Conference PaperPDF Available

Energy saving in smart homes based on consumer behavior: A case study

Authors:

Abstract and Figures

This paper presents a case study of a recommender system that can be used to save energy in smart homes without lowering the comfort of the inhabitants. We present an algorithm that uses consumer behavior data only and uses machine learning to suggest actions for inhabitants to reduce the energy consumption of their homes. The system mines for frequent and periodic patterns in the event data provided by the Digitalstrom home automation system. These patterns are converted into association rules, prioritized and compared with the current behavior of the inhabitants. If the system detects an opportunities to save energy without decreasing the comfort level it sends a recommendation to the residents.
Content may be subject to copyright.
Energy saving in smart homes based on consumer
behavior: A case study
Michael Zehnder, Holger Wache, Hans-Friedrich Witschel
Institute of Business Information Systems
University of Applied Sciences and Arts Northwestern Switzerland
FHNW Olten, Switzerland
mzehnder@gmx.net
{holger.wache; hansfriedrich.witschel}@fhnw.ch
Danilo Zanatta, Miguel Rodriguez
Research and Development
digitalSTROM AG
Zurich, Switzerland
{danilo.zanatta; miguel.rodriguez}@digitalstrom.com
AbstractThis paper presents a case study of a recommender
system that can be used to save energy in smart homes without
lowering the comfort of the inhabitants. We present an algorithm
that mines consumer behavior data only and applies machine
learning to suggest actions for inhabitants to reduce the energy
consumption of their homes. The system looks for frequent and
periodic patterns in the event data provided by the
digitalSTROM home automation system. These patterns are
converted into association rules, prioritized and compared with
the current behavior of the inhabitants. If the system detects
opportunities to save energy without decreasing the comfort
level, it sends a recommendation to the inhabitants.
The system was implemented and deployed to a set of test
homes. The test participants were able to rate the impact of the
recommendations on their comfort. This feedback was used to
adjust the system parameters and make it more accurate during
a second test phase. The historical data set provided by
digitalSTROM contained 33 homes with 3521 devices and over 4
million events. The system produced 160 recommendations on
the first phase and 120 on the second phase. The ratio of useful
recommendations was close to 10%. We found out that a
recommender system that uses an algorithm that mines patterns
based on their confidence, independent of their frequency and
periodicity, might achieve better results and a higher acceptance
by users.
Keywordssmart cities; smart homes; energy saving;
recommender system; association rules; unsupervised learning;
internet of things; IoT
I. INTRODUCTION
King defines a smart home as a “dwelling incorporating a
communications network that connects the key electrical
appliances and services” and which can be “remotely
controlled, monitored or accessed” [1]. Blumendorf enhances
the “smart” part of the definition with a home which does act
in a smart way, is a system which is autonomously operating,
based on artificial intelligence [2]. But there would be many
ways to measure the smartness of such a home. Harper
acknowledged that “smart house technologies that most people
are pleased with are connected with saving energy or
money” [3].
Previous studies state that energy savings can only be
achieved by involving the inhabitants [4]. We believe
otherwise and, in this paper, we investigate the use of
unsupervised learning to achieve energy savings without
lowering the comfort level of inhabitants. We consider user
comfort as being a soft characteristic that cannot be objectively
measured as it is the case with savings in energy consumption.
In this sense, user comfort is uniquely defined by each user and
can include, but not be limited to, visual comfort and thermal
comfort.
In this paper we propose a recommender system, which
relies on the opinion of the users to approve a recommendation.
We evaluate the smartness of the recommender system
designed in this project by the energy savings created without
reduction in comfort. A long-term goal is for the algorithm to
learn which kind of patterns would be likely to be accepted by
the user because it does not impact their comfort.
This paper is organized as follows. Chapter II provides
some background about smart homes and unsupervised
learning. Chapter III describes the pattern-mining algorithm
used and Chapter IV describes the design of the
recommendation system. The results from the evaluation of the
recommender system in real life are presented in Chapter V,
while conclusions and future work are described in Chapter VI.
II. BACKGROUND
A. Smart home technology
digitalSTROM products provide connectivity to electrical
devices in the home over the existing power cables. This
includes every lamp, light switch, blinds and any plugged in
device. This network of devices is connected through a server
mounted in the electrical cabinet to a local area network. The
result is a network of connected devices, bringing the internet
of things (IoT) to each home. digitalSTROM components are
based on a high volt IC in a small size module. Each
digitalSTROM module can switch, dim, measure electricity
and communicate its status. The products are available through
Europe with its larger installed base in Germany and
Switzerland.
A digitalSTROM system is based on concentrators that
reside in the electrical distribution panel, acting as power
meters for the individual distribution circuits and
communicating with individual nodes installed within a home
over differently modulated up and downstream channels [5].
