Content uploaded by Hale Bakır
Author content
All content in this area was uploaded by Hale Bakır on May 14, 2019
Content may be subject to copyright.
Konut Tipi Şebeke Bağlantılı Fotovoltaik Sistemde Adaptif Nöro Bulanık
Arayüz Sistemi Tabanlı Enerji Yönetimi Uygulaması
Hale Bakir1, Umair Younas2, Derya Betul Unsal Celimli3, Ahmet Afsin Kulaksiz4*
1,2,4Konya Teknik Üniversitesi, Mühendislik ve Doğa Bilimleri Fakültesi, Elektrik Elektronik Mühendisliği
Bölümü, Konya / Türkiye
3Cumhuriyet Üniversitesi, Mühendislik Fakültesi, Elektrik Elektronik Mühendisliği Bölümü, Sivas / Türkiye
* Sorumlu Yazar: aakulaksiz@ktun.edu.tr
Özet
Talep tarafı yönetiminin enerji piyasalarında aktif kullanımı ile ülke genelindeki elektrik
şebekesi üzerindeki yükü azaltmak ve şebekenin kesintilerinden kaçınmak mümkündür.İsteğe
bağlı hassasiyet analizi, elektrik güç sistemlerinin doğru çalışmasını sağlar.Ancak, yalnızca
talep tarafındaki düzenlemeler bir çözüm için yeterli değildir; arz tarafının çevre ile daha
sürdürülebilir ve daha uyumlu bir enerji arz-talep dengesi oluşturmak için geliştirilmesi
gerekmektedir. Bunun nedeni gelecekteki güneş ve rüzgar enerjisi teknolojilerinin ve diğer
yenilenebilir enerji türlerinin fosil yakıt bazlı üretim ile birlikte yer almasıdır. Böylece, akıllı
şebekeden yararlanılarak güç dağıtımı geliştirilebilir. Bu çalışmada, Adabtif Nöro-Bulanık
Çıkarım Sistemine (ANFIS) dayanan yeni bir enerji yönetimi yöntemi önerilmiştir. Önerilen
model için öncelikle talebi etkileyen faktörler belirlenir ve verilen modelin veri tabanı
oluşturulur. Sistem, prototip bir şebekeye bağlı fotovoltaik sistemde uygulandı ve üretilen güç
gün boyunca ölçüldü ve tasarruf edildi. Yöntemde müşteri talep gücü ve elektrik fiyatları da
kullanılmış ve sunulan yöntemin geçerliliği test edilmiştir. Bir kış günü verileriyle birlikte bir
yaz günü verileri alınmış ve sistemde belirli bir zamanda yüklerle tüketilmesine izin
verilebilecek referans güç yüzdesi elde edilmiştir. Elde edilen sonuçlar, doğruluk ve verimin,
bu araştırma alanında kullanılan geleneksel yöntemlerle karşılaştırıldığında arttığını
göstermektedir.
Anahtar kelimeler: Adabtif Nöro-Bulanık Çıkarım Sistemi, Talep tarafı yönetimi, Enerji
yönetimi, Akıllı şebeke
Adaptive Neuro-Fuzzy Interface System-based Energy Management
Application for a Residential Photovoltaic On-Grid System
Hale Bakir1, Umair Younas2, Derya Betul Unsal Celimli3, Ahmet Afsin Kulaksiz4*
1,2,4Konya Technical University, Faculty of Engineering and Natural Sciences, Department of Electrical &
Electronics Engineering, Konya/Turkey
3Cumhuriyet University, Faculty of Engineering, Department of Electrical & Electronics Engineering,
Sivas/Turkey
*Corresponding Author: aakulaksiz@ktun.edu.tr
Abstract
With the active use of demand-side management in energy markets, it is possible to reduce
the load on the electricity network around the country and avoid the interruptions of the
mains. On-demand sensitivity analysis ensures that electrical power systems operate properly.
However, the regulations only on the demand side are not sufficient for a solution; the supply
side needs to be developed to create a more sustainable and more harmonious energy supply-
demand balance with the environment. This is because the future solar and wind energy
technologies and other renewable energy types take its place along with fossil fuel-based
production. Thus, delivery of power can be improved by taking advantage of the smart grid.
