Conference PaperPDF Available

Towards Buildings Energy Management: Using Seasonal Schedules Under Time of Use Pricing Tariff via Deep Neuro-Fuzzy Optimizer

Authors:
  • Institute of Space Technology KICSIT Campus

Abstract and Figures

Management of increasing amount of the electricity information provided by the smart meters is becoming more valuable and a very challenging issue in modern era, especially in residential sector for maintaining the records of consumers' consumption patterns. It becomes the necessity of retailers and utilities to provide the consumers more effective demand response programs for handling the uncertainties of their consumption patterns. In order to deal with the unceratian behaviours of the consumers and their unprecedented high volume of data, this work introduces the deep neuro-fuzzy optimizer for effective load and cost optimization. Three premises parameters: energy consumption, price and time of the day and two consequents parameters: peak reduction and cost are used for the optimization process of the optimizer. The dataset is taken from the Pecan Street Incorporation site and Takagi Sugeno fuzzy inference system is used for the evaluation of the rules developed from the memebership functions of the parameters. Membership Functions (MFs) are choosen as Guassian MFs for continuously monitoring the consumers' behaviours. Performance of this proposed energy optimizer is validated through the simulations which shows the robustness of optimizer in cost optimization and energy efficiency. Index Terms-Energy management, seasonal schedules, time of use price, smart grid, deep neuro-fuzzy optimizer, takagi sugeno fuzzy inference system, residential buildings.
Content may be subject to copyright.
Towards Buildings Energy Management: Using
Seasonal Schedules Under Time of Use Pricing
Tariff via Deep Neuro-Fuzzy Optimizer
Sakeena Javaid1, Muhammad Abdullah1, Nadeem Javaid1,, Tanzeela Sultana1, Jawad Ahmed2, Norin Abdul Sattar3
1COMSATS University Islamabad, Islamabad 44000, Pakistan
2Capital University of Science and Technology, Isamabad, Pakistan
3Pounch University, Rawalakot, Azad Kashmir
Corresponding author: nadeemjavaidqau@gmail.com
Abstract—Management of increasing amount of the electricity
information provided by the smart meters is becoming more
valuable and a very challenging issue in modern era, especially
in residential sector for maintaining the records of consumers’
consumption patterns. It becomes the necessity of retailers and
utilities to provide the consumers more effective demand response
programs for handling the uncertainties of their consumption
patterns. In order to deal with the unceratian behaviours of
the consumers and their unprecedented high volume of data,
this work introduces the deep neuro-fuzzy optimizer for effective
load and cost optimization. Three premises parameters: energy
consumption, price and time of the day and two consequents
parameters: peak and cost reduction are used for the opti-
mization process of the optimizer. The dataset is taken from
the Pecan Street Incorporation site and Takagi Sugeno fuzzy
inference system is used for the evaluation of the rules developed
from the memebership functions of the parameters. Membership
Functions (MFs) are chosen as Guassian MFs for continuously
monitoring the consumers’ behaviours. Performance of this
proposed energy optimizer is validated through the simulations
which shows the robustness of optimizer in cost optimization and
energy efficiency.
Index Terms—Energy management, seasonal schedules, time of
use price, smart grid, deep neuro-fuzzy optimizer, takagi sugeno
fuzzy inference system, residential buildings.
I. INTRODUCTION
Aggregation of fine-grained information regarding electricity
is made possible via the use of smart meters for individual
consumers in any sector [
1
]. This information is very helpful in
extracting the demands of consumers which is used for enhanc-
ing the services, upgrading the energy efficiency programs and
are useful for the improvement of the smart grid functionalities.
Various artificial intelligence methodologies have been pre-
sented for the extraction of the smart meter information which
are also applicable for the following applications: consumers
grouping and load profiling [
2
], energy estimation [
3
],[
4
],[
5
],
various demand response programs [
6
],[
7
], development of new
pricing tariffs [
8
],[
9
], and users’ energy consumption patterns
identification [
10
]. These applications are evaluated through
the supervised or unsupervised learning methodologies, i.e.,
clustering, regression, classification, etc. [11].
The proposed study has got motivation from the optimized
resource allocation for every individual consumer’s consumtion
patterns through the utility supply or the supply from the
renewable energy resources such as wind and solar. All
of these resources have stochastic nature, either they are
from consumption side or from generation side. Various
optimization techniques are already proposed in literature:
Particle Swarm Optimization (PSO), Binary PSO (PSO), fuzzy
logic, game theory, etc. [
12
]-[
18
]. However, these techiques
are not sufficient to handle the large amount of the information
in the real databases. More precisely, these techniques take a
lot of time to manipulate the whole data in order to determine
the optimal solution which is not the optimized way to tackle.
In order to overcome this limitation, variety of deep learning
methodologies are presented to extract the users’ consumption
patterns for their efficient control and management. The list
of these techniques is: auto-encoder, Convolutional Neural
Network (CNN), Recurrent Neural Network (RNN), Deep
Belief Network (DBN), Restricted Boltzman Machines (RBM)
and deep reinforcement learning.
Deep learning provides a lot of emergent techniques for
training deep neural networks through intelligent computation
capabilities [
19
],[
20
]. One solution from the literature is on-line
building energy optimization technique based on reinforcement
learning which is presented in [
1
] for analyzing the 4 years
data (approximately). However, it is based on the binary values
for maintaing its action and reward vectors which cannot
process the minute stochastic occurences in users’ consumption
patterns efficiently. In order to detect the small occurences in
the consumers’ behaviours, we present the deep neuro-fuzzy
logic technique. It uses fuzzy logic to identify the degree of
truth or uncertainty in power consumption patterns. Fuzzy logic
is based on the following steps: fuzzification, rule base, Fuzzy
Inference Systems (FISs) and defuzzification. In fuzzification,
it defines the linguistic variables for defining the degrees of
occurences of certain parameter, i.e., low, medium or high. On
the basis of fuzzification; rule base is developed and then FIS is
used for evaluation of the rules. After the evaluation of the rules,
defuzzification is performed. Defuzzification gives the concrete
solution for the identified problems. On the other hand, neural
978-1-5386-7747-6/19/$31.00 ©2019 IEEE 1594
network provides efficient training for optimization problems.