The system includes a Linux server application with a JSON
API. Moreover, real-time data from test homes are collected by
a logging system, parsed and stores in a database, being
available for processing. From this logging system, historical
data can easily be obtained and processed.
Please note that for this approach no motion detectors or
other location sensors where used. Only activities like turning
on (or off) light, TV etc. are considered.
B. Unsupervised learning and mining for patterns
Many supervised activity recognition methods for smart
homes have been published in the literature in the last
years (see e.g. [6]). Supervised learning is used in methods
like decision trees [7], Markov models [8] and dynamic Bayes
networks [9]. Although these classifiers rely on conditional
independence of the events, the classifiers achieve good
accuracy with large amounts of training data [10].
However, these methods have the major disadvantage that
the training data needs manual annotation. This manual
annotation is a very tedious and time-consuming task and
limits the scalability of the system. Furthermore, the training
data can only be used for the household for which they were
made for [11].
Due to the limitation listed above, in this work, an
unsupervised approach for pattern mining is used. In the
unsupervised approach, the user is not required to scan his
data for activities and the classifier is able to find the patterns
autonomously.
III. PATTERN MINING
The recommender system proposed in this work is based on
the mining of patterns from historical event data from a home
in order to produce recommendations for its users. This
chapter describes the choice of the pattern mining algorithm
and the adaptations required in order that it can be used in the
smart home context.
A. Data analysis
Finding frequent and periodic patterns in event data is
usually referred to as “discovering frequent episodes in
sequences” [12], “sequence mining” [6, 13] or “activity
recognition” [9, 14, 15, 16, 17]. The starting point for this
research was the Apriori algorithm [23], which was later
implemented and enhanced by other projects. The Apriori
algorithm in its original form is not able to mine periodic
patterns in sequential data and was therefore improved by
different people for the use in smart home automation
systems, for example for the CASAS System [10]. In this
approach, the algorithm starts by mining patterns with a
minimum length of two activities, which is extended until the
algorithm is not able to find frequent patterns anymore.
B. Data set the events
The basic structure of the house is the hierarchical model
shown by Fig. 1. The root node represents a home, one
household. A home has several dSMs (dSM stands for
digitalSTROM Meter”), one per electric circuit in the home,
and several zones, where each zone represents a physical room
in the house.
Each zone contains one or many scenes, where each scene
represents a pre-defined configuration of the devices, e.g. each
light dimmed to a pre-defined state, shades closed and
television turned-on.
Each dSM controls the devices that are connected to its
electrical circuit. Power measurements are recorded per dSM.
Events are either inhabitant generated button clicks or sensor
events, such as temperature sensors.
The historical training dataset used contains 33 homes with
3521 devices, which are related to 4,331,443 events and 6829
unique scenes. These events extend over a period between
08/12/2002 and 25/06/2014.
Fig. 1. Hierarchy of components in a digitalSTROM smart home
C. Definition and identification of patterns that would
reduce energy usage
Analysis of the data revealed that the power measurement
data is not recorded as precisely as necessary for obtaining
information about the power consumption of a single device.
In contrast, the event data is well suited and therefore eligible
to find possible actions.
We defined rules to identify relevant energy saving actions
from the underlying data. We considered the following type
of events as energy saving, and therefore as an action for the
system:
Absent (a predefined scene which turns off all
devices when leaving the house)
Dim (scenes to dim devices)
Off (scenes to turn off devices)
Sleep (scenes to turn off devices before going to
sleep)
Standby
In the training data this rule results in a list of 20 actions,
which are called by 1,283,756 unique events.
D. Identification of relevant types of behavior patterns
Because not all frequent or periodic patterns result in
energy savings, we defined some characteristics to identify the
relevant behavior patterns.
To ensure that a relevant pattern can be used to suggest
actions, it must be composed of two main components:
(1) A relevant pattern must contain at least one action to
lower energy usage (actions are a subset of normal
events).
(2) The pattern must consist of normal events, which
serve as condition to suggest the action at the right
time.
Suppose that such patterns occur regularly, they reflect the
normal behavior of the inhabitants. The main idea now is to
discover situations where the user forgot the action. Therefore
the recommender would suggest the user to perform this action
to lower energy usage.
Because a one-event condition is insufficient as evidence to
suggest an action, the overall length of a relevant pattern must
be at least three events (one action and two normal events).
Therefore, a relevant pattern can be defined as a pattern that is
longer than two events and contains at least one action to lower
energy usage.
Because a relevant pattern consists of normal events, which
represents the condition, and an action, this pattern can be
interpreted using an association rule. An association rule is an
implication of the form
X → Y, where X, Y I,
where X is a sequence of normal events, Y is a single action,
and I is the set of all possible events. The association rule
above states that when X occurs, Y occurs with certain
probability [22]. This rule is also depicted in Fig. 2.
