You are on page 1of 7

Evaluation of Automatic Voltage Regulator’s PID

Controller coefficients using Python


Pasala Gopi* M. Ramesh M. Padma Lalitha
Electrical and Electronics Engineering, Electrical and Electronics Engineering, Electrical and Electronics Engineering,
Annamacharya Institute of Technology Annamacharya Institute of Technology Annamacharya Institute of Technology
and Sciences, Rajampet, India. and Sciences, Rajampet, India. and Sciences, Rajampet, India.
pasala.epe07@gmail.com ramesh206365@gmail.com padmalalitha_mareddy@yahoo.co.in

Abstract—Python is a flexible programming tool for all kinds oscillations, (ii) Poor robustness. These severe issues are
of technological and engineering tasks of academicians, overcome by a PID controller tuned by Internal Model control
researchers, etc. The variations in the terminal voltage will affect concept. A closed-loop transfer function data is needed to
the relative power of the synchronous generator, these variations apply this tuning method. The PID coefficients are evaluated
in the terminal voltage will mitigate by controlling the generator analytically to reach the closed-loop set-point response with
field excitation. This goal will achieve with the help of an Automatic the desired response [5].
Voltage Regulator (AVR), but unfortunately, the AVR solely cannot
control the variations in terminal voltage effectively. It needs a Lots of closed-loop tuning methods are available
supplementary control like a PID controller to achieve the goal. In nowadays. In [6-7], the authors estimated the model of the
this research paper, the PID coefficients are evaluated, in a closed-loop system from the data by operating the PID
Python environment, using Good gain and Pattern search tuning controller block as P-controller, and then the controller is
algorithms (proposed tuning methods). The robustness of the
2021 IEEE Madras Section Conference (MASCON) | 978-1-6654-0405-1/21/$31.00 ©2021 IEEE | DOI: 10.1109/MASCON51689.2021.9563458

tuned using the estimated closed-loop model. In [8] the author


proposed controllers is evaluated by finding the maximum tuned PI controller for different time-delayed processes based
allowable uncertainties in nominal parameters for different first on the experimental set-point step response data with a P-
peak overshoots. The simulation results indicate that the controller and also derived a correlation between the set-point
proposed tuning methods give a good voltage stability response data and the SIMC setting. The Good gain method is
performance in terms of time-domain specifications (including
also similar and simpler than Shams’s set-point method [9].
parameter uncertainties and external disturbances) than the
The Pattern search algorithm is an effective heuristic method
other tuning methods.
proposed by Dingyu xue in 2012 for solving linear or non-
Keywords—Python control toolbox, PID tuning algorithms, linear control problems [10]. This algorithm is the modified
Automatic Voltage Regulator, Parameter uncertainties. version of a simple search algorithm i.e the Pattern search
algorithm, to define a mesh, requires a random selection of
vectors, but whereas a simple search algorithm requires fixed
I. INTRODUCTION direction vectors. For the convergence of the solution, the
In a power system network, maintaining voltage profile as Pattern search algorithm requires fewer iterations compare to
constant under any circumstances is a big challenge because the direct search algorithm. Good literature on different
power system voltage profile has a close relation to power tuning algorithms is given in [11, 13-14, 17]. In this work,
quality, grid reliability, and control. When the gird voltage Good gain and Pattern search tuning methods are proposed for
profile deviates, it leads to a lot of changes in system tuning of controller.
dynamics and the life of the electrical equipments which are
associated with this power grid will drop quickly(because the II. TRANSFER FUNCTION MODEL OF AVR
electrical equipment can operate efficiently when the voltage
profile is constant)[1]. Another difficulty of not maintaining The generator terminal voltage and reactive power generation
the voltage profile as constant is more reactive power are controlled by its excitation. The real power is affected by the
consumption at the load. To overcome these difficulties a variations in generation frequency, whereas the reactive power is
voltage regulator has to install in the power system. The affected by the variations in the terminal voltage. The generator’s
primary function of the AVR is to keep the generator terminal reactive power is controlled by its field excitation. Practically, to
voltage at desired voltage level by controlling the alternator improve the transmission line voltage profile, load-tap changing
field excitation. But unfortunately, the AVR cannot do this transformers, step-voltage regulators, static VAR control
function solely and it needs another control like PI/PID equipment, etc are used. Nowadays AVR is used to mitigate the
controller [2]. The Proportional + Integral + Derivative variations in reactive power by controlling the generator field
controller is the most commonly used control function in excitation. The major important function of the AVR is to
industrial applications. maintain the alternator voltage profile at a desired level. Fig. 1
shows the schematic diagram of AVR. With the help of a
A very popularly used algorithm for process control is ZN voltage sensor, the voltage magnitude is sensed continuously
method [3] and C & C method [4], but these methods suffer on one phase. This voltage is rectified and filtered before it is
from two severe drawbacks. Those are (i) More overshoot and compared with the reference set-point voltage and generates a