In our work, we are proposing the deep neuro-fuzzy optimizer
to handle the large amount of data (average hourly data of 5 five
years: 2012 to 2016) regarding electricity consumption patterns
in residential area. Upto the best of our knowledge, no such
similar work exists in this domain. Our major contributions
are enlisted below.
A. List of Contributions
There are following contributions which are described below:
We have proposed a deep neuro-fuzzy optimizer for energy
management in the residential buildings.
For the efficient training and testing of the system, we have
used backpropogation algorithm with the deep nuero-fuzzy
optimizer for efficient identification of the consumers’
behaviours during each interval of the day.
The deep neuro-fuzzy optimizer is integrated with the
Takagi Sugeno FIS which is used for evaluation of the
rule base having 8 membership functions and 15 rules for
maintaining the energy consumption and cost of users’
schedules. Dataset for two seasons: summer and winter
is considered under Time of Use (TOU) pricing scheme.
This work is validated through the simulations which
proves that the proposed neuro-fuzzy optimizer gives
optimal cost and peak reduction.
II. LITERATURE REVIEW
Dariush et al. in [
21
] has presented a new intelligent EMS
(iEMS) for a smart home which is based on two subsystems:
a fuzzy subsystem and an intelligent lookup table. The first
subsystem uses fuzzy rules and inputs which generates the
feasible results for the intelligent lookup table, whereas, second
subsystem is used for mapping the inputs to desired outputs
using the associated neural network architecture. An intelligent
lookup table considers three inputs which are derived from
the fuzzy subsystem, outside sensors, and feedback outputs.
This system considers that whatever trained in lookup table is
diverse according to the scenario and this system is sufficient in
determining the best energy-efficient solution in all scenarios.
In [
22
], another neuro-fuzzy approach is developed to predict
the energy utilization in the buildings by considering the
buildings’ physical properties, i.e., thickness and insulation
K-value. They have conducted more than 180 simulations
by considering several thickness and inculation values using
the EnergyPlus simulator. The Non-Intrusive Load Monitoring
(NILM) system is presented using the hybrid technique for
classification [
23
]. This hybrid technique is based on the
combination of Fuzzy C-means clustering-piloting PSO with
neuro-fuzzy classification. The household appliances can be
detected via electrical signatures in real scenario. Whereas,
certain anomalies are still found in case of load recongnition
which are resolved using the fuzzy logic. The proposed system
is validated in the real laboratory and home environments by
considering various uncertainties. In [
24
], a review of the fuzzy
logic controllers regarding the power saving services in smart
buildings of Australia is presented. In addition, authors have
also mentioned the drawbacks, conceptual foundations and
capabilities.
A new bluetooth based low energy oriented scheme is presented
to identify the sleeping duration of the devices in any home
area [
25
]. This scheme works on the battery level and their
throughput. In this work, authors have considered fixed sleeping
duration of each appliance and it saves their time upto
30%. In addition, three energy management and controlling
techniques are introduced for household devices in [
26
]. Home
EMSs (HEMSs) track and schedule three types of devices:
1) heating, 2) storage, and 3) energy storage devices in
the considered scenario. Mixed Integer Linear Programming
(MILP), Continuous Relaxation (CR), and Fuzzy Logic are
used for the energy management along with three types of
approaches: heat-related fuzzy logic controller, task-related
fuzzy logic controller and fuzzy logic controller for battery.
Each technique is investigated for optimization of the cost,
computing resources and pratical scenarios. Fuzzy logic is used
for improving the consumers’ comfort standards by including
the humidity as an input from the consumers’ environment
[
27
]. They have also used the room temperature as a feedback
for the FIS in order to improve the energy efficiency. Authors
of this work claim that manual increment in defining the rules
generates more error as compared to automatic rule generation.
For minimizing this error, they have proposed a new automatic
way of defining the rule base via the combinatorial method. This
work is based on Mamdani and Sugeno FISs for inferrring
the rules from the rule base. The proposed technique has
maintained the thermal comfort by facilitating the consumers
with the flexibility and efficiency in decision making.
A worldwide adoptive thermostat scheme is presented for
controlling the temperature setpoints of the residential area
using fuzzy logic [
28
]. Setpoint optimization has been done
for both the hot and cold regions of the world. Authors have
validated their approach using the datasets of Russia and Al-
Azizia cities. This scheme reduces the energy consumption upto
18% and 35% using two FISs: fuzzy mamdani inference system
and fuzzy sugeno inference system. However, this scheme
compromises the user comfort during peak hours because it
applies load curtailment during peak hours. A comprehensive
review of the standalone hybrid and MG-oriented hybrid
approaches regarding energy management is discussed in [
29
].
Authors in [
30
] have considered flexible power requirement
and generation coordination issue. In [
31
], reinforcement
learning is used for the dynamic price management and
load scheduling. Using re-inforcement learning, consumers
learn without apriori knowledge and it helps in cost saving
through learning. Building energy prediction has performed
by Conditional Restricted Boltzmann Machine (CRBM) and
Factored Conditional Restricted Boltzmann Machine (FCRBM)
in [
32
] which is tested on the benchmark dataset of four
years. Performance of these schemes shows that they have
outperformed the previous schemes: ANN, Support Vector
Machine (SVM), and RNNs. In addition, authors in [
33
]
have presented the long term load forscasting using the
neural network and fuzzy logic with multilayer perceptron
1595
for obtaining the better training results.
III. SYS TE M MOD EL
The proposed system model is elaborated via the set of the
constraints and proposed energy optimizer descriptions. First,
system constraints are elaborated and then proposed optimizer
is described in detail.
A. System Constraints
The proposed learning algorithm has considered the follow-
ing mandatory constraints in order to guarantee the character-
istics of this deep neuro-fuzzy system:
Fuzzy sets for the defined membership functions are
considered to stay in normalized form.
Fuzzy sets for membership functions have not assumed
to exchange their defined boundaries.
The defined fuzzy sets (or membership values) are
considered to overlap each other for every interval.