Fig. 2. Sequence of events and the association rule
To summarize, the types of behavior patterns that are
required to recommend some action must fulfill the following
criteria:
The pattern must occur frequently and/or periodically
in the data
The pattern must be relevant for action prediction and
contain at least one action to lower energy usage
The pattern must have a minimum length of three
events, including normal events and actions
E. Evaluation of pattern-mining algorithms
Three established frequent sequential pattern-mining
algorithms PrefixSpan, BIDE+ and GapBIDE as well as their
adaptations were evaluated by Schweizer in [19] for the usage
on smart home event data. They were benchmarked against
each other and against a self-developed algorithm named
WSDD. All four algorithms mined a reasonable amount of
frequent sequential patterns. Under the following parameters,
nearly thousand patterns were found in the historical test data:
pattern length: between 3 and 7 events
appear quite often (i.e. minimum support between
0.01 and 0.001)
patterns may overlap
wildcarding disabled, i.e. without a place holder in
between the sequence of events
The traditional frequent sequential pattern mining
algorithms like PrefixSpan, BIDE+ or GapBIDE require pre-
and post-processing to be used for mining smart home event
data. Furthermore, if different minimum and maximum
lengths of patterns shall be mined, those algorithms need to
run multiple times to report the correct support count.
The run times of the different algorithms showed large
deviations, as shown in Fig. 3. While BIDE+ needed the
longest to mine the same patterns as the other three
algorithms, both GapBIDE and PrefixSpan run significantly
faster. However, they were all outperformed by the WSDD
algorithm.
The relevant patterns mined in the underlying data
showed lengths of 3 to 6 events. Whereas the minimum of
three events is a predefined parameter, the maximum was not
defined. However, testing on the underlying data showed that
no relevant pattern was longer than six events.
Fig. 3. Benchmark of run times for data mining algorithms
The following table presents three examples of relevant
patterns mined in the underlying event data by the relevant
pattern-mining algorithm. The grey background of the cell
indicates the action, which must be part of a relevant pattern.
Pattern
Events
1
3
1
Turn off light in
kitchen
Motion detector
garage
2
Turn on light in
laundry room
Turn off light in
basement
3
Turn on light
outside of basement
Turn off lamp
in basement
It should be noted that the order of the events/actions is not
important. As the examples above show, an action can appear
at the first position. The two next events for the condition
might appear after the action but can be used to discover the
missing action.
IV. DESIGN OF THE RECOMMENDER SYSTEM
A. Architecture
The architecture of the recommender system developed in
this project is shown in Fig. 4 and can be divided in three main
parts:
The storage of the association rules
The event stream of the current behavior data inside
the smart home
The matching algorithm
Fig. 4. Recommender system architecture
The rule database stores the association rules, which were
obtained from the mined relevant patterns. The event stream
contains the real-time events from the smart home, ordered by
time of their occurrence.
The matching algorithm is the core component of the
recommender system. It matches the rules and the event
stream. The most common existing rule matching algorithm is
RETE by Forgy [21]. We used a deterministic finite state
machine (FSM) approach as depicted in Fig. 5, which reflects
the order of the events better than RETE. A new instance of
the FSM is created for each new event in the stream. If there is
no matching in the first attempt, the instance is removed from
memory. If the condition did match and the next event is not
the action itself, the machine sends a recommendation.
The design of the recommender system allows more than
one rule to be matched at the same time. In order to avoid
multiple conflicting recommendations, we propose to weight
the rules and use the weights as a prioritization criterion.
Furthermore, the prioritization criterion is also used to exclude
rules under a certain threshold, i.e. rules that are weakly
matched. The criteria where defined as follows:
Confidence of the rule
Support of the pattern
The length of the pattern
The position of the action
Date when the pattern was mined
Fig. 5. Example of a matching rule in the algorithm (FSM)
B. Confidence
The confidence of a rule denotes how often the action Y
appears in patterns that contain the sequence of events X
(condition). If a rule has a confidence of 100%, no occurrence
of the pattern without the action was mined. The confidence is
calculated using the support (count) of the pattern. The
confidence is thus expressed as
where denotes the action, denotes a normal event,
denotes the pattern containing the action and denotes the
pattern without the action.
Fig. 6. Recommender deployment view
C. Implementation
The recommender system was implemented on a cloud
based Microsoft® Azure VM (Virtual Machine). The VM was
set up with Ubuntu 14.04. The deployment view of the VM is
shown in Fig. 6.
The smart homes event data is parsed from the log files of
the digitalSTROM system. The files are uploaded by a script
installed on the digitalSTROM infrastructure in the houses and
made accessible for this project on a file server. The files are
copied every 5 minutes by remote synch to the VM where the
recommender system is running. They are parsed by a Python
script and stored in a MySQL database.