In this study, a new energy management method is proposed based on Adaptive Neuro-Fuzzy
Inference System (ANFIS). For the proposed model, firstly the factors affecting the demand
are determined and the database of the given model is established. The system has been
implemented on a prototype grid-connected photovoltaic system and generated power was
measured and saved throughout the day. Costumer demand power and electricity prices are
also employed in the method and validity of the introduced method was tested. A summer day
data, along with a winter day data, were taken and the percentage of reference power that can
be allowed to consume by loads at a given time in the system was obtained. The obtained
results show that the accuracy and the efficiency are improved compared to conventional
methods that have been used in this research field.
Keywords: Adaptive Neuro-Fuzzy Inference System, Demand side management, Energy
management, Smart grid
Introduction
The increase in energy prices, global warming, and climate change, increase in population
and living standards, increase in energy demand in line with developments in industry and
technology, the dependence on rapidly depleting fossil fuels which will continue in the near
future, foreign dependency and security of supply, and developments in new energy
technologies are leading countries to new quests. These concerns suggest the need for using
renewable energy sources at a higher rate for electricity generation. With the spread of
distributed production from renewable energy sources, a reduction in the use of fossil fuels
and a transformation of the electric system to reduce costs and pollution is expected.
With the evolution of concepts of microgrid and smartgrid, the traditional power system is
advancing to a new era. Smart grid offers the possibility of control, communications, and
monitoring in the power system. Integration using a suitable converter defines the connection
of heterogeneous types of energy sources with AC or DC supply. As the power output of
distributed renewable energy sources is dependent on climatic conditions, energy
management systems become crucial. For the effective operation of the smart grid, it is also
necessary to disperse generation and storage communication between the various control
nodes. With this system, the consumer can have the opportunity to play a role in the
optimization of the electrical system. However, in order to have the opportunity to purchase
electricity, consumers must have more dynamic pricing automation. As a result, users can
manage their consumption habits according to the price signals of the smart grid (Giordano
and Fulli, 2012).
Several classical or statistical techniques have been carried out in literature but this
interesting field still needs further research to improve the accuracy in a satisfactory manner.
A literature search of studies on the Adaptive Neuro-Fuzzy Inference System (ANFIS) based
algorithms and demand-side energy management variants show that a range of methods on
this topic has been proposed in recent years. ANFIS has been widely used in many fields of
engineering. Ekici and Aksoy have proposed an algorithm to forecast building energy
consumption in a cold region adopting an ANFIS model (Ekici and Aksoy, 2011). The
objective of their study was to study the feasibility and applicability of ANFIS in load
forecasting for the energy of a building. The authors observed that ANFIS is an effective tool
with an energy prediction of 96.5 and 83.8% for heating and cooling, respectively, in the pre-
design stage of energy efficient buildings for choosing the best combinations. In a study by
Bogaraj and Kanakaraj proposed an energy management method for a hybrid renewable
energy system supplying AC loads using ANFIS (Bogaraj and Kanakaraj, 2016). The
developed algorithm is designed to increase the power transfer capability between the source
side and load side. From the analysis made by using different environmental and load test
conditions, the method confirmed that control method is effective in prediction of energy
required for the next instant and manages the energy flow among hybrid power sources and
energy storage devices.
Sarduy et al. proposed linear and non-linear methods for prediction of peak load at the
University of São Paulo. The obtained results confirm that the predicted and real peak load for
all developed models agree well (Sarduy et. al., 2016). However, the accuracy was better for
nonlinear models, especially for Adaptive Neural Networks Inference Systems. Pérez-Romero
et al. presented a local energy management system to optimize energy consumption of a home
by absorbing or delivering energy to energy storage system according to the residential
consumption (Perez-Romero et. al., 2013). It utilizes different types of communication to help
them to adopt efficient energy behavior in a prototype system. Ozturk et al. proposed a
personalized home energy management system for residential demand response (Öztürk et.
al., 2013). The Master Energy Controller implements a branch and bound scheduling
algorithm for managing appliance runtime schedules. For a personalized home energy
management system, the described Master Energy Controller in the paper provides a user-
friendly, inexpensive, and easy to install/maintain architecture for both customer and utility
provider.