Authorized licensed use limited to: Sri Sivasubramanya Nadar College of Engineering. Downloaded on May 06,2022 at 06:36:40 UTC from IEEE Xplore. Restrictions apply.
voltage error signal. Now the error signal is amplified and fed Kg
as input to the field winding to increase the field excitation Gg(s)  (4)
sτg  1
voltage as well as current. The increase in the field excitation
will increase the reactive power generation and results in raise In this work, for the design of various controllers, the
in the terminal voltage to the desired voltage profile. This nominal parameters are taken from [1]. The nominal values of
research work mainly focuses on the achievement of the desire the components of the AVR system are as shown in TABLE I.
voltage profile. TABLE I. NOMINAL PARAMETERS OF AVR SYSTEM
Range of nominal Proposed nominal
Component
values value
Exciter Amplifier Kvs : 1.0 Kvs = 1.0
Vref(t) Voltage Sensor
- τvs : 0.01 to 0.06sec τvs = 0.01 sec
If
Vf(t) Ka : 10 to 40 Ka = 10.0
Alternator Amplifier
τa: 0.02 to 0.1 sec Τa = 0.1 sec
(3-Ph) Voltage sensor with Ke : 1.0 to 10.0 Ke = 1.0
Exciter
rectifier and filter Τe : 0.4 to 1.0 sec Τe = 0.4 sec
Alternator Kg : 0.7 to 1.0 Kg = 1.0
Transformer
(Syn. generator) Τg : 1.0 to 2.0 sec Τg = 1.0 sec
(3Ph)
3-Ph Bus Bar From the block diagram (see Fig. 2), Closed-Loop
Transfer Function of the AVR,
Fig. 1. Schematic diagram of Automatic Voltage Regulator (AVR). Vt(s) 0.1s  10
 (5)
The dynamic performance of the AVR is investigated by Vref(s) 0.0004s  0.0454s3  0.555s2  1.51s  11
4
developing the mathematical transfer function of the AVR
(neglecting the saturation/non-linearity) [15-16]. The closed- Similarly, Open-Loop Transfer Function of the AVR,
loop block diagram of an AVR is as shown in Fig. 2. The
components of AVR are modeled as first order transfer Vt(s) 10
 (6)
function models. Vref(s) 0.0004s4  0.0454s3  0.555s 2  1.51s  1