V. RESULTS
The evaluation was conducted in households with
inhabitants producing real event data. The test households were
equipped with the smart home automation system
digitalSTROM. Overall, 33 households were considered for the
evaluation. The historical event data of the homes was mined
by the recommender system in advance. According to the
number of relevant patterns found in the houses, they were
ranked and the owners of the 15 most promising households
where requested to take part in the evaluation. From these, 8
houses agreed to participate in the evaluation of this work
including both single- and multi-inhabitant houses.
Recommendations were sent per SMS to the mobile devices of
the inhabitants. An example of such recommendation is shown
in Fig. 7.
We ran the evaluation in two phases, which are described in
the following sections.
Fig. 7. Recommendation SMS
TABLE I. KEY RESULTS OF EVALUATION
Parameter
Phase
1
2
# days evaluated
14
34
Recommendations
sent
160
120
Answered
recommendations
76
55
Voted useful
7
5
Voted not useful
69
50
Ratio
useful/answered
9.21%
9.10%
Number of active
rules
54
46
Number of rules
that resulted in
recommendations
23
17
Number of rules
with 10 negative
feedbacks
5
3
A. Phase 1
The aim of the first phase was to provide a large basis of
data for evaluation and further improvement of the system.
The analysis of the data collected during phase 1 should help
to improve the recommender system in terms of decreasing the
negatively rated recommendations in phase 2, while holding
the positives at the highest amount possible. The results for
phase 1 are summarized in TABLE I.
After running the evaluation for phase 1 for 2 weeks, the
inhabitants were interviewed and their feedback revealed two
main findings:
Absent scene should not be recommended as action
it doesn’t make sense to recommend a user to leave
home
The low response rate of 47.5% was caused by
ambiguous recommendations some users have
similar or equal names for different devices inside
their household. Consequently, a recommendation to
turn off the “spot light” could apply for more than
one device and the users could not decide if the
recommendation was useful and ended up not
responding at all
We did a regression analysis of the results using the
weighted feedback as the dependent variable and the following
prioritization served as explanatory variables:
The length of the pattern
The position of the action
Support of the pattern
Confidence of the rule
Because repeating the pattern mining on the event data
barely produced new patterns, the date when the rule was
mined was ignored as explanatory variable.
Patterns with high confidence and high pattern-length tend
to receive better feedback than the other patterns, as shown in
Fig. 8. On the other hand, support and position of the action
did not show any significance to describe the feedback a rule.
It is worth to notice that support was the major attribute for
mining frequent patterns.
Fig. 8. Scatter plot of confidence with line of best fit
B. Phase 2
The aim of the second phase of the evaluation was to
increase the ratio of answered recommendations and the ratio
of useful recommendations compared to the first phase. The
analysis of the data collected during phase 1 was used to adapt
the system, which should lower negative rated and
unanswered recommendations. The prototype was improved in
the following points:
All rules which were excluded during phase 1, with
10 negative feedbacks in a row, were removed from
phase two. The feedback count from phase 1 was
reset for all patterns
All rules which recommended the action “absent”,
where excluded from the second phase
To reduce the problem of the low response rate
caused by ambiguous recommendations, we enriched
the text with the name of the room of the device.
Since no manual matching was necessary, this feature
is still in the scope of this work
As result of the regression analysis, confidence and
pattern-length of each rule where multiplied with
their estimate to calculate a coefficient which gives
indication about the usefulness of a rule. A threshold
is defined and 19 rules out of 54 with a coefficient
below this value where excluded from the second
phase (35 rules remained).
Hi Michael Zehnder,
I would recommend to turn-off device Bed
side lamp (on 2014-11-16 23:19:16).
Is this recommendation useful?
Yes:
http://snurl.com/29fgj3d?r=2663&c=4f=1
No:
http://snurl.com/29fgj3d?r=2663&c=4f=0
The results from phase 2 are summarized in TABLE I. as
well. The results show a similar ratio of useful
recommendations as in phase 1, as well as a similar response
rate (45.8%). However, a significant improvement can be
observed in the number of recommendations sent: 0.44
recommendations/day/home in phase 2 versus 1.43
recommendations/day/home in phase 1. Note that, for the
same ratio of useful recommendations, a lower number of
recommendations per day per home means less noise for the
user and a better comfort level.
We consider the 10% ratio of useful recommendations a
promising good start for a first version of the system which
has not been optimized nor has it seen a large amount of usage
data.
VI. CONCLUSION AND FUTURE WORK
In this work, we presented a case study of a system that
generates recommendations to save energy in smart homes
without reducing the comfort of the inhabitants. The results
show that such a system works in real life and achieved a ratio
of useful recommendations of about 10%, while sending 0.44
recommendations/day/home.
Several points of improvement were identified during the
evaluation phases of this work. A follow-up research project is
already ongoing and will build upon the findings of this work.
