable energy generation and leveraging energy stor-
age units. Furthermore, with views to ensuring dwellers’ engagement
with the project, the proposed DR events are aimed to generate the
least disturbance in their every-day activities.
Figure 5. Effects of DR events on the demand profile at the Aarhus pilot.
This is a complex multi-objective problem and RESPOND pro-
poses an AI system that comprises four blocks: Measurement, Fore-
casting, Optimization and Control. The Measurement block aims at
acquiring and storing all the necessary ambient, energy and other rel-
evant data. The Forecasting block focuses on exploiting this data to
estimate the short-term generation of PV panels and STCs as well
as the demand for each dwelling. This information is later on used
by the Optimization block to calculate the optimal demand curve.
Finally, the Control block generates and notifies the adequate DR
events to dwellers in order to achieve this optimal demand curve.
The implementation of this AI system has been demonstrated in
a real-world scenario involving both electric and thermal loads. This
demonstration shows the seamless interaction of the four blocks that
comprise the AI system, and its effectiveness to achieve the desired
energy efficiency without neglecting the user comfort.
This work has received support from H2020 project RESPOND (in-
tegrated demand REsponse Solution towards energy POsitive Neigh-
bourhooDs) with grant agreement number 768619 and from the Min-
istry of Education, Science and Technological Development of the
Republic of Serbia.
[1] Xiaodong Cao, Xilei Dai, and Junjie Liu. “Building energy-
consumption status worldwide and the state-of-the-art tech-
nologies for zero-energy buildings during the past decade”.
In: Energy and buildings 128 (2016), pp. 198–213.
[2] T Abergel, B Dean, and J Dulac. “Global Status Report 2017:
Towards a zero-emission, efficient, and resilient buildings
and construction sector”. In: United Nations Environment
Programme 48 (2017).
[3] International Energy Agency. Transition to Sustain-
able Buildings. 2013, p. 284. DO I:10 . 1787 /
[4] Lyle D Collins and Richard H Middleton. “Distributed de-
mand peak reduction with non-cooperative players and min-
imal communication”. In: IEEE Transactions on Smart Grid
10.1 (2017), pp. 153–162.
[5] Jakub Jurasz and Pietro Elia Campana. “The potential of pho-
tovoltaic systems to reduce energy costs for office buildings
in time-dependent and peak-load-dependent tariffs”. In: Sus-
tainable cities and society 44 (2019), pp. 871–879.
[6] Longbo Huang, Jean Walrand, and Kannan Ramchandran.
“Optimal demand response with energy storage manage-
ment”. In: 2012 IEEE third international conference on
smart grid communications (SmartGridComm). IEEE. 2012,
pp. 61–66.
[7] Christos D Korkas et al. “Occupancy-based demand response
and thermal comfort optimization in microgrids with renew-
able energy sources and energy storage”. In: Applied Energy
163 (2016), pp. 93–104.
[8] Mohamed H Albadi and Ehab F El-Saadany. “Demand re-
sponse in electricity markets: An overview”. In: 2007 IEEE
power engineering society general meeting. IEEE. 2007,
pp. 1–5.
[9] Peter Warren. “A review of demand-side management policy
in the UK”. In: Renewable and Sustainable Energy Reviews
29 (2014), pp. 941–951.
[10] David Parra et al. “Optimum community energy storage for
renewable energy and demand load management”. In: Ap-
plied energy 200 (2017), pp. 358–369.
[11] Nicholas Good, Keith A Ellis, and Pierluigi Mancarella. “Re-
view and classification of barriers and enablers of demand
response in the smart grid”. In: Renewable and Sustainable
Energy Reviews 72 (2017), pp. 57–72.
[12] JJ C´
ardenas et al. “A multi-objective GA to demand-side
management in an automated warehouse”. In: ETFA2009, the
IEEE 2009 Conference on Emerging Technologies and Fac-
tory Automation, Palma de Mallorca, Spain. Citeseer. 2009.
[13] Leehter Yao, Wen-Chi Chang, and Rong-Liang Yen. “An it-
erative deepening genetic algorithm for scheduling of direct
load control”. In: IEEE Transactions on Power Systems 20.3
(2005), pp. 1414–1421.
[14] Amir-Hamed Mohsenian-Rad et al. “Autonomous demand-
side management based on game-theoretic energy consump-
tion scheduling for the future smart grid”. In: IEEE transac-
tions on Smart Grid 1.3 (2010), pp. 320–331.