e(t) Vt(t)
+ Kamp Kex Kg III. TUNING OF PID CONTROLLER
Vref(t) - samp  1 sex  1 sg  1 The important objective of this section is PID controller
Vf(t)
design and testing its performance with set-point change. A
Kvs PID controller is a control loop mechanism that is widely used
svs  1 in industrial control systems. When the PID controller is used,
the overshoot increases and the error at steady-state will
Fig. 2. Closed loop block diagram representation of AVR.
decrease (but not entirely) with the increase in proportional
gain. However, with the addition of derivative, the overshoot
A. Transfer function model of AVR system decreases but there is no change in steady-state error. When
The AVR system mainly consists of a voltage sensor to the integral term is added, the controller will mitigate the
sense voltage magnitude on any one phase, an Amplifier (to steady state error completely/partially. In this research, the
strengthen the error signal), Exciter (to adjust the field error signal, e(t) (controller input) is defined by taking the
excitation), and a Generator to feed the power to the 3-Ph difference between the set-point voltage (Vr) and the generator
utility grid through a 3-Ph transformer. The transfer function terminal voltage (Vt).
model representation of these components is as follows:
The voltage sensor is modeled 1st order transfer function A. Internal Model Control – Direct Synthesis (IMC - DS)
with gain Kvs and time period τvs.
In this tuning method, by proper section of required
Kvs closed-loop response and time delay using Pade´
Gvs(s)  (1)
sτvs  1 approximation, PID controller coefficients are derived. Let a
The first order transfer function model of an Amplifier required closed-loop transfer function with delay time (θ) and
with gain KAmp and time constant TAmp is given as tuning parameter (τc) as
KAmp y(s) 1
GAmp(s)  (2)  e θs (7)
sτAmp  1 ref(s) τcs  1
The modern AC generator uses a brushless excitation system. The Closed Loop Transfer Function for set point variations is
The transfer function model representation of an exciter is as
below with a gain Kex, time constant τex. y(s) Gp(s)Gc(s)
 (8)
Kex ref(s) Gp(s)Gc(s)  1
Gex(s)  (3)
sτex  1 Re-arranging the equation (8) for the feedback controller,
The first order transfer function model of an AC generator
with a gain Kg, time constant τg is given as below.

Authorized licensed use limited to: Sri Sivasubramanya Nadar College of Engineering. Downloaded on May 06,2022 at 06:36:40 UTC from IEEE Xplore. Restrictions apply.
y(s) stability is poor, reduce the controller gain (i.e kc0/4) and
ref(s) increase the integral time constant proportionally (i.e τi/4),
Gc(s)  (9) repeat procedure from Step-III.
 y(s) 
Gp(s)1   1.5
τou
 ref(s)  Set-point
Comparing and rearranging equations (8) and (9)

Plant response
1
1
Gc(s)  e θs (10)
Gp(s)(τc  θ)s Plant response
0.5
But the standard form of 2nd order + time delay model is
k
Gp(s)  e θs (11) 0
(τ 1  1)(τ 2 s  1) 0 1 2 3 4 5 6
Time (sec)
Comparing equations (10) & (11), the coefficients of the PID
controller are Fig. 3. Plant response for P-mode controller (Good gain tuning)

1  τ  τ2   τ τ 
 , Ki  Kp τ1  τ 2  and K d  Kp  1 2 
D. Pattern Search Algorithm (PSA)
Kp   1 τ τ 
k  θ  τ c   1 2  This algorithm is initiated by creating mesh (set of points)
around the given point (called as current point) for the
B. Skogestad Internal Model Control (S IMC) minimization of the objective function. The starting point set
by the user or set from the previous step of the algorithm is
This tuning method is preferred when a fast response with called the current point. The mesh is created by the addition
good robustness is required. The tuning procedure of this of the current point to a pattern. When the improvement in the
method is similar to tuning procedure of IMC-DS tuning, but objective function at a current point is identified by a point in
in this method the tuning parameter τc = θ. The coefficients of mesh, then this current point is treated as a new point for the
PID controller are, where τi = min(τi, 8θ). next iteration and so on. This search algorithm will continue
0.5τ1 until an optimal solution is obtained with minimized objective
Kp  , Ki  Kp/τi and Kd  Kp τ 2
kθ function [10, 12]. Fig. 4 shows the flow chart of Pattern
Search Algorithm.
C. Good Gain tuning
In this method of tuning, the controller coefficients Start

evaluation purely depends on closed-loop experiments,


Initialize mesh size, Max. generations (Gen),
keeping the controller in proportional mode (P-mode). This mesh expansion factor, mesh contraction factor
method is simpler than other closed-loop tuning methods. It
does not require any advanced data processing; it needs data
from simple experiments conducted by the user. The following Set the initial/starting point

algorithm is used to tune the controller.


Construct pattern vectors
Step-I: Make sure that the controller is P-controller by putting and create mesh points

any ‘kc0’ value as Kp, keep Ki and Kd as ‘0’.


