112 IEEE TRANSACTIONS ON GREEN COMMUNICATIONS AND NETWORKING, VOL. 1, NO.

1, MARCH 2017

An Innovative Approach for Forecasting of Energy

Requirements to Improve a Smart Home
Management System Based on BLE
Mario Collotta, Member, IEEE, and Giovanni Pau, Member, IEEE

Abstract—Smart grid applications are becoming increasingly major research topic for sensor networks, cellular networks [4]
popular, as they aim to meet the energy requirements with inno- and mobile devices, where minimizing both battery size and
vative solutions by integrating the latest digital communications weight is a critical issue [5].
and advanced control technologies to the existing power grid.
In smart metering management systems, several incentives, such The need for greener and more energy efficient communica-
as demand response programs, time-of-use, and real-time pric- tions technologies [6] has been well recognized by the research
ing, are applied by utilities in order to encourage customers to community, as demonstrated by the significant research efforts
reduce their load during peak load hours. However, it is usu- in this area during the last ten years [7]. In detail, there has
ally a hassle for residential customers to manually respond to been a considerable research on green wireless communica-
prices that vary over time. To overcome this limitation, this paper
presents an artificial neural network (ANN) as a support for a tion technologies [8], [9] because, as mentioned previously,
home energy management (HEM) system based on Bluetooth they are expected to help in reducing their impact on envi-
low energy, called BluHEMS. The proposed mechanism is able ronment. Therefore, it is necessary to design energy-efficient
to forecast the energy consumption conditions, i.e., to predict schemes that can reduce CO2 emissions and cost-effective
the home energy requirements at different times of the day or energy management also in the smart grids. A smart grid is
on different days of the week. This paper provides a detailed
description of the ANN configuration, an analytical analysis on conceptualized as a combination of underlay electrical network
the embedding parameters for the derivation of best performance and overlay communication network. In detail, a smart grid
conditions values, and simulative assessments, obtained through is a modernized power transmission and distribution network,
MATLAB and NS-2 simulations. which uses two-way data communications, distributed comput-
Index Terms—Smart home, home energy management, wireless ing technologies and smart sensors in order to improve safety,
sensor networks, fuzzy logic, neural network, BLE. reliability and efficiency of power delivery and use. An appli-
cation field is surely the Home Area Network (HAN) [10],
where a communication path is created among smart
I. I NTRODUCTION meters, home appliances and, eventually, plug-in electric
HE RESEARCH of new technologies, that can enable vehicles.
T a transition towards a more sustainable society, is one
of the most urgent challenges of the 21st century. Certainly,
The HAN enables consumers to collect information about
their consumption behaviors and the electricity usage costs
Information and Communications Technology (ICT) is one of via in-home display devices. The main purpose of the tra-
the keys to a future sustainable society. The goal of communi- ditional electric energy metering is to achieve the electric-
cations technologies is to achieve a large-scale energy savings, ity charge calculation, whose precision is not enough on
in order to reduce greenhouse gas emissions and to attain the customer measurement data acquisition. On the con-
effective environmental protection in several contexts, includ- trary, the smart metering management system, for Home
ing communications, manufacturing, transportation, buildings, Energy Management (HEM), collects more comprehensive
electricity generation and distribution. As a consequence, there and detailed measurement information, by installing smart
is a critical need for new ways to make communication meters for residential, industrial and commercial users. The
and information systems more efficient. Furthermore, it is smart metering management system is assisted by Time-Of-
necessary to highlight that energy costs [1] are significant Use (TOU) power price measures [11], in order to inhibit peak
in scenarios where there is a broad range of communica- loads thus reducing their growth. Furthermore, it is possible
tions networks, ranging from wired networks [2] to wireless to obtain both a more detailed and a real-time control of load
ones [3]. For this reason, the reduction of energy use is also a cases.
The introduction of the smart metering technology can
Manuscript received September 7, 2016; revised December 28, 2016; strengthen the management on demand side [12]. The main
accepted February 10, 2017. Date of publication February 17, 2017; date of
current version March 17, 2017. The associate editor coordinating the review aim is to inform the consumers of energy consumed costs
of this paper and approving it for publication was C. Chaudet. at any time, thus helping them to make adjustments accord-
The authors are with the Faculty of Engineering and Architecture, Kore ingly, such as shutting down some home appliances or using
University of Enna, 94100 Enna, Italy (e-mail: mario.collotta@unikore.it;
giovanni.pau@unikore.it). the energy from high price hours to low price ones. As a
Digital Object Identifier 10.1109/TGCN.2017.2671407 consequence, the consumers are benefited heavily from such
2473-2400 c 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: VIT University. Downloaded on July 06,2020 at 17:48:50 UTC from IEEE Xplore. Restrictions apply.
COLLOTTA AND PAU: INNOVATIVE APPROACH FOR FORECASTING OF ENERGY REQUIREMENTS 113