[15] Zahra Baharlouei et al. “Achieving optimality and fairness
in autonomous demand response: Benchmarks and billing
mechanisms”. In: IEEE Transactions on Smart Grid 4.2
(2013), pp. 968–975.
[16] Azim Keshtkar et al. “Smart residential load reduction via
fuzzy logic, wireless sensors, and smart grid incentives”. In:
Energy and buildings 104 (2015), pp. 165–180.
[17] Petra Mesari´
c and Slavko Krajcar. “Home demand side man-
agement integrated with electric vehicles and renewable en-
ergy sources”. In: Energy and Buildings 108 (2015), pp. 1–
[18] Aras Sheikhi et al. “Integrated demand side management
game in smart energy hubs”. In: IEEE Transactions on Smart
Grid 6.2 (2015), pp. 675–683.
[19] Mads Almassalkhi and Ian Hiskens. “Optimization frame-
work for the analysis of large-scale networks of energy hubs”.
In: Power Systems Computation Conference. 1. 2011.
[20] Marco Aiello and Schahram Dustdar. “Are our homes ready
for services? A domotic infrastructure based on the Web ser-
vice stack”. In: Pervasive and Mobile Computing 4.4 (2008),
pp. 506–525.
[21] Hannu J¨
arvinen and Petri Vuorimaa. “Semantic Interoper-
ability for Web Services based Smart Home Systems.” In:
ICAART (1). 2015, pp. 141–148.
[22] Li Jiang, Da-You Liu, and Bo Yang. “Smart home research”.
In: Proceedings of 2004 international conference on machine
learning and cybernetics (IEEE Cat. No. 04EX826). Vol. 2.
IEEE. 2004, pp. 659–663.
[23] Eiji Tokunaga et al. “A framework for connecting home
computing middleware”. In: Proceedings 22nd International
Conference on Distributed Computing Systems Workshops.
IEEE. 2002, pp. 765–770.
[24] Vittorio Miori et al. “An open standard solution for domotic
interoperability”. In: IEEE Transactions on Consumer Elec-
tronics 52.1 (2006), pp. 97–103.
[25] Shengwei Wang et al. “A middleware for web service-
enabled integration and interoperation of intelligent build-
ing systems”. In: Automation in Construction 16.1 (2007),
pp. 112–121.
[26] Thinagaran Perumal et al. “Middleware for heterogeneous
subsystems interoperability in intelligent buildings”. In: Au-
tomation in Construction 19.2 (2010), pp. 160–168.
[27] Xiaohang Wang et al. “Semantic space: An infrastructure
for smart spaces”. In: IEEE Pervasive computing 3 (2004),
pp. 32–39.
[28] Daqing Zhang, Tao Gu, and Xiaohang Wang. “Enabling
context-aware smart home with semantic web technologies”.
In: International Journal of Human-friendly Welfare Robotic
Systems 6.4 (2005), pp. 12–20.
[29] Liming Chen et al. “Semantic smart homes: towards knowl-
edge rich assisted living environments”. In: Intelligent Pa-
tient Management. Springer, 2009, pp. 279–296.
[30] Eunhoe Kim and Jaeyoung Choi. “An ontology-based con-
text model in a smart home”. In: International Conference on
Computational Science and Its Applications. Springer. 2006,
pp. 11–20.
[31] Jingjing Xu et al. “Ontology-based smart home solution and
service composition”. In: 2009 International Conference on
Embedded Software and Systems. IEEE. 2009, pp. 297–304.
[32] Vinayak V Ranade and Jacob Beal. “Distributed control
for small customer energy demand management”. In: 2010
Fourth IEEE International Conference on Self-Adaptive and
Self-Organizing Systems. IEEE. 2010, pp. 11–20.
[33] Bhargav Dave et al. “A framework for integrating BIM and
IoT through open standards”. In: Automation in Construction
95 (2018), pp. 35–45.
[34] Pieter Pauwels and Anna Roxin. “SimpleBIM: From full if-
cOWL graphs to simplified building graphs”. In: Proceedings
of the 11th European Conference on Product and Process
Modelling (ECPPM). 2016, pp. 11–18.