Step-II: Change the kc0 value until some small overshoot Calculate the objective function
(generally, 5 to 20% of set-point range) is obtained. Now
Gen = Gen +1 Gen = Gen +1
measure the period between the 1st peak overshoot, and
undershoot of the response and is represented as τou. Is desirable Yes
Condition Stop
Step-III: The proportional gain constant (Kp) is obtained by reached

taking 80% of P-controller gain (kc0). Because the closed


system stability may decrease with the addition of integrator No

term with P-mode controller, hence to compensate for this, the


Yes No
Kp should decrease. Therefore Kp= 0.8kc0. Expand Mesh size
Is the pole
Construct Mesh size
is successful?
Step-IV: Calculate integral and derivative time constants as τi
= 1.5τou and τd = 0.25τi.
Step-V: Now observe the closed-loop system stability(i.e 1st Fig. 4. Pattern Search Algorithm (PSA) – Flow chart
peak over-shoot and ts) with the above PID settings, if system
stability is poor, reduce the controller gain (i.e kc0/4) and In the design of the Pattern search algorithm based PID
increase the integral time constant proportionally (i.e τi/4), controller, the objective function is defined based on the
repeat procedure from Step-III. desired specification. In this work, the desired specification is
the minimization of voltage error. The performance criteria
Step-V: Now observe the closed-loop system stability (i.e 1st usually considered in the control design is the Integral of Time
peak over-shoot and ts) with the above PID settings, if system multiplied Absolute Error (ITAE).

Authorized licensed use limited to: Sri Sivasubramanya Nadar College of Engineering. Downloaded on May 06,2022 at 06:36:40 UTC from IEEE Xplore. Restrictions apply.
IV. PYHTON CONTROL TOOLBOX: EVALUATION OF PID 10
Set-point
COEFFCIENTS AVR response without controller

Terminal voltage (Vt)


8
(Open loop)
Python is the most popular and general-purpose 6
programming language for scientists, engineers, and many
because of its adjustable, and well-built analysis packages 4

such as Numpy and Scipy. It is a good tool for solving all 2


technological and mathematical issues of academicians,
researchers, etc. The main advantage of Python control 0
0 5 10 15
toolbok over MATLAB toolbox is that it is a free-ware and is Time(sec)
available on standard platforms (like Windows, MacOS, and Fig. 5. AVR Open-loop step response (without controller)
Linux). Clear and simple syntax is another advantage of 2
Set-point
Python’s language. AVR response without controller

Terminal voltage(Vt)
1.5

A. Introduction to Python Control Toolboxs


1
The Python control system toolbox consists of python
classes and functions that are used for the design and analysis 0.5
of feedback control systems. This toolbox has a MATLAB
compatibility module that provides many functions same as 0
0 5 10 15
functions in MATLAB control system toolbox [18]. The Time(sec)
python control package takes the help of NumPy and SciPy. It Fig. 6. AVR Closed-loop step response (with controller)
is a good programming language than MATLAB i.e
MATLAB’s syntax is more difficult than NumPy for the same From Fig. 6, without a controller, the AVR system has an
array manipulations. overshoot of 50.4%, the response reaches the steady-state at
The Python control system toolbox can be installed using 5.87sec with steady-state error of 0.091(pu) i.e utility grid
‘pip’, ‘conda’ setup tools mechanisms. This package required terminal voltage fall by 9.1%. But, practically ±6% of terminal
numpy, scipy and matplotlib (for plots). NumPy can perform voltage variations (maximum) is allowed. If the value ±8% is
many things than the MATLAB just cannot. Once this
allowed, it leads to problems for motor loads and also
toolbox is installed, use the following two ways to use the
package. represents continuous power loss. Now this voltage drop is
mitigated by controlling the field excitation as explained in
>>> import control - Default interface in function reference section 2. In this section, the coefficients of various PID
>>> from control.matlab import * - for MATLAB controllers (which are discussed in section 3) are evaluated
using the Python control toolbox.
Code 1 and Code 2 depict the Python codes to get the step
response of open-loop and closed-loop Automatic Voltage B. Evaluation of PID coefficients in Python
Regulator (AVR) system respectively. Fig. 5 and Fig. 6 show
the AVR’s open-loop and closed-loop step responses a) Internal Model Control –Direct Synthesis tuning: To
respectively. evaluate the PID coefficients using this method, it is required
to approximate the 4th order OLTF of the AVR, in equation
1 importnumpy as np (6), as 2nd order + time delay model using the half rule. Using
2 importmatplotlib.pyplotasplt the half rules
3 # Transfer function representation
4 num=[10]
System Gain, k = 10, Delay time θ (theta) = 0.06, Time
5 den=[0.0004, 0.0454, 0.555, 1.51, 1] constants τ1 = 1sec and τ2 = 0.45sec
6 sys=signal.TransferFunction(num,den) The approximated 2nd order + time-delay model of the AVR
7 t,y = signal.step(sys) # Step response
system transfer function is
8 plt.plot(t,y)
Vt(S) 10
e 0.06s
9 plt.xlable(‘Time(sec)’)
10 plt.ylable(‘Terminal voltage (Vt)’)
 (12)
Vref(S) (s  1)(0.45s 1)
Code 1. Python code for step response (open loop) of AVR Code 3 is used to evaluate the PID coefficients with the
1 importnumpy as np proper selection of tuning parameter τc.
2 importmatplotlib.pyplotasplt
3 # Transfer function representation 1 defpid(k, theta, tau1, tau2, tauc): #Define controller
4 2 # PID calculation
num=[0.1, 10]
5 3 Kc = (1/k)*(tau1 + tau2)/(tauc + theta)
6 den=[0.0004, 0.0454, 0.555, 1.51, 11] 4 Kp = Kc
7 sys=signal.TransferFunction(num,den) 5 Ki = Kc /(tau1 + tau2)
8 t,y = signal.step(sys) # Step response 6 Kd = Kc *((tau1 * tau2)/(tau1 + tau2))
9 plt.plot(t,y)
10 7 a = pid(10, 0.06, 1, 0.45, 0.15) # Initialization
plt.xlable(‘Time(sec)’) 8 print(a)
Code 2. Python code for step response (closed loop) of AVR
Code 3. PID coefficients evaluation: Internal Model Control – Direct

