You are on page 1of 7

Michael Faraday IET International Summit: MFIIS-2015, September 12 – 13, 2015, Kolkata, India

(Paper ID: 135)

A COMPARISON BETWEEN MPC AND OPTIMAL PID


CONTROLLERS: CASE STUDIES
Fawzan Salem* and Mohamed I. Mosaad **
* Electronics Research Institute (ERI), on leave to YIC-KSA, fawzan@lycos.com
** Higher Technological Institute (HTI), on leave to YIC-KSA, m_i_mosaad@hotmail.com

Keywords: Model Predictive Control, Optimal PID Controller, In recent years, MPC has also been used in power
Bacterial Foraging Optimization and Performance Indices. system field. In [7], MPC is applied in load frequency
control to improve the dynamic performance of power
Abstract system subject to several disturbances. Effectiveness of MPC
This paper presents a comparative study between the controller is validated upon a comparison with a fuzzy logic
performance of Model Predictive Control (MPC) and optimal controller. Another MPC-based load frequency controller of
PID controllers in terms of various performance measures. a multi-area power system including wind turbines is
Parameters of the PID controller are adjusted based on introduced in [8]. In such a system, each local controller is
Bacterial Foraging Optimization (BFO) algorithm and its considered separately so that stability of the overall closed
results are compared with those obtained using MPC loop system is assured.
controller in terms of some control parameters. These control Moreover, MPC has been employed in many applications
parameters are rise time, settling time, peak overshoot of electrical drive systems, including but not limited to,
percentage, integral of absolute errors and integral of time speed control of DC motors [9], position control of DC
multiplied absolute of error for both MPC and PID servomotors [10], torque ripple reduction of brushless DC
controllers. Many case studies are considered in this article. motors [11], field oriented control of permanent magnet
First, second, third, fourth and fifth order systems are synchronous motors (PMSM) [12,13], and torque control of
investigated. Simulation results indicate that MPC controller induction motors [14,15].
outperforms optimal PID controller in all cases. Recently, renewable energy sources have been also
controlled with the aid of MPC [16]. MPC is also applied to
1. Introduction grid-tied photovoltaic storage system yielding satisfactory
real-time dynamic response [17]. Advantages of MPC in
such a case include smoother power output while respecting
Model predictive control (MPC) is an advanced and maintaining the functional requirements of the storage
technique applied originally in industry for chemical units and power converters.
plants and other dynamical systems [1]. MPC systems
depend on obtaining control values for plant inputs by Comparing to conventional PID controllers, it should be
solving an optimization problem that is formulated using a noted that MPC consumes excess time for on-line
plant model and online measurements [2]. The most computations when the constraints intervene. Parameters of
important merit of MPC is that it allows the current time slot MPC are designed based on successive iterations, where no
to be optimized, while keeping future time slots into mathematical forms have been developed yet to determine
consideration. In other words, a finite time horizon is the best configuration of the parameters. To be realistic and
optimized, whereas the current time slot, merely, is more credible, some modern optimization algorithms have to
implemented. Therefore, MPC has the capability to predict be applied to the process for fine tuning the PID control
future behavior and can implement control actions parameters. A large volume of literature considers
accordingly [3]. performance comparison between MPC and conventional
PID control, where the results usually point out to noticeable
Basically, MPC is related to linear optimal control MPC superiority [18, 19].
systems. However, MPC can be applied to linear and In this paper, a comparison between the performances of
nonlinear time varying systems. MPC algorithm is presented Model Predictive Control (MPC) and optimal PID controllers
in [4] for a linear time varying system to verify its ability in is presented in terms of different performance indices. The
navigation under obstacle avoidance conditions. Two main intent is to maintain minimum values for rise time,
nonlinear models for predictive control algorithms are settling time, peak overshoot percentage, integral of absolute
implemented and compared in [5]. In [6], two chemical errors and integral of time multiplied absolute errors for both
processes are modeled and a thorough procedure for MPC MPC and PID controllers.
control is conducted along with system identification,
To guarantee a fair comparison, the PID controller
controller scheme and parameter adjustment.
parameters are optimized by using Bacterial Foraging
Optimization (BFO) algorithms as one of the most promising