[35] Ekaterina Petrova et al. “In search of sustainable design pat-
terns: Combining data mining and semantic data modelling
on disparate building data”. In: Advances in Informatics and
Computing in Civil and Construction Engineering. Springer,
2019, pp. 19–26.
[36] Iker Esnaola-Gonzalez and Francisco Javier Diez. “Inte-
grating Building and IoT data in Demand Response solu-
tions”. In: Proceedings of the 7th Linked Data in Architecture
and Construction Workshop (LDAC 2019). Vol. 2389. 2019,
pp. 92–105.
[37] Sobrina Sobri, Sam Koohi-Kamali, and Nasrudin Abd
Rahim. “Solar photovoltaic generation forecasting methods:
A review”. In: Energy Conversion and Management 156
(2018), pp. 459–497.
[38] Florian Barbieri, Sumedha Rajakaruna, and Arindam Ghosh.
“Very short-term photovoltaic power forecasting with cloud
modeling: A review”. In: Renewable and Sustainable Energy
Reviews 75 (2017), pp. 242–263.
[39] Alberto Dolara, Sonia Leva, and Giampaolo Manzolini.
“Comparison of different physical models for PV power out-
put prediction”. In: Solar energy 119 (2015), pp. 83–99.
[40] Peder Bacher, Henrik Madsen, and Bengt Perers. “Short-term
solar collector power forecasting”. In: proceedings of ISES
Solar World Conference. 2011.
[41] John A Duffie and William A Beckman. Solar engineering of
thermal processes. John Wiley & Sons, 2013.
[42] X. M. Zhang et al. “Forecasting Residential Energy Con-
sumption: Single Household Perspective”. In: 2018 17th
IEEE International Conference on Machine Learning and
Applications (ICMLA). 2018, pp. 110–117. DO I:10.1109/
[43] Marko Bati´
c et al. “Combined energy hub optimisation and
demand side management for buildings”. In: Energy and
Buildings 127 (2016), pp. 229–241.
[44] Anca-Diana Barbu, Nigel Griffiths, Gareth Morton, et al.
“Achieving energy efficiency through behaviour change:
what does it take”. In: Publications Office (2013).
[45] Reinhilde D’hulst et al. “Demand response flexibility and
flexibility potential of residential smart appliances: Experi-
ences from large pilot test in Belgium”. In: Applied Energy
155 (2015), pp. 79–90.
[46] Chen Wang and Martin De Groot. “Managing end-user pref-
erences in the smart grid”. In: Proceedings of the 1st Interna-
tional Conference on Energy-Efficient Computing and Net-
working. 2010, pp. 105–114.
... More specifically, RESPOND aims to deploy an interoperable energy automation, monitoring and control solution to deliver DR programs at a dwelling, building and district level to neighborhoods across Europe. To do so, RESPOND proposes an Artificial Intelligent (AI) system to assist residential consumers that would like to make use of DR and incorporate it into their energy management systems [3]. The proposed system considers the forecast energy consumption and production based on the data acquired by the deployed IoT equipment and looks for modifications that would mitigate potential instabilities in the energy supply network by applying optimal energy use and load shifting. ...
Full-text available
RESPOND proposes an Artificial Intelligent (AI) system to assist residential consumers that would like to make use of Demand Response (DR) and incorporate it into their energy management systems. The proposed system considers the forecast energy consumption based on the data acquired. This work compares the results obtained by different forecasting methods using the Root Mean Square Error (RMSE) as a measure of the forecast performance. The ARIMA, Linear Regression (LR), Support Vector Regression (SVR) and k-Nearest Neighbors (KNN) models are tested, and it is concluded that the results achieved with the KNN obtain a better fit. In addition to obtaining the lowest RMSE, KNN is the algorithm that spends less time in obtaining the forecasts.
Conference Paper
Full-text available
DR (Demand Response) programs have a big potential in the residential sector to reduce peak energy demands. However, the poor user-engagement is one of the main barriers of their adoption and success. The RESPOND H2020 project aims to bring DR programs to neighbourhoods across Europe and in this article, focus is placed on the approach implemented to solve the challenging integration of building topological data and data produced by IoT systems within houses including sensors , meters and actuators. In this regard, RESPOND leverages Semantic Technologies to represent building data, while it uses Time Series Databases (TSDB) to store IoT data. The combination of these technologies is expected to enhance the data querying, which is of utmost importance for its display in the RESPOND mobile app.