Authorized licensed use limited to: Sri Sivasubramanya Nadar College of Engineering. Downloaded on May 06,2022 at 06:36:40 UTC from IEEE Xplore. Restrictions apply.
The open-loop response of the original and 2nd order + performance will occurs for τc = 0.15sec, the corresponding
time-delay AVR model is as shown in Fig.7. The time domain coefficients are Kp = 0.6905, Ki = 0.4762 and Kd = 0.2143.
performance of the AVR system for various value of τc are
tabulated in TABLE II. b) Skogestad Internal Model Control tuning: This
10
tuning method is used when the fast response with good
robustness is required. The process for identification of system
Original model
8
parameters is same as IMC-DS tuning, hence the system gain
Terminal voltage (Vt)

2nd order + time delay model


6 k = 10, delay time θ(theta) = 0.06, and time constants τ1 = 1sec
and τ2 = 0.45sec.
4
1 defpid(k, theta, tau1, tau2) : # Define PID controller
2 3 Kc = tau1/(2*k*theta)
4 Kp = Kc
0
0 5
Time (sec)
10 15 5 taui = 8*theta# taui = min[τ1, 8θ],τc = θ
6 Ki = Kc /taui
nd
Fig. 7. Response of the original and 2 order + time-delayed AVR models 7 taud = tau2
8 Kd = Kc *taud
TABLE II indicates that, with the increase in tuning
9 a = pid(10, 0.06, 1, 0.45) # Initialize parameters
parameter (τc), the overshoot and settling time (ts) decreases 10 print(a)
but the rise time (tr) increases. A better time domain
Code 4. PID coefficients evaluation – Skogestad Internal Model Control

TABLE II. PID COEFFICIENTS AND TIME DOMAIN RESPONSE OF AVR FOR DIFFERENT VALUES OF ΤC

Controller Coefficients Time-domain specifications


Tuning parameter
Kp Ki Kd tr(sec) %Overshoot ts (sec) *
τc = 0.1 0.9062 0.625 0.2813 0.31 12.1 0.84
τc = 0.15 0.6905 0.4762 0.2143 0.41 5.30 0.64
τc = 0.2 0.5577 0.3846 0.1731 0.55 1.03 0.55
*
± 3% variations allowed