a system, as they can monitor the real-time energy used by the wireless network proposed in these works obtain good
the individual appliances. Moreover, they can compare energy performance, the authors do not bother to choose the most
efficiency of different appliances and control various of them, appropriate wireless protocol. In fact, the choice of a wireless
like air conditioning and ventilation systems, smart appliances, protocol with respect to another one may involve signifi-
smart plugs, etc. In addition, the peak load management can cant benefits and quite different results. For this reason, in
be employed using the available data obtained by the HEM other literature works, focused both on smart grids and on
system. Whereby, the dynamic demand response management, smart homes research fields, also a comparison among wire-
which is a desirable feature of smart grid system, can be inte- less protocols has been carried out. For instance, in [27]
grated with individual household profiles using a HEM system. IEEE 802.15.4/ZigBee is compared with Wi-Fi and Bluetooth
To this end, in several literature works [13], [14] a smart HEM Classic, showing its performance. In [28] Wi-Fi technology
system, based on optimization techniques in the context of for smart grid sensor networks is explored and results high-
smart grid for effective demand side management, have been light the advantages of Wi-Fi over other wireless protocols
introduced, highlighting that these techniques mainly rely on when large data rate is required. In [29] a thorough analysis
matching present generation values with demand by control- of wireless protocols for HANs is presented, comparing and
ling the energy consumption of appliances and optimizing their discussing their advantages and downsides.
operation at the user side. Among wireless protocols, in these last years a novel
Due to the low-bandwidth requirements of HAN appli- research activity has focused on BLE because it has a high
cations, it is necessary to use cost-effective communication potential in becoming a key technology for smart homes in low
technologies [15]. Consequently, it is important to propose an power, low cost and small devices. In spite of its installed base
energy-efficient technique for smart metering towards green in smart energy, home and building automation applications,
communications in the smart grid. Recently, the Wireless IEEE 802.15.4 faces competition in BLE in these applications,
Networks (WNs) have been widely recognized as a technol- as well. However, BLE has a lower energy consumption than
ogy promising to improve several aspects [16] also of smart IEEE 802.15.4 and, for this reason, may be a fine choice for
grids [17], especially those that deal with power generation, home automation applications. On the contrary, IEEE 802.11
bidirectional delivery, utilization and seamless monitoring, is used in devices where cost and low power are less impor-
providing an energy efficient, reliable and low-cost solu- tant [30], when large data rate is required and as a wireless
tion for control management [18]. The existing and potential backbone combined with the other wireless technologies.
WN applications for smart grid include advanced metering, This paper shows an improvement to BluHEMS (Bluetooth
demand response and dynamic pricing, equipment fault diag- Low Energy Home Energy Management System) [31] and
nostics, fault detection, load control, distribution automation to its later version [32]. In detail, BluHEMS is a novel
and remote power system monitoring and control. The smart energy management approach based on the communication
grid is also considered as a data communications network. among home appliances, a central energy management unit,
As a consequence, the communication capabilities among the a smart meter and an energy storage unit inside a smart home,
elements of an electrical power system play a key role on through a wireless network based on BLE. The main limit of
the efficient performance of any WN-based smart grid appli- BluHEMS consists that it lacks in a system able to manage the
cation. Furthermore, it is necessary to note that the selection scheduling of home appliances in a smart way, that is related
for the most appropriate wireless communication technology to crisp variables in which the switching on/off choice may be
varies depending on the requirements of WN-based smart grid taken based on the degree of membership of these variables
applications. For instance, smart metering applications require in certain intervals (membership functions). This, along with
latency-tolerant information exchange between the meters and the ability to use a mechanism able to manage the feedback of
utility management center. consumers, since they are Involved in the choice of switching
A number of wired as well as wireless communications on/off of home appliances, has led the authors to propose a
technologies for smart grid applications have been identi- new version of their approach [32] introducing a Fuzzy Logic
fied in recent years. However, in many of them the sheer Controller (FLC). However, the approach proposed in [32]
number of communications links makes the use of wired solu- has the limitation that it is not able to predict the home
tions economically and/or physically prohibitive. On the other energy consumption and then bases its choices only taking
hand, wireless technologies offer benefits such as lower cost into account the current value (recorded at run-time) of home
of equipment and installation, quick deployment, widespread loads (energy requirements). This, for instance, does not allow
access and greater flexibility [19], [20]. To this end, sev- the consumers to use more energy at this time and then save
eral efforts have been made in order to explore suitability of it later.
specific wireless communications technologies, such as IEEE This paper extends the work in [32] introducing an Artificial
802.15.4/ZigBee [21], IEEE 802.11/Wi-Fi [22], Bluetooth Neural Network (ANN) able to forecast the energy consump-
Low Energy (BLE) [23], for smart grid applications. tion conditions, i.e., to predict the energy requirements at
Several approaches, based on wireless networks capabilities, different times of the day or on different days of the week.
have been introduced in the literature, with the aim to make The novelty introduced in this work is the combined use of
intelligent decisions in residential buildings, in order to link BluHEMS [31], [32] with the ANN, for forecasting of energy
demand response, TOU, and/or real-time pricing incentives requirements. The synergy between the approach proposed in
with energy efficiency aspects [24]–[26]. However, although this work and BLE allows the HEM scheme, introduced in

Authorized licensed use limited to: VIT University. Downloaded on July 06,2020 at 17:48:50 UTC from IEEE Xplore. Restrictions apply.
114 IEEE TRANSACTIONS ON GREEN COMMUNICATIONS AND NETWORKING, VOL. 1, NO. 1, MARCH 2017

Fig. 1. Architecture of the proposed energy management system.

BluHEMS, to make its decision taking into account not only

the current situation, but also the probable short-term evolu-
tion of the energy consumption conditions. The benefits of the
improvements introduced in this work for BluHEMS are con-
Fig. 2. Wireless HAN architecture based on BLE.
firmed by validations, whereas, to the best of our knowledge,
there are no works in the literature in which BLE is applied
in a wireless network that supports a HEM scheme based on
a cooperation with an ANN. the electricity bill of the consumer by shifting the appliances’
This paper is organized as follows. Section II introduces operation from peak demand hours to off-peak ones. In fact,
the proposed system model for smart home, presenting both the electrical appliances that are part of the home system can
the architecture of the energy management system and the be used at various time periods every day depending mainly
wireless HAN based on BLE. Section III presents the ANN on the day of the week. It is known that there are substan-
for forecasting of energy requirements while Section IV shows tial differences in the usage time of appliances throughout the
simulation results. Finally, Section V concludes the paper. day, especially between working days and the weekend. For
this reason, as shown in Figure 1, the Energy Management
block take as input parameter the difference E between the
II. S YSTEM M ODEL value Ei (w) of the predicted energy requirement at time w
The proposed system for smart home is characterized by and the value Ei (w − n) of the energy requirement measured
the architecture depicted in Figure 1. The main elements of the at time (w − n), where n = 60, 55, . . . , 5 minutes. In this way
system are BluHEMS [31], for monitoring and controlling the the proposed system makes its decision taking into account
electrical appliances, planning a convenient start time for them, not only the current situation, but also the probable short-term
a FLC [32], to manage both the scheduling of home appliances evolution of the energy consumption conditions.
and the feedback of consumers, and an ANN, for forecasting The network architecture, on which the proposed solution
of energy requirements, described in detail in Section III of is based, is depicted in Figure 2 and is composed by sev-
this work. Since the main limitation of [32] is the lack of an eral independent Wireless Smart Cells (WSCs), managed by a
automated system capable to make choices based on both the BLE Master device that acts as a decision-making center of the
actual energy consumption values and of predicted ones, the entire HEM system. The whole wireless network is composed
ANN proposed in this paper has the aim to overcome this of all of the WSCs within which are the Field Devices (FDs),
limitation. In fact, the system proposed in this paper is similar which are Bluetooth Low Energy devices dealing with a spe-
to the one introduced in [32], that involves communications cific task, or are associated with home appliances using smart
among smart appliances and BluHEMS through a wireless plugs. Through wireless communication, the Master device
network. However, it is placed in series to that shown in [32] receives and processes data detected by FDs placed inside
(Figure 1). The consumer may turn on any home appliance at the WSC. It also allows sending user commands or system
any moment of the day without taking into account that, at that commands to FD nodes. Moreover, wireless links, based on
moment, it can be also in peak hours. BluHEMS, assisted by BLE, allow the communication among mobile devices (such
the FLC, allows the switching on of the appliance or suggests as smart phones or tablets) and the WSCs. As a consequence,
to the consumer which is the more appropriate start time, tak- people can authenticate themselves inside the home automa-
ing into account both the available stored energy in the storage tion system and subsequently monitor data detected by sensor
system and the updated prices in that time slot. However, the nodes and, if needed, send commands.
consumer can decide whether to accept the schedule proposed In home automation environments the blocking effects
by BluHEMS; for instance, the time period falls during peak between BLE master and slaves, due to walls or obstacles
hours or the level of stored energy is not enough, or to ignore in rooms, is not a real protocol limitation. In fact, BLE has
it, switching on the appliance. an approximately 3 dB better link budget compared to Classic
This paper aims to highlight the influence that different Bluetooth, thanks to a modified modulation. As a consequence,
parameters may have on the models proposed in [31] and [32]. a BLE device can offer a range of 200-300 m in the line
Compared to the work presented in [31] and, mainly, in [32], of site without the need for an additional power amplifier.
in this paper several parameters, such as the day of the week, This is an important feature concerning the robustness and
the hour of the day and the home load, are taken into account the reliability, because, eventually, it is possible to have an
in order to train an ANN model aiming at forecasting the appropriate placement of several devices in order to bridge
energy requirements. The output of the ANN is used to feed temporary obstacles and interference. In smart homes con-
BluHEMS, and as a consequence the FLC, which has the goal texts, sensors and actuators often need a range of 20-50 m,
of reducing home electricity consumption charges, decreasing whereby it would be not necessary to use power amplifiers.