The following ideas for further research materialized during
the design, implementation or evaluation of the recommender
system:
Using confidence and pattern length instead of
support or periodicity as criteria for the mining
algorithm, resulting in more and better patterns
The time between two events (or the action) is
considered neither by the mining algorithm nor by
the recommender system. Using this information will
improve the accuracy of the suggestions made by the
system
Other attributes could be introduced to decide if a
rule is relevant or not. Such attributes might be:
Time of day when the pattern occurs most
Weekday when the pattern occurs most
Season when the pattern occurs most
The recommender should learn from the feedback of
the inhabitants in order to prioritize the rules, instead
of just excluding a rule after 10 negative feedbacks in
a row
Look into estimating the amount of energy that
would be saved by the recommendations
Test other machine learning algorithms and
frameworks such as Torch and Coffe
VII. REFERENCES
[1] King,N. “Smart home. A definition” Intertek Research and Testing
Center, pp. 1-6, 2003
[2] Blumendorf, M. “Building Sustainable Smart Homes” Proceedings of
the First International Conference on Information and Communication
Technologies fro Sustainability ICT4S pp 151-158, Feb 2013.
[3] Harper, R.”Inside the Smart Home” Springer-Verlag ISBN 978-1-
85233-688-2, 2003
[4] Baumann, C. et al. “Smart energy case study” Proceedings ACM
Workshop on Embedded Sensing Systems for Energy-Efficiency in
Buildings. BuildSys pp 36-38, Nov 2012.
[5] Dickmann, G. digitalSTROM: A centralized PLC Topology for Home
Automation and Energy Management” IEEE International Symposium
on Power Line Communications and its Applications (ISPLC), pp 352-
357, April 2011.
[6] Rashidi, P. & Cook, D.J. Mining and Monitoring Patterns of Daily
Routines”, 2010, pp.336345..
[7] Augusto, J. & Nugent, C. Designing smart homes: the role of artificial
intelligence, pp 1-15, Springer Verlag ISBN 978-3540359944,2006
[8] Gopalratnam, K. & Cook, D. Online Sequential Prediction via
Incremental Parsing: The Active LeZi Algorithm. IEEE Intelligent
Systems, 22(1), 2007, pp.5258.
[9] Du, Y. et al. Recognizing Interaction Activities using Dynamic
Bayesian Network. 18th International Conference on Pattern
Recognition (ICPR’06), 2006, pp.618621
[10] Rashidi, P. et al. Discovering Activities to Recognize and Track in a
Smart Environment. IEEE transactions on knowledge and data
engineering, 23(4), 2011, pp.527539.
[11] Rashidi, P. & Cook, D. Mining sensor streams for discovering human
activity patterns over time. IEEE Data Mining (ICDM), 2010, pp. 431-
440.
[12] Mannila, H., Toivonen, H. & Verkamo, A.I. Discovery of frequent
episodes in event sequences. Data Mining and Knowledge Discovery,
ISSN 1384-5810, 1997, pp.259289
[13] Youngblood G.M, et al.. Automation Intelligence for the Smart
Environment.” 2005
[14] Crandall, a. S. & Cook, D. Attributing events to individuals in multi-
inhabitant environments. 4th International Conference on Intelligent
Environments (IE 08), 2008, pp.4B44B4
[15] Nazerfard, E., Rashidi, P. & Cook, D. Using association rule mining to
discover temporal relations of daily activities. Toward Useful Services
for Elderly and People with Disabilities. ISBN: 978-3-642-21534-6
2011, pp.4956
[16] Krishnan, N., Cook, D. & Wemlinger, Z. Learning a taxonomy of
predefined and discovered activity patterns. Journal of Ambient
Intelligence andSmart Environments, vol5, no.6, 2013
[17] Rohini, P. & RajKumar, R. “A New Approach to Behavioral Reasoning
in Smart Homes using DVSM Algorithm. ” International Journal of
Advanced Research in Computer Science and Software Engineering. ,
4(2), 2014, pp.334341
[18] Achar, A., Laxman, S. & Sastry, P.S. A unified view of the apriori-
based algorithms for frequent episode discovery. Knowledge and
Information Systems, 31(2), 2011, pp.223250
[19] Schweizer, D. Learning frequent and periodic usage patterns in smart
homes. Master’s thesis, School of Business, University of Applied
Sciences and Arts Nortwestern Switzerland (FHNW) Feb 2014.
[20] Aztiria, A. et al. “Discovering frequent user--environment interactions in
intelligent environments. Personal and Ubiquitous Computing, 16(1),
2011, pp.91103.
[21] Forgy, C.L. Rete : “A Fast Algorithm for the Many PatternIMany Object
Pattern Match Problem. , 19(1982), pp.1737
[22] Deenadayalan, E., Hanirex, D. & Kaliyamurthie, K.,. “Association rule
mining in distributed database system”. 3(4), 2014, pp.727732.