c) Good Gain tuning: As explained in section 3, the The objective function is prepared in time and error
response is obtained with the controller in P-mode, the signal (difference in set-point and terminal voltage) as
proportional gain (kc0) is identified as 0.625 to achieve the t 
overshoot as 20% approximately. The period between the 1st Objective function (ITAE) =   t.Vref - Vt dt (13)
peak overshoot and undershoot of the step response is 0 
represented as τou, from the response τou = 0.86sec.
V. SIMULATION RESULTS AND DISCUSSIONS
1 defpid(kc0, tauou) : # Define PID controller
2 # PID calculation
Fig. 8(a) and 8(b) shows the change in terminal voltage
3 Kp = 0.8*kc0
w.r.t time for various PID controllers.
4 taui = 1.5* tauou
5 Ki = Kp/taui
Change in Terminal voltage

6 taud = 0.25*taui 1
7 Kd= Kp*taud 0.8
8 a = pid(0.625, 0.861) # Initialize parameters
0.6
9 print(a)
0.4 Set-point
Code 5.PID coefficients evaluation – Good gain method
0.2 IMC (DS)
Python code 5 is used to evaluate the PID coefficients. The 0
S-IMC
PID coefficients are Kp = 0.50, Ki = 0.3407 and Kd = 0.1421. 0 1 2 3 4 5
Time (sec)
d) Pattern Search Algorithm: As discussed in section 3, Fig. 8(a). Change in terminal voltage (IMC-DS, S-IMC tuning)
the following objective function is considered for tuning the
Change in Terminal voltage

PID controller coefficients using this method. In this research 1


work, to implement this algorithm number of generations 0.8
(Iterations) are 100, mesh size is initiated as 1, mesh 0.6
expansion and mesh contraction factors are initiated as 2 and 0.4 Set-point
0.5 respectively, number of objective function evaluations are
0.2 Good Gain
taken as 100. After successful implementation of this Pattern Search
algorithm, the coefficients of the PID controller are Kp = 0
0 1 2 3 4 5
0.582, Ki = 0.42 and Kd = 0.181. Time (sec)
Fig. 8(b). Change in terminal voltage (Proposed tuning methods)

Authorized licensed use limited to: Sri Sivasubramanya Nadar College of Engineering. Downloaded on May 06,2022 at 06:36:40 UTC from IEEE Xplore. Restrictions apply.
TABLE III shows the comparison in overshoot (1 st peak), S-IM C
rise and settling times for different controllers. 0.8

Error signal (Vt-Vref)


IM C(DS)
TABLE III. COMPARISON OF TIME DOMAIN PERFORMANCE OF AVR SYSTEM 0.6 Pattern Searh
Good Gain
Time domain performance 0.4
Type of controller
%Overshoot tr (sec) ts (sec) * 0.2

Internal Model control - DS 5.36 0.41 0.667 0

Skogestad-IMC 14.86 0.26 1.788 -0.2


0 1 2 3 4 5
Good Gain method 1.10 0.482 0.682 Time (sec)
Pattern Search algorithm 1.72 0.54 0.597 Fig. 9(b). Variations in error signal e(t)
*
±3% variations allowed
From Fig. 9 (b), the proposed controllers provide a smooth
The results shows that the proposed controllers possess a control signal for control of field current of the AC generator.
small overshoot (1st peak) and settling time even though the
rise time is more compare to other methods. Fig. 9(a) and 9(b) A. Controller robustness verification
shows the variations in control signal (PID output) and error
The robustness of the proposed controller is studied for
signal (deviation of terminal voltage from reference voltage)
parameter’s uncertainties and External noise/ disturbances.
and respectively.
40
The proposed controller’s robustness is analyzed by finding
Good gain the uncertainties of the parameter (within the range shown in
Control Signal (PID output)

30 IM C (DS) TABLE I) for various first peak overshoots because the


20
Pattern Search stability of the system mainly influenced by the first peak
S-IM C overshoot. Usually, the overshoot is associated with a period
10 that requires reaching its steady-state i.e settling time.
0 In this study, the settling times are measured by allowing
±3% terminal voltage variations from set-point at steady state.
-10 TABLE IV - VI show the maximum allowable uncertainty
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Time(sec) factors (compare to its nominal values) and its corresponding
Fig. 9(a). Variations in control signal (controller output) rise time and settling times of Amplifier, Exciter, and
Alternator for Overshoots of 25%, 35%.
TABLE IV. COMPARISON OF UNCERTAINTY FACTOR, CORRESPONDING RISE TIME, SETTLING TIME FOR 25% OVERSHOOT (AMPLIFIER PARAMETERS VARIATIONS)

