Sie sind auf Seite 1von 8

CSA based PID Controller Design Technique for

Optimizing Various Integral Errors


1 3
Amrit Kaur, 2Ranjit Kaur Swati Sondhi
Department of Electronics and Communication Department of Electrical and Instrumentation Engineering,
Engineering, Thapar Institute of Engg. and Tech,
Punjabi University, Patiala, Punjab
3
Patiala, Punjab Swati.sondhi@thapar.edu
2
1
amritameek@gmail.com, ranjit24_ucoe@pbi.ac.in

Abstract— Control design plays a significant role in almost has time delay. Bialkowski reported an audit of paper mills
all types of industries. Proportional-integral-derivative (PID) in Canada showing that a typical mill has more than 2000
controllers are an integral part of process control loops. PID control loops and 97% used Proportional-Integral-Derivative
controllers are popular for their simplicity of implementation control. From these, only 20% of the controllers were found
and broad applicability. In recent years, various metaheuristic to work well, 30% were found to be poorly tuned [2].
algorithms and modified hybrid algorithms have been applied Lot of research work has been done to find a more
to design the controllers. The aim of this paper is to design a
systematic approach for tuning the PID parameters efficiently
controller with high versatility, accuracy and good control
quality. In this research paper, first, a novel tuning method and make it more robust and powerful. With the development
based on Crow Search Algorithm (CSA) is proposed to optimize and application of fast and improved optimization algorithms
parameters of PID controller: Kp, Ki and Kd. Each crow in the past decades, tuning of PID controller has reached
represents a feasible solution for the PID parameters. Second, another level. To enhance the capabilities of traditional PID
four objective functions have been explored and the controller tuning techniques, various intelligent approaches
effectiveness and convergence rates of CSA-PID controller is have been suggested as in [3] PID controller has been
evaluated therein for two different control problems. Last, optimized with Ant algorithm, [4] presented a study for the
comparison has been carried out between CSA optimized PID methodology and application of Swarm intelligence for the
The main advantage of CSA is its simplicity, faster convergence
tuning of the PID controllers, [5] treated tuning of PID DC
rate, ease of implementation and easy understanding. As per
findings based on statistical analysis, Crow search Algorithm Motor Speed controllers method using Multi-Objective
(CSA) has been found to be more reliable. Simulation results Differential Evolution (MDE), Multi-objective Ant Colony
based on two control problems and four evaluation functions (MACO) and a hybrid method based on Multi-objective Ant
have been tested for set point tracking, load rejection capability, Colony Optimization and Differential Evolution,
noise suppression and modelling errors. (MACODE). [6] tuned PID parameters using modified Bat
Algorithm, [7] improved the Velocity Update Relaxation
Keywords—Crow Search Algorithm (CSA), Integral square Particle Swarm Optimization algorithm (VURPSO), called
error (ISE), Integral time square error (ITSE), Integral average Adaptive VURPSO (AVURPSO) algorithm.
error (IAE), Integral time average error (ITAE).
Current trend is to utilize nature-inspired metaheuristic
I. INTRODUCTION algorithms to tackle difficult problems and are found to be
surprisingly very efficient [8]. Crow Search Algorithm
Controllers play an important part in process industries. (CSA) was proposed by Askarzadeh in 2016 [9], which is a
Since invention of Proportional-Integral-Derivative (PID) user-friendly metaheuristic technique based on Crows
control in 1910 and the Ziegler–Nichols (Z-N) straight intelligence and memorization power.
forward tuning methods in 1942 [1], the popularity of PID This paper attempts to apply Crow Search Algorithm
controllers has grown tremendously. Despite a simple (CSA), a nature inspired algorithm, to optimize the PID
structure they can effectively control a very large number of parameters for two different control problems with two
process control loops and processes with uncertainty [2]. different objective functions. This algorithm has two
When applied to a certain process, the efficiency of controller adjustable parameters: Awareness Probability (AP) and
depends on its effective tuning. Tuning refers to “best” Flight length (fl). So, to find the effectiveness of proposed
adjustment of the controller parameters i.e. Kp, Ki and Kd for controller, the results of CSA-PID based for integral
PID controller. There are variety of performance criteria for evaluation functions are compared in terms of time response
tuning of PID controllers like rise time, settling time characteristics, Integral index values, set point tracking, load
overshoot, steady state error and integral indices. Classical rejection capability, noise suppression and modelling errors.
methods, like Zeigler-Nichols (Z-N) and Cohen Coon (C-C) for the four objective functions as well as on two different
methods have been used widely for tuning of PID controller. control problems.
But prior knowledge of plant model is a major drawback of The main structure of the paper consists of three Sections.
these techniques. Further, tuning a PID controller manually Section I introduces the work done on optimization based
is difficult as the plant order goes high, becomes nonlinear or PID controller; Section II gives brief overview of CSA for