Authorized licensed use limited to: VIT University. Downloaded on July 06,2020 at 17:48:50 UTC from IEEE Xplore. Restrictions apply.
COLLOTTA AND PAU: INNOVATIVE APPROACH FOR FORECASTING OF ENERGY REQUIREMENTS 115

On the contrary, considering that the range of BLE radio may of this neural network to handle problems involving the
be optimized according to the application, BLE manufacturers modeling of nonlinear dynamic systems, such as dependen-
may choose to optimize the range, because there is no limit cies among time series [34]. The proposed ANN is built by
imposed by the BLE specification. using a MultiLayer Perceptron (MLP) topology and is trained
One of the major advantages of the technology for the by using the duration and the energy consumption of home
smart grid area is the existing Bluetooth ecosystem, which appliances presented in [35]. The used input data in the ANN
makes it seamless for consumers to start using BLE in their (Figure 1) are: day of the week (D); hour of the day (H);
smart grid devices. Bluetooth is already in mobile phones home load (L) and energy requirement (Ei ).
and many personal computers, which can provide the nec- The input parameters (D, H, L, Ei ) of the ANN are given in
essary user interface for smart energy applications. Thanks to the NNARX model at the previous time moments. Therefore,
Bluetooth, the integration of a smart ecosystem throughout the the NNARX model computes the expected energy requirement
home could be made very easy and very fast. BLE is very reli- at time t one step ahead (t + 1). Subsequently, the difference
able with its support for adaptive frequency hopping (AFH) (E) of the energy requirement between t and t + 1 represents
and other features inherited from Bluetooth Classic. Its trans- the input parameter of BluHEMS that allows the switching on
mission rate, 1 Mbps, is suitable for all common applications of the appliance or suggests to the consumer which is the
in a smart home. As a consequence, BLE has a high potential more appropriate start time. Furthermore, the time instants of
in becoming an important technology for smart homes in low inputs parameters have not been determined arbitrarily, but
power, low cost and small devices. Regarding the integration using a formulation of the embedding parameters, shown in
of BLE in home appliances, it is possible to chose differ- Section III-A. The equation used in order to develop the ANN
ent solutions. In fact, the easiest way is to use smart plugs. model is the following [36]:
Another solution is to use COTS (Commercial Off-The-Shelf)
ŷ(t|θ ) = ŷ(t|t − 1, θ ) = g(ϕ(t), θ ) (1)
hardware that has to be programmed by a geek user. In this
case, the use difficulty may be higher, because the user has to where ŷ is the y value at time t predicted by the model, θ
develop his BLE network from scratch. is a vector containing the weights of the neural network, g
is the function realized by the neural network and g(ϕ) is
the vector of regressors. The selection of past used signals
III. ANN FOR F ORECASTING OF E NERGY as regressors are carried out by using a formulation of the
C ONSUMPTION C ONDITIONS embedding parameters described in the following subsection.
The soft computing techniques have been proposed for The network has been trained using the Levenberg-
the construction of new generation artificial intelligence Marquardt algorithm [36] and the NNARX model has been
(high machine intelligence quotient, human-like information built with 40 neurons in the Hidden Layer (HL). This value
processing) and for solving non-linear and mathematically has been chosen considering the results obtained by the ANN
unmodeled systems. In addition, soft computing techniques shown in Section III-B. In the ANN proposed in this work, the
can be implemented at low cost and can be appropriate for sev- early stopping technique is applied with the aim to ensure good
eral engineering problems, especially for complex problems, performance and the best generalization of the implemented
where classical control methods do not achieve comparatively model, as well as avoiding the over-fitting phenomenon. In
favorable results. For instance, in home automation scenarios, fact, this technique envisages that the data set is split into
where the behavior of consumers can be many times repetitive three groups: training set (it computes the gradient and then
and almost never unpredictable, it is possible to formulate an updates the weights and the bias of the network), validation
appropriate ANN model able to learn (as it is easy to obtain set and test set. During the training process, the error on the
a large number of data on which to perform the training) validation set is calculated. Usually, both the validation set
and then to predict the behavior of consumers. This means error and the training set error decrease in the initial phase
that the ANN can be a useful tool for the evaluation and of training. When the over-fitting starts, the error on the val-
the management of energy consumption. There are some fea- idation set typically begins to rise. If the validation set error
tures that make useful the application of ANNs, such as they continues to rise for a given number of epochs, the training
can handle large amount of data sets; they have the ability is stopped and returns the corresponding weights and bias to
to implicitly detect complex nonlinear relationships between the minimum validation error.
dependent and independent variables; they have the ability to In the proposed ANN, 60% of the data have been used for
detect all possible interactions between predictor variables; the training set, 20% for the validation set and the remaining
they are a powerful data-driven, self-adaptive, flexible com- 20% for the test set. Furthermore, before starting the simula-
putational tool having the capability of capturing nonlinear tions, all data have been normalized between the minimum
and complex underlying characteristics of any physical process and maximum of the data values. In the training set, the
with a high degree of accuracy. performance of the network have been evaluated by Mean
An ANN model with a non-linear-autoregressive Squared Error (MSE). The prediction performance of the
NNARX-type (Neural Network Auto Regressive eXter- network have been evaluated in the test set, using the fol-
nal input model) [33] structure is proposed in this work for lowing parameters: Mean squared error (MSE), Root square
forecasting of energy consumption conditions. The reason for mean error (RMSE), Mean absolute error (MAE), Mean abso-
the choice of NNARX network is due to the good ability lute percentage error (MAPE). Considering that Yi is the value

