Artificial Bee Colony based Optimal PMU Placement in Power

System State Estimation

Mohammad Shoaib Shahriar1*, Farhan Ammar Ahmad1, Ibrahim Omar Habiballah1, Mohammed Afzal Asif2,
Shagorika Mukherjee2
1
Department of Electrical Engineering, King Fahd University of Petroleum & Minerals (KFUPM), Dhahran, Saudi Arabia
2
Department of Electrical and Electronic Engineering, Independent University, Bangladesh
*[email protected]

Abstract— Advanced power system nowadays demands to system, fault detection, wide area protection, transmission line
ensure accurate estimate of the system states (bus voltages and thermal monitoring etc. [5-7].
angles) in order to ensure protection, monitoring, controlling and
smooth running of the system. The conventional state estimator For proper running of any estimator, it must provide
estimates the power system states based on the measurements sufficient amount of measurements that make the system
obtained from the Supervisory Control and Data Acquisition observable. The minimum number of measurements provided
(SCADA) system. Besides, the use of synchronized phasor must be (2N-1) if N indicates the number of buses. Lots of
measurement unit (PMU) has become a popular choice of the work has already been done on this field and researchers has
time in this field because of its ability of providing the real time used both the heuristic and mathematical algorithms as solution
phasors of voltage and currents. For a better phase estimation methodologies [8-9]. Integer programming and exhaustive
accuracy of the system, proper site selection of placing PMU is a search technique was used to search for best placement
must. Keeping an eye on system observability and system meter location of PMUs which has been explained in detail in the
economy, it is important to identify the optimal location and taxonomy work on PMU placement done by Nikolas et al. [10].
number of PMUs to be used in state estimators. This paper Among the heuristic techniques, Tabu search, Simulated
presents an optimal solution of PMU’s placement by using one of Annealing, Genetic Algorithm, Particle Swarm Optimization,
the most widely used intelligent technique, Artificial Bee Colony Differential Algorithm, Bee colony, Spanning Tree Search,
(ABC). State estimation is then carried out by placing PMUs on Immune Algorithm, Recursive Security N algorithm and
the resultant optimum locations. Both the voltage and current several hybrids of such intelligent techniques were used in
phasors has been taken from the PMUs installed in the buses.
Optimum PMU placement (OPP). The recent most survey on
The impact of bad data in the measurement series is also
OPP by Nazari-Heris et al. [11] presented all such techniques
investigated. IEEE 14-bus and 30-bus systems has been taken as
test system and Weighted Least Square (WLS) Algorithm has
along with the solution sets for each as well as the objective
been considered as estimator algorithm to carry out the functions considered in those problem formulation. All of those
estimation process. works mainly focused on minimizing the number of PMUs to
be installed in the system which will make the system
Keywords— State Estimation, Phasor Measurement Unit, observable as well as will fulfill several other specified
Optimal PMU placement, Artificial Bee Colony. objectives.
This paper presents a technique to find out the optimal
I. INTRODUCTION PMU locations in a power system. As it is known, that a bus if
State estimation is such kind of an imperative process installed with a PMU, it can provide the voltage phasors of that
which does the very important duty of ensuring power system bus as well as the current phasors of all or some of the adjacent
security by monitoring it precisely. A state estimator estimates branches connected to that bus depending upon the
the values of state variables (voltage magnitudes & phase communication facility availability in that bus location. A
angles) at the buses after it is provided with the measurements MATLAB based optimization tool along with Artificial Bee
like magnitudes voltages and power injections in the buses, Colony (ABC) is used to solve the optimization problem.
power flows and current flows in the branches [1-2]. Different solution sets are found for OPP which has been
Conventional Supervisory Control and Data Acquisition further investigated by doing state estimation. Performance of
(SCADA) of remote terminal units (RTU) installed at the the estimator is checked by placing PMUs in the optimal
substations provides the estimator with these measurement solution sets keeping the measurement redundancy level same
values [3]. After introducing Phasor Measurement Unit (PMU) for all which will enable us to compare between the solution
around the year 1990, it has become the most reliable tool in sets. Test cases with and without presence of bad data are
wide area monitory system (WAMS) which provides real time considered both for estimation. Widely used WLS algorithm is
bus voltage and branch current phasor measurements. The very used to carry out the state estimation for IEEE 14-bus system.
high refreshing rate of PMU has made it highly acceptable in In this paper, section II will present the details of
power system control compared with conventional SCADA optimization problem of PMU placement. Section III will
systems [4]. Besides of state estimation, PMU has got some present the two optimization tools used in getting optimal
more applications like transient stability analysis of power solutions. WLS algorithm will be briefly presented in section
IV. Part V will present all the simulation results with necessary Where AnXn is denoted as the connectivity matrix which
analysis and followed by the conclusion in section VI.
defines the connection pattern between the buses.
II. PROBLEM FORMULATION FOR OPTIMAL PMU LOCATION  1, if i  j