Full-text available
The built environment provides significant opportunities for IoT (Internet of Things) deployment, and can be singled out as one of the most important aspects for IoT related research. While the IoT deployment in the built environment is growing exponentially, there exists a gap in integrating these two in a systematic way through open standards and systems. From technological perspective, there is a need for convergence of diverse fields ranging from Building Information Systems and Building Services to Building Automation Systems, and IoT devices and finally the end user services to develop smart, user oriented applications. This paper outlines the efforts to develop a platform that integrates the built environment data with IoT sensors in a campus wide, web based system called Otaniemi3D that provides information about energy usage, occupancy and user comfort by integrating Building Information Models and IoT devices through open messaging standards (O-MI and O-DF) and IFC models. The paper describes the design criteria, the system architecture, the workflow and a proof of concept with potential use cases that integrate IoT with the built environment. Initial results show that both the end users and other research groups can benefit from such platforms by either consuming the data in their daily life or using the data for more advance research.
Full-text available
Cross-domain analytical techniques have made the prediction of outcomes in building design more accurate. Yet, many decisions are based on rules of thumb and previous experiences, and not on documented evidence. That results in inaccurate predictions and a difference between predicted and actual building performance. This article aims to reduce the occurrence of such errors using a combination of data mining and semantic modelling techniques, by deploying these technologies in a use case, for which sensor data is collected. The results present a semantic building data graph enriched with discovered motifs and association rules in observed properties. We conclude that the combination of semantic modelling and data mining techniques can contribute to creating a repository of building data for design decision support.
Electrical power generation across the world is facing dramatic changes for a variety of reasons related to reliability, economics and environmental concerns. Over recent years a significant increase has been observed in installed capacity of photovoltaic systems. Due to their typical seasonal and diurnal energy conversion patterns their integration into power systems creates new opportunities as well as threats. This paper intends to show how photovoltaics can contribute to reducing peak load in office buildings and thereby minimise expenditure on electricity during time- and peak-load-dependent energy prices/tariffs. An additional benefit is also provided to the national power system by reducing the need for peaking power stations. The calculations are performed for energy tariffs commonly used for commercial buildings in Poland. The simulation relies on climatic and price data for 2016. The results show significant potential for photovoltaics to reduce the peak load (from almost 60 kW to slightly over 44 kW) whilst simultaneously minimising energy costs to the building (from 1.2% up to 5.8% depending on the selected tariff). This study demonstrates the economic benefits of using PV system for reducing peak loads. A sensitivity analysis with regard to photovoltaics investment costs is carried out showing that the increasing investment costs have different impact on total energy cost depending on the considered energy tariff.
Demand response technology offers the exciting potential to reduce peak energy demand, electricity infrastructure expenditure, and household electricity bills. In this paper, a pricing mechanism that relies on non-cooperative heterogeneous loads knowledgeable of future energy consumption—such as electric vehicles—transferring minimal amounts of information to achieve peak demand response in a distributed fashion, whilst maintaining the privacy of the players. The existence of a Nash equilibrium is proven, as well as convergence conditions proving uniqueness of a Nash equilibrium and the stability of an “Iterated Synchronous Best Response Algorithm.” The price of anarchy (PoA) is proven to approach 1 as the number of homogeneous players approaches infinity, indicating there is no advantage to cooperation for a large number of similar players. Finally, simulation results are presented which suggest that the PoA for a system with heterogeneous players is likely to be proportional to the spread of energy consumption constraints.
While the management of PV generation is the prime application of residential batteries, they can deliver additional services in order to help systems to become cost-competitive. They can level-out the demand and potentially reduce the cost and emissions of the energy system by reducing demand peaks. In this study, community energy storage (CES) is optimised to perform both PV energy time-shift and demand load shifting (using retail tariffs with varying prices blocks) simultaneously. The optimisation method obtains the techno-economic benefits of CES systems as a function of the size of the community ranging from a single home to a 100-home community in two different scenarios for the United Kingdom: the year 2020 and a hypothetical zero emissions target. It is demonstrated that the levelised cost and levelised value of CES systems reach intermediate values to those achieved when both applications are performed independently. For the optimal performance of a battery system being charged from both local PV plants and the grid, our results suggest that the battery should be sized suitable to ensure it can fully discharge during the peak period.