Summation of the membership degrees are considered
equal to 1.
B. Proposed Deep Neuro-Fuzzy Optimizer
In this system, we have considered the set of residential
buildings comprising of several consumers. The energy con-
sumtpion patterns of the consumers are recorded through a
dataset available on Pecan Street Incorporation site [
1
] for the
cost and peak reduction optimization during the 24 hours a
day. Deep neuro-fuzzy is the hybrid of deep neural network
and fuzzy logic. For the efficient optimization of the cost and
peak reduction, we have applied the deep neuro-fuzzy optimier.
It is a hybrid approach where deep neural network has been
applied first and then fuzzy logic is applied at second step
for computing the system objectives. The whole system is
comprised of following layers: input layer, number of hidden
layers for learning and validation and output layer. Input layer is
based on number of the inputs parameters and their fuzzification
values in a system [
34
]. Number of hidden layers are three:
rule layer, normalization and defuzzification layer. Output layer
is defuzzification layer. The most significant parameters of this
system are: consequents and premises of the system. Here,
the number of the premise parameters include the following:
pricing tariff, load and time of the day, whereas, number of
the consequent parameters are: cost and peak reduction as
shown in Fig. 1.In this system, premises are foundations for
the membership functions in the fuzzification at input layer
which define the levels of the occurrences in the consumers’
behaviours or energy consumption patterns. The consequent
parameters are related to the defuzzification process. This
work addresses two major objectives: cost minimization and
peak reduction. The equations for the computing the objective
functions are taken from [
1
]. Neuro-fuzzy contains the takagi
sugeno FIS for rule base evaluation where we have used the
grid partitioning method for FIS rule generation purpose. This
is the default method available in the neuro-fuzzy system
which we have integrated with deep neuro-fuzzy optimizer.
The mathematical description of these parameters is described
below.
Each input or output parameter is mapped to the specific
entity or node (information processing unit) in the neuro-fuzzy
network for each layer. Degree to each input is assigned
between 0 and 1 as per criterion defined by the fuzzy
system. Each entity in the first layer is followed by an output
value. Assume that there are two premises:
u
and
v
and one
consequent:
w
, then these are represented in the next equation:
L1,j =µXj(u)or L1,j =µYj2(v),j= 1,2,3,4.(1)
In Eq. 1,
u
and
v
denote the inputs to each
jth
entity,
µX
and
µYj2
depicts the antecendent membership functions, whereas,
L1,j
represents the degree of membership. The membership
functions are represented by the bell shaped funtions (known
as guassian membership functions) which are assigned with
the maximum ”1” and minimum ”0” values.
µXj(u) = 1
1 + |usj
tj|2pi
(2)
Here,
pj
,
sj
, and
tj
shows the membership functions of the
premise parameters which are optimized through training.
Layer 2 is known as rule base layer which is used for describing
the set of rules. Every entity in this layer multiply the linguistic
veriables’ values to satisfy the degree of memberhship. The
product of membership variables values shows the firing
strength of the rule as described by the equation below.
L2,j =ωj=µXj(u)µYj2(v),j= 1,2.(3)
Layer 3 deals with normalization where every entity com-
putes the ratio of the firing strength of
jth
rule with the
summation of the firing strength of all rules.
ωj
represents
the generic network parameter: weight. The result of every
rule is then normalized via firing strength of the rule which is
described as under.
L3,j = ¯ωj=ωj
ω1+ω2,j= 1,2.(4)
Fourth layer is the defuzzification layer where each rule’s
consequents are computed to represent their overall effect on
the output. This phenomenon is mathematically described by
the equation below.
L4,j = ¯ωjzi= ¯ωj(aju+bjv+cj),j= 1,2.(5)
Here,
a
,
b
, and
c
are the consequent parameters set. Af-
terwards, last layer is considered as summation layer which
computes the summation of the previous layers outcomes. Next
equation describes the process of the final result computation.
L5,j =X
j
¯ωjzj=Pjωjzj
Pjωj
(6)
For evaluating the operation of these layers, Takagi sugeno
FIS is used; as mentioned above, for inferring the number of
1596
Fig. 1: System Model for the Optimal Energy Consumption.
rules defined by the IF-THEN statements using the linguistic
membership variables of the antencendents and consequents.
These parameters are initially assigned with the random values,
then they are tunned through the training algorithm for best
values optimization. In this case, backpropogation algorithm
is used for training of the system parameters. In this system,
number of hidden layers are used for the making the system
training efficient for the large set of data. Three types of
appliances are considered in this work: time scaling, time
shifting and time shifting and scaling appliances. Time scaling
appliances are air conditioners, which cloud be switched-on
or off for any time interval. Secondly, we consider the time
shifting appliances, like dishwashers. These appliances are
deferrable appliances and can be shifted to the other time
intervals. Thirdly, we consider the EVs as time shifting and
scaling appliances of the building.
IV. SIMULATION RESULTS
In this section, description of the dataset is explained first
and then simulation results of the proposed optimizer with
respect to price and peak load reduction are discussed. Two
scenarios are discussed in the simulations: peak load and cost
reduction for summer season and peak load and cost reduction
for winter season.
A. Dataset and System Parameters Description
We have used the dataset of 5 years (2012 to 2016) regarding
energy consumption of residential buildings of city Austin
for conducting the simulations of proposed optimizer. For
determining the patterns of the residential users, we have
collected the average hourly data of each year. This data is
recorded from the Pecan Street Incopration site [
1
]. Firstly,
75% of dataset is categorised for the training of the network.
Secondly, 25% of the dataset is used for the training of the
network. Other system parameters are taken from [
35
]. Pricing
tariff used in this work is TOU for the residential area which
is developed according to the consumers’ living patterns in
the city of Austin. Price tariff is desinged for both summer
and winter seasons, where summer tariff includes the on-peak,
mid-peak and off-peak tariffs, and winter tariff is only having
the off-peak and mid-peak tariff [1].