Authorized licensed use limited to: VIT University. Downloaded on July 06,2020 at 17:48:50 UTC from IEEE Xplore. Restrictions apply.
116 IEEE TRANSACTIONS ON GREEN COMMUNICATIONS AND NETWORKING, VOL. 1, NO. 1, MARCH 2017

of the i-th actual observation and Ŷi is its predicted value, the These methods, used to calculate the embedding dimen-
forecast error is calculated as follows: sion and the time delay, have been chosen, as they allow one
to obtain good results with the univariate time series {xi }N i=1
ei = Yi − Ŷi (2) taken into account in this paper. It useful to note that a direct
while pi is determined with the following equation: comparison with other approaches in the literature has not
carried out in this work, but after a deep analysis of other
Yi − Ŷi methods in the literature, it has been deduced that, in the con-
pi = × 100. (3)
Yi text considered in this paper and if an univariate time series is
taken into account, the autocorrelation function (for the time
A. State Reconstruction: Embedding Parameters delay) and the nearest neighbor technique (for the embedding
Considering the Takens’ embedding theorem [37], the dimension) are the most appropriate. The nearest neighbor
dynamical behavior of a state x = x(t) can be reconstructed has been chosen for several reasons: in the considered con-
by means of a m-dimensional vector, such as: text there are lots of training data; it has a very fast training;
learn complex target functions; do not lose information; at
[x(t); x(t − τ ); . . . ; x(t − (m − 1)τ )] (4) the tending of the amount of data to infinity, the algorithm
never exceeds two times the minimum error due to the dis-
where τ is the delay time, and m is the number of
tribution of the data. On the contrary, the auto-correlation
element also known as the embedding dimension [38].
function measures the correlation of a signal with itself shifted
Even if their values are not uniquely determined, these
by some time delay. So, in its nature, it is designed to take into
two parameters are crucial in the algorithm efficiency and
account a time delay and then can be used to detect repeats or
result accuracy, during the reconstruction state. In liter-
periodicity.
ature works two perspectives can be distinguished [39].
The embedding dimension (m) and the time delay (τ ),
In the first one, the embedding parameters may be con-
obtained for this study, for all of the regressors are the
sidered as independent each other (according with Takens
following: m ⇒ D = 5; H = 6; L = 6; Ei = 7 and
theorem), thus, different approaches can be proposed, such
τ ⇒ D = 4; H = 3; L = 3; Ei = 5. The respective vector
as Grassberger-Procaccia (GP) algorithm [40], for m cal-
ϕ(t) containing the regressors corresponding to the NNARX
culation; series correlation approaches, [41], approaches of
model is:
phase space extension [42], multiple autocorrelation and non-
bias multiple autocorrelation [43], for τ evaluation. On the ϕ(t) = [D(t − 4); . . . D(t − 20); H(t − 3); . . . H(t − 18);
other hand, if m and τ are closely related, combined method
L(t − 3); . . . L(t − 18); Ei (t − 5); . . . Ei (t − 35)].
are small-window solution [44], C-C (Correlation-Correlation
integral) method [45], and automated embedding [46]. (10)
In this paper an univariate time series {xi }N i=1 is taken into
account. The time delay is determined by using the autocorre- B. ANN Validation
lation approach [47], while the embedding dimension comes
In this subsection the results obtained by the proposed
out from the nearest neighbour technique [48]. Let xi , i = 1..N
NNARX model through MATLAB simulations are shown.
be a N-length discretization of the state x(t). The value of
These simulations have been performed in order to obtain the
the time delay τ is selected by examining the autocorrelation
configuration of the ANN that achieves the best performance.
function:
N−τ It is necessary to highlight that a very low error (a varia-
(xi − x)(xi+τ − x) tion of the order of tenths of a degree) may not affect the
r(τ ) = i=1N (5) performance of the proposed system. However, it is useful to
i=1 (xi − x)
2

 note that the proposed ANN aims to get good performance

where x is the mean value, i.e., x = N i=1 xi /N. As proposed through a low level of implementation complexity. This has
in [47] a good choice for the time delay is setting r(τ ) ≈ 0.97. been possible thanks to the method described in Section III-A.
Let us define: The inputs of the neural network are given by the sum of the
  embedding dimensions (m), specifically 24 = 5 + 6 + 6 + 7.
Vn = xn , xn−τ , . . . , xn−(m−1)τ (6) A higher number of inputs leads to a significant accumu-
for all n = (m − 1)τ + 1, . . . , N and let us denote J0 = lation of data in memory, and considering that the BLE
(m − 1)τ + 1 and Master, where the ANN is implemented, could have reduced
   capacity in terms of memory, the reduction in the num-
Vη(n) = argmin Vn − Vj 2 , j = J0 , . . . , N j = n . (7) ber of inputs also improves memory management in the
device.
The optimal embedding dimension m comes out by minimiz- The training parameters used in the simulations are the
ing the error function E(m), that is following:

N
  • performance goal: 7 × 10−3 ;
1 xn − xη(n) 
E(m) = (8) • learning rate: 0.4;
N − J0
n=J0 • maximum failure number for validation: 30;
m = {E(m) : m ∈ N} (9) • Marquardt adjustment parameter: 0.07.