A PMU is different from the conventional meters in the Ai , j  1, if i and j are connected , (4)
sense that it has got the ability to provide not only the voltage 
phasors of the connected bus but also the current phasors of the  o, otherwise
branches which are attached with the bus [12-13]. In this
problem formulation, it has been considered that the buses are
X nX 1  [ x1 x2 x3 ... xn ]T ; a vector with decision variables
all equipped with proper communication facilities which will
 nx1
T
allow an installed PMU on a particular bus to make all its And b  1 1 1....1 (5)
neighboring buses observable. So, the problem formulation of
OPP has got the objective to get the minimal set of PMUs so For IEEE 14 bus system, the formulation is presented in figure
that each of the buses are at least reached once by the installed 4.
PMUs.
1
The objective function of the optimization problem for an .
n-bus system is [12]:
Subject to [ A]14 X 14 [ x]14 X 1  [b]14 X 1   
n .
min  xi ci (1) 
i 1 114 X 1
Where xi is the decision variable which denotes that whether a The connectivity matrix can be seen from the figure and can be
bus is installed with a PMU or not. For a 14 bus system, we formulated from equation (2) as follows:
will have 14 values of x starting from x1 upto x14 . So, xi can
1 1 0 0 1 0 0 0 0 0 0 0 0 0
be represented as: 1
 1 1 1 1 0 0 0 0 0 0 0 0 0 
1, if PMU is installed 0 1 1 1 0 0 0 0 0 0 0 0 0 0
xi   (2)  
0, Otherwise 0 1 1 1 1 0 1 0 1 0 0 0 0 0
1 1 0 1 1 0 0 0 0 0 0 0 0 0
Ci is the cost needed to install a PMU in a particular bus. It is  
0 0 0 0 0 1 0 0 0 0 1 1 1 0
considered that the installation costs will be same for each 0 0 0 1 0 0 1 1 1 0 0 0 0 0
PMUs wherever it is installed. A 
0 0 0 0 0 0 1 1 0 0 0 0 0 0
0 0 0 1 0 0 1 0 1 1 0 0 0 1
 
0 0 0 0 0 0 0 0 1 1 1 0 0 0
 
0 0 0 0 0 1 0 0 0 1 1 0 0 0
0 0 0 0 0 1 0 0 0 0 0 1 1 0
 
0 0 0 0 0 1 0 0 0 0 0 1 1 1
 0 0 0 0 0 0 0 0 1 0 0 0 1 1 

From the constraint equation (3), all of the 14 equations can be

formulated which will help the system to make the system
observable.
Suppose for bus 1 observability equation will be:

x1  x2  x5  1 ; which means that at least one PMU must

have to be placed in any of these three buses (1,2,5) so that
other two buses could reach the installed PMU with proper
Fig 1: IEEE 14 bus test system
communication channel and make the zone around bus 1
Min ( x1  x2  x3  ...  xn ) observable. Like the same way all other equations are formed
for each of the buses which will make the respective bus
Constraints of this optimization problem will be used to make “reachable” from the neighboring buses. Thus the entire system
the network observable which can be expressed as: will become observable by following all the 14 constraint
equations.
AnXn X nX 1  bnX 1 (3)
 III. OPTIMIZATION TOOLS FOR OPP Onlooker Bees Phase
For each onlooker bee:
a) CVX:
 Food source (solution) with better probability value pm,
To solve the convex optimization problems, CVX is widely calculated based on the fitness values provided by the
used which works in Matlab environment. It transforms employed bees. The probability of an individual being
Matlab into a modeling structured language, allow to selected by an onlooker bee is given by [16]:
incorporate objective functions and constraints using standard fitm
Matlab expression syntax which allows the Matlab users to pm  SN
(9)
deal with the tool very easily[14].
 fit
m 1
m