59
Michael Faraday IET International Summit: MFIIS-2015, September 12 – 13, 2015, Kolkata, India
(Paper ID: 135)

evolutionary algorithms. Several case studies are conducted of computed future control referring to the equation: ǻX NM
to validate such comparison and carried out using = 0 for j •1X w(k + j) signifies the reference trajectory for
MATLAB/SIMULINK environment. Both qualitative the future horizon N. Constraints for the control signal, the
comparison (via time domain step response) and qualitative control signal change and the outputs form the following cost
comparison (via certain performance indices) are used to function:
evaluate the behaviors of both controllers. Results show
distinct superiority in favor of the MPC controller. u min d u (k ) d u max
'(u min ) d 'u (k ) d '(u max ) (2)
2. Model Predictive Control y min d y (k ) d y max
Step 3: The current control signal u (t ) is assigned to the
MPC is a modern control scheme, which was proved to plant. At the next interval, y (t  1) is measured and step 1 is
powerfully control an extensive range of industrial
applications. These applications include unstable systems, repeated according to the receding horizon strategy to
multi-input multi-output systems, systems with delay, compute u (t  1) . Accordingly, at each interval, the horizon is
constrained and hybrid systems [2]. shifted towards the future maintaining the same length.
Unlike most control algorithms, MPC has the ability to MPC can be tuned easily apart from the fact that it handles
predict the future response of the plant. At each control constraints properly. Constraints can be either on the output
interval, MPC attempts to predict future plant behaviour by of the controlled processes (control variable) or on the
means of an on-line optimization process, which maximizes control signals that are inputs to the process (manipulated
the tracking performance under certain constraints [3]. Fig. 1 variables). The constraints are in the form of saturation
depicts a block diagram describing MPC. characteristics, e.g., valves with a finite range of adjustment,
control surface with limited deflection angles, etc. Input
constraints also appear in the form of rate constraints: valves
and other actuators with limited rates.

Fig. 1. A block diagram describing MPC


MPC is characterized by the following strategy which is
represented in Fig. 2:
Step 1: An explicit model is used to predict the process
output through a future time horizon. Predicted outputs
 Fig. 2. Strategy of Model Predictive Control
y (t  k ), k 1, ..., N for the prediction horizon are
calculated at each instant t depending on the past inputs and
I. OPTIMAL PROPORTIONAL-INTEGRAL-DERIVATIVE
outputs as well as the future control
CONTROLLER
signal u (t  k ), k 0, ..., N  1 .
Step 2: A chain of future control signals is computed to The ideal continuous PID controller is described as
optimize a performance criterion by minimizing an objective t de(t )
function. The objective function to be minimized is generally u K P e(t )  K I ³ e(t )dt  K D (3)
a weighted summation for square of predicted errors and 0 dt
square of future control values. Where u is the controller output, KP, KI and KD are the
proportional, integral and derivative gains, respectively, and
N2 2
e(t) is the error between reference and actual output values.
¦

J ( N1 , N 2 , N u ) E ( j )> y (k  j )  w(k  j )@ To guarantee a fair comparison between MPC and PID
j N1
(1) controllers, the PID controller parameters (KP, KI and KD) are
Nu 2 optimized by using BFO (Appendix A) as one the most
 ¦ O ( j)>u(k  j  1)@
j 1
promising evolutionary algorithms. Four performance terms
are applied for optimization to compromise both static and
Where ȕ M  Ȝ M are weighting factors, N1 and N2 lower and dynamic responses. These terms are; maximum overshoot
upper prediction horizons over the output, Nu is the control percentage (Mp), settling time (ts), rise time (tr) and steady
horizon. The control horizon allows to decrease the number state error (ess). The optimization function is defined as [20]:

60
Michael Faraday IET International Summit: MFIIS-2015, September 12 – 13, 2015, Kolkata, India
(Paper ID: 135)

Some constraints have to be identified to the toolbox, for


Minimize J T
(4) each plant, such as mminimum and maximum constraints on
J T (K P , K I , K D ) (1  e  E ) ( M p  e ss )  e  E (t s  t r ) manipulated variables, minimum and maximum constraints
on output variable. To fine-tune the MPC controller, other
Where E is a weighting factor (E < 0.7 to reduce ts and tr and parameters should be carefully selected; these parameters are:
E >0.7 to reduce Mp and ess). In order to normalize the BFO control interval, prediction horizon, control horizon, overall
fitness function (F), some medications are applied to tuning factor, input weights and output weights.
optimization function (JT) as in (5).
Case Study 1. A First Order System with Delay:
Temperature control process can be described by a first order
1 system with delay caused by a sensor and/or an actuator [21]
Maximize F (5)
( J T  1) as modelled in equation (8)
1  0.2 s
eG ( s) (8)
s 1
3. Simulation Results The time delay introduces nonlinearity to the first order
Several case studies are considered to compare the system. Zeigler-Nichols tuning method couldn’t lead to a
performances of MPC and optimal PID controllers. These satisfactory response in this case. However, the evolutionary
case studies cover a wide range of practical processes with algorithm (BFO) could reach a fitness function (F) of 0.561.
different orders. Both qualitative and quantitative With saturation limits ranging from 0 to 10 for the control
comparisons of MPC and optimal PI controllers are carried signal, the optimal values obtained for KP and KI are 2.51 and
out having regard to certain computable performance indices. 2.19, respectively.
The indices include the rise time (tr), settling time (ts), One the other hand, the MPC controller has many parameters
integral absolute error (IAE), and integral time-multiplied that need to be initialized. After some trials, the following
absolute error (ITAE). The integral errors, IAE and ITAE, are initial design parameters are assigned to MPC controller
performance measures to assess how far the actual response using MATLAB/MPC toolbox:
is from a desired ideal response, where both depend on the control interval = 0.001s, prediction horizon = 10, control
time integration of an error function. Large errors contribute horizon = 2, minimum and maximum constraints on
more to IAE, whereas ITAE penalizes more the error which manipulated variables = [0, 10], minimum and maximum
occurs late in time. In other words, IAE and ITAE reflect the constraints on output variable = [0, 1], overall tuning factor =
transient and steady-state characteristics of the system, 0.8, input weights = 0.1 and output weights =1. Step
respectively. Mathematical expressions for IAE and ITAE are responses of both optimal PID controller and MPC controller
given as are shown in Fig. 3.
f The results of the performance measures for both optimal
IAE ³ e(t ) dt (6) PID controller and MPC controller are illustrated in Table I.
t 0 It is revealed from Fig. 1 and Table I that MPC controller
outperforms optimal PI controller in terms of all performance
f measures.
ITAE ³ t u e(t ) dt (7) 1.4
t 0 1.2

To validate the two controllers, a unit step function is taken 1


Output, (p.u.)

as a reference value for the different cases. Both


0.8
MATLAB/SIMULINK and MATLAB/MPC Toolboxes are
used, with a variable step Domand-Prince (ODE45) solver, 0.6
to conduct all numerical studies. 0.4
BFO algorithm is utilized to get the PID controller MPC
parameters with improved performance by maximizing the 0.2
Optimal PID
fitness function F. 0
MATLAB/MPC toolbox is a collection of functions 0 0.5 1 1.5 2
time, (sec.)
developed for the analysis and design of Model Predictive
Control (MPC) systems. While MPC is suitable for almost Fig. 3. Comparison between optimal PI and MPC controllers
any kind of problems, it displays its main power when for Case Study 1
applied to plants with a number of manipulated and TABLE I
controlled variables. In, MPC Toolbox transfer functions, COMPARISON BETWEEN OPTIMAL PI AND MPC PERFORMANCES IN
state-space matrices, or a combination of the two can be CASE STUDY 1
employed. Moreover, time delay can also be imposed, which
are a common feature of industrial plants. Performance Controller
Measure Optimal PI MPC