[23] Agrawal, R. & Srikant, R.,. “Fast algorithms for mining association
rules.” Proc. 20th Int. Conf. Very Large Data Bases, VLDB, Volume
1215, 1994, pp. 487-499.
... The authors of [102] concluded that the two major energy preservation methods are energy saving and energy harvesting. Most of the energy saving techniques are implemented through the estimate and control of the uptime of end devices [100,103,104,105,106,107,108]. The rest of this subsection focuses on machine learning-based energy saving techniques. ...
... The authors from [104] pointed out that the manual labeling of training data is time consuming in supervised learning algorithms. Therefore, the authors proposed an energy saving scheme based on unsupervised learning. ...
Article
Full-text available
Applications and technologies of the Internet of Things are in high demand with the increase of network devices. With the development of technologies such as 5G, machine learning, edge computing, and Industry 4.0, the Internet of Things has evolved. This survey article discusses the evolution of the Internet of Things and presents the vision for Internet of Things 2.0. The Internet of Things 2.0 development is discussed across seven major fields. These fields are machine learning intelligence, mission critical communication, scalability, energy harvesting-based energy sustainability, interoperability, user friendly IoT, and security. Other than these major fields, the architectural development of the Internet of Things and major types of applications are also reviewed. Finally, this article ends with the vision and current limitations of the Internet of Things in future network environments.
... The behavior of consumers has been described as soft characteristic which proves difficulties in measuring objectively in a similar way as energy consumption (Zehnder et al., 2015). Note that not all consumer behaviors result into energy conservation. ...
... Note that not all consumer behaviors result into energy conservation. It is therefore important to identify relevant behavioral patterns that result into energy savings (Zehnder et al., 2015). Important patterns must result into at least an action which results into energy savings and comprises of normal events. ...
Article
Full-text available
Innovative solutions targeting improvements in the behavior of energy consumers will be required to achieve desired efficiency in the use of energy. Among other measures for stimulating consumers’ behavior changes based on attention triggers, personalized recommendations are essential to enhance sustainable progress towards energy efficiency. In light of this challenge, the current study focuses on innovative energy services that are based on intelligent recommendation systems and digital twins. We review several trends associated with the modeling and diffusion of energy services, taking into account the positive interrelationships existing between recommendation provisions and demand-side consumer energy behavior. This is achieved by means of a content analysis of the state-of-the-art works, focusing on the IEEE Xplore and Scopus databases. Based on this review, we present new empirical evidence to validate data-driven twin technologies as novel ways of implementing consumer-oriented demand-side management via sophisticated abstraction of consumers energy behaviors, and identify various barriers associated with the adoption of energy services, especially as they relate to the implementation and overall adoption of the digital-twins concept. Lastly, we use the review to summarize a coherent policy recommendations related to the wide-spread adoption of the digital-twins concept, and demand-side management solutions in general.
... Most of the research works existing in the literature focus on automatic monitoring and controlling of home appliances [30][31][32], effective energy management [33][34][35], integration of different protocols [36], and security [37]. However, the technological solutions reported so far for EOL management for PCBs and mechatronics products for domestic use are limited. ...
Article
Full-text available
Due to rapid changes in consumer demand and electronic technology advancements, electronics products in general, and specifically home automation segment products, are becoming obsolete at a very high rate. Therefore, strategies for End-of-Life (EOL) management need to be adapted. However, EOL management of these products is not globally approached due to the lack of autonomous, integrated, modular, and flexible systems to cope with the wide variety, and heterogeneous complicate structure of products. This paper reviews current practices and issues related to EOL management of printed circuit boards (PCBs) and mechatronics products in the home automation segment, focusing on their different material compositions as well as on the uses of these raw materials in the circular economy perspective. In addition, this paper provides possible Cyber-Physical System (CPS) implementation scenarios using physical tools available at the STIIMA-CNR de-manufacturing pilot plant. Eventually, this paper highlights challenges and guidelines to develop a new generation of advanced EOL management through the Cyber-Physical System technique.
... By moving insight from a center incorporated cloud to an edge of an enterprise can be made nearer to where they have happened. Edge computing can help to illuminate latency challenges and empower organizations to open doors utilizing a distributed computing architecture [5][6][7][8]. A smart IoT network incorporates policies to push information preparation and management of application data to the edge of the net. ...
Article
Full-text available
In today’s era, the acceptance of IoT-based edge devices is growing exponentially, which creates challenges of data acquisition, processing, and communication. In the edge computing paradigm, intelligence is shifted from the center to the edge by performing specific processing and prediction locally. Existing methods are not well computed with timely manner and it reduces the security level by overfitting problem. A strategy based on reducing communication resources between sensors and edge devices is the prime focus of this investigation. It uses a predictive model-based policy at edge devices for the reconstruction of not delivered context vector. A new hybrid Averaged Exponential Smoothening policy proposed is based on the current context vectors as well as a smoothing vector to reduce reconstruction error and improve the percentage of communication. It is observed that if we send data only when there is a marginal change in data then we can reduce communication overhead as well as keep reconstruction error low. This policy would be suitable for IoT-based edge computing applications for the smart city such as Smart Home, Healthcare, and Intelligent traffic to delivers the power of AI.