Most of these studies use assumed consumption values and some of them use simulated
power generation values in their studies. In order to improve the viability of the proposed
algorithm, this study uses the actual values of generation and consumption. In this paper, an
energy management algorithm has been realized by reducing the peak load by obtaining an
output according to three inputs using ANFIS, according to a certain calculation of the actual
consumption data of a house belonging to the PECAN STREET data which is focused on
advancing university research and accelerating innovation in water and the electricity energy
generation obtained from an on-grid PV system and electricity prices obtained from electricity
market (Anonymous, 2016). The paper is organized as follows. Following the introduction
section, Section 2 describes the material and method used in this work. Here, PV modules,
on-grid inverter, and energy analyser are briefly explained. This section also addresses the
ANFIS-based energy management application. Application of the proposed method and
results are given in Section 3 and the final conclusions are addressed in Section 4.
Material and Method
A. Photovoltaic modules
Solar energy systems provide many benefits to the environment in today's world. Energy
from solar photovoltaic is used to convert solar energy into electrical energy by means of the
unit referred to as a solar cell. The DC power is converted to AC power to be used in
residential and commercial buildings (Giordano and Fulli, 2012). Peak power of each PV
module used in this study is 250Wp. Five modules were connected in series to obtain bus
voltage for the DC input of the inverter and 1250Wp power was obtained (Bakır and
Kulaksız, 2016).
B. Grid-connected Inverter
Grid-connected photovoltaic systems can meet some or all of the electricity demand in the
housing. In the system, the excess energy can be sold to the grid. Inverter, converting the DC
voltage into AC voltage is a device that can operate fully compatible with the network
(Giordano and Fulli, 2012). The advantages of the grid-connected system are as follows; ease
of installation, high efficiency, reliability, flexible/modular scalable structure. The model of
the grid-connected inverter used in this study is Fronius Galvo 1.5. Schematic diagram of the
grid-connected PV system architecture employed in this study is shown in Fig. 1. The on-grid
PV power system was considered in evaluating the proposed algorithm (Bakır and Kulaksız,
2016).
C. Energy analyzer
In the system, energy analyzer is used to measure the electrical parameters of the plant. The
device used is ENTES Power and Energy Meter. The device can measure three-phase active,
reactive and apparent power values in the power system (Bakır and Kulaksız, 2016).
Fig. 1. Grid-connected PV System diagram
D. Adaptive Neuro-Fuzzy Inference System
One of the most important research topics of the smart grid is energy management
applications. In particular, energy management techniques such as load-side management and
the demand-side response is common (Shao et. al., 2011). Users and electricity companies
deploy some of the benefits provided by this energy management applications. The energy
management method presented in this study is based on ANFIS. On the basis of the ANFIS
method, Fuzzy inference systems have Takagi-Sugeno-Kang fuzzy inference system.
Jang developed the ANFIS method and system is used in the modeling of nonlinear
functions, in estimating nonlinear components in the control system and in estimating chaotic
time series (Jang, 1993). ANFIS model is shown in Fig. 2.
A1
A2
B1
B2
N
N
Layer 2 Layer 3 Layer 4 Layer 5
w2f2
w1f1
x y
x y
x
y
w1
w2
1
2
Layer 1
Fig. 2 . ANFIS model for two input-one output
In this model, the values of A1 and B1 are used in the first layer to express verbal variables.
The values of A1 and B1 are derived from the layer by taking a membership grade through a
membership function. On the 2nd layer, the inputs are multiplied and removed from a node.
In the third layer, the ignition power values are normalized to the total ignition power values.
After the fourth layer, the Takagi-Sugeno-Kang model is operated (Şen, 2017).
The definition of each layer are given below (Ekici and Aksoy, 2011).
Layer 1: A membership function µ(x) is chosen to be a member of determined verbal
variables. A and B are the linguistic labels, µAi(x) and µBi(x) are the membership function
outputs obtained from these nodes.
, , (1)
Layer 2: The firing strength of each rule is calculated by this formula.
, (2)
Layer 3: The ith node calculates the ratio of the ith rule’s firing strength to all rules firing
strength.
, (3)
Layer 4: Where f1 and f2 are the fuzzy if-then rules as follows the outputs written as given
in this formula.