Authorized licensed use limited to: VIT University. Downloaded on July 06,2020 at 17:48:50 UTC from IEEE Xplore. Restrictions apply.
COLLOTTA AND PAU: INNOVATIVE APPROACH FOR FORECASTING OF ENERGY REQUIREMENTS 117

TABLE I
P ERFORMANCE OF THE ANN

Fig. 4. Peak load reduction comparison.

note that it might be the IEEE 802.15.4. In the second scenario,

the network parameters are evaluated between the approach
proposed in this paper and the one of Keshtkar et al. Since
both the works in [31] and [32] and the proposed system are
based on a BLE architecture, in Section IV-B a comparison
is carried out among the results obtained by the approach
of Keshtkar et al. by using IEEE 802.15.4 (which it is the
protocol suggested by them), and the proposed approach,
Fig. 3. Measures of forecasting error.
implemented both with IEEE 802.15.4 and BLE. The results in
Section IV-B are shown in order to assess the implementabil-
Table I shows the values obtained by performance evalu- ity of the proposed approach in a HAN based on a wireless
ation in terms of error values (MSE, RMSE, MAE, MAPE). protocol. To this end, parameters of network performance will
The measures have been carried out by varying the number of be shown.
neurons in the HL from 10 to 100. In detail, the values shown
in Table I are the best obtained from all trained networks. The A. Comparison Among HEM Solutions
results show that the best performance has been obtained with The simulation scenario has been composed by several
40 neurons in the hidden layer. Moreover, regarding the vali- loads, represented by a washer, a dryer, a dishwasher, a
dation of the trained ANN, Figure 3 shows the errors obtained hairdryer, a dehumidifier and a coffeemaker. It’s obvious that
by the difference between the real energy requirement and the the duration (minutes) and the energy consumption (kWh)
best performance of predicted one within a year (365 days). As of these appliances are vendor specific; however, the refer-
it is possible to see, the ANN proposed in this paper obtains ence values for average load per cycle given in [35] have
a lower error, that is an average of about 0.1. been used. In addition, in order to develop a more real sim-
ulation scenario, an extra load has been supposed with an
IV. P ERFORMANCE E VALUATION electricity consumption varying between 0 kWh and 5 kWh
To assess the performance of the proposed model, a suitable randomly. Regarding the load, 80% of it has been considered
simulation campaign has been developed using the Network as miscellaneous while the remaining 20% has been related to
Simulator Version-2 (NS-2) [49]. In Section IV-A the proposed standby appliances. Just like in the simulation scenario devel-
approach is evaluated in terms of HEM performance and oped in [31] and [32], in this paper the peak hours fall from 8
in Section IV-B the network performance in a typical home AM to 2 PM, the switching on of an appliance has been con-
automation scenario, as depicted in Figure 2, are tested. In the sidered as a Poisson distribution and the requests have been
first evaluation scenario the simulations have been performed generated randomly. Regarding to the configuration parame-
making a comparison among the original BluHEMS [31], ters, the threshold value of power has been set to 1 kWh, the
its first improvement through the introduction of a FLC threshold value of delay has been set to 24 hours, simulations
(BluHEMS-FLC) [32] and the proposed model (BluHEMS- duration has been between 5 days and 365 days (1 year) and
ANN). Moreover, a comparison has been made with the the first 5 days are spared for warm up.
solution proposed by Keshtkar et al. [26], in which a fuzzy The electricity consumption pattern measured in a generic
logic approach, supported by wireless sensors and smart single day is depicted in Figure 4. The percentage of load
grid incentives, is presented for smart residential load reduc- in peak hours is a ratio between the amount of load in peak
tion. The main differences between the approach presented hours to the total load. High value of this ratio results in high
in [26] and the proposed approach are that the solution of electricity charges due to pricing tariffs. As it is shown in
Keshtkar et al. does not take into account of the consumers’ Figure 4, the approach of Keshtkar et al. [26] try to reduce
feedback, can not make any predictions on energy require- the percentage of load in peak hours in order to provide mone-
ments and the wireless protocol of their considered wireless tary benefits to the consumer. However, in general, BluHEMS,
network is not specified, although, only at the end, the authors BluHEMS-FLC and BluHEMS-ANN (here proposed) obtain

Authorized licensed use limited to: VIT University. Downloaded on July 06,2020 at 17:48:50 UTC from IEEE Xplore. Restrictions apply.
118 IEEE TRANSACTIONS ON GREEN COMMUNICATIONS AND NETWORKING, VOL. 1, NO. 1, MARCH 2017

Fig. 5. Load of the appliances ratio during peak hours. Fig. 7. Electricity consumption costs.

terms of energy requirements, and this represents a step for-

ward in determining the delay when the consumer turns on a
home appliance. Secondly, the consumer’s feedback, managed
through a FLC [32], represents a clear indicator on its desired
comfort level. This feedback is stored in order to improve the
calculation of the delay and a proper value can be obtained if
the consumer has already given one or more feedback in the
past on the same appliance. However, it is necessary to note
that the delays shown in Figure 6 can be tolerable for several
appliances, such as washers or dishwashers. On the contrary,
Fig. 6. Average delay experienced by the consumer in HEM schemes. for other appliances, such as coffeemakers or hairdryers, the
consumer, possibly, wants to start them immediately. For this
reason, the described smart system allows the consumer to
best performances since they are able to shifts loads from peak deny or to accept the scheduling proposal.
hours to off peak ones efficiently. Moreover, the best results Finally, the economic benefits obtainable by applying the
have been obtained by the approach proposed in this paper. proposed solution has been analyzed. It is known that across
This highlights the efficiency and the usefulness in forecasting the world the electricity distributors have implemented various
household energy requirements. tariff schemes for peak load management. In the simulation
In Figure 5 the contribution of the appliances on the aver- scenario developed in this paper, a TOU tariff has been taken
age peak load is shown. In this case, with the approach of into account and, as a consequence, tariff pricing rates are
Keshtkar et al. almost 0.1 of the load generated by the appli- different for different hours. In detail, the considered tar-
ances takes place during peak periods. This is a good result, iff is one proposed by an Italian provider, whose costs are
although it would be desirable to have better values. Even 0.0365 e/kWh in off peak hours and 0.1725 e/kWh in peak
in this case, BluHEMS, BluHEMS-FLC and BluHEMS-ANN hours (from 8 a.m. to 7 p.m.). The electricity consumption
shift loads from peak times in a more efficiently way and, cost of a single home has been simulated and the results for a
as a consequence, almost 0.06 of the total load is left in the twelve month electricity bill are shown in Figure 7. The elec-
peak hours. However, by using the solution here described, tricity bill increases with increasing months because the bill is
this value drops to 0.04. Whereby, even in this case, the best calculated cumulatively, month after month. Even in this case,
performance are those obtained by the proposed approach. the best results are obtained by using the ANN that achieves
During simulations, the average delay experienced by the a higher monetary cost reduction.
consumer has been measured and the obtained results are
depicted in Figure 6. It’s obvious that without an energy man-
agement scheme an appliance starts as soon as the consumer B. A Scenario of Simulated HANs
turns it on and, as a consequence, the delay is almost zero. On During simulations the performance of the wireless
the contrary, using an HEM scheme some delay is introduced. networks have been measured. In detail, the proposed system is
Using the Keshtkar et al. scheme, the delay fluctuates from based on BLE, while, as mentioned previously, in the solution
about 3.5 to 4.5 hours, while using BluHEMS, BluHEMS-FLC introduced by Keshtkar et al. [26] they note that IEEE 802.15.4
and BluHEMS-ANN, the delay fluctuates from about 2.5 to could be a suitable wireless protocol. For this reason, in order
3.5 hours, and in most of the simulation is less than 3 hours. to make a direct comparison and to show the benefits that can
Moreover, through the ANN as a support to BluHEMS (the be obtained by using BLE than IEEE 802.15.4, the wireless
solution shown in this work), the delay experienced by the con- networks of the proposed approach have been implemented
sumer is about 2.5 hours, which is the lowest value obtained both with BLE and with IEEE 802.15.4. The performance has
in the simulations. This is due to several reasons. First of all, been evaluated in terms of packet delivery ratio, end-to-end
thanks to low errors values measured during its validation, delay and jitter. It is useful to highlight that the delivery ratio
the neural network is able to make appropriate predictions in is the ratio of the number of successfully received packets to