B. Peak Load and Cost Reduction for Summer Season
The total energy consumption and reduction is shown in
Fig. 2(a) using our proposed scheme, where data is collected
for 24 hours and it is categorised for testing and training as
mentioned above. Initial 18 hours data is used for training
and the remaining six hours data is used for testing. The
training phase of the proposed scheme is shown for initial
18 samples which shows the total observed consumption and
reduction patterns. During the training phase, maximum 8
kWh conumsption is observed during any hour of the day,
whereas, 0.3 kWh reduction is observed per hour. The reason
is that the use of ANN tranining and evaluation of fuzzy
rules through takagi sugeno FIS helps in efficient training and
evaluation. Mean Square Error (MSE), Root Mean Square Error
(RMSE) are introduced during the cost computation which
shows robustness of the system by considering the uncertainties
as shown in Fig. 2(b). The maximum cost procured in any
hour during training phase is equal to 1$. Takagi sugeno FIS
uses guassian membership functions which uses the mean and
standard deviation as inputs for computing the number of rules
based on these parameters and are displayed in Fig. 2(c).
After training, testing is performed for the remaining 25%
of the selected dataset considering the remaining 6 hours of the
1597
Fig. 2: Training Results for Total Energy Utilization and
Reduction.
day. Fig. 3(a) shows the testing results for the remaining 6 hours
which shows that the maximum 1.6 kWh energy is consumed
per hour and 0.2 kWh energy is reduced per hour. Fig. 3(b)
shows the cost considering the MSE and RMSE which shows
the robustness of sytem in uncertainity-oriented environment.
During the testing phase, the obtained cost is shown below
0.15$ which proves that system has been trained intelligently.
Fig. 3(c) shows the participation of membership fucntions using
the mean and standard deviation of the membership functions
as an input.
Fig. 3: Testing Results for Total Energy Utilization and
Reduction.
C. Peak and Cost Reduction for Winter Season
In this section, the proposed optimizer is tested using the
winter TOU pricing tariff. For training phase, again dataset is
categorised for 75% of the total dataset (similar to the summer
season scenario). In this phase, initial 18 hours of the day are
considered where maximum 10 kWh consumption is observed
during any hour of the day. Maximum 4 kWh peak reduction
is observed because appliances used in winter season consume
more load as compared to the appliances used in summer season
[
36
]. These appliances are having high power rating and if
there are load reduction strategies applied, more reduction is
also observed. Fig. 4(a) shows the training results for energy
consumption and reduction results, whereas, Fig. 4(b) and Fig.
4(c) show the cost and number of rules’ participation of the
system. Cost is more high as maximum 4 kWh is observed
during any hour because of the heavy load appliances used in
the winter season.
Fig. 4: Training Data Results for Total Energy Utilization and
Reduction.
After training phase, testing is performed for the last six
hours of the day. In Fig. 5(a), maximum 3.5 kWh consumption
is observed and there is no reduction observed there. Surplus
power is consumed in the night hours because residents are
staying at home and heating systems are on most of the time
while consumers stay at home. However, they mostly shift their
heating systems to the storage devices or renewable energy
systems. At night time, prices are normally low because of the
off-peak hours. So, the cost obtained is very low as displayed
in Fig. 5(b). Fig. 5(c) shows the number of rules used in testing
phase for cost computation, energy consumption and energy
reduction processes.
Fig. 5: Tested Data Results for Total Energy Utilization and
Reduction.
TABLE I: Comparison of Performance Parameters.
Season
Number
of Rules
Cost
Energy
Consumption
Peak Reduction
Summer 15 1.0$ 8.0kWh 0.3kWh
Winter 10 4.0$ 12.5kWh 4.0kWh
V. CONCLUSION
In this work, a deep neuro-fuzzy optimizer has been proposed
for the efficient optimization of the cost and energy for two
seasons using the large dataset (2012 to 2016). This optimizer
1598
combines the functionalities of fuzzy logic and deep neural
network. Fuzzy logic deals with the handling of uncertainties
of data, whereas, neural network helps in efficient training by
tuning the FIS parameters and computation of large scale data.
This optimizer has been used for the testing and training data of
5 years. Guassian membership functions are used for efficient
monitoring of the system’s states for every hour. From the
simulation results, it provides efficiency in energy reduction for
the consumers using the TOU tariffs’ rates for both summer and
winter seasons. MSE and RMSE metrices have been introduced
for computing the cost of the system in current scenarios
for proving the system robustness. It is concluded from the
simulation results that using the same set of rules for the same
parameters’ set do not improve the performance; however, by
altering the number of epochs and error tolerance, system
performance can be improved.
-REFERENCES
[1]
Mocanu, E., Mocanu, D. C., Nguyen, P. H., Liotta, A., Webber, M.
E., Gibescu, M., and Slootweg, J. G. (2018). On-line building energy
optimization using deep reinforcement learning. IEEE Transactions on
Smart Grid. (Accepted).
[2]
Wang, Y., Chen, Q., Kang, C., Zhang, M., Wang, K., and Zhao, Y.
(2015). Load profiling and its application to demand response: A review.
Tsinghua Science and Technology, 20(2), 117-129.
[3]
Li, R., Li, F., and Smith, N. D. (2017). Load characterization and low-
order approximation for smart metering data in the spectral domain.
IEEE Transactions on Industrial Informatics, 13(3), 976-984.
[4]
Taieb, S. B., Huser, R., Hyndman, R. J., and Genton, M. G. (2016).
Forecasting uncertainty in electricity smart meter data by boosting
additive quantile regression. IEEE Transactions on Smart Grid, 7(5),
2448-2455.
[5]
Zhang, P., Wu, X., Wang, X., and Bi, S. (2015). Short-term load
forecasting based on big data technologies. CSEE Journal of Power
and Energy Systems, 1(3), 59-67.
[6]
Hsiao, Y. H. (2015). Household electricity demand forecast based on
context information and user daily schedule analysis from meter data.
IEEE Transactions on Industrial Informatics, 11(1), 33-43.
[7]
Sijie, C. H. E. N., and Chen-Ching, L. I. U. (2017). From demand
response to transactive energy: state of the art. Journal of Modern Power
Systems and Clean Energy, 5(1), 10-19.
