See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/308815443

Comparison of input signal choices for a fuzzy logic-based power system

stabilizer

Conference Paper · October 2015

DOI: 10.1109/NAPS.2015.7335158

CITATIONS READS
2 1,291

2 authors, including:

Ahmed Sheikh
Florida International University
3 PUBLICATIONS   42 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Microgrid Control using MultiAgent Systems View project

All content following this page was uploaded by Ahmed Sheikh on 24 November 2017.

The user has requested enhancement of the downloaded file.

 Comparison of Input Signal Choices for a Fuzzy
Logic-based Power System Stabilizer

Ahmed Faizan Sheikh and Shelli K. Starrett

Electrical and Computer Engineering Department
Kansas State University
Manhattan, Kansas, USA
Email: [email protected], [email protected]

Abstract—Variability in power systems in increasing due to complex in nature [2]. A fuzzy neural network based on Phasor
pushing the system to limits for economic purposes, the inclusion Measurement Units (PMUs) has been applied to predict power
of new energy sources like wind turbines and photovoltaic, and swing stability [3].
the introduction of new types of loads such as electric vehicle
chargers. In this new environment, system monitoring and Wide Area Measurement (WAM) technologies using
control must keep pace to insure system stability and reliability PMUs can provide control signals at exceptionally high speed.
on a wide area scale. Phasor measurement unit technology PMUs can be deployed at particular locations in the grid to
implementation is growing and can be used to provide input measure voltages and currents at different locations of the grid
signals to new types of control. Fuzzy logic based controllers to a get coherent picture of the whole network in real time [4].
have also been shown effective in various studies. This paper PMUs gather data from various locations in the power system
considers several choices of input signals, composed assuming with Global Positioning Systems and transmit it to centralized
phasor measurement availability, for a common fuzzy logic- locations for processing and analysis. This processed data can
based controller. Nonlinear transient simulation results for two be used to determine generator variables such as angles,
fault locations for a two-area system are used to compare the speeds, accelerations, and powers from time-stamped voltages
effects of input choice on damping of system oscillations. and currents [5]. These voltage and current signals can be used
as feedback inputs for a power system stabilizer (PSS) control
Keywords—power system stabilizer; phasor measurement units;
design in order to enhance damping of oscillations [3]. The
wide area control; fuzzy logic control; transient simulations
effects of time delay and data uncertainty have been studied by
I. INTRODUCTION Kamwa [6].
Modern day electrical power systems are interconnected The center of inertia (COI) angle and speed of a system or
and tend to have low frequency electromechanical oscillation region of a system has been used in the development of energy
modes that have been a key concern in planning and operation function methods fo analysis for power system dynamics [7].
of power systems [1]. These low frequency modes are The COI can be used as a reference to allow the visualization
unfavorable to maximum power transfer and system security. of the machine dynamics with respect to a dynamic center of
In order to operate the power system securely, damping of the system as opposed to a fixed synchronous reference or a
these inter-area oscillations is of dynamic concern. Power single large machine. The COI speed for a system or sub-
system stabilizers (PSS) are typically added to the automatic system is defined as:
voltage regulator on generators to damp these oscillations and
to improve system dynamic performance. ∑ 𝜔𝑖 ∗𝐻𝑖
𝑠𝑦𝑠𝑡𝑒𝑚_𝐶𝑂𝐼_𝑠𝑝𝑒𝑒𝑑 = (2)
𝐻𝑡𝑜𝑡𝑎𝑙
Most PSS used in today’s power utilities are devised from
classical control theory based on a linearized model of the where the i are the machine angles, Hi are generator inertias
power system. They are adjusted for one set of operating and Htotal is the sum of the generator inertias for the system or
conditions and may not be effective for different operating sub-system. COI angle is defined similarly. The COI machine
conditions and configurations. Conventional PSS can be speed or angle for a single machine is defined by taking the
represented by following transfer function. system COI speed or angle as a reference. Namely:
1+𝑇1 𝑠 1+𝑇3 𝑠
𝐺(𝑠) = 𝐾𝑠 (
1+𝑇2 𝑠
)(
1+𝑇4 𝑠
) (1)  COI ii – system_COI_speed 