Authorized licensed use limited to: VIT University. Downloaded on July 06,2020 at 17:48:50 UTC from IEEE Xplore. Restrictions apply.
COLLOTTA AND PAU: INNOVATIVE APPROACH FOR FORECASTING OF ENERGY REQUIREMENTS 119

TABLE II
P ERFORMANCE OF HAN S week. The proposed approach is suitable for smart grid appli-
cations for effective demand side management, as it is able to
rely on matching present generation values with demand by
controlling the energy consumption of appliances and optimiz-
ing their operation at the user side. The paper proposed a deep
analysis for the configuration of the ANN in order to obtain the
one that achieves the best performance. As far as the imple-
the number of sent packets. The end-to-end delay represents mentation complexity, the embedding parameters have been
the interval between sending a packet from the application analytically assessed and their values for best performance
layer of the source and receiving it at the application layer of conditions have been derived. The paper also has pro-
the destination. The jitter is the difference between the delays vided extensive simulative assessments, performed through the
experienced by the packets. The packet size of IEEE 802.15.4 Network Simulator Version-2 (NS-2), in terms of electricity
can be varied from 32 bytes to 128 bytes and during simu- consumption, load percentage and delay experienced by con-
lations the considered size has been set at 32 bytes, in order sumers for the proposed HEM solution, and in terms of packet
to obtain the closest size to that of the BLE packets. In fact, delivery ratio, delay, and jitter for the wireless networks.
BLE 4.0 supports only very short data packets, whose maxi-
mum size is 27 bytes. The traffic model has been considered
as Poisson and the NS-2 radio propagation model has been R EFERENCES
chosen as Shadowing [49]. [1] O. Orhan, D. Gündüz, and E. Erkip, “Energy harvesting broadband com-
The obtained results are shown in Table II. Regarding to munication systems with processing energy cost,” IEEE Trans. Wireless
Commun., vol. 13, no. 11, pp. 6095–6107, Nov. 2014.
the packet delivery ratio, the best result has been obtained [2] A. P. Bianzino, C. Chaudet, D. Rossi, and J. Rougier, “A step towards
using BLE protocol, which has allowed to achieve a value energy-efficient wired networks,” in Green Networking. Hoboken, NJ,
about 94.2%. On the contrary, using IEEE 802.15.4, both USA, 2013, pp. 17–58.
[3] P. Serrano, A. De La Oliva, P. Patras, V. Mancuso, and A. Banchs,
the approach of Keshtkar et al. and the proposed system “Greening wireless communications: Status and future directions,”
get roughly the same results, about 78%. In fact, as it is Comput. Commun., vol. 35, no. 14, pp. 1651–1661, 2012.
known, BLE data packets are more efficient when sending [4] K. Davaslioglu and E. Ayanoglu, “Quantifying potential energy effi-
ciency gain in green cellular wireless networks,” IEEE Commun. Surveys
small amounts of data infrequently. The proposed system, Tuts., vol. 16, no. 4, pp. 2065–2091, 4th Quart., 2014.
based on BLE, has obtained the best performance even in [5] E. Björnson, J. Hoydis, M. Kountouris, and M. Debbah, “Massive
terms of end-to-end delay, about 0.16 seconds. On the con- MIMO systems with non-ideal hardware: Energy efficiency, estima-
tion, and capacity limits,” IEEE Trans. Inf. Theory, vol. 60, no. 11,
trary, the approaches based on IEEE 802.15.4 have obtained an pp. 7112–7139, Nov. 2014.
end-to-end delay value of about 0.80 seconds. Finally, the jit- [6] J. M. H. Elmirghani et al., “Greentouch greenmeter core network power
ter parameter has been measured this analysis has been useful consumption models and results,” in Proc. IEEE Online Conf. Green
Commun. (OnlineGreencomm), Tucson, AZ, USA, Nov. 2014, pp. 1–8.
because high jitter values mean that consumers may experi- [7] M. Meo et al., “The TREND experimental activities on ‘green’ com-
ence variable end-to-end delays when they are communicating munication networks,” in Proc. 24th Tyrrhenian Int. Workshop Digit.
with the HEM system. In both wireless protocols the val- Commun. Green ICT (TIWDC), Genoa, Italy, Sep. 2013, pp. 1–6.
ues are low, less than 0.05, which are negligible considering [8] B. Wang, Y. Wu, F. Han, Y.-H. Yang, and K. J. R. Liu, “Green wire-
less communications: A time-reversal paradigm,” IEEE J. Sel. Areas
the human response times. However, even in this case, BLE Commun., vol. 29, no. 8, pp. 1698–1710, Sep. 2011.
obtains better results than IEEE 802.15.4 and the lowest value [9] J. Tang, D. K. C. So, E. Alsusa, K. A. Hamdi, and A. Shojaeifard,
of jitter has been measured with the proposed system based “Resource allocation for energy efficiency optimization in heterogeneous
networks,” IEEE J. Sel. Areas Commun., vol. 33, no. 10, pp. 2104–2117,
on BLE (0.01). Oct. 2015.
The obtained results clearly show that by using the IEEE [10] Y.-K. Hiew, N. M. Aripin, and N. M. Din, “Performance of cognitive
802.15.4 the approach of Keshtkar et al. and the solution smart grid communication in home area network,” in Proc. IEEE 2nd
Int. Symp. Telecommun. Technol. (ISTT), Nov. 2014, pp. 417–422.
introduced in this work achieve poorer network performance [11] P. Samadi, H. Mohsenian-Rad, V. W. S. Wong, and R. Schober,
than the approach proposed in this paper by using BLE. “Tackling the load uncertainty challenges for energy consumption
This analysis, as well as validate the proposed model from scheduling in smart grid,” IEEE Trans. Smart Grid, vol. 4, no. 2,
pp. 1007–1016, Jun. 2013.
the implementation point of view of the wireless network, [12] L. Song, Y. Xiao, and M. van der Schaar, “Demand side management
increases the goodness of the proposed approach, which com- in smart grids using a repeated game framework,” IEEE J. Sel. Areas
pared to the HEM scheme of Keshtkar et al. obtains better Commun., vol. 32, no. 7, pp. 1412–1424, Jul. 2014.
[13] J. Han, C.-S. Choi, W.-K. Park, I. Lee, and S.-H. Kim, “Smart home
performance, as seen in Section IV-A. energy management system including renewable energy based on ZigBee
and PLC,” IEEE Trans. Consum. Electron., vol. 60, no. 2, pp. 198–202,
V. C ONCLUSION May 2014.
[14] S. Chen, N. B. Shroff, and P. Sinha, “Heterogeneous delay tolerant task
In this work an Artificial Neural Network (ANN) for scheduling and energy management in the smart grid with renewable
BluHEMS to deal with the problem of peak load manage- energy,” IEEE J. Sel. Areas Commun., vol. 31, no. 7, pp. 1258–1267,
Jul. 2013.
ment using the available data obtained by the Home Energy [15] S. Tombaz et al., “Is backhaul becoming a bottleneck for green wireless
Management (HEM) system has been presented. The proposed access networks?” in Proc. IEEE Int. Conf. Commun. (ICC), Sydney,
mechanism provides the possibility to forecast the energy con- NSW, Australia, Jun. 2014, pp. 4029–4035.
[16] M. Collotta, “FLBA: A fuzzy algorithm for load balancing in IEEE
sumption conditions, i.e., to predict the home energy require- 802.11 networks,” J. Netw. Comput. Appl., vol. 53, pp. 183–192,
ments at different times of the day or on different days of the Jul. 2015.