b) Artificial Bee Colony (ABC)  Determine new neighboring food source positions using
Artificial Bee Colony (ABC) is a meta-heuristic global equation (7)
optimization algorithm based upon the smart hunting and  Calculate the fitness value, fitm
practical behavior of drone flock. It is a promising method for  Similarly, as in employed bee phase, the best individual
engineering applications and since its invention in 2005 by D. is being updated after each iteration based on fitness
Karabog, it is being used for numerical optimization, values.
combinatorial optimization problems as well as unconstrained
and constrained optimization problems [15]. User defined size Scout Bees Phase
of population and cycle number are used as control The employed bees which cannot improve its solution by
parameters. Abandonment criteria is also predetermined by the using preset number of trials (known as limits or terminating
user. The ABC algorithm progresses in mainly four steps: condition) become scout and as such their solutions get
discarded. Thus, the scouts get busy looking for new solutions
Initialization in a random manner and is defined by equation (6) [18].
A fixed set of solutions or food source positions are
initialized by the scout bees which is denoted as x m, where the
value of m is bounded within 1 to SN (population size). In this
step, the controlling limits are also set. Arbitrary food sources
produced initially within the predefined range can be acquired
from the following equation [16]:

xmi  li  rand (0,1)*(u i  li ) (6)

u i and li are the lower and upper bound of xmi . After
initialization, the set of solutions are forced to undergo
through an iterative search process of the following three
phases [17].

Employed Bees Phase

For each employed bee:
 Determine new food source position, vm using the
expression:
vmi  xmi  mi ( xmi  xki ) (7)
where xk is a food source and i is a parameter index.
mi is a random number with the range [-a, a].
 Calculate the value fitm using: Figure 2: Flowchart demonstrating ABC algorithm
 1
 fm  0
fitm  1  f m (8) IV. STATE ESTIMATION USING WLS ALGORITHM
1  abs( f ) Weighted Least square represents an estimation problem that
 m fm 0
selects the criterion of a solution to the over-defined matrix
The m-th fitness value is the fitm and fm is the objective
equations when less states and more measurements are
function value of the m-th solution.
available in the power system [1-2]. Weighted least square
 vm’s with better fitness values than the corresponding
have the objective function that minimizes the square of the
xm’s are used in updating the xm’s. It is done by using a
error. Traditional state estimator utilizes SCADA
greedy selection.
measurements whose relationship with system states:
zi  hi  x   ei  x f  x 
 2 H   R H 
(10) T 1
where
x
zi is SCADA measurement vector and its size is
hi  x  correspond to a nonlinear function vector which relates And the complete formulation for updating states becomes
measurements to states
ei is the error vector between estimated and measured values  z1  h1  x  
 
of size  z2  h1  x  
Power System State Estimation is a system of overdetermined  
1 1  
xk 1  xk  x  xk   H   R   H   H   R  
 
T 1 T
nonlinear equations and must be solved as unconstrained WLS 
 
problem. For weighted least square we need to minimize sum  z  h  x 
of square of residuals:  m m

 
N  
 
min f ( x)  Wi 2 | hi ( x)  zi |2 (11)
∆x is the measurement mismatch, which is used as iteration
i 1
step for next iteration, for to exist non-singularity for gain
matrix is must. R (error covariance matrix of
1
min f  x    z  h  x   R 1  z  h  x  
T
(12) SCADA measurements), iterative procedure terminates when
2 goes below a certain low threshold value.

By taking partial derivative of with respect to state

vector , Jacobian matrix [H] will be obtained. V. SIMULATION RESULTS

 Pflow ij  Pflow ij   A) IEEE 14 bus system:

  As discussed above, simulation is carried out on the IEEE 14
  Vm  bus test system. In solving the optimization problem in placing
 Pflow  ji  Pflow  ji   PMUs in the proper location, MATLAB based optimization
 
  Vm  tool CVX is applied for the comparison with the proposed
 Pinj Pinj  Artificial Bee Colony (ABC) algorithm. Three different
  optimal solution sets are found from ABC optimization with
  Vm 
 Vai Vai  same number of PMUs (four) which can meet the objective
  Vm  function. The minimum number of PMUs required and the
  best bus locations for PMU placements are presented in table I
 Q flow ij  Q flow ij  
below for both the optimization tools.
 