... A clustering phase takes place, to identify which consumers are similar, and historical retail plan data are used to train the model. In [104] and [105], the authors use pattern mining techniques to extract energy-saving patterns from household smart meter (appliance level) data and create association rules. These rules are then used as part of a recommender system that matches the incoming real-time household data with the association rules pool, and if a pattern is matched, the relevant action of the rule is recommended to the user. ...
Article
Full-text available
Consumers lie at the epicenter of smart grids, since their activities account for a large portion of the total energy demand. Therefore, utility companies, governmental agencies, and various other entities with environmental concerns aim at lowering and shaping energy consumption patterns to achieve peak load reduction, load smoothing, and hence carbon emission curtailment. In this survey paper, we present an overview of approaches for engaging smart grid consumers and for providing them with information, motivation, and recommendations for energy efficiency through mobile apps. Our focus is to bring machine learning approaches closer to smart grid mobile apps so as to optimally manage consumer flexibility and enhance energy savings through detailed consumer profiling and modeling, since an increasing amount of energy consumer data is becoming available. A novel survey and analysis of prior work in the area is conducted in order to identify gaps from this perspective. We consider both recent research project outputs and commercial products and we discuss various aspects of the designs, such as state-of-the-art technologies, extrinsic and intrinsic motivation techniques, gamification, consumer profiling, and the role of machine learning and recommender systems in this context. Furthermore, different mobile apps are presented and compared based on the most important features that affect consumer energy efficiency and sustainability, such as data visualization, gamification, flexibility, consumer profiling methods, feedback mechanisms, recommendations, social media, and machine learning integration. The main goal of this work is to identify how mobile apps incorporate these features to engage energy consumers in energy-efficient behavior, assess the current state-of-the-art in the area, and highlight future research directions.
... In the literature, several authors have addressed the use of recommender systems for energy efficiency purposes. In the context of smart homes and with the aim to save resources (mainly energy) and reduce consumption, we refer the reader to (Shah et al., 2010;González Alonso et al., 2011;Bhattacharjee et al., 2014;Zehnder et al., 2015;Streltov and Bogdan, 2015;Palaiokrassas et al., 2017;Ayres et al., 2018;Schweizer et al., 2015;Chen et al., 2016;García et al., 2017;Teoca and Ciuciu, 2017;Nakamura et al., 2016;Matsui, 2018;Li et al., 2013). Similarly, but in the context of smart buildings, we found (Fotopoulou et al., 2017;Pinto et al., 2019) as relevant papers. ...
Article
Among other conceptualizations, smart cities have been defined as functional urban areas articulated by the use of Information and Communication Technologies (ICT) and modern infrastructures to face city problems in efficient and sustainable ways. Within ICT, recommender systems are strong tools that filter relevant information, upgrading the relations between stakeholders in the polity and civil society, and assisting in decision making tasks through technological platforms. There are scientific articles covering recommendation approaches in smart city applications, and there are recommendation solutions implemented in real world smart city initiatives. However, to the best of our knowledge, there is not a comprehensive review of the state of the art on recommender systems for smart cities. For this reason, in this paper we present a taxonomy of smart city features, dimensions, actions and goals, and, according to these variables, we survey the existing literature on recommender systems. As a result of our survey, we do not only identify and analyze main research trends, but also show current opportunities and challenges where personalized recommendations could be exploited as solutions for citizens, firms and public administrations.
Article
in the recent era of digital revolution the life becomes simple and more comfortable than it was before. Today, we are witnessing a revolution in the Information Technology domain where the Internet of Things is employed in various aspects of our life. The impact of this revolution is clearly seen in our homes where appliances are controlled and managed via voice commands, inspiration, hand signals, or by a command which results from analyzing the human behavior. Power conservation is one of the significant issues in smart homes and smart cities where buildings consume about 40% of the total energy. Another crucial issue is the user satisfaction. Achieving a balance between power conservation and the user satisfaction is a challenge. In this paper, our contributions are: 1) A survey that sheds light on various techniques (the state-of-the-art) used for reducing the power consumption based on monitoring the occupant's behavior. 2) A comparison between these techniques based on various factors elicited from the literature review. 3) This study reveals the following gaps in the previous work; A) Lack of integrity between the IoT systems. B) Lack of auto measuring of the user satisfaction. C) Lack of achieving a balance between the user satisfaction and the power saving. As an attempt to close the above gaps, this paper proposes a smart and integrated IoT framework for auto measuring of the user satisfaction and thus achieves a balance between the power conservation and the user satisfaction. Besides, it suggests future research directions for researcher.