Authorized licensed use limited to: VIT University. Downloaded on July 06,2020 at 17:48:50 UTC from IEEE Xplore. Restrictions apply.
120 IEEE TRANSACTIONS ON GREEN COMMUNICATIONS AND NETWORKING, VOL. 1, NO. 1, MARCH 2017

[17] Q. Wang and F. Granelli, “An improved routing algorithm for wireless [39] H.-G. Ma and C.-Z. Han, “Selection of embedding dimension and delay
path selection for the smart grid distribution network,” in Proc. IEEE Int. time in phase space reconstruction,” Front. Elect. Electron. Eng. China,
Energy Conf. (ENERGYCON), Cavtat, Croatia, May 2014, pp. 800–804. vol. 1, no. 1, pp. 111–114, 2006.
[18] Z. Feng, Q. Li, W. Li, T. A. Gulliver, and P. Zhang, “Priority-based [40] A. M. Albano, J. Muench, C. Schwartz, A. I. Mees, and
dynamic spectrum management in a smart grid network environment,” P. E. Rapp, “Singular-value decomposition and the Grassberger–
IEEE J. Sel. Areas Commun., vol. 33, no. 5, pp. 933–945, May 2015. Procaccia algorithm,” Phys. Rev. A, vol. 38, no. 6, pp. 3017–3026,
[19] A. Mahmood, N. Javaid, and S. Razzaq, “A review of wireless com- Sep. 1988.
munications for smart grid,” Renew. Sustain. Energy Rev., vol. 41, [41] A. M. Fraser, “Information and entropy in strange attractors,” IEEE
pp. 248–260, Jan. 2015. Trans. Inf. Theory, vol. 35, no. 2, pp. 245–262, Mar. 1989.
[20] D. G. Costa, M. Collotta, G. Pau, and C. Duran-Faundez, “A fuzzy- [42] T. Buzug and G. Pfister, “Comparison of algorithms calculating opti-
based approach for sensing, coding and transmission configuration of mal embedding parameters for delay time coordinates,” Phys. D Nonlin.
visual sensors in smart city applications,” Sensors, vol. 17, no. 1, p. 93, Phenom., vol. 58, nos. 1–4, pp. 127–137, 1992.
2017. [43] J. Lin, Z. Huang, Y. Wang, and Z. Shen, “Selection of proper embed-
[21] Institute of Electrical and Electronics Engineers, “IEEE standard for ding dimension in phase space reconstruction of speech signals,”
information technology–Telecommunications and information exchange J. Electron. (China), vol. 17, no. 2, pp. 161–169, 2000.
between systems–Local and metropolitan area networksŮ Specific [44] D. Kugiumtzis, “State space reconstruction parameters in the analysis
requirements part 15.4: Wireless medium access control and physi- of chaotic time series—The role of the time window length,” Phys. D
cal layer specifications for low-rate wireless personal area networks,” Nonlin. Phenom., vol. 95, no. 1, pp. 13–28, 1996.
Inst. Elect. Electron. Eng., Tech. Rep., 2006. [Online]. Available: [45] H. S. Kim, R. Eykholt, and J. D. Salas, “Nonlinear dynamics, delay
http://dx.doi.org/10.1109/ieeestd.2006.232110 times, and embedding windows,” Phys. D Nonlin. Phenom., vol. 127,
nos. 1–2, pp. 48–60, 1999.
[22] “IEEE standard for information technology-Telecommunications and
[46] M. Otani and A. J. Jones, “Automated embedding and the creep
information exchange between systems-Local and metropolitan area
phenomenon in chaotic time series,” 2000.
networks-Specific requirements–Part 11: Wireless LAN medium access
[47] K.-H. Kim, J. J. Jeong, S. J. Lee, S. Moon, and S. W. Kim, “Modeling
control and physical layer specifications,” Tech. Rep., Dec. 2007.
of DC electric arc furnace using chaos theory and neural network,”
[23] Bluetooth Specification Version 4.0, Bluetooth SIG, Kirkland, WA, USA, in Proc. 12th Int. Conf. Control Autom. Syst. (ICCAS), Oct. 2012,
2010. pp. 1675–1678.
[24] M. Erol-Kantarci and H. T. Mouftah, “Wireless sensor networks for [48] S. Liyun, L. Ruihua, L. Fenglan, and L. Jiaojun, “Prediction compara-
cost-efficient residential energy management in the smart grid,” IEEE tive study of complex multivariate systems with AGA-BP,” in Proc. Int.
Trans. Smart Grid, vol. 2, no. 2, pp. 314–325, Jun. 2011. Conf. Comput. Appl. Syst. Model. (ICCASM), vol. 10. Taiyuan, China,
[25] S. Ulukus et al., “Energy harvesting wireless communications: A review Oct. 2010, pp. V10-50–V10-54.
of recent advances,” IEEE J. Sel. Areas Commun., vol. 33, no. 3, [49] L. Breslau et al., “Advances in network simulation,” Computer, vol. 33,
pp. 360–381, Mar. 2015. no. 5, pp. 59–67, May 2000.
[26] A. Keshtkar, S. Arzanpour, F. Keshtkar, and P. Ahmadi, “Smart resi-
dential load reduction via fuzzy logic, wireless sensors, and smart grid
incentives,” Energy Build., vol. 104, pp. 165–180, Oct. 2015. Mario Collotta (M’12) received the Ph.D. degree
[27] Q. Zhang, Y. Sun, and Z. Cui, “Application and analysis of ZigBee tech- from Catania University, Italy, in 2011, on the topic
nology for smart grid,” in Proc. Int. Conf. Comput. Inf. Appl. (ICCIA), of factory automation networks. Since 2010, he has