 Vm
H  
 Q flow  ji  Q flow  ji   Table I: Results for Optimal PMU locations
  Optimized with Number of PMUs PMU locations
  Vm  CVX 5 2, 6, 7, 9, 13
 Qinj Qinj  4 Set1: 2, 6, 7, 9
  Artificial Bee Colony 4 Set2: 2, 7, 11, 13
  Vm  (ABC) 4 Set3: 2, 7, 10, 13
 Vmi Vmi 
  Vm 
  It can be observed from table I that ABC is giving better
 I F , Real ij  I F , Real ij   results with keeping the number of PMUs only 4 even though
  Vm  they are having some different sets of solutions. CVX
  optimized the problem with 5 PMUs. In all three sets of
 I F , Imag ij  | I F , Imag ij  | 
  solutions given by ABC, the system has become observable.
  Vm 
1) Without any Bad data:
According to Newton method to minimize a function f(x),
k 1
The test case has been made of measurement readings taken
where x  x k  x is from the conventional meters. PMUs are then placed in the
obtained optimal locations as presented in table I and state
1
1   f  x   estimation is carried out then to see their impact on the
x   f   x   f  x    x    x f  x  estimation performance. This will lead to (8)find out the optimal
 x  solution set. It has been done for each of the solution sets with
same redundancy level of measurement values so that a
comparison can be made between the solution sets. To make
the test cases, total of 42 measurements are taken with a Table III: SE indicator with readings from PMUs after affected with bad data
redundancy level of 1.55. To represent the estimator Optimize PMU locations Measurement Estimator Indicator
d with Redundancy
performance, an indicator has been used which is actually the Voltage Voltage
sum of the differences between all the estimated values and from PMU and
the base case values. Lower value of indicator indicates the current
from
better performance of the estimator. PMU
Initially, only the voltage magnitudes and phasors from PMUs CVX 2, 6, 7, 9, 13 1.55 7.3238 6.4680
are considered to run the estimation. So, for the case of 4 Set1: 2, 6, 7, 9 1.55 5.2336 5.0382
PMUs, we are considering 8 measurement values taken from Artificial Set2: 2, 7, 11, 1.55 5.2862 5.0217
Bee 13
PMU meters and rest are from conventional SCADA meters. Colony Set3: 2, 7, 10, 1.55 5.2014 4.9914
After that the estimation is carried out with considering the (ABC) 13
current phasors from PMUs. To keep the redundancy level
same, 8 conventional power flow readings are replaced by B) IEEE 30 bus System:
same amount of current measurements of PMUs. The results The work is then extended to the larger power system with 30
obtained are presented in table II below. buses. The superiority of ABC over CVX in optimizing
problems is being proved already in previous section.
Table II: SE indicator with readings from PMUs
Optimized PMU locations Measurement Estimator Indicator
Different optimal solution sets by Artificial Bee Colony
with Redundancy algorithm for placing PMUs are presented below in table IV.
Voltage Voltage
from and Table IV: Results for Optimal PMU locations by ABC
PMU current Number of PMUs PMU locations
from
PMU 9 Set1: 3, 5, 8, 10, 12, 18, 24, 25, 29
CVX 2, 6, 7, 9, 13 1.55 3.4668 2.8523 9 Set2: 1, 7, 8, 10, 12, 19, 23, 26, 29
Artificial Set1: 2, 6, 7, 9 1.55 3.4096 2.8256 9 Set3: 1,7, 8, 10, 12, 19, 24, 26, 30
Bee Colony Set2: 2, 7, 11, 1.55 3.4283 2.8016 9 Set4: 1, 5, 10,11, 12, 19, 24, 26, 27
(ABC) 13 9 Set5: 3, 5, 8, 10, 12, 18, 23, 26, 29
Set3: 2, 7, 10, 1.55 3.3529 2.7705 9 Set6: 3, 5, 8, 10, 12, 19, 23, 25, 29
13
9 Set7: 3, 5, 8, 10, 12, 15, 20, 25, 30
9 Set8: 1, 7, 8, 10, 12, 15, 19,25, 29
It is seen from the results that the estimator performance for
9 Set9: 1, 5, 10, 11, 12, 19, 23, 25, 27
CVX solution set is not better than ABC even if the PMU
numbers are higher. Inclusion of current phasors improved the
estimator performance for all the cases as the indicators are