61
Michael Faraday IET International Summit: MFIIS-2015, September 12 – 13, 2015, Kolkata, India
(Paper ID: 135)

Tr (sec.) 0.3900 0.1230 In this example, the ability of the two controllers to
Ts (sec.) 0.7596 0.4212 compensate the load change effect on their performances is
Mp % 0.7000 0.000 checked. A reference step, followed by a load disturbance at
IAE 0.4387 0.0928 15 seconds, is applied. Zeigler-Nichols tuning method could
ITAE 0.1262 0.0242 be also applied with the values 4.8, 2.56 and 2.25 for KP, KI
and KD, respectively. However, about 50% as an overshoot is
Case Study 2. A Small Damped Second Order System: A resulted by using these values.
mass spring damper system is an Case Study [22] By applying BFO, a good performance is obtained with a
fitness function (F) of 0.21 and a saturation limit of [-10 10]
1 for the control signal. The optimal values achieved for KP, KI
G( s) (9)
s 2  0.02 s  1 and KD are 1.44, 0.4736 and 0.7875, respectively.
Zeigler-Nichols tuning method gives poor response also for On the other side, the initial values for the MPC controller
that system. Due to the small damping ratio, the derivative are found to be: control interval = 0.3s, prediction horizon =
action is necessary. The integral action also helps to 10, control horizon = 2, minimum constraints on manipulated
eliminate the steady state error. BFO could reach a fitness variables = -10, maximum constraints on manipulated
function (F) of 0.512 while keeping a saturation range of [- variables = 10, minimum constraint on output variable = 0,
10 10] for the control signal. The optimal values achieved for maximum constraint on output variable = 1, overall tuning
KP, KI and KD are 8.65, 1.65 and 6.76, respectively. factor = 0.8, input weights = 0.1 and output weights =1.
Moreover, the initial values for the MPC controller are found Responses of (10) due to set point as well as load disturbance
to be: control interval = 0.08s, prediction horizon = 10, are shown in Fig. 5. Table III provides the quantitative
control horizon = 2, minimum and maximum constraints on performance analysis of both optimal PID controller and
manipulated variables are -10 and 10, respectively, minimum MPC controller. From the results, we find that the overall
and maximum constraints on output variable are 0 and 1, performance of MPC is also better in presence of disturbance
respectively, overall tuning factor = 0.8, input weights = 0.1 signals.
and output weights =1. The step responses of both optimal 1.4
PID controller and MPC controller are shown in Fig. 4.
1.2 1.2

1
1
Output, (p.u.)

0.8
0.8
Output, (p.u.)

0.6
0.6
0.4
0.4 MPC
0.2
Optimal PID
0.2
0
0 5 10 15 20 25 30
0 time, (sec.)
0 0.5 1 1.5 2 MPC
2.5 3
time, (sec.) Optimal PID Fig. 5. Comparison between optimal PID and MPC
Fig. 4. Comparison between optimal PID and MPC controllers for Case Study 3
controllers for Case Study 2
TABLE III
TABLE II COMPARISON BETWEEN OPTIMAL PI AND MPC PERFORMANCES IN
COMPARISON BETWEEN OPTIMAL PI AND MPC PERFORMANCES IN CASE STUDY 3
CASE STUDY 2
Performance Controller
Performance Controller Measure Optimal PID MPC
Measure Optimal PID MPC Tr (sec.) 2.4400 1.3745
Tr (sec.) 0.9578 0.6142 Ts (sec.) 4.3500 2.4735
Ts (sec.) 1.3266 1.0332 Mp % 1.5800 0.0000
Mp % 3.8600 0.0000 IAE 2.3070 1.2270
IAE 0.6579 0.4964 ITAE 6.0230 1.9080
ITAE 0.3205 0.1522

The results of the performance measures for both optimal Case Study 4. Fourth Order System: Automatic voltage
PID controller and MPC controller are illustrated in Table II. regulator of a power system shown can be represented by a
fourth order system [24] as shown in Fig. 6.
Case Study 3. Third Order System [23]: Position control of
DC motors is an example for third order systems
1
G(s) (10)
s 3  3s 2  3s  1

62
Michael Faraday IET International Summit: MFIIS-2015, September 12 – 13, 2015, Kolkata, India
(Paper ID: 135)

TABLE IV
COMPARISON BETWEEN OPTIMAL PI AND MPC PERFORMANCES IN
CASE STUDY 4
Performance Controller
Measure Optimal PI MPC
Tr (sec.) 0.7363 0.3868
Ts (sec.) 1.5335 0.7223
Mp % 0.0000 0.0000
Fig. 6. A Block diagram for the AVR of Case Study 4 IAE 0.5617 0.3330
without a controller ITAE 0.2513 0.0676