978-1-7281-2791-0/20/$31.00 2020
c IEEE 61
tuning of PID controller. Section III discusses comparative memorizing the location of hidden food for months. It is also
Analysis of CSA-PID for four objective functions as applied a population-based algorithm which increases the probability
on two different control problems. Section IV concludes this of finding a good solution and escaping from local optima.
paper. CSA have only two adjustable parameters as compared to
three adjustable parameters in Harmony Search, four
II. TUNING OF PID USING OPTIMIZATION adjustable parameters in Particle Swarm Optimization (PSO)
METHODS and six adjustable parameters in Genetic Algorithm (GA)
In comparison to traditional methods of PID tuning, which makes it simpler to use and apply to control problems.
optimization algorithms have been found to give better Pseudo code of CSA-PID is given as under:
results when applied to PID. However, every method has CSA-PID
some disadvantage or limitation [10]. For example, if the INPUT:
controller provides faster response then stability is affected x Number of iterations (itermax)
and vice versa. Hence, a compromise is made between
x Flock size of Crows (N)
acceptable stability and medium speed of response. Overall,
x Awareness Probability (AP)
researchers have not been able to achieve both faster response
x Flight length (fl)
and good stability simultaneously in relation to the tuning of
x Dimensions (d)
PID controller.
INITIALIZATION:
The objective function plays an important role in
effective design of an optimized PID controller. There are x Randomly generate initial position of N crows in d
often four indices to depict the system performance dimensions i.e. the values of Kp, Ki and Kd are generated for
considering the entire closed loop response: integral of square N crows.
error (ISE), integral of absolute error (IAE), integral of time x Put these initial positions in memory of the crows.
square error ITSE and integral of time absolute error (ITAE). x Calculate the performance of control system with N different
Literature suggests [11], that single objective optimization PID controllers based on the fitness function value.
algorithm may not give the best solution for the controller x Use equation (1) - equation (4) to calculate fitness function
optimization. Better performance can be achieved by value (F) and save minimum value as fmin.
integrating the individual objective functions with proper x For t=1:itermax
weights which is used as a weighted aggregation method for x Generate random candidate crows for chasing.
multi-objective optimization problems. Consequently, CSA x For i=1:N
optimized PID have been simulated for following four multi- x if random ≥ AP
objective functions given by equations (1) and (2): ‫ܭ‬௣௜ǡ௜௧௘௥ାଵ ൌ ‫ܭ‬௣௜ǡ௜௧௘௥ ൅ ‫݁ݑ݈ܽݒ݉݋݀݊ܽݎ‬
‫ܭ‬௜௜ǡ௜௧௘௥ାଵ ൌ ‫ܭ‬௜௜ǡ௜௧௘௥ ൅ ‫݁ݑ݈ܽݒ݉݋݀݊ܽݎ‬
‫ܨ‬ଵ ൌ ‫ ܧܵܫ כ ܤ‬൅ ሺͳ െ ‫ܤ‬ሻ ‫ܶ כ‬௦ ‫ܯ כ‬௣ (1) ‫ܭ‬ௗ௜ǡ௜௧௘௥ାଵ ൌ ‫ܭ‬ௗ௜ǡ௜௧௘௥ ൅ ‫݁ݑ݈ܽݒ݉݋݀݊ܽݎ‬