New technologies such as FACTS devices and phasor Undeniably, electrical power systems are exceedingly non-
measurement systems make it possible to monitor rotor angle linear and demonstrate conditions that are stochastic and
stability of a power system. This is considered to be a dynamic in nature. Conventional PSS (CPSS) performance
challenging task as power system dynamics become heavily drops, as system configuration and operating conditions change
from one to another, because of fixed stabilizer parameters [8]. membership function parameters that best permit the associated
With CPSS, gain and other time constants may not suit the fuzzy inference system to track the certain data set [16].
varied operations. The definitive goal of deploying PSS in ANFIS uses either hybrid or back-propagation learning
Power Systems is to ensure performance and dynamic stability algorithm to smooth learning and adaption. The hybrid
under an extensive range of operating conditions and network algorithm is a combination of back-propagation and least-
configurations. square estimation [19].
With the development of control technology, authors have The basic purpose of fuzzy systems is to substitute operator
been developing modern controllers based on more experience with a rule-based system. The fuzzy controller
sophisticated algorithms. Among these refined methods, converts a linguistic strategy based on experience into an
artificial intelligence-based approaches have been proposed to automatic control strategy. In fuzzy logic, the idea is to fuzzify
design effective PSS. These approaches include Fuzzy Logic given inputs then infer control decisions based on control rules.
[1], [9]-[11], neural networks [12] and genetic algorithms [8]. The FLC output is obtained by defuzzifying these inferred
Fuzzy Logic based PSS shows great potential in damping inter- control decisions. A fuzzy system, as shown in Figure 1,
area generator oscillations [13]. consists of a total of four steps namely: fuzzification interface,
an expert knowledge, an inference engine, and defuzzification
Unlike classical control theory, which demands very interface [8].
profound and comprehensive understanding of the physical
system, fuzzy logic does not require a multifaceted
mathematical system model. It allows developers to utilize
knowledge with subjective concepts such as big, moderate and
small, which are mapped to numerical ranges [12]. It is,
therefore, compatible when the system is highly non-linear,
complex, and difficult to model mathematically. Since it
requires a lower computation burden and deals with
uncertainties with substantial ease, Fuzzy Logic PSS’s are
considered to be suitable amongst different PSS
implementation schemes [8]. The performance of Fuzzy Logic
PSS predominantly depends on power system operating
conditions; however, they are less sensitive to changing Fig. 1. General Fuzzy Logic Structure [8]
conditions than CPSS [9].
A Fuzzy Logic Controller (FLC) endures uncertainty, The fuzzification interface is the process of transforming
imprecision or change of input parameters and additionally numerical input variables into fuzzy variables that can be
gives an opportunity to present expert knowledge in control observed as defined fuzzy sets. Fuzzy sets can be characterized
rules. It provides reliable output under changing operating by various membership functions including bell-shaped, linear,
conditions and time-varying input signals as normally triangular and exponential functions. A variables’ degree of
experienced in power systems [14]. Hsu and Cheng proposed a membership in a given set is defined by these membership
fuzzy logic based PSS that used speed deviation and functions.
acceleration as inputs that were converted into linguistic The knowledge base represents knowledge of the
variables using some membership functions [15]. Lotfi and application domain using a set of linguistic rules. The inference
Tsoi tuned membership functions of the fuzzy outputs engine operates on different conditions from fuzzy sets using
associated with control rules of an FLC [15]. An adaptive the rule base obtained from experience. The defuzzification
neuro-fuzzy inference system (ANFIS) combines the FLC inference obtains numerical output from the fuzzy output sets
structure with neural network (NN) learning aspects to according to output membership functions.
determine parameters of the controller. Initially, a rule base,
relating controller parameters and operating conditions, is The rules are stated to define relationship between input
established and then parameters of the controller are tuned in a and output of FLC and are demarcated using linguistic
neural network based training process. Fuzzy parameter variables. The knowledge required to yield fuzzy rules can be
estimation is acquired by utilizing different arrays of input- developed from an offline simulation. However, as electrical
output data [16]. Jang deployed ANFIS to generate power systems are highly non-linear and complex in operation
membership functions and manage rules of FLC [15]. O.P. so operator experience plays crucial role in describing the rules
Malik designed an Adaptive Neuro Fuzzy Controller by [8].
adapting input link weights to obtain optimum performance The Mamdani type defuzzification inference method is
[17]. established by the centroid method. In this method, the
ANFIS has turned out to be excellent function estimation weighted average of membership functions, or the center of
tool. It does not require an exact mathematical model of the gravity of the area within the membership function curves, is
physical system. Furthermore, speed and robustness are attained to yield a numerical value of the fuzzy quantity [20].
prominent properties of this scheme [18]. MATLAB ANFIS The Takagi-Sugeno defuzzification method uses
gives a method for the fuzzy modeling procedure to learn polynomial functions in place of membership functions. It has
information about a given data set for calculating the an ability to accurately model nonlinear system behavior with
relatively few linear equations. Smooth transitions are Maximum and minimum values defining range for input
obtained by implementing fuzzy rules [21]. Takagi-Sugeno and output variables are mentioned in Table 2. Open loop
method is computationally efficient and performs well with simulation is performed for different initial conditions in order
linear techniques, optimization, and adaptive techniques [22]. to obtain minimum and maximum values for stabilizer [20].
To give a fair comparison of the various input signals, the same
In this paper, we use different signals such as machine active ranges are used for each variable. Inputs above or below
speed, bus frequency, COI machine speed, and COI machine the active range are are taken to be LN and LP respectively.
speed estimations as inputs to a fuzzy logic controller to
observe their effect on PSS output individually. Machine speed TABLE I. RANGE OF FUZZY VARIABLES
and machine angle responses versus time results are compared
to determine the effectiveness of various input choices for a Fuzzy State Minimum Maximum
Fuzzy Logic based PSS. Variables Value (pu) Value (pu)
Input -0.012 0.012
II. FUZZY LOGIC-BASED PSS Vpss -0.05 0.2
The design process may be divided into following steps:
state variable selection, membership function definition, rule C. Rule Building
building, and fuzzy inference and defuzzification strategy The rules are stated to define relationship between input
selection. and output of FLC and are demarcated using linguistic
variables. The knowledge required to yield fuzzy rules can be
A. State Variable Selection developed from an offline simulation. However, as electrical
State variable selection involves the choice of input and power systems are highly non-linear and complex in operation
output variables. The selection of FLC inputs is the primary so operator experience plays crucial role in describing the rules
research objective of this paper. This paper uses PMU-type [8].
data to formulate various FLC input signals for a fairly
standard FLC in order to illustrate the effect on input on Fuzzy For the proposed FLPSS, the inference mechanism is
PSS performance. The output is taken as the standard PSS characterized by a decision table. The set of rules is expressed
voltage output signal used for CPSS. The input values are in Table 1.
normalized and converted into fuzzy variables. So, the fuzzy
TABLE II. FLPSS DECISION TABLE
power system stabilizer has one input and one output as shown
in Figure 2. If Input is: Output Signal is:
LN SN
MN SN
SN SN
ZE ZE
SP SP
MP MP
Fig. 2. Basic Block Diagram of a Fuzzy Logic PSS LP LP