The results of the performance measures for both optimal


The following closed loop transfer function can be easily PID controller and MPC controller are illustrated in Table IV.
obtained
0.1s  10 Case Study 5. Fifth Order System: A fifth order system
G(s) 4 3 2 (11) is investigated in this case to show the capability of both
0.0004 s  0.0454 s  0.555s  1.51s  11
controllers to deal with high order systems. An example for
By applying a PID controller, the closed loop transfer
such systems can be described as [25]:
function becomes
N (s) 1
G(s)
D( s)
(12) G(s) (13)
s  5s 10 s  10 s 2  5s  1
5 4 3
Where N(s) = With a saturation limit ranging from -10 to 10, the following
0.1K D s 3  (0.1K P  10 K D ) s 2  (0.1K I  10 K P ) s  10 K I optimal PID controller parameters are obtained using BFO:
and D(s) = KP = 0.906, KI = 0.323 and KD = 0.575. Regarding the MPC
controller, the following initial values are assumed: control
0.0004 s 5  0.0454 s 4  0.555s 3  (1.51  10 K D ) s 2
interval = 1s, prediction horizon = 10, control horizon = 2,
 (1  10 K P ) s  10 K I minimum and maximum constraints on manipulated
variables = -10 and 10, respectively, minimum and
The BFO could reach a fitness function (F) of 0.24 with a maximum constraints on output variable = 0 and 1,
saturation limit of [0 2] for the control signal. The optimal respectively, overall tuning factor = 0.8, input weights =
values achieved for KP, KI and KD are 0.5473, 0.1778 and 0.045 and output weights = 2.23. Step responses of both
0.1668, respectively. optimal PID and MPC controllers are shown in Fig. 7.
Moreover, the initial values for the MPC controller are found The results of the performance measures for both optimal
to be: control interval = 0.1s, prediction horizon = 6, control PID controller and MPC controller are illustrated in Table V.
horizon = 2, minimum constraints on manipulated variables
= 0, maximum constraints on manipulated variables = 1.1, 4. Conclusion
minimum constraint on output variable = 0, maximum This paper presents qualitative and quantitative
constraint on output variable = 1, overall tuning factor = 0.8, comparison between the performances of Model Predictive
input weights = 0.1 and output weights = 1. Control (MPC) and optimal PID controllers based BFO in
The step responses of both optimal PID controller and MPC terms of various performance measures. Five case studies are
controller are shown in Fig. 7. introduced representing a wide range of practical systems.
Simulation results indicate that MPC controller outperforms
1.4 optimal PID controller in all cases. However, MPC includes
too many parameters that have to be initialized carefully to
1.2 achieve the best performance.
1
1.4
Output, (p.u.)

0.8 1.2

0.6 1
Output, (p.u.)

0.8
0.4
0.6
0.2 MPC
0.4
Optimal PID
0 0.2 MPC
0 0.5 1 1.5 2 2.5 3 Optimal PID
time, (sec.) 0
0 5 10 15 20 25 30
Fig. 7. Comparison between optimal PI and MPC controllers time, (sec.)
for Case Study 4 Fig. 7. Comparison between optimal PI and MPC controllers
for Case Study 5