‫ܨ‬ଶ ൌ ‫ ܧܵܶܫ כ ܤ‬൅ ሺͳ െ ‫ܤ‬ሻ ‫ܶ כ‬௦ ‫ܯ כ‬௣ (2) **else


‫ܭ‬௣௜ǡ௜௧௘௥ାଵ ǡ ‫ܭ‬௜௜ǡ௜௧௘௥ାଵ ǡ ‫ܭ‬ௗ௜ǡ௜௧௘௥ାଵ ൌa random position of search space
‫ܨ‬ଷ ൌ ‫ ܧܣܫ כ ܤ‬൅ ሺͳ െ ‫ܤ‬ሻ ‫ܶ כ‬௦ ‫ܯ כ‬௣ (3) **end
CALCULATE:
‫ܨ‬ସ ൌ ‫ ܧܣܶܫ כ ܤ‬൅ ሺͳ െ ‫ܤ‬ሻ ‫ܶ כ‬௦ ‫ܯ כ‬௣ (4)
If fitness function value of the PID controller with new positions is coming
Where, TS is the settling time, Mp is the maximum overshoot, better than the previous value
and B is the weighting factor. From the findings in literature x set [bestfood bestindex]=min(F)
it has been established that though ISE penalizes large errors x if bestfood is better than fmin
better than the smaller ones as it considers the square of error x update ‫ܭ‬௣௡௘௪ ൌ ‫ܭ‬௣ ሺ„‡•–‹†‡šሻ, ‫ܭ‬௜௡௘௪ ൌ ‫ܭ‬௜ ሺ„‡•–‹†‡šሻܽ݊݀ ‫ܭ‬ௗ௡௘௪ ൌ
into consideration of a plant output, IAE takes the average ‫ܭ‬ௗ ሺ„‡•–‹†‡šሻ
error over time, ITSE and ITAE, are more intensive in x update memory
reducing error when time t is high. So, it helps to damp out ௜ǡ௜௧௘௥ାଵ ௜ǡ௜௧௘௥ାଵ ௜ǡ௜௧௘௥ା
݉௜ǡ௜௧௘௥ାଵ ൌ ‫ܭ‬௣௡௘௪ ‫ܭ‬௜௡௘௪ ‫ܭ‬ௗ௡௘௪
oscillations faster. As our aim is to reduce error, ܶ௦ and end
maximum overshoot ‫ܯ‬௣ . So, in this study, four independent
objective functions consisting of two parts are joined with a
Pseudo code of CSA-PID
weighing factor B whose value is taken as 0.75. The first part
of objective function consists of integral index term Crows move in the environment and search for better food
(ISE/IAE/ITAE/ITSE) and second part includes settling time sources. Each crow memorizes the position or location of its
ܶ௦ and maximum overshoot ‫ܯ‬௣ . best hiding place. CSA randomly generates the initial
position and memory of each crow. As shown in Fig.1, in
A. Overview of Crow Search Algorithm (CSA) CSA Awareness Probability (AP) controls the intensification
Crow Search Algorithm (CSA) proposed by Askarzadeh and diversification, while the value of flight length (fl) affects
in 2016 [9] is a new method for finding global optimization the search capability of the crows.
based on intelligence of crow and its capability of

62 10th International Conference on Cloud Computing, Data Science & Engineering (Confluence)
B. Tuning of PID based on CSA (CSA-PID) With these parameters the performance criterion in terms of
In this section, CSA has been used to tune the PID MP, TR and TS are recorded and corresponding response is
parameters: Kp, Ki and Kd. For each crow the quality of its plotted. The results, figures and discussions are as follows:
position is computed by inserting the decision variable values
into the objective function. New solutions are generated and A Comparison in terms of MP, TR and TS
the feasibility of each crow position is checked. If it is Fine tuning of controller parameters is highly difficult, since
feasible, the crow updates its position i.e. solutions are increase in time delay narrows down the performance of
updated otherwise, the crow stays in the current position and closed loop system under control. First, the open loop
does not move to new position. The fitness function value for response is calculated and time domain specifications in
the new solutions is computed. If this value is better than the terms of MP, TR and TS of both the systems presented in
fitness function value of the memorized position, the crow equations (7)-(8) without PID controller are recorded in
updates its memory by the new position. Table II. Next, the results are optimized using CSA-PID
When maximum iteration is reached, the best based on four fitness functions to get the best response in
position of the memory in terms of minimum value of fitness terms of speed and accuracy. The optimized values of PID
function, fmin is reported as the solution for the optimization are presented in Table II in terms of TR, TS, Peak value and MP
process. Corresponding to this fmin the position of the crow for both the plants and their corresponding characteristics are
in d dimensions is taken as the final optimized values for the presented in Figure 1-2 and values are tabulated in Table II.
PID controller and the time response of the plant with It is found that CSA-ISE and CSA-IAE are giving the best
optimized PID controller is calculated. values in terms of time domain characteristics as well as
The parameters for this comparison study are taken integral indices values.
as listed in Table I. For all four controllers based on equations
(1)-(4), the position values corresponding to fmin
individually are taken as the values of PID controller and time
response is found for two plants whose open loop response is
first calculated and the results are continuously compared to
get the best response in terms of speed and accuracy. For both
algorithms, maximum of 100 iterations are taken for 40 Magnitude
independent trials of both these algorithms individually.
TABLE I. CSA PARAMETERS
Parameter Range Parameter Range/Value
Kp [0 10] flock size (N) 10
Ki [0 10] Awareness 0.4
Probability (AP)
Kd [0 10] flight length (fl) 1