For example, the first row of the table means “if the input is
COI estimated machine speeds are calculated using large negative, the output will be small negative.”
moment of inertia and bus frequency for all machines as:
D. Fuzzy Inference and Defuzzification Method Selection
∑ 𝐵𝑢𝑠_𝐹𝑟𝑒𝑞𝑖 ∗𝐻𝑖
𝑠𝑦𝑠𝑡𝑒𝑚_𝐶𝑂𝐼_𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒 = (4) A Takagi-Sugeno inference method is used for all FLC PSS
𝐻𝑡𝑜𝑡𝑎𝑙
designs presented in this paper. This simplified the fuzzy
system by utilizing constants for each output membership
Bus frequency is calculated using bus voltage angles as: function. The values of the output membership functions were
distributed evenly within the active range given in Table I.
𝑏𝑢𝑠_𝑎𝑛𝑔𝑙𝑒(𝑖,𝑘)−𝑏𝑢𝑠_𝑎𝑛𝑔𝑙𝑒(𝑖,𝑘−1)
𝐵𝑢𝑠 𝐹𝑟𝑒𝑞𝑖 = (5) The values are SN = -0.05, ZE = 0, SP = 0.1, MP = 0.15, and
𝑡(𝑘)−𝑡(𝑘−1)
LP = 0.20.
In above equation, i indicates the bus number of a generator III. SIMULATION RESULTS
while k indicates time step involved in numerical integration.
Thus bus frequency is an estimate of the time derivative of the Simulations were performed using transmission line fault
bus angle. on a Two Area Multi-Machine System. Machine responses
subjected to fault are obtained by non-linear simulation using
B. Membership Function Definition Power System Toolbox for Matlab [23]. System responses with
Linear triangular and sigmoidal & z-shaped FLPSS for various operating conditions are then compared
membership functions are used for the input variables. The with conventional PSS. Matlab and its fuzzy logic toolbox
following fuzzy sets are chosen for input signal: SN (Small have been used for system simulations.
Negative), ZE (Zero), SP (Small Positive), MP (Medium
Positive), LP (Large Positive).
A. Test System
The Kundur system [24] comprises of two largely
symmetrical areas connected together by two 230KV lines of
220 km length. Each has two identical round rotor generators
rated 20KV/900MVA. The synchronous machines’ parameters
are all the same except for inertia which is H=6.5s for Area 1
machines and H=4s for Area 2 machines. The base-case load-
flow (with generator 2 as slack machine) is such that all
generators in the system are producing about 700 MW each.
The loads are assumed to be constant impedance load models
everywhere while area 1 and 2 loads are 976 MW and 1765
MW respectively. In order to improve the voltage profile, 187
Mvar capacitors were added in each area as shown in Fig 3. A
three phase fault of 0.1 sec duration was simulated at line 3-
101. Power system stabilizer under study is located at Bus 1,
while conventional stabilizers are placed on the other three
generators.