much lower when current phasors are included. Results show 1) Without any Bad data: Test case has been made with
that the solution set3 for ABC with PMU locations in buses 2, the same redundancy level (1.55) as done for 14 bus system.
7, 10, 13 gives the best performance indicator with lowest Results are presented below in table V.
value. Table V: SE indicator with readings from PMUs
Estimator Indicator
2) With bad data: PMU locations
Voltage from Voltage and
A single bad data is then considered in the measurements PMU current from
series. Power flow of branch 4 to 7 has been taken -0.3526 PMU
now which was 0.3526 previously. Because of the presence of Set1: 3, 5, 8, 10, 12, 18, 24, 25, 29 12.0423 7.7182
Set2: 1, 7, 8, 10, 12, 19, 23, 26, 29 12.4724 8.0055
this bad data, indicator becomes 8.2919 which was 4.8916
Set3: 1,7, 8, 10, 12, 19, 24, 26, 30 12.3997 8.6287
before. Table III clearly indicates that the third solution set of Set4: 1, 5, 10,11, 12, 19, 24, 26, 27 11.9904 8.4069
ABC algorithm with PMU locations 2, 7, 10, 13 gives the best Set5: 3, 5, 8, 10, 12, 18, 23, 26, 29 12.3182 8.7552
solution with lowest indicator among all. All the results are Set6: 3, 5, 8, 10, 12, 19, 23, 25, 29 12.0081 8.2738
presented below in table III. Set7: 3, 5, 8, 10, 12, 15, 20, 25, 30 12.0463 8.2971
Set8: 1, 7, 8, 10, 12, 15, 19,25, 29 12.1249 8.2745
Set9: 1, 5, 10, 11, 12, 19, 23, 25, 27 11.7615 8.0818
Inclusion of current phasors in the place of power flows
improved performance as it happened without bad data too.
Only the voltage phasors are considered initially from PMUs.
The solution set of CVX, even providing an extra PMU to be
Current phasors are then considered by replacing same amount
installed, could not perform that well in estimation. Placing an
of power flows to keep the redundancy same. It is seen that
extra PMU will not be an economical choice too.
inclusion of current phasors improves estimation indicator for
all cases and it also changes the optimal solution set. PMU
Similarly, like 14 bus, state estimation is then carried out by
location set 9 was best among all when only voltage phasors
WLS algorithm after placing PMUs in each of the optimal
were considered. Inclusion of current phasors from PMUs
location sets.
made solution set 1 the best one with lowest indicator 7.7182.
 2) With Bad data: [3] Ali Abur, “Power System State Estimation: Theory and
Implementation,” 2004.
Bad data is then applied in one of the power flow values.
[4] A Phadke, J Thorp, “History and applications of phasor measurements”,
Estimator performance deteriorates as expected. Then the In: IEEE PES power systems conference and exposition, 2006
PMU measurements are considered as above and the results (PSCE’06):IEEE; 2006. p.331–5.
are presented below in table V. [5] Lin Y-H, Liu C-W, Chen C-S, “A new PMU-based fault
detection/location technique for transmission lines with consideration of
Table V: SE indicator with readings from PMUs after applying bad data arcing fault discrimination-Part I: Theory and algorithms”, IEEE Trans
Estimator Indicator Power Delivery 2004;19:1587–93.
PMU locations [6] Liu C-W, Thorp JS, “New methods for computing power system
Voltage from Voltage and
dynamic response for real-time transient stability prediction”, IEEE
PMU current from Trans Circuits Syst I: Fundam Theor Appl 2000;47:324–37.
PMU
Set1: 3, 5, 8, 10, 12, 18, 24, 25, 29 17.7770 14.3027 [7] Skok S, Ivankovic I, Cerina Z, “Applications based on PMU technology
for improved power system utilization”, In: IEEE power engineering
Set2: 1, 7, 8, 10, 12, 19, 23, 26, 29 18.2584 14.6064
society general meeting, 2007: IEEE; 2007. p. 1–8.
Set3: 1,7, 8, 10, 12, 19, 24, 26, 30 18.1936 15.0873
Set4: 1, 5, 10,11, 12, 19, 24, 26, 27 17.6991 14.8926 [8] P Yang, Z Tan, A Wiesel, A Nehorai, “power system state estimation