III. EXPERIMENTAL RESULTS AND Fig.1. Nominal output for Plant 1


DISCUSSION
As part of the methodology, to compare the
advantages of the proposed integral error-based CSA-PID
technique, these four controllers are implemented on two
different control problems in order to study the effect on
response characteristics with optimized PID.
Magnitude

The best values of PID parameters are taken when


optimized by CSA-ISE, CSA-ITSE, CSA-IAE and CSA-
ITAE for third order system named as Plant-1 given by
(7)[12] and resonant time delay system named as Plant-2
given by (8) [13].
Plant 1: Third order system given by [12]:

ͶǤʹʹͺ Fig. 2. Nominal output for Plant 2



ሺ•ሻ ൌ (7)
• ଷ ൅ ʹǤͳͶ• ଶ ൅ ͻǤʹ͹• ൅ ͶǤʹʹͺ
B Load disturbance rejection and Control Effort
Plant 2: Resonant time delay system given as [13]:
To check the load disturbance rejection capability, all four
controllers based on CSA-PID are provided with a load
‡ሺି଴Ǥଵୱሻ disturbance at t=1 sec of 0.5V magnitude in Plant-1and Plant-
‫ܩ‬ሺ‫ݏ‬ሻ ൌ (8)
•ଶ ൅ ͲǤͲʹ• ൅ ͳ 2. It is illustrated (in Figure 3a and 3b) that systems controlled
by robust ITAE-CSA-PID controller return back to the set-
point value quickly after the appearance of external