, (4)
Layer 5: This node computes the overall output of ANFIS as the summation of all incoming
signals from the 4th layer.
, (5)
Research Findings and Discussion
In this study, for the operation of electrical appliances at home, the ANFIS method was
used to determine total reference power in percent to be consumed by appliances for the next
instant depending on the consumer demand for any given time. This is done to reduce present
load demand by enabling each consumer to employ their own strategy by delaying flexible
loads (Bakır and Kulaksız, 2016). The input variables introduced for the ANFIS control,
shown in Fig. 3, are explained below:
PV power generation: The system has been implemented on a 1.25 kWp grid-connected
photovoltaic system and generated power was measured and saved throughout the day.
Costumer demand power: Daily consumption data for specific loads from PECAN
STREET dataport have been taken to evaluate the algorithm using real data.
Electricity price: Retail Electricity Tariffs Table in Turkey in 2017 have been used. Three-
time residential electricity tariff was considered here (Anonymous, 2017). Figures 4 and 5
show the daily consumption data for certain flexible loads from the PECAN STREET
dataport.
Fig. 3. Three input-one output ANFIS structure of the system design
(a)
(b)
(c)
(d)
Fig. 4. The consumption values on 14.11.2017 (a) daily consumption dataport for
waterheater, (b) for airconditoner, (c) for microwave, (d) for clotheswasher_drying
(a)
(b)
(c)
(d)
Fig. 5- The consumption values on 16.07.2017 (a) daily consumption dataport for
waterheater, (b) for airconditoner, (c) for microwave, (d) for clotheswasher_drying
The fuzzy rules are the sets of “If” and “then” statements. The three input parameters are
fuzzified as per the membership function of the respective variable. These, in addition to
the membership function curves, are utilized to obtain a solution (Hatagar and Halase,
2015). This value is important to assign priorities to each load in a home and delay times
can be determined for each of them (Bakır and Kulaksız, 2016).
ANFIS output produces a result that can be measured. This is called the defuzzification.
This unit interprets the degree of membership of the real valuable input. The centroid
method was used for defuzzification to get a scalar output value (Hatagar and Halase,
2015). The method was tested on a sunny day. The profiles of PV power, customer
demand power of the current time, and electrical price, which are the inputs of ANFIS and
percent power of electrical appliances, which is the output of ANFIS, respectively, are
shown in Figs. 6-7 for a day in winter and in summer, respectively. For the day in winter
(Fig. 6), the percentage of electricity power output is lower because the energy generated
from the PV system is less. Consequently, the output percentage for electrical applications
is %75, all loads cannot be switched on, and one or two loads will be switched on at the
subsequent phase depending on the new results. For the day in summer (Fig. 7), the
percentage of electric power is higher because the energy generated from the PV system is
larger and the algorithm encourages the consumption. Consequently, the output
percentage for electrical applications is %100 and all of the loads can be switched on.
Based on the results, the output responds accordingly depending on the input values.
For example, from the results, it can be reflected that in the early morning the percentage
power was at the maximum level. This can be justified as the PV power production is high
at that time. In the late afternoon, the result of reference power gets the smallest value as
electricity price is high, PV power is low and customer demand power is high. Thus, by
identifying the reference power, flexible loads are either connected as demanded by the
consumer or they were disconnected. It can be concluded that the method can be
employed to reduce peak loads, use the PV generation more efficiently and reduce the
cost of electric invoice of the consumer.
Results and Suggestions
In this paper, a new energy management system was proposed and it was confirmed
that by the inclusion and exclusion of flexible loads demanded by the consumer, it can
possible to achieve a reduction in cost and the peak load for the consumer can be
decreased. The electrical power generated by PV modules, customer demand power, and
electrical price were used as input variables in ANFIS and the results proved that the
method can be employed to spread the peak load throughout the day as high customer
demand inhibits the output in ANFIS. Electricity price was another factor inhibiting
ANFIS-output when the price is high. ANFIS also encouraged operating the electrical
appliances when PV generation is high. A summer and winter day data were taken and the
percentage of output power in the system was compared as a reference value. For summer
day in the system, it was seen that there is a higher percentage of electric power and the
delay in the operation of loads is less because the power generated from the PV system is
larger. For a winter day, the percentage of electric power output is lower and the delay to
the more economic price time occurs more because the power generated from the PV
system is smaller.