[8]
Wang, Y., Chen, Q., Kang, C., and Xia, Q. (2016). Clustering of electricity
consumption behavior dynamics toward big data applications. IEEE
transactions on smart grid, 7(5), 2437-2447.
[9]
Li, R., Wang, Z., Gu, C., Li, F., and Wu, H. (2016). A novel time-of-use
tariff design based on Gaussian Mixture Model. Applied energy, 162,
1530-1536.
[10]
Chen, S., Love, H. A., and Liu, C. C. (2016). Optimal opt-in residential
time-of-use contract based on principal-agent theory. IEEE Transactions
on Power Systems, 31(6), 4415-4426.
[11] Chicco, G. (2016, October). Customer behaviour and data analytics. In
2016 International Conference and Exposition on Electrical and Power
Engineering (EPE) (pp. 771-779). IEEE.
[12]
Alam, M. R., St-Hilaire, M., and Kunz, T. (2016). Computational methods
for residential energy cost optimization in smart grids: A survey. ACM
Computing Surveys (CSUR), 49(1), 1-22.
[13]
Barbato, A., and Capone, A. (2014). Optimization models and methods
for demand-side management of residential users: A survey. Energies,
7(9), 5787-5824.
[14]
Loukarakis, E., Dent, C. J., and Bialek, J. W. (2016). Decentralized multi-
period economic dispatch for real-time flexible demand management.
IEEE Transactions on Power Systems, 31(1), 672-684.
[15]
Mohsenian-Rad, A. H., and Leon-Garcia, A. (2010). Optimal residen-
tial load control with price prediction in real-time electricity pricing
environments. IEEE Trans. Smart Grid, 1(2), 120-133.
[16]
Paterakis, N. G., Erdin
c¸
, O., Pappi, I. N., Bakirtzis, A. G., and Catalao, J.
P. (2016). Coordinated operation of a neighborhood of smart households
comprising electric vehicles, energy storage and distributed generation.
IEEE Transactions on smart grid, 7(6), 2736-2747.
[17]
Vardakas, J. S., Zorba, N., and Verikoukis, C. V. (2015). A survey
on demand response programs in smart grids: Pricing methods and
optimization algorithms. IEEE Communications Surveys & Tutorials,
17(1), 152-178.
[18]
Hurtado, L. A., Mocanu, E., Nguyen, P. H., Gibescu, M., and Kling, W.
L. (2015, May). Comfort-constrained demand flexibility management
for building aggregations using a decentralized approach. In 2015
International Conference on Smart Cities and Green ICT Systems
(SMARTGREENS) (pp. 1-10). IEEE.
[19]
Mocanu, E., Nguyen, P. H., Gibescu, M., Larsen, E. M., and Pinson, P.
(2016, June). Demand forecasting at low aggregation levels using factored
conditional restricted boltzmann machine. In 2016 Power Systems
Computation Conference (PSCC) (pp. 1-7). IEEE.
[20]
Ryu, S., Noh, J., and Kim, H. (2016). Deep neural network based demand
side short term load forecasting. Energies, 10(1), 1-20.
[21]
Shahgoshtasbi, D., and Jamshidi, M. M. (2014). A new intelligent
neuro–fuzzy paradigm for energy-efficient homes. IEEE Systems Journal,
8(2), 664-673.
[22]
Naji, S., Shamshirband, S., Basser, H., Keivani, A., Alengaram, U.
J., Jumaat, M. Z., and Petkovi
´
c, D. (2016). Application of adaptive
neuro-fuzzy methodology for estimating building energy consumption.
Renewable and Sustainable Energy Reviews, 53, 1520-1528.
[23]
Lin, Y. H., and Tsai, M. S. (2014). Non-intrusive load monitoring by novel
neuro-fuzzy classification considering uncertainties. IEEE Transactions
on Smart Grid, 5(5), 2376-2384.
[24]
Ghadi, Y. Y., Rasul, M. G., and Khan, M. M. K. (2016). Design and
development of advanced fuzzy logic controllers in smart buildings
for institutional buildings in subtropical Queensland. Renewable and
Sustainable Energy Reviews, 54, 738-744.
[25]
Collotta, M., and Pau, G. (2015). Bluetooth for Internet of Things: A fuzzy
approach to improve power management in smart homes. Computers &
Electrical Engineering, 44, 137-152.
[26]
Wui, Z., Zhangi, X. P., Brandt, J., Zhoui, S. Y., and Li J. N. (2015). Three
Control Approaches for Optimized Energy Flow With Home Energy
Management System. IEEE Power and Energy Technology Systems
Journal, 2(1), 21-31.
[27]
Ain, Q. U., Iqbal, S., Khan, S. A., Malik, A. W., Ahmad, I. and Javaid,
N. (2018). IoT Operating System Based Fuzzy Inference System for
Home Energy Management System in Smart Buildings. Sensors, 18(9),
1-30.
[28]
Javaid, S., Javaid, N., Iqbal, S., Guizani, M., Almogren, A., and Alamri,
A. (2018). Energy Management with a World-wide Adaptive Thermostat
using Fuzzy Inference System. IEEE Access, 33489-33502.
[29]
Olatomiwa, L., Mekhilef, S., Ismail, M.S., Moghavvemi, M. (2016).
Energy management strategies in hybrid renewable energy systems: A
review. Renewable and Sustainable Energy Reviews 62, 821–835.
[30]
Paterakis, N. G., Erdin
c¸
, O., Pappi, I. N., Bakirtzis, A. G., and Catal
˜
ao, J.
P. (2016). Coordinated operation of a neighborhood of smart households
comprising electric vehicles, energy storage and distributed generation.
IEEE Transactions on smart grid, 7(6), 2736-2747.
[31]
Kim, B. G., Zhang, Y., Van Der Schaar, M., and Lee, J. W. (2016).
Dynamic pricing and energy consumption scheduling with reinforcement
learning. IEEE Transactions on Smart Grid, 7(5), 2187-2198.
[32]
Mocanu, E., Nguyen, P. H., Gibescu, M., and Kling, W. L. (2016). Deep
learning for estimating building energy consumption. Sustainable Energy,
Grids and Networks, 6, 91-99.