Tianjin, China, Dec. 2010, pp. 171–174. been serving as an Assistant Professor with tenure
[28] L. Li, H. Xiaoguang, C. Ke, and H. Ketai, “The applications of WiFi- with the Faculty of Engineering and Architecture,
based wireless sensor network in Internet of Things and smart grid,” Kore University of Enna, Italy, and in 2011 he
in Proc. 6th IEEE Conf. Ind. Electron. Appl. (ICIEA), Beijing, China, became a Principal Researcher and the Director of
Jun. 2011, pp. 789–793. the Computer Engineering and Network Laboratory.
He has published two book chapters, and over 60 ref-
[29] T. D. P. Mendes, R. Godina, E. M. G. Rodrigues, J. C. O. Matias, and
ereed international journals and conference papers.
J. P. S. Catalão, “Smart home communication technologies and applica-
His research interests concern the realization of
tions: Wireless protocol assessment for home area network resources,”
strategies and innovative algorithms in order to ensure a flexible (the ability of
Energies, vol. 8, no. 7, pp. 7279–7311, 2015.
the system to dynamically adapt to changes in its operating conditions, with-
[30] P. Salvador, V. Mancuso, P. Serrano, F. Gringoli, and A. Banchs, out need for reconfiguration) management of resources (band/CPU or resource
“VoIPiggy: Analysis and implementation of a mechanism to boost capac- allocation and battery consumption) in real-time systems and networks. He
ity in IEEE 802.11 WLANs carrying VoIP traffic,” IEEE Trans. Mobile has served on several committees of distinguished journals and international
Comput., vol. 13, no. 7, pp. 1640–1652, Jul. 2014. IEEE conferences. He has collaborated with several organizing and program
[31] M. Collotta and G. Pau, “A novel energy management approach for committees. He is currently an Associate Editor of some Elsevier and Springer
smart homes using Bluetooth low energy,” IEEE J. Sel. Areas Commun., journals. He has also served as a guest editor and a lead guest editor of several
vol. 33, no. 12, pp. 2988–2996, Dec. 2015. special sections and special issues focused on the study of real-time networks,
[32] M. Collotta and G. Pau, “A solution based on Bluetooth low energy systems, and applications.
for smart home energy management,” Energies, vol. 8, no. 10,
pp. 11916–11938, 2015.
Giovanni Pau (M’17) received the bachelor’s degree
[33] J. M. P. Menezes, Jr., and G. A. Barreto, “Long-term time
in telematic engineering from the University of
series prediction with the NARX network: An empirical evaluation,”
Catania, Italy, and the master’s (cum Laude) degree
Neurocomputing, vol. 71, nos. 16–18, pp. 3335–3343, 2008.
in telematic engineering and the Ph.D. degree from
[34] R. H. Inman, H. T. C. Pedro, and C. F. M. Coimbra, “Solar forecasting the Kore University of Enna, Italy, where he is
methods for renewable energy integration,” Progr. Energy Combust. Sci., a Professor with the Faculty of Engineering and
vol. 39, no. 6, pp. 535–576, 2013. Architecture. He has published over 35 papers
[35] R. Stamminger, “Synergy potential of smart appliances,” A report in journals and conferences and authored a book
prepared as part of the EIE project—Smart domestic appliances in chapter. His research interests include wireless sen-
sustainable energy systems (Smart-A), D2.3 of WP 2, Nov. 2008. sor networks, soft computing techniques, home
[36] M. T. Hagan and M. B. Menhaj, “Training feedforward networks with automation, and real-time systems. He serves as
the Marquardt algorithm,” IEEE Trans. Neural Netw., vol. 5, no. 6, an Associate Editor of the Journal of Network and Computer Applications
pp. 989–993, Nov. 1994. (Elsevier), Wireless Communications and Mobile Computing (Hindawi), and
[37] F. Takens, “Detecting strange attractors in turbulence,” in Dynamical Human-Centric Computing and Information Sciences (Springer). Moreover,
Systems and Turbulence, Warwick 1980 (Lecture Notes in Mathematics), he serves/served as a lead guest editor of several special issues for
vol. 898, D. Rand and L.-S. Young, Eds. Heidelberg, Germany: Springer, Energies (MDPI), Sensors (MDPI), and Wireless Communications and Mobile
1981, pp. 366–381. Computing (Hindawi). He collaborates/collaborated with the organizing and
[38] G. Kember and A. C. Fowler, “A correlation function for choosing time technical program committees of several conferences in order to prepare con-
delays in phase portrait reconstructions,” Phys. Lett. A, vol. 179, no. 2, ference activities and is serving as a reviewer of several international journals
pp. 72–80, 1993. and conferences.

Authorized licensed use limited to: VIT University. Downloaded on July 06,2020 at 17:48:50 UTC from IEEE Xplore. Restrictions apply.