Fig. 4. Fault comparision of CPSS to Fuzzy PSS with speed and COI speed
inputs

Fig. 3. Kundur 4 Machine, 2 Area System [24]

B. Comparing CPSS to FLPSS with Speed Inputs

Machine angle responses for the generator on which the
PSS was installed (Generator 1) are shown in the various
figures. Fig. 4 compares the machine angle responses for a
CPSS with machine speed as input, and a Fuzzy PSS with
machine speed as input, and a Fuzzy PSS with COI machine
speed as input. The Fuzzy PSS shows a lower peak angle
swing with better damping than the CPSS.
C. Comparing FLPSS with Speed and COI Speed Inputs
In observing Fig. 4, it is seen that the COI speed input
Fig. 5. Fault responses for Fuzzy PSSs with bus frequency input
FLPSS swings further in the positive direction than the FLPSS
with speed input. Damping appears to be quite similar in both E. Comparing Estimates of COI Speed as FLPSS Input
cases.
Similarly, Fig. 6 shows Generator 1’s machine angle
D. Comparing FLPSS with Bus Frequencies as Inputs comparison between a Fuzzy PSS with COI machine speed as
Here we approximate the use of PMUs to measure bus input and a Fuzzy PSS using COI machine speed estimated two
voltage angles which are then used to estimate bus frequency different ways as input. The first estimate uses bus frequencies
for use as input to the FLPSS. In this work, no time delay in in place of the machine speed as given in Eq. (4). For the
the measurements is considered, because it has been shown to second estimate, Area 1’s speed is approximated as the
be manageable by other authors [6]. Fig. 5 depicts the machine frequency at Bus 6 and is multiplied by the sum of the inertias
angle comparison of several Fuzzy PSSs with bus frequencies of Generators 1 and 2. Similarly, Area 2’s speed is taken as the
as inputs. Comparing Fig. 4 and Fig. 5 shows that using bus bus frequency at Bus 10 and multiplied by the sum of the
frequency at the generator terminals produces similar but Generator 3 and 4’s inertias.
slightly better results than seen when using speed as input to
the FLPSS. As seen in Fig. 5, when frequency at Busses 1 and 𝐵𝑢𝑠𝐹𝑟𝑒𝑞6 (𝐻1 +𝐻2 )+𝐵𝑢𝑠𝐹𝑟𝑒𝑞10 (𝐻3 +𝐻4 )
𝐶𝑂𝐼_𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒_2 = (6)
6 are used as inputs, the responses are quite similar, but when 𝐻𝑡𝑜𝑡𝑎𝑙
the input is taken from Bus 8 or 10 the swings are larger and
less well damped. Thus, as expected, measurements taken near The responses are again quite similar to those seen
the generator being controlled yield better results. previously, with the FLPSS’s with estimated COI calculations
having slightly lower maximum swing values, but exhibiting
some higher frequency oscillations later in the simulations.
 A second case is seen in Fig. 8 with the inertias of the
generators changed from (6.5, 6.5, 4, and 4) to (8, 7, 5, and 6)
respectively. Since the inertias of the system are increased,
lower frequency oscillations are expected. As seen in Fig. 8,
the FLPSS with speed input has the smallest maximum swings
and good damping. The FLPSS with the 2nd estimated COI
speed input is well damped but has a larger maximum swing.
The FLPSS with Bus Frequency input swings further and is the
least damped.