[33]
Miranda, S. T. d., Abaide, A., Sperandio, M., Santos, M. M., Zanghi,
E. (2018). Application of artificial neural networks and fuzzy logic to
long-term load forecast considering the price elasticity of electricity
demand. Wiley, 1-17.
[34]
Semero, Y. K., Zhang, J., and Zheng, D. (2018). PV power forecasting
using an integrated GA-PSO-ANFIS approach and Gaussian process
regression based feature selection strategy. CSEE Journal of Power and
Energy Systems, 4(2), 210-218.
[35]
Azar, A. T. (2011). Adaptive Neuro-Fuzzy Systems. Electrical Communi-
cation & Electronics Systems Engineering department, Modern Science
and Arts University (MSA), City, Egypt.
[36]
Michael Sivak. Hot cities more sustainable than
cold ones, study says, (2013). Available online:
http://science.nbcnews.com/news/2013/03/27/17490775-hot-cities-
more-sustainable-than-cold-ones-study-says?lite,(accessed on 25 Jan.
2019).
1599
Article
Full-text available
The intrusion detection system identifies the attack through the reputation and progression of network methodology and the Internet. Moreover, conventional intrusion recognition techniques usually utilize mining association rules for identifying intrusion behaviors. However, the intrusion detection model failed to extract typical information of user behaviors completely and experienced several issues, including poor generalization capability, high False Alarm Rate (FAR), and poor timeliness. This paper uses a hybrid optimization-based Deep learning technique for the multi-level intrusion detection process. First, the fisher score scheme is applied to extract the important features. Then, in the data augmentation the data size is increased. In this model, Rider Optimization Algorithm-Based Neural Network (RideNN) is employed for first level detection, where the data is categorized as normal and attacker. Besides, the RideNN classifier is trained by devised Rider Social Optimization Algorithm (RideSOA). Additionally, the Deep Neuro Fuzzy network (DNFN) is utilized for the second level classification process in which attack types are categorized. Besides, the DNFN classifier is trained through devised Social Squirrel Search Algorithm (SSSA). The introduced intrusion detection algorithm outperformed with maximum precision of 0.9254, recall of 0.8362, and F-measure 0.8718.
Article
Cardiovascular disease (CVD) represents an emerging death reason worldwide. CVD is based on the capability to discover the high-risk individuals before designing overt events. An effective technique for CVD risk prediction is developed using retinal fundus images. Initially, the retinal fundus images are subjected to pre-processing using grayscale conversion. The optic disc is detected with binarization and circle fixing. Then, the blood vessel segmentation uses deep joint segmentation, wherein dice coefficient and binary cross-entropy are integrated. After that, the feature extraction is done for mining convenient features that include several statistical features. Meanwhile, features like Local Directional Texture Pattern (LDTP) and Local Gabor Binary Pattern (LGBP) are mined from the inputted image. Then, the cardiovascular risk prediction is made by a Deep neuro-fuzzy network (DNFN), such that the risks are classified into normal and hypertensive. Finally, the DNFN is trained using the developed Fractional Calculus-Horse Herd Optimization Algorithm (FCHOA), which is devised by combining Fractional Calculus (FC) and the Horse Herd Optimization algorithm (HOA). The proposed FCHOA-based DNFN offered enhanced efficiency with the highest accuracy, sensitivity and specificity of 91.6, 92.3 and 91.9%.
Article
The earlier analysis of plant diseases has the prospective to control the growth of diseases on several crops with the focus of improving the production and superiority of crops. Various techniques are devised for crop disease classification, but the automatic classification of disease is a key constraint in agricultural production. Therefore, an effective optimization enabled deep learning framework, called Exponential Rider Henry Gas Solubility Optimization (ExpRHGSO) algorithm is introduced for rice leaf disease detection and classification. Here, the rice leaf disease detection is performed using Deep Neuro-Fuzzy Network (DNFN) where healthy and unhealthy plants are detected. However, DNFN is trained by the devised ExpRHGSO, which is the incorporation of Exponentially Weighted Moving Average (EWMA) with Rider Henry Gas Solubility Optimization (RHGSO). Meanwhile, RHGSO is the hybridization of Rider Optimization Algorithm (ROA) and Henry gas solubility optimization (HGSO). Finally, rice leaf disease classification is done using Deep Residual Network (DRN) to classify the rice leaf diseases into Bacterial Leaf Blight (BLB), Blast, and Brown spot. The developed ExpRHGSO-based Hybrid Deep Learning method achieved superior testing accuracy, sensitivity, and specificity with the maximum values of 0.916, 0.923, and 0.919.
Article
Full-text available
Fog computing is a framework, which expands services of the cloud to network for addressing inherent issues of the cloud. Intrusion Detection Systems (IDSs) represent a major unit of the security model for fog networks for facilitating service quality. This paper devises a hybrid optimization-driven ensemble classifier in a fog computing platform. In fog computing, the complete processing is done utilizing trio layers that include the cloud layer, end point layer, and fog layer. In the cloud layer, three procedures, like data transformation, selection of features, and classification processes were carried out. A data transformation is conducted with log transformation. A feature is chosen using Kolmogorov–Smirnov correlation-based filter. Then, the classification is done utilizing ensemble classifiers, named RideNN, Deep Neuro–Fuzzy Network (DNFN), and Shepard convolutional neural network (ShCNN). The tuning of the ensemble classifier is conducted utilizing the developed Rider Sea Lion Optimization (RSLO) algorithm. RSLO algorithm is conceived by amalgamating the Rider optimization algorithm (ROA) and Sea Lion Optimization Algorithm (SLnOA). At an end point layer, a physical process is done. In the fog layer, intrusion detection is done based on a trained ensemble classifier. The proposed RSLO-based ensemble approach provides enhanced performance with higher precision of 88%, recall of 88.4%, F-measure of 88.2%, and accuracy of 97.2%.