63
Michael Faraday IET International Summit: MFIIS-2015, September 12 – 13, 2015, Kolkata, India
(Paper ID: 135)
Trans. on Industrial Informatics, vol. 9, no. 2, May 2013, pp. 909-
TABLE V 919.
COMPARISON BETWEEN OPTIMAL PI AND MPC PERFORMANCES IN [16] J. Momoh, F. Zhang and W. Gao, “Optimizing Renewable Energy
CASE STUDY 5 Control for Building using Model Predictive Control,” North
American Power Symposium (NAPS), 7-9 Sept., 2014, pp. 1-6.
Performance Controller [17] T. Wang, H. Kamath and S. Willard, “Control and Optimization of
Measure Grid-Tied Photovoltaic Storage Systems Using Model Predictive
Optimal PI MPC
Control,” IEEE Trans. Smart Grids, vol. 5, no. 2, 2014, pp. 1010–
Tr (sec.) 6.0850 2.9955 1017.
Ts (sec.) 10.6590 5.7333 [18] S. Thomsen, N. Hoffmann and F. Fuchs, “PI Control, PI-Based State
Mp % 1.7200 0.0000 Space Control, and Model-Based Predictive Control for Drive
Systems With Elastically Coupled Loads—A Comparative Study,”
IAE 5.8100 3.0460 IEEE Transactions on Industrial Electronics, vol. 58, no. 8, August
ITAE 20.3900 5.3170 2011, pp. 3647-3657.
[19] K. Mapok, T. Zuva, H. Masebu and K. Zuva, “Performance
Comparison of Two Controllers on a Nonlinear System,”
References International Journal of Chaos, Control, Modelling and Simulation
(IJCCMS), vol.2, no.3, September 2013, pp. 17-30.
[20] Ehab Bayoumi and Fawzan Salem, “PID controller for series-parallel
[1] R. Sivakumar and Shennes Mathew, “Design and Development of
resonant converters using bacterial foraging optimization”,
Model Predictive Controller for Binary Distillation Column,”
Electromotion, volume 19, Number 1-2, June 2012, pp. 64-79.
International Journal of Science and Research (IJSR), vol. 5, no. 12,
[21] Baogang Hu, George K. I. Mann, and Raymond G. Gosine, “New
Dec. 2014, pp. 445-451.
Methodology for Analytical and Optimal Design of Fuzzy PID
[2] K. Holkar and L. Waghmare, “An Overview of Model Predictive
Controllers”, IEEE Transactions on Fuzzy Systems. Vol. 7, No. 5
Control,” International Journal of Control and Automation, vol. 3
October 1999.
No. 4, December, 2010, pp. 47-63.
[22] H. A. Malki, H. Li, and G. Chen, “New design and stability analysis
[3] P. Orukpe, “Model Predictive Control Fundamentals,” Nigerian
of fuzzy proportional-derivative control systems,” IEEE Trans. Fuzzy
Journal of Technology (NIJOTECH), vol. 31, No. 2, July, 2012, pp.
Syst., vol. 2, pp. 245–254, 1994.
139–148.
[23] Jan Tantzen, Technical University of Denmark, Department of
[4] A. Bemporad and C. Rocchi, “Decentralized Linear Time-Varying
Automation, Bldg 326, DK-2800 Lyngby, DENMARK. Tech. report
Model Predictive Control of a Formation of Unmanned Aerial
no 98-H 871 (fpid), April 16, 1999.
Vehicles,” 50th IEEE Conference on Decision and Control and
[24] B. K. Sahu, P. K. Mohanty, S. Panda, S. K. Kar and N. Mishra,
European Control Conference (CDC-ECC) Orlando, FL, USA,
“Design and Comparative Performance Analysis of PID Controlled
December 12-15, 2011, pp. 7488-7493.
Automatic Voltage Regulator tuned by Many Optimizing Liaisons”,
[5] A. Bamimore, O. Taiwo, and R. King, “Comparison of Two
IEEE conference on Advances in Power Conversion and Energy
Nonlinear Model Predictive Control Methods and Implementation on
Technologies (APCET), Mylavaram, Andhra Pradesh, Aug. 2012,
a Laboratory Three Tank System,” 50th IEEE Conference on
pp. 1-6.
Decision and Control and European Control Conference (CDC-
[25] M. Zhuang and D. P. Atherton, “Automatic tuning of optimum PID
ECC) Orlando, FL, USA, December 12-15, 2011, pp. 5242-5247.
controllers,” Proc. Inst. Elect. Eng.–Control Theory Appl., vol. 140,
[6] N. Danesh Pour, A. Montazeri, J. Poshtan and M.R. Jahed Motlahgh,
pp. 216–223, 1993.
“Two Case Studies for Applying Model Predictive Controllers on
[26] Sambarta Dasgupta, Swagatam Das, Ajith Abraham, Arijit Biswas,
Chemical Processes,” The 3rd Annual Conference of the IEEE
“Adaptive Computational Chemotaxis in Bacterial Foraging
Industrial Electronics Society (IECON), Taipei, Taiwan, Nov. 5-8,
Optimization: An Analysis”, IEEE Transactions on Evolutionary
2007, pp. 580-585.
OMPUTATION, VOL. 13, NO. 4, pp. 919 - 941, Aug. 2009.
[7] A. M. Y. Ali, “Model Predictive Control Approach Based Load
[27] S. Mishraand C. N. Bhende “Bacterial Foraging Technique-Based
Frequency Controller,” Journal of Engineering Sciences, Assiut
Optimized Active Power Filter for Load Compensation”, IEEE
University, Egypt, vol. 38, no. 5, September 2010, pp.1259-1273.
Transactions on Power Delivery, vol. 22, NO. 1, pp.457 – 465, Jan.
[8] T. H. Mohamed, J. Morel, H. Bevrani and T. Hiyama, “Model
2007.
Predictive Based Load Frequency Control Design Concerning Wind
[28] M.I.Mosaad and Mohamed G. Ashmawy “Short Term Load
Turbines,” Electrical Power and Energy Systems, vol. 43, 2012, pp.
Forecasting Using Curve Fitting Prediction Optimized By Bacterial
859–867.
Foraging Optimization” International Review on Modelling and
[9] R. Singh, G. Onwubolu, K. Singh and R. Ram, “DC Motor Control
Simulations (IREMOS), Italy, Vol 6,no.3, pp1150-1154, August
Predictive Models,” American Journal of Applied Sciences, vol. 11,
2013.
no. 3, 2006, pp. 2096-2102.
[10] V. Naik, D. Sonawane, D. Ingole1 and D. Ginoya, “Model Predictive
Control of DC Servomotor using Active Set Method 2013 IEEE
Multi-conference on Systems and Control,” Conference on Control Appendix A: Bacterial Foraging Optimization
Applications, Hyderabad, India, August 28-30, 2013, pp. 820-825. (Bfo)
[11] M. Doss, S. Dash, D. Mahesh and V. Marthandan, “A Model
Predictive Control to Reduce Torque Ripple for Brushless DC Motor BFO is a new division of metaheuristic algorithm. It is a population
with Inbuilt Stator Current Control,” Universal Journal of Electrical based optimization technique advanced by excitation of the foraging
and Electronic Engineering, vol. 1, no. 3, 2013, pp. 59-67. manners of E. coli bacteria that maximize their energy consumption
[12] S. Sivaranjani and R. Rajeswari, “Simulation of MPC Based Speed (E) per time (t) spent in foraging. In contrary to other optimization
Control of Permanent Magnet Synchronous Motor Drive,” Journal of techniques, individuals in BFO could converge rapidly without
Theoretical and Applied Information Technology, vol. 58, no. 2, Dec. shared information among each other [26]. The key four principal
2013, pp. 237-242. steps of the bacteria algorithm are chemotaxis, swarming,
[13] M. Fan, H. Lin and T. Lan, “Model Predictive Direct Torque Control
reproduction, and elimination-dispersal [26] that are briefly
for SPMSM with Load Angle Limitation,” Progress In
Electromagnetic Research B, vol. 58, 2014, pp. 245-256 elaborated below.
[14] Y. Zhang, and H. Yang, “Model Predictive Torque Control of The movement of E. coli bacteria in human intestine searching
Induction Motor Drives with Optimal Duty Cycle Control,” IEEE for nutrient-rich location away from toxic environment is
Trans. on Power Electronics, vol. 29, no. 12, Dec.2014, pp. 6593- accomplished with the help of locomotory organelles known as
6603. flagella by chemotactic movement in either swimming or tumbling
[15] J. Scoltock, T. Geyer, and U. Madawala, “A Comparison of Model [ 26-27].
Predictive Control Schemes for MV Induction Motor Drives,” IEEE