10th International Conference on Cloud Computing, Data Science & Engineering (Confluence) 63
disturbance. However, the corresponding recovering time is Similarly, ± 50% error is introduced in delay time of
quite long for the systems controlled by ITSE, IAE and ISE Plant-2 (equation (8)) to get equation (11) and equation (12).
based CSA-PID controller. As the magnitude of control effort
is also less (less than -9V for Plant-and -180V for Plant-2) in ݁ ሺି଴Ǥଵହ௦ሻ
‫ܩ‬ଶ௔ ൌ (11)
bringing back the disturbed output to set-point value as ‫ݏ‬ଶ ൅ ͲǤͲʹ‫ ݏ‬൅ ͳ
compared to largest magnitude of IAE-CSA-PID which is ݁ ሺି଴Ǥ଴ହ௦ሻ
more than -100V for Plant 1 and ISE-CSA-PID which has ‫ܩ‬ଶ௕ ൌ (12)
-300V for Plant 2 (Figure 4a and 4b). The Table III shows ‫ ݏ‬ଶ ൅ ͲǤͲʹ‫ ݏ‬൅ ͳ
that CSA-PID needs lesser TR, MP, IAE and Ess as compared A system is said to be robust to modeling errors if the
to CSA-PID in steady state for both the plants and AVR optimized values of Kp, Ki, and Kd parameters obtained from
system. The same observation is also supported by the equation (7) and (8) also hold good for controlling their
Figures 4a and 4b. From these figures, the significance of modified variants (i.e. equations (9) - (12)). To demonstrate
CSA-PID in terms of integral indices is clearly depicted. the robustness of CSA-PID, the same optimized PID
Plant-2 has a time delay and slower output but after addition parameters (as given in Table II) are employed on controlling
of CSA-PID controller it has achieved faster and stable the plants given by equations (9) - (12) and the responses of
output. The integral indices of Plant-1 and Plant-2 are both the algorithms are shown in Figure 6a-b and Figure 7a-
tabulated in Table III and CSA-ISE has been found to be b.
having the best disturbance rejection capability. The best results with the optimized parameters are recorded
C Noise Rejection in Table IV for modelling errors. As illustrated in Figures 6-
7 and supported by data from the Table IV there is improved
Robustness of the Controlled system with respect to noise response in relation to settling time, rise time and overshoot
suppression is tested in this subsection. White noise with of the systems with controllers. For Plant-1 as well as Plant-
noise power of 0.0001 and sampling time 0.1 is added into 2, CSA-ITSE gave best steady state values in relation to
the feedback path of Plant 1 and Plant 2. The effect of the minimum settling time and rise time as compared to other
noise inputs are demonstrated in Figure 5a and 5b, which applied algorithms of CSA-PID for -50% modelling errors
shows Robust ISE-CSA-PID controller to be better than other for PLANT-1 while CSA-IAE gave best results for +50%
CSA-PID controllers with relatively lesser fluctuation for modeling error of Plant-1 and -50% modeling error of Plant-
Plant-1 and ITAE-CSA-PID for Plant-2 has been found to be 2. For Plant-1, CSA-ITAE for -50% modeling error have
better amongst four types of output. Hence noise suppression become unstable and showed worst response but for +50%
ability of the ISE and ITAE based CSA-PID controller have modeling error of Plant-2 it showed best response.
been best for plant-1 and Plant-2 respectively. A system is
said to be robust to modeling errors if the optimized values
of Kp, Ki, and Kd parameters obtained from equation (7) and
(8) also hold good for controlling their modified variants (i.e.
equations (9) - (12)). To demonstrate the robustness of CSA-
PID, the same optimized PID parameters (as given in Table
2) are employed on controlling the plants given by equations
Magnitude

(9) - (12) and the responses of both the algorithms are shown
in Figure 6a-b and Figure 7a-b. Integral Indices for all four
controllers have been tabulated in Table III and CSA-IAE has
been found to be the best.
D Set-Point tracking with modeling errors
To test the robustness of the proposed methods, all four
controllers based on optimized CSA-PID were applied on
modified models (with error introduced). As a controller a
tuning method should be robust to errors in plant models
tuned by same optimized values for original plant models.
Modeling errors were introduced by adding a ± 50% error in
damping ratio of Plant-1 (equation (7) to get equation (9) and
equation (10).
ͶǤʹʹͺ
‫ܩ‬ଵ௔ ൌ (9)
ሺ‫ ݏ‬൅ ͳሻሺ‫ ݏ‬ଶ ൅ ͲǤͷ͹ͷ͸ͺ‫ ݏ‬൅ ͶǤʹʹͺሻ
ͶǤʹʹͺ
‫ܩ‬ଵ௕ ൌ (10)
ሺ‫ ݏ‬൅ ͳሻሺ‫ ݏ‬ଶ ൅ ͳǤ͹ʹ͹‫ ݏ‬൅ ͶǤʹʹͺሻ

64 10th International Conference on Cloud Computing, Data Science & Engineering (Confluence)
MAGNITUDE
a
b
Fig. 3. Comparison of input with Load Disturbance for a) Plant-1 b) Plant-2

MAGNITUDE

b
Fig.5. Noise Suppression capability for a) Plant-1 b) Plant-2

a
Magnitude
Magnitude

b
Fig. 4. Comparison of Controller effort for a) Plant-1 b) Plant-2.