Article
The discovery of voice pathology is an emerging domain in biomedical engineering. The assessment helps to offer information for effective diagnosis. An optimization-enabled deep model for pathological disorder classification is devised in this paper. The speech signal is employed as the input of the pre-processor which subsequently helps to eliminate noise considering the Hanning window. The initial improvement is done by exploiting multi-band spectral subtraction for generating frames. The enhancement of speech is performed using Deep Residual Network (DRN) classifier. The training of DRN is performed by exploiting adopted Feedback Artificial Tree Firefly algorithm (FATFA) that is attained by combining Feedback Artificial Tree Algorithm (FAT), and Firefly algorithm (FA). The features, like Amplitude modulation spectrogram (AMS), and frequency-domain features namely tonal power ratio, pitch chroma, spectral spread, spectral flux, as well as spectral autocorrelation is mined from the improved speech. The classification is done with Deep Neuro-Fuzzy Network (DNFN), which is tuned with the devised Verse FATFA. The proposed Verse FATFA is conceived by merging FATFA and Multiverse Optimizer (MVO). Therefore, the devised Verse FATFA-based DNFN categories voice disorders into Laryngitis, Dysphonie, spasmodischedysphonie, Leukoplakie, Cyste, Dish-Syndrom, and Monochorditis. The proposed Verse FATFA-based DNFN provides enhanced performance with maximum values of 94.93% accuracy, 95.94% sensitivity with 91.93% specificity.
Article
Glaucoma represents dangerous ailment, which affected the nervous model and causes loss of vision. Several researchers developed automated discovery of glaucoma, but redundancy elimination is still challenging. Hence, this research study introduces an effective method for detecting glaucoma with deep neurofuzzy network (DNFN). Initially, the retinal image is input for preprocessing to remove the noises. Then, the optic disc (OD) detection and blood vessel segmentation are employed using the blackhole entropy fuzzy clustering algorithm and the DeepJoint model, respectively. Finally, the obtained OD and blood vessels are fed to the DNFN, wherein DNFN training is performed with newly devised MultiVerse Rider Wave Optimization (MVRWO). The newly developed MVRWO integrates the Water Wave Optimization, Rider Optimization Algorithm, and MultiVerse Optimizer. Finally, the output is classified based on the loss function of the DNFN. The developed MVRWO–DNFN obtained an elevated accuracy of 92.214%, a sensitivity of 93.422%, and a specificity of 92.34%.
Article
Full-text available
The rising rates of costs in labor and growth of computerization in fabric industries made defect detection in fabric a promising domain. For a huge time, manual discovery is extensively utilized in textile industries by trained staff that results in high cost. Meanwhile, the strict quality assessment is done by modern textile industries, which made automatic fabric defect detection a reliable choice. Since defect detection is an important and challenging aspect of modern industrial manufacturing, it is necessary to determine the quality and acceptability of garments and to reduce the cost and time waste caused by defects. Different methods are in practice for effective detection of fabric defects, however, they limit due to many reasons. Thus we proposed a new method named Competitive Cat Swarm Optimizer (CCSO) based Deep neuro-fuzzy network (DNFN) for effective fabric detection. Here, the pre-processing is performed with a median filter for eliminating noise contained in the image. Furthermore, features are extracted that involves Tetrolet transform-based features, statistical features, like energy, entropy, homogeneity, contrast, correlation, and texture feature, like Local gradient pattern. The data augmentation is carried out based on obtained features to make it apposite for processing. The defect detection is carried out using RideNN and DNFN. Here, the training of DNFN is done using the proposed CCSO, which is devised by combining Cat Swarm Optimization and Competitive Swarm Optimizer. The correlation is performed on the outputs of RideNN and DNFN for generating the final output of fabric defect detection. The performance of the proposed CCSO-based DNFN is compared with the different existing methods and the proposed CCSO-based DNFN outperforms with the highest specificity of 0.920, the accuracy of 0.919, and sensitivity of 0.916.
Article
Full-text available
Energy consumption in the residential sector is 25% of all the sectors. The advent of smart appliances and intelligent sensors have increased the realization of home energy management systems. Acquiring balance between energy consumption and user comfort is in the spotlight when the performance of the smart home is evaluated. Appliances of heating, ventilation and air conditioning constitute up to 64% of energy consumption in residential buildings. A number of research works have shown that fuzzy logic system integrated with other techniques is used with the main objective of energy consumption minimization. However, user comfort is often sacrificed in these techniques. In this paper, we have proposed a Fuzzy Inference System (FIS) that uses humidity as an additional input parameter in order to maintain the thermostat set-points according to user comfort. Additionally, we have used indoor room temperature variation as a feedback to proposed FIS in order to get the better energy consumption. As the number of rules increase, the task of defining them in FIS becomes time consuming and eventually increases the chance of manual errors. We have also proposed the automatic rule base generation using the combinatorial method. The proposed techniques are evaluated using Mamdani FIS and Sugeno FIS. The proposed method provides a flexible and energy efficient decision-making system that maintains the user thermal comfort with the help of intelligent sensors. The proposed FIS system requires less memory and low processing power along with the use of sensors, making it possible to be used in the IoT operating system e.g., RIOT. Simulation results validate that the proposed technique reduces energy consumption by 28%.
Article
Full-text available
This paper presents a hybrid approach for the forecasting of electricity production in microgrids with solar photovoltaic (PV) installations. An accurate PV power generation forecasting tool essentially addresses the issues resulting from the intermittent and uncertain nature of solar power to ensure efficient and reliable system operation. A day-ahead, hourly mean PV power generation forecasting method based on a combination of genetic algorithm (GA), particle swarm optimization (PSO) and adaptive neuro-fuzzy inference systems (ANFIS) is presented in this study. Binary GA with Gaussian process regression model based fitness function is used to determine important input parameters that significantly influence the amount of output power of a PV generation plant; and an integrated hybrid algorithm combining GA and PSO is used to optimize an ANFIS based PV power forecasting model for the plant. The proposed modeling technique is tested based on power generation data obtained from Goldwind microgrid system found in Beijing. Forecasting results demonstrate the superior performance of the proposed method as compared with commonly used forecasting approaches. The proposed approach outperformed existing artificial neural network (ANN), linear regression (LR), and persistence based forecasting models, validating its effectiveness.