Uncertainties in Amplifier gain (Kamp) Uncertainties in Amplifier Time constant (τamp)


Controller type
Uncertainty factor tr (sec) ts (sec)* Uncertainty factor tr (sec) ts (sec)*
IMC (DS) 2.012 0.218 0.80 2.95 0.506 1.877
S-IMC 1.39 0.198 1.57 1.67 0.30 1.688
Good Gain 2.773 0.227 0.836 4.01 0.72 2.695
Pattern search 2.375 0.217 0.796 3.57 0.61 2.236
TABLE V. COMPARISON OF UNCERTAINTY FACTOR, CORRESPONDING RISE TIME, SETTLING TIME FOR 25% OVERSHOOT (EXCITER PARAMETERS VARIATIONS)

Controller Uncertainties in Exciter gain (Ke) Uncertainties in Exciter Time constant (τe)
type Uncertainty factor tr (sec) ts (sec)* Uncertainty factor tr (sec) ts (sec)*
IMC (DS) 2.01 0.218 0.80 3.95 0.853 3.86
S-IMC 1.39 0.198 1.57 2.17 0.422 3.24
Good Gain 2.78 0.227 0.836 3.85 1.07 4.68
Pattern search 2.375 0.217 0.796 3.832 0.945 4.31

TABLE VI. COMPARISON OF UNCERTAINTY FACTOR, CORRESPONDING RISE TIME, SETTLING TIME FOR 35% O1VERSHOOT (AMPLIFIER PARAMETERS VARIATIONS)

Controller Uncertainties in Exciter gain (Ke) Uncertainties in Exciter Time constant (τe)
type Uncertainty factor tr (sec) ts (sec)* Uncertainty factor tr (sec) ts (sec)*
IMC (DS) 2.01 0.218 0.80 3.95 0.853 3.86
S-IMC 1.39 0.198 1.57 2.17 0.422 3.24
Good Gain 2.78 0.227 0.836 3.85 1.07 4.68
Pattern search 2.375 0.217 0.796 3.832 0.945 4.31
*
±3% variations allowed

Authorized licensed use limited to: Sri Sivasubramanya Nadar College of Engineering. Downloaded on May 06,2022 at 06:36:40 UTC from IEEE Xplore. Restrictions apply.
B. External Noise/Disturbances References
In this case, to study the robustness of the controller, [1] Celik Durgut R, “Performance enhancement of automatic voltage
regulator by modified cost function and symbiotic organisms search
external noise/ disturbances are applied to the AVR as shown algorithms”, Engineering science and Technology, International Journal,
in Fig. 10. Issue, Vol.21, pp. 1104-1111,Oct 2018.
External disturbances [2] Kansit S, Wudhicahai, “Optimization of PIDcontroller based PSOGSA
for an Automatic Voltage Regulator system”, 2016 International
-
Vt(t) e(t) + Vref(t) Electrical Engineering Congress, iEECON2016, pp.87-90, Chiang Mai,
+ PID
AVR System Thailand, 2-4 March 2016.
Controller
[3] Ziegler J.G, Nichols N.B, “Optimum Settings for Automatic
-
Controllers”, Trans. ASME, Vol. 64, pp. 759-768, November 1942.
Vf(t)
Voltage [4] Cohen G.H, Coon G.A, “Theoretical Consideration of Retarded
sensor Control”, ASME, Vol.75, pp. 827-834, 1953.
Fig. 10. Closed-loop AVR system with disturbances/noise [5] Dan C, Dale E.S, “PI/PID Controller Design based on Direct Synthesis
and Disturbance Rejection”, Ind. Eng. Chem. Res. Vol. 19, pp. 4807-
Fig.11 shows the response of change in terminal voltage of 4822,August 2002.
a closed-loop AVR system with Band limited noise/ [6] Lee J, Cho W, and Edgar T, “An Improved Technique for PID
disturbance. The Band limited noise/disturbance is defined Controller Tuning from Closed- Loop Tests”, AIChE Journal, Vol. 36,
pp. 1891-1895, September 1990.
with noise power as ‘0.075’and sample time as ‘0.25sec’. [7] Yuwana M. and Seborg D, “New Method for On-Line Controller
From Fig. 10, it is clear that when a band limit noise type Tuning”, AlChE Journal, Issue 3, Vol. 28, pp. 434- 440,1982.
disturbance is applied, both proposed PID controllers rejects [8] Mohammad S, Sigurd S, Ivar J.H, “A simple approach for on-line PI
the disturbances smoothly and quickly. controller tuning using closed-loop setpoint responses”, 20th European
Symposium on Computer Aided Process Engineering, ESCAPES20,
1.5 pp.619-624, Ischia, 6-9 May 2010.
Change in Terminal voltage