Fig. 6. Fault responses for FLPSSs with COI speed and estimated COI speed
inputs

F. Effects of Significant Changes in Generator Inertia

To compare the robustness of various PSS designs to
significant changes in system parameters, this case shows
Generator 1’s angles for the same fault case, but with the
inertias of the generators changed from (6.5, 6.5, 4, and 4) to
(3, 4, 5 and 6) respectively. Though the inertia of an individual
generator would not typically change, if each generator in this Fig. 8. Fault comparision of CPSS to Fuzzy PSS with speed, bus frequency
system is taken to represent a group of generators making up a and COI speed inputs
large portion of the system, inertia of a region could change
due to the replacement of large a synchronous machine with a G. Observing Effects of Changes in Generator Power
comparably sized windfarm. As seen in Fig. 7, the FLPSS’s
Another parameter change studied is to modify the
and the CPSS respond similarly for this large change in system
scheduled generator active power. The power are changed from
parameters. The FLPSS with speed input again has the lower
all being approximately 7 pu to (8, 8, 6.53, and 6) respectively.
maximum angle swings and appears most well-damped.
Fig. 9 illustrates that CPSS and Estimated COI PSS become
However, the FLPSSs show low-amplitude, higher frequency unstable for this case due to larger first swings. The FLPSS
oscillations not seen in the CPSS response. In particular, the speed and bus frequency input cases show large swings but are
bus frequency input produces sustained high frequency still well damped.
oscillations.