The adoption of ANFIS method in this study was that it required less time to prepare
the model than conventional methods. Because this method does not require the definition
of the relation between the data, a function should be established. They learn by
themselves through the relevant training dataset between the data. As a result, within the
scope of this study, a solution to the demand estimation problem was searched with
ANFIS methods.
As a future study, priorities will be assigned to each load in a home and the results of
this study will be used to determine delay times for each of them. The results of our
system will be tested in a test bed having based communication network and more
experiments will be carried out.
(a) (b)
(c) (d)
Fig. 6. For 14.11.2017 (a) Profile of power produced by the PV panel (b) Profile of total
customer demand power for water heater, for airconditioner, for microwave, for
clotheswasher_drying (c) Electricity price 2017 in Turkey (d) Profile of percent power of
electrical appliances during per 15 sec for water heater, for airconditioner, for microwave,
for clotheswasher_drying
(a) (b)
(c) (d)
Fig. 7. For 16.07.2107 (a) Profile of power produced by the PV panel (b) Profile of total
customer demand power for water heater, for airconditioner, for microwave, for
clotheswasher_drying (c) Electricity price 2017 in Turkey (d) Profile of percent power of
electrical appliances during per 15 sec for water heater, for airconditioner, for microwave,
for clotheswasher_drying
Acknowledgment: Authors would like to thank Pecan Street Project Inc.
(http://www.pecanstreet.org) for allowing us to access electrical usage database.
References
Giordano, V., Fulli, G., "A business case for Smart Grid technologies: A systemic perspective", Energy Policy
40, pp. 252-259, 2012.
Ekici, B.B., Aksoy, U.T. ‘’Prediction of building energy needs in early stage of design by using ANFIS’’,
Expert Systems with Applications 38, pp. 5352–5358, 2011.
Bogaraj, T., Kanakaraj, J., “A Novel Energy Management Scheme using ANFIS for Independent Microgrid”,
International Journal of Renewable Energy Research, vol.6, no.3, pp. 735-746, 2016.
Sarduy, J.R.G., Di Santo, K.G., Saidel, M.A., ‘’Linear and non-linear methods for prediction of peak load at
University of São Paulo’’, Measurement 78, pp. 187–201, 2016.
Pérez-Romero, M. J., Gallardo-Lozano ,. Romero-Cadaval, E., Lozano-Tello, A., “Local Energy Management
Unit for Residential Applications”, Elektronika ir Elektrotechnika, ISSN 1392-1215, vol. 19, no. 7, pp. 61-64,
2013.
Özturk, Y., Jha, P., Kumar, S., Lee, G., ‘’A Personalized Home Energy Management System for Residential
Demand Response’’, 4th International Conference on Power Engineering, Energy and Electrical Drives, pp. 13-
17, 2013.
Anonymous, Pecan Street, http://www.pecanstreet.org/,2016.
Bakır, H., Kulaksız, A.A., ‘’Energy Management Application to Reduce Peak Load in a Photovoltaic Power
Integrated Grid’’, 8th Ege Energy Symposium, Afyonkarahisar, 2016.
Shao, S.N., Pipattanasomporn, M., Rahman, S., "An Approach for Demand Response to Alleviate Power System
Stress Conditions", IEEE Power and Energy Society General Meeting, 2011.
Jang, J.S. R., “ANFIS: Adaptive-Network-Based Fuzzy Inference System”, IEEE Transactions on Systems, Man
and Cybernetics, vol. 23, no. 3, pp. 665-685, 1993.
Şen, Z., ‘’Modeling Principles with Fuzzy in Engineering’’, Water Foundation Publications , İstanbul, 2004.
Anonymous, EPDK, http://www3.epdk.org.tr/index.php/elektrikpiyasasi/tarifeler?id=133, 2017.
Hatagar, S., Halase, H., ‘’Three Input – One Output Fuzzy logic control of Washing Machine’’, International
Journal of Scientific Research Engineering & Technology, vol. 4 no. 1, 2015.