64
Michael Faraday IET International Summit: MFIIS-2015, September 12 – 13, 2015, Kolkata, India
(Paper ID: 135)

Let și( j, k, l) be the ith bacterium at jth chemotactic, kth


reproductive. Also lth elimination-dispersal step. Then the
bacterium’s chemotactic movement enables bacteria for optimum
local search and can be represented mathematically by equation (1).
The term, C(i) in equation (1) represents the size of the unit step
taken in an arbitrary direction, and ǻ L specifies a vector in random
direction whose elements are between íDQG>28].
(A1)

Swarming is considered in many free collections of bacteria.


These cells release an attractant aspartate when they are stirred up
by a high level of succinate.
The cell-to-cell coding in bacterial clusters via attractant and
repellent may be modelled as (2) where Jcc(ș(i, j, k, l)) is the
objective function added to the actual objective function, J that will
be optimized to present a time varying objective function, S
specifies the number of bacteria in the population, p is the number
of variables to be optimized, and ș= [ș1, ș2, . . , șp]T is a point in
the p-dimensional search domain. The coefficients dattractant, wattractant,
hrepellant and wrepellant are the measures of quantity and diffusion rate
of the attractant signal and the repellent effect magnitude,
respectively [26-28].

Fitness value for ith bacterium after travelling Nc chemotactic steps


can be evaluated by the following equation [2]:
(A3)
where represents the health of ith bacterium.
The least vigorous bacteria constituting half of the bacterial
population are eventually removed while each of the vigorous
bacteria asexually fragmented into two parts located in the same
location. Hence, ultimately the population remains constant [28].
Each bacterium X contains three members (KP, KI and KI) that
represent the control parameters. The process starts by setting a zero
value for the three control parameters, generates random positions
of each bacterium and evaluates the objective function of each
bacterium. Then the position of each bacterium is modified using
tumbling/swimming process that is followed by reproduction and
elimination operation. This process is repeated till the maximum
specified steps have been reached.

65

You might also like