Article
This work is devoted to the method of determining the effective thermal parameters of heating sources in a smart home, which involves a combination of algorithms for data analysis and the equation of the physical process of heat transfer. The use of such parameters allows one to create software and hardware solutions for modeling the thermal map of the house, as well as to analyze energy consumption using the machine learning models. Since, for the most part, the total consumption of heating energy is known, it is of interest to determine the part of the energy that corresponds to the individual heating sources. To this end, the article proposes a mathematical model and algorithm for estimating the effective thermal characteristics of heating sources based on the heat transfer equation and data analysis approaches that can be used to obtain information about individual heating sources. The task of determining such parameters is reduced to two stages. At the first stage, using the finite-difference approach to the heat transfer equation, the effective thermal parameter of the heating sources is determined. Further, according to the data of energy consumption and distributions of room temperatures and temperatures on the surface of heating elements, by applying data analysis methods, an algorithm for estimating individual effective thermal characteristics of heating elements installed in rooms is proposed.
Article
Full-text available
Many intelligent systems that focus on the needs of a human require information about the activities that are being performed by the human. At the core of this capability is activity recognition. Activity recognition techniques have become robust but rarely scale to handle more than a few activities. They also rarely learn from more than one smart home data set because of inherent differences between labeling techniques. In this paper we investigate a data-driven approach to creating an activity taxonomy from sensor data found in disparate smart home datasets. We investigate how the resulting taxonomy can help analyze the relationship between classes of activities. We also analyze how the taxonomy can be used to scale activity recognition to a large number of activity classes and training datasets. We describe our approach and evaluate it on 34 smart home datasets. The results of the evaluation indicate that the hierarchical modeling can reduce training time while maintaining accuracy of the learned model.
Article
Full-text available
Frequent episode discovery framework is a popular framework in temporal data mining with many applications. Over the years, many different notions of frequencies of episodes have been proposed along with different algorithms for episode discovery. In this paper, we present a unified view of all the apriori-based discovery methods for serial episodes under these different notions of frequencies. Specifically, we present a unified view of the various frequency counting algorithms. We propose a generic counting algorithm such that all current algorithms are special cases of it. This unified view allows one to gain insights into different frequencies, and we present quantitative relationships among different frequencies. Our unified view also helps in obtaining correctness proofs for various counting algorithms as we show here. It also aids in understanding and obtaining the anti-monotonicity properties satisfied by the various frequencies, the properties exploited by the candidate generation step of any apriori-based method. We also point out how our unified view of counting helps to consider generalization of the algorithm to count episodes with general partial orders. KeywordsFrequent episode mining–Serial episodes–Apriori-based–Frequency notions
The area of smart homes is fast developing as an emergent area which attracts the synergy of several areas of science. This volume offers a collection of contributions addressing how artificial intelligence (AI), one of the core areas of computer science, can bring the growing area of smart homes to a higher level of functionality where homes can truly realize the long standing dream of proactively helping their inhabitants in an intelligent way. After an introductory section to describe a smart home scenario and to provide some basic terminology, the following 9 sections turn special attention to a particular exemplar application scenario (provision of healthcare and safety related services to increase the quality of life) exploring the application of specific areas of AI to this scenario.
Article
Conceptions of the Home.- Inside the Smart Home: Ideas, Possibilities and Methods.- Conceptions of the Home.- Smart Homes: Past, Present and Future.- Households as Morally Ordered Communities: Explorations in the Dynamics of Domestic Life.- Time as a Rare Commodity in Home Life.- Emotional Context and "Significancies" of Media.- Designing for the Home.- Paper-mail in the Home of the 21st Century.- Switching On to Switch Off.- The Social Context of Home Computing.- Design with Care: Technology, Disability and the Home.- The Home of the Future.- Towards the Unremarkable Computer: Making Technology at Home in Domestic Routine.- Daily Routines and Means of Communication in a Smart Home.- Living Inside a Smart Home: A Case Study.- Smart Home, Dumb Suppliers? The Future of Smart Homes Markets.
Conference Paper
As the world moves towards renewable energy production, the importance of involving energy-consumers into the process of scheduling energy increases. In this paper we investigate which devices should be used to balance energy consumption to match its availability. We classify electrical devices in categories that can either start consuming energy when available or can be stopped when there are low levels of energy available. We built a prototype system that enables any household to use its electrical appliances to load balance its consumption using information on how a user has configured its appliance available to a central energy distribution system. We simulated two algorithms that could be used at an energy provider to make use of the information on how users have configured their appliances for load balancing. We present our results and a recommendation on the algorithm that can be used for load balancing on the consumer side.