Fig. 9. Fault comparision of CPSS to Fuzzy PSS with speed, bus frequency
Fig. 7. Fault comparision of CPSS to Fuzzy PSS with speed, bus frequency and COI speed inputs
and COI speed inputs
 IV. CONCLUSIONS [7] Fouad, A. A., V. Vittal (1992). Power System Transient Stability
Analysis Using the Transient Energy Function Method. New Jersey:
FLPSS with machine speed as input performed best in most Prentice-Hall.
of the cases considered. The PMU-based inputs, bus [8] Patel, H. D. and C. Majmudar (2011). “Fuzzy logic application to single
frequencies and COI estimates, performed similarly to the machine power system stabilizer.” Engineering (NUiCONE), 2011
speed-based measurements. The COI estimates tended to have Nirma University International Conference on, IEEE.
high frequency oscillations. Better results are seen with bus [9] Dasgupta, S., et al. (2013). "Real-time monitoring of short-term voltage
frequency input closer to generator being controlled. stability using PMU data." Power Systems, IEEE Transactions on 28(4):
3702-3711.
When considering changes in system parameters, the [10] Momoh, J. A., et al. (2008). “Voltage stability enhancement using
FLPSS with speed input continued to out-perform the CPSS. Phasor Measurement Unit (PMU) technology.” Power Symposium,
2008. NAPS'08. 40th North American, IEEE.
The PSSs with bus frequency based input signals also showed
robustness to changes in generator inertia and power [11] Cvetkovic, M. and M. Ilic (2011). “PMU based transient stabilization
using FACTS.” Power Systems Conference and Exposition (PSCE),
generation balance changes. 2011 IEEE/PES, IEEE.
The CPSS did not show higher frequency oscillations as [12] Dotta, D. and I. Decker (2007). “Power system small-signal angular
seen in some of the FLPSS responses. CPSS response is stability enhancement using synchronized phasor measurements.” Power
Engineering Society General Meeting, 2007. IEEE, IEEE.
frequency dependent based on its transfer function and does
[13] You, R., et al. (2003). "An online adaptive neuro-fuzzy power system
not show the higher frequency oscillations seen in some of the stabilizer for multimachine systems." Power Systems, IEEE
other cases. Many FLPSS designs presented in the literature Transactions on 18(1): 128-135.
use both speed and acceleration (or power) signals to gain [14] Mokhtari, M., et al. (2013). "Wide-area power oscillation damping with
frequency dependence. In this work simple, single input Fuzzy a fuzzy controller compensating the continuous communication delays."
systems were used, which may explain the presence of the Power Systems, IEEE Transactions on 28(2): 1997-2005
higher frequencies even in the speed input case. [15] Mrad, F., et al. (2000). "An adaptive fuzzy-synchronous machine
stabilizer." Systems, Man, and Cybernetics, Part C: Applications and
Future work includes the testing of input choices in a larger Reviews, IEEE Transactions on 30(1): 131-137.
system with more modes of oscillation and more complex [16] Feilat, E., et al. (2006). “Adaptive neuro-fuzzy technique for tuning
responses. Also of interest is the use PMU-based signals to power system stabilizer.” Universities Power Engineering Conference,
synthesize damping controller inputs to control other dynamic 2006. UPEC'06. Proceedings of the 41st International, IEEE.
devices such as static VAR compensators. [17] Ramirez-Gonzalez, M. and O. Malik (2008). "Power system stabilizer
design using an online adaptive neurofuzzy controller with adaptive
ACKNOWLEDGMENT input link weights." Energy Conversion, IEEE Transactions on 23(3):
914-922.
The authors appreciate the Fulbright Program’s support of [18] Fraile-Ardanuy, J. and P. J. Zufiria (2005). “Adaptive power system
this research. stabilizer using ANFIS and genetic algorithms.” Decision and Control,
2005 and 2005 European Control Conference. CDC-ECC'05. 44th IEEE
REFERENCES Conference on, IEEE.
[1] Dobrescu, M. and I. Kamwa (2004). “A new fuzzy logic power system [19] Gholipour, A., et al. (2009). “Performance of a ANFIS based PSS with
stabilizer performances.” Power Systems Conference and Exposition, tie line active power deviation feedback.” Power Electronics and
2004. IEEE PES, IEEE. Intelligent Transportation System (PEITS), 2009 2nd International
Conference on, IEEE
[2] Yan, J., et al. (2011). "PMU-based monitoring of rotor angle dynamics."
Power Systems, IEEE Transactions on 26(4): 2125-2133. [20] Toliyat, H. A., et al. (1996). "Design of augmented fuzzy logic power
system stabilizers to enhance power systems stability." Energy
[3] Chompoobutrgool, Y. and L. Vanfretti (2012). A fundamental study on Conversion, IEEE Transactions on 11(1): 97-103.
damping control design using PMU signals from dominant inter-area
oscillation paths. North American Power Symposium (NAPS), 2012, [21] Tecec, Z., et al. (2010). "A takagi-sugeno fuzzy model of synchronous
IEEE. generator unit for power system stability application." AutomaticaVol
51: 127-137
[4] Erlich, I., et al. (2011). “Selective damping of inter area oscillations
using phasor measurement unit (PMU) signals.” PowerTech, 2011 IEEE [22] Hassan, L. H., et al. (2010). "Takagi-sugeno fuzzy gains scheduled pi
Trondheim, IEEE. controller for enhancement of power system stability." American Journal
of Applied Sciences 7(1): 145.
[5] Monchusi, B., et al. (2008). “Power system stability assessment based
on synchronized phasor measurements.” Power and Energy Conference, [23] Power System Toolbox ver 3.0 Cherry Tree scientific software,1991-
2008. PECon 2008. IEEE 2nd International, IEEE. 2008
[6] Dobrescub, I. K. M., et al. "A Fundamental Study of Wide-Area [24] Kundur, Prahba (1994). Power System Stability and Control. New York:
Damping Controllers with Application to Fuzzy-Logic Based PSS McGraw-Hill, Inc
Design for Dynamic Shunt Compensators."

View publication stats