Article
Full-text available
In the smart grid, one of the most important research areas is load forecasting; it spans from traditional time series analyses to recent machine learning approaches and mostly focuses on forecasting aggregated electricity consumption. However, the importance of demand side energy management, including individual load forecasting, is becoming critical. In this paper, we propose deep neural network (DNN)-based load forecasting models and apply them to a demand side empirical load database. DNNs are trained in two different ways: a pre-training restricted Boltzmann machine and using the rectified linear unit without pre-training. DNN forecasting models are trained by individual customer's electricity consumption data and regional meteorological elements. To verify the performance of DNNs, forecasting results are compared with a shallow neural network (SNN), a double seasonal Holt-Winters (DSHW) model and the autoregressive integrated moving average (ARIMA). The mean absolute percentage error (MAPE) and relative root mean square error (RRMSE) are used for verification. Our results show that DNNs exhibit accurate and robust predictions compared to other forecasting models, e.g., MAPE and RRMSE are reduced by up to 17% and 22% compared to SNN and 9% and 29% compared to DSHW.
Article
Full-text available
This paper reviews the state of the art of research and industry practice on demand response and the new methodology of transactive energy. Demand response programs incentivize consumers to align their demand with power supply conditions, enhancing power system reliability and economic operation. The design of demand response programs, performance of pilot projects and programs, consumer behaviors, and barriers are discussed. Transactive energy is a variant and a generalized form of demand response in that it manages both the supply and demand sides. It is intended for a changing environment with an increasing number of distributed resources and intelligent devices. It utilizes the flexibility of various generation/load resources to maintain a dynamic balance of supply and demand. These distributed resources are controlled by their owners. However, the design of transaction mechanisms should align the individual behaviors with the interests of the entire system. Transactive energy features real-time, autonomous, and decentralized decision making. The transition from demand response to transactive energy is also discussed.
Conference Paper
Full-text available
The electrical demand forecasting problem can be regarded as a non-linear time series prediction problem depending on many complex factors since it is required at various aggregation levels and at high resolution. To solve this challenging problem, various time series and machine learning approaches has been proposed in the literature. As an evolution of neural network-based prediction methods, deep learning techniques are expected to increase the prediction accuracy by being stochastic and allowing bi-directional connections between neurons. In this paper, we investigate a newly developed deep learning model for time series prediction, namely Factored Conditional Restricted Boltzmann Machine (FCRBM), and extend it for demand forecasting. The assessment is made on the EcoGrid EU dataset, consisting of aggregated electric power consumption, price and meteorological data collected from 1900 customers. The households are equipped with local generation and smart appliances capable of responding to real-time pricing signals. The results show that for the energy prediction problem solved here, FCRBM outperforms the benchmark machine learning approach, i.e. Support Vector Machine.
Article
Over the past few decades, the behavior of electricity consumption has been changing, especially because of improvements in the distributed generation segment and technological innovations presented by smart grids. The use of microgeneration and the availability of electricity pricing in real time allow consumers to control their consumption, or generation, according to market conditions. This new dynamic tends to increasingly change the price elasticity of electricity demand, by indicating the need to readjust load forecasting models. In this market environment, in addition to providing robust estimates for the planning and operation of electric power systems, load forecasting models have become fundamental in the context of demand management. Thus, this paper proposes to develop an artificial neural network and fuzzy logic for load forecasting to perform an efficiency analysis. This system is able to provide estimates of the elasticity of electricity demand behavior with more satisfactory results. To do so, improvements in the neural network with multilayer perceptron are proposed. In this case, the adaptation of parameters to correlate variations in consumption with the changes in electricity tariffs was developed. The addition of this new structure produced better results compared with the conventional neural network. Computer tests were conducted using historical data from the ISO New England Inc and PJM Interconnection. Price elasticity estimates of electricity demand showed a sharp increase of demand in relation to the elasticity behavior.
Article
Energy management of residential buildings plays an important role in a smart grid. Region specific fuzzy logic strategies are proposed recently. However, no such approach exists that covers all regions of the world. A fuzzy logic based strategy for the construction of fuzzy controller covering the entire globe would be cost effective due to the increasing power of micro-controllers. Results show that our proposed approach achieves a minimum energy savings of 6.5%, irrespective of where it is used around the world. This research will provide a model for extending the region specific solutions for a worldwide adaption.
Article
Unprecedented high volumes of data are becoming available with the growth of the advanced metering infrastructure. These are expected to benefit planning and operation of the future power system, and to help the customers transition from a passive to an active role. In this paper, we explore for the first time in the smart grid context the benefits of using Deep Reinforcement Learning, a hybrid type of methods that combines Reinforcement Learning with Deep Learning, to perform on-line optimization of schedules for building energy management systems. The learning procedure was explored using two methods, Deep Q-learning and Deep Policy Gradient, both of them being extended to perform multiple actions simultaneously. The proposed approach was validated on the large-scale Pecan Street Inc. database. This highly-dimensional database includes information about photovoltaic power generation, electric vehicles as well as buildings appliances. Moreover, these on-line energy scheduling strategies could be used to provide real-time feedback to consumers to encourage more efficient use of electricity.
Conference Paper
The present evolution of the electrical sector is paying increasing attention to the customer needs. This paper deals with a number of aspects addressed in the present discussions on the characteristics of individual and aggregate consumers, referring to the shape and flexibility of the demand. Specific points addressed include the role of metering, how to obtain knowledge from load pattern disaggregation and clustering, customer-related effects of microgrids development, and active users' participation in demand response initiatives.
Article
Smart metering data are providing new opportunities for various energy analyses at household level. However traditional load analyses based on time-series techniques are challenged due to the irregular patterns and large volume from smart metering data. This paper proposes a promising alternative to decompose smart metering data in the spectral domain, where i) the irregular load profiles can be characterized by the underlying spectral components, and ii) massive amount of load data can be represented by a small number of coefficients extracted from spectral components. This paper assesses the performances of load characterization at different aggregated levels by two spectral analysis techniques, using the discrete Fourier transform (DFT) and discrete wavelet transform (DWT). Results show that DWT significantly outperforms DFT for individual smart metering data while DFT could be effective at a highly aggregated level.