1 [9] Finn Haugen, “The Good Gain method for PI(D) controller tuning. Tech
Teach”, July 2010.
0.5
[10] Nahvi H, Silani S, “ Using Pattern Search algorithm and Finite element
0 method to detect rotor cracks”, International Journal of Engineering,
-0.5 Transactions A: Basics, Issue 2, Vol.22, pp. 195-204, June 2009.
Disturbances/Noise
[11] S. Ciba, I. Iskender, H.A Ariani, “Comparing control performance of
-1 Good Gain
Pattern search
MRAC and PID applied on a brushless DC motor”, International Journal
-1.5 on Technical and Physical Problems of Engineering (IJTPE), Issue 44,
0 2 4 6 8 10
Vol. 12, No. 3, pp. 10-15, September 2020.
Time (sec)
[12] Zewail I. et al. “Maximization of Total throughput using Patter Search
Fig. 11. AVR system response with Band limited noise algorithm underlay Cognitive radio network”, Menofia journal of
Electronics Engineering Research, Issue 2, Vol. 26, pp. 307-319,
Conclusions November 2017.
[13] Mohamma A.S et al “Closed loop tuning of cascde controllers based on
Python is a user-friendly programming tool for solving all set point experiment”, Journal of Engineering Research, Issue 4, Vol.8,
kinds of control engineering issues. Its syntaxes and usage are pp.117-138, December 2020.
simpler than similar programming tools. The controller [14] Ercan K. “Optimal control of AVR system with Tree seed algorithm-
coefficients are evaluated in the Python environment. In this based PID controller”, IEEE Access, Vol.8, pp. 89457-89467, May
work, a new tuning method (Good gain, Pattern Search) has 2020.
been introduced for tuning the coefficients of a PID controller [15] Mohanty P.K., Sahu B.K., Panda S, “Tuning and assessment of
for the Automatic Voltage Regulator. It is observed that even proportional integral–derivative controller for an automatic voltage
regulator system employing local unimodal sampling algorithm”,
though S-IMC tuning’s rise time is better than the proposed Electric Power Components and Systems, Issue 9, Vol. 42, pp. 959–969,
methods, the proposed methods are less sensitive to parameter May 2014.
uncertainties and external disturbances. [16] Shayeghi H, Younesi A, Hashemi Y, “Optimal design of a robust
discrete parallel FP + FI + FD controller for the Automatic Voltage
From the simulation results, the overshoots and settling Regulator system”, International Journal of Electrical Power and Energy
times of the controlled system are optimized with the Good Systems, Vol.67, pp.66–75, May 2015.
Gain method and Pattern search algorithm superior to other [17] M. Topal, I. Iskender, N. Genc, “Sensor less speed control of BLDC
tuning methods. Also, the proposed controllers provide good motor using improved sliding mode observer technique”, International
Journal on Technical and Physical Problems of Engineering (IJTPE),
robustness for parameter uncertainties and external Issue 38, Vol. 11, No. 1, pp. 1-9, March 2019.
disturbances. [18] https://python-control.readthedocs.io/en/0.8.4/index. html.

Authorized licensed use limited to: Sri Sivasubramanya Nadar College of Engineering. Downloaded on May 06,2022 at 06:36:40 UTC from IEEE Xplore. Restrictions apply.

You might also like