10th International Conference on Cloud Computing, Data Science & Engineering (Confluence) 65
Magnitude
b
Fig. 6. Modelling error for plant 1 for a) -50% error b) +50% error in ζ b
Fig. 7. Modelling error for Plant 2 for a) -50% error b) +50% error in delay time.
Magnitude

TABLE II COMPARISON OF TIME DOMAIN CHARACTERISTICS FOR PLANT-1 and PLANT-2

Kp Ki Kd MP Peak Ts Tr ITSE ISE IAE ITAE


PLANT1
CSA_ITSE 2.0902 0.9511 1.3863 1.1468 1.0116 55.6964 22.5427 0.3501 0.5148 1.12 1.688
CSA_ISE 1.0724 0.69188 0.22269 1.7349 1.0218 50.8231 29.4605 0.6567 0.9189 1.607 2.39
CSA_IAE 2.6068 0.92057 2.379 0.0587 1.0018 58.9159 4.3372 0.3896 0.4585 1.1 2.023
CSA_ITAE 1.3266 0.80813 0.18584 2.5081 1.0261 71.1632 29.252 0.4736 0.8095 1.361 1.715
WITHOUT - - - 2.3435 0.5131 64.874 11.6881 13.07 3.236 5.548 25.48
CONTROLLER
PLANT2
CSA_ITSE 0.48954 4.5728 4.5083 3.034 1.0314 11.5479 2.1664 0.02827 0.1874 0.3635 0.6475
CSA_ISE 0.38611 7.2449 6.5217 3.2647 1.0334 6.4682 0.7941 0.02209 0.1699 0.3266 0.492
CSA_IAE 0.41548 3.7558 4.3592 3.0129 1.0279 88.2869 2.0999 0.04238 0.1927 0.4324 1.00
CSA_ITAE 0.56441 3.7587 3.789 3.0326 1.0260 61.3001 2.4135 0.039925 0.2044 0.4297 0.873
WITHOUT 0 2.629E+14 2.918E+03 191.728 28 4.92 5.757 31.22
- - -
CONTROLLER

66 10th International Conference on Cloud Computing, Data Science & Engineering (Confluence)
TABLE III INTEGRAL INDICES FOR LOAD DISTURBANCE AND NOISE SUPPRESSION FOR PLANT-1 and PLANT-2
ITSE ISE IAE ITAE ITSE ISE IAE ITAE
PLANT1 WITH LOAD DISTURBANCE PLANT 1 WITH NOISE
CSA_ITSE 0.6664 0.1142 0.5196 3.418 0.4 0.5297 1.243 2.607
CSA_ISE 0.7147 0.1113 0.5596 4.034 0.7092 0.939 1.69 3.026
CSA_IAE 0.5411 0.0978 0.4119 2.582 0.4588 0.4774 1.236 3.074
CSA_ITAE 1.128 0.2022 0.6521 3.92 0.5316 0.832 1.488 2.586
PLANT 2 WITH LOAD PLANT 2 WITH NOISE
CSA_ITSE 184.1 25.41 9.863 72.41 0.1461 0.2194 0.592 1.926
CSA_ISE 207.6 28.92 10.46 76.48 0.154 0.2068 0.624 2.101
CSA_IAE 188.9 25.63 10.17 75.93 0.123 0.219 0.5805 1.805
CSA_ITAE 167.2 23.04 9.416 69.17 0.119 0.2298 0.581 1.717
TABLE IV MODELLING ERROR ±50% FOR PLANT-1 and PLANT-2
MP Peak Ts Tr ITSE ISE IAE ITAE

PLANT-1
CSA_ITSE 1.34 1.012 84.82 7.9877 0.3123 0.4712 1.055 1.886
CSA_ISE 0.7829 0.999 97.79 18.09 0.5209 0.8198 1.439 2.22
+50% CSA_IAE 10.22 1.09 78.911 4.8 0.3595 0.4207 1.136 2.572
CSA_ITAE 6.7 1.06 163.42 17.13 0.5214 0.7828 1.408 2.557
CSA_ITSE 26.87 1.2656 253.23 4.9 2.64 1 2.49 9.7
CSA_ISE 37.39 4.46e+24 386.98 12.12 5.53 1.52 3.21 14.7
-50% CSA_IAE 31.32 1.3176 182.198 4.498 1.69 0.8316 2.09 6.9
CSA_ITAE 1.4e+03 1.247e+12 3.3189e+03 280.98 27.25 4.31 5.56 31.82
PLANT-2
CSA_ITSE 20.07 1.2 10.5 1.59 0.04517 0.2419 0.4565 0.6581
CSA_ISE 48.24 1.49 24.37 1.2 0.07436 0.2875 0.5665 0.6785
+50% CSA_IAE 15.37 1.15 67.95 1.36 0.059 0.2518 0.5363 1.055
CSA_ITAE 12.76 1.12 84.84 2.64 0.053 0.2574 0.5022 0.9014
CSA_ITSE 1.7612 1.014 10.61 4.1 0.02115 0.1452 0.3215 0.6031
CSA_ISE 0 1.009 128.8 62.9 0.01332 0.1149 0.2525 0.5
-50% CSA_IAE 2.24 1.02 80.18 3.19 0.0349 0.1517 0.3994 0.958
CSA_ITAE 2.59 1.02 61.24 3.55 0.03313 0.1653 0.398 0.8303