Set5: 3, 5, 8, 10, 12, 18, 23, 26, 29 18.0779 15.0889 using pmus with imperfect synchronization”, IEEE Transactions on
Power Systems, Volume:28 Issue:4, 2013.
Set6: 3, 5, 8, 10, 12, 19, 23, 25, 29 17.7401 14.7951
Set7: 3, 5, 8, 10, 12, 15, 20, 25, 30 17.7970 14.8547 [9] F Aminifar, M Fotuhi-firuzabad, A Safdarian, A Davoudi, M
Set8: 1, 7, 8, 10, 12, 15, 19,25, 29 17.9509 14.8538 Shahidehpour, “Synchrophasor Measurement Technology in Power
Set9: 1, 5, 10, 11, 12, 19, 23, 25, 27 17.4219 14.6587 Systems: Panorama and State-of-the-Art,” IEEE transactions,
10.1109/Access, 2015.
[10] Nikolas M. Manousakis, George N Korres, Pavlos S Georgilakis,
The optimal solution sets are similar as the cases of without “Taxonomy of PMU placement methodologies”, IEEE transactions on
applying bad measurement. Voltage measurements only from powers systems, Vol 27, No 2, May 2012.
PMUs give optimal solution set 9 as the best solution with [11] M. Nazari-Heris, B. Mohammadi-Ivatloo, “Application of heuristic
least indicator of 17.4219. Incorporation of PMU current algorithms to optimal PMU placement in electric power systems: An
updated review”, Renewable and Sustainable Energy Reviews 50 (2015)
measurements results solution set 01 as the best optimal 214–228.
locations as it gives the minimum indicator 7.7182. [12] Bei Gou, “Generalized Integer Linear Programming Formulation for
Optimal PMU Placement”, IEEE transactions on powers systems, Vol
VI. CONCLUSION 23, No 3, August 2008.
This paper focused on two main issues: to place PMUs [13] M Gol, A Abur, “Optimal PMU placement for state estimation
optimally which will not only reduce the installation cost but robustness”, Innovative Smart Grid Technologies Conference Europe
(ISGT-Europe), 2014 IEEE PES.
also make the overall system observable and to check the
[14] Grant, Michael, Stephen Boyd, and Yinyu Ye. "CVX: Matlab software
performance of the state estimator if the PMUs are placed in for disciplined convex programming." (2008).
the resultant optimal positions irrespective of the presence of [15] Rana, Madhurima, Saurabh Bilgaiyan, and Utsav Kar,"A study on load
bad data. It has been found that Artificial Bee Colony based balancing in cloud computing environment using evolutionary and
optimal position provides the best estimation performance. For swarm based algorithms", International Conference on Control
Instrumentation Communication and Computational Technologies
the case of running estimation, redundancy has been taken (ICCICCT), 2014.
considerably above 1 which will allow the estimator to work [16] Murugan, R. and Mohan, M. R.. "Artificial Bee Colony Optimization
even if some meters fail to provide service. Besides of PMU For The Combined Heat And Power Economic Dispatch Problem",
readings, some measurement readings has been taken from Journal of Engineering & Applied Sciences, 2012.
conventional meters with some sort of normal noises. The [17] Bahriye Akay. "Parameter Tuning for the Artificial Bee Colony
detail formulation of optimization problem has been presented Algorithm", Lecture Notes in Computer Science, 2009
as well as the formulation of the estimator. It has been [18] P. Mathiyalagan. "Hybrid enhanced ant colony algorithm and enhanced
bee colony algorithm for grid scheduling", International Journal of Grid
observed that the performance of the estimation process was and Utility Computing, 2011
improved when using ABC by indicating the optimal
placement of PMUs in the system. The results carried out
using the IEEE 14-bus and 30 bus system.

REFERENCES
[1] A. Abur and A. G. Exposito, “Power system state estimation: theory and
implementation,” CRC Press, 2004.
[2] A. Monticelli, “Electric power system state estimation,” Proc. IEEE, vol.
88, no. 2, pp. 262–282, Feb. 2000.