IV. CONCLUSION as objective function plays a major role in optimizing the


controller so the objective function should be chosen wisely.
With only two adjustable parameters AP and fl to get the best
results out of CSA, makes it more attractive and less time REFERENCES
consuming to apply in different application areas as well as
different kinds of problems. There are four types of integral [1] G. Ziegler, N. B. , “Optimum settings for automatic controllers,” Trans.
ASME, vol. 64, pp. 759–768, 1942.
performance indicators which govern the performance of
[2] S.Bhanot, Process Control Principles and Applications. New Delhi:
output of any process: IAE, ISE, ITSE and ITAE. So, in the Oxford University Press, 2008 .
present paper, four multi-objective functions based on these [3] Bialkowski., W. L. “Dreams versus reality: A view from both,” Pulp and
integral indices have been chosen, optimized using CSA and Paper Canada, vol. 94, No. 11 , pp. 19–27, 1993.
[4] H.A.Varol, Z. B., “A new PID Tuning Technique using Ant Algorithm,”
then compared for their robustness on the basis of integral
American Control Conference, pp. 2154-2160, 2004.
index values and time response characteristics when these are [5] M. H. T. Omar, W. M., “Auto Tuning of PID Controller Using Swarm
subjected to set point tracking, load disturbance rejection Intelligence,” International Review of Automatic Control, Vol. 4, No. 3, pp.
capability, noise suppression ability and robustness towards 319-327, 2011.
[6] I. Chiha, H. L., “A Hybrid Method Based On Multi-objective Ant Colony
modeling errors. CSA-IAE and CSA-ITAE have been found
Optimization and Differential Evolution to Design PID DC Motor Speed
to have best load disturbance rejection capability and noise Controller,” International Review of Modelling and Simulations (IREMOS),
suppression ability for third order system and resonant time vol.5, No.2, 2012 .
delay system respectively. For modeling errors, CSA-ISE [7] Ujjwal Arora, M. E., “PID Parameters tuning Modified BAT Algorithm,”
Journal Of Automation and Control Engineering, pp. 347-352, 2016.
have been found to be best for +50% for third order and -50%
for resonant time delay system. As, these indices when taken

10th International Conference on Cloud Computing, Data Science & Engineering (Confluence) 67
[8] M.M. Zirkohi, “Optimal PID Controller design using Adaptive Conference on Control and Automation, Montreal, Canada: IEEE, pp. 942-
VURPSO algorithm,” DE Gruyter, pp. 179-185, 2015. 946, 2003.
[9] XS, Y., Nature-inspired metaheuristic algorithms. Luniver Press, 2008 [12] Y.T Hsiao, C., "Ant colony optimization for designing of PID
[10] Askarzadeh, A., "A Novel metaheuristic method for solving constrained controllers," International symposium on Computer aided Control systems
engineering optimization problems:Crow Search Algorithm," Elsevier: design, Taipei, Taiwan: IEEE, pp. 321-326, 2004.
Computers and Structures, pp. 1-12, 2016. [13] B.Kristiansson, B., "Optimal PID controllers for unstable and resonant
[11] Saeed Tavakoli, M. T., "Optimal tuning of PID controllers for first order plants," 37th IEEE conference on Decision and Control, Tampa, USA: IEEE,
plus time delay model using dimensional analysis," The Fourth International pp. 4380-4381, 1998.

68 10th International Conference on Cloud Computing, Data Science & Engineering (Confluence)

Das könnte Ihnen auch gefallen