1

Determination of ZIP Parameters with Least

Squares Optimization Method
M.Sadeghi, G.Abdollahi sarvi

identify parameters for more detailed load models. The data is

Abstract—An appropriate operation of a power system in obtained either from measurements in-site, and includes
steady state situation and the equal values of bus voltages and the voltage and frequency variations, and the corresponding
nominal ones, depends on the probable changes in the load, so the variations in active and reactive load, either to intentional
exact description of the load is very important, on the other hand
disturbances, test measurements, or to natural events, normal
the load characteristics have a significant effect on the
performance of a power system. Until now the constant operation data. The main advantage of using a measurement-
representation of the loads was assumed in load flow programs based approach is the availability of actual data from the
and equations. Because of changing in power, current and system under study, and the possibility to track seasonal
impedance of the load of the power system along the day, the variations but also deviations from normal operation.
constant power model for the power system can not be correct. Measurement-based techniques have the advantage of
This paper presents the algorithm of least squares to determine
obtaining data directly from the actual system. However, there
the parameters of “ZIP” load model. This algorithm is tested on a
2 bus test system to show the applicability of this method. It is are several disadvantages, including [11]:
noted that this method is a measurement based approach. ¾ Application of data gathered for load models at one
substation may not be possible for other substations .
Keywords—Load Modelling, Dynamic Model, Static Model, ¾ Determination of characteristics over a wide range of
Least Squares, ZIP Model. voltage and frequency may be impractical.
¾ Accounting for variation of load characteristics due to
I. INTRODUCTION daily, seasonal, weather and end-use changes requires

A CCURATE power system models are necessary in order

to reduce power system operational uncertainty.
Decisions concerning system reinforcements and system
on-going measurements under these varying
conditions.
The component-based approach involves developing a
performance are based in large on the results of power flow composite load model from information of its constituent
and stability simulation studies. In performing power system parts, i.e, mix of classes at the substation, composition of each
analysis, accurate models must be developed for all of the of those classes, and main characteristics of each single load
system components including load components. Load component, so the purpose of component-based approach is to
characteristics have a significant effect on system performance develop load models by aggregating models of the individual
and the results of load flow, voltage and transient stability components forming the load. Individual components
simulations are known to be highly dependent on the type of characteristics have been determined by theoretical and
load characteristics assumed. Accurate models of different laboratory analysis.
complexity for generators, lines and transformers are available The component-based approach has the advantage of not
today, whereas load models are usually simplified. requiring field measurements and of being adaptable to
Two basic approaches to load model development have different systems and conditions and it is much easier to put
been considered by the electric utility system. into use since load characteristics and load composition data
• Measurement-based Models. should not vary widely over a particular system, they can be
• Component-based Models. developed once for the entire system. The load class mix data
The measurement-based approaches involves direct varies from bus to bus and is dependent on weather and time,
measurements at representative substations and feeders by it is necessary to often determine and update the load class
placing monitors at various load substations to determine the mix data for each bus of the system and this is the main
sensitivity of the load active and reactive power to voltage, disadvantage of the method. Fig.1 shows the terminology for
frequency and weather variations to be used directly or to component-based load modeling [11] , [12].

M.Sadeghi is with the Center of Excellence for Power System Automation

and Operation , Electrical Engineering Department , Iran University of
Science and Technology (IUST) , Narmak , Tehran , Iran (email:
[email protected]),
G.Abdollahi sarvi is with Electrical an Engineering Department of Khaje
Nasiredin Toosi University(Kntu), Seyed Khandan, Tehran, Iran
(email:[email protected])

2009 IEEE Electrical Power & Energy Conference

978-1-4244-4509-7/09/$25.00 ©2009 IEEE
 2

based bibliography on load modeling papers complement the

IEEE Task Force paper analyzing and organizing standardized
load models.

II. THE FAMOUS LOAD MODELS

A. PQ Load Model
In most load flow programs and simulations, the active and
reactive consumption of the load is modeled as costant powers
as shown in equations (1) , (2) [13].

Fig.1. Terminology for the component-based load P = PL (1)

modeling
(2)
The issue of load modeling is addressed in various Q = QL
references. Ref [1] introduces a dynamic reactive load model As the voltage of the buses does not exceed from the
for power system dynamic voltage stability studies. This special limits, the above equations are correct. But when the
model is a measurement-based and incorporates the effect of voltages exceed from the limitations, the load consumption
voltage regulating equipment, specifically, switched will change to the constant impedance that is shown in
distribution capacitor banks and subtransmission substation equations (3) , (4) [13].
on-load tap changer transformers. A set of parameters that
reflect these delayed-action devices are involved in the P.V 2
proposed model. A parameter estimation scheme is introduced P= 2
(3)
with the aim of minimizing errors between model and Vlim
collected data at a real test system. In Ref [2] the mixture data
obtained from billing information has been used with the (4)
Q.V 2
LOADSYN program to develop dynamic load models for Q= 2
large scale studies. Ref [3] has developed a method for Vlim
preparing load models for power flow and stability. The load
modeling (LOADMOD) computer software transforms data Vlim is the limitation of the voltage.
on load class mix, composition and characteristics into the
B. General Static Load Model
form required for commonly-used power flow and transient
stability simulation programs. Ref [4] recommends standard Equations (5) , (7) recommended a model consisting of ZIP
load models for power flow and dynamic simulation terms plus two voltage/frequency dependent terms [4].
programs. The goal of this paper is to promote better load
P V 2 V
modeling and advanced load modeling, and to facilitate data = K pz ( ) + K pi ( )+ K pc +
P frac . P0 V0 V0
exchange among users of various production-grade simulation
programs. In this paper, for transient stability, longer term V npv 1
K P1 ( ) (1 + n pf 1 Δ f ) + (5)
dynamics, and small-disturbance stability programs, is V0
recommended the structure of multiple load types connected V npv 2
to a load bus. For longer-term dynamics programs, a model for K p2 ( ) (1 + n pf 2 Δf )
V0
LTC transformers is also recommended. Ref [5] deals with the
calculation of representative static and dynamic load models
based on component-based approach for computer simulation (6)
packages. Dias and Hawary in Ref [6] considered the problem K pz = 1 − ( K pi + K pc + K p1 + K p 2 )
of estimating the parameters of static power system load
models intended for use in load flow studies that incorporate Where Pfrac is the fraction of the bus load represented by the
the variation of active and reactive power with busbar static model [4].
voltages. Ref [7] described laboratory measurements and Q V 2 V
= K ( ) + K ( )+ K +
derived models of modern loads subjected to large voltage Q frac .Q 0
qz
V0
qi
V0
qc

changes and their effect on voltage stability studies. Ref [8]

V nqv 1
presents some field data illustrating the response of several K q1 ( ) (1 + n qf 1 Δ f ) +
types of load to small voltage deviations. Model parameters V0 (7)
were derived from these measurements for each load type V nqv 2
K q2 ( ) (1 + n qf 2 Δ f )
using curve-fitting techniques for both of its static and V0
dynamic components. Ref [9] presents the voltage dependence
of composite loads in power systems. In Ref [10] a broad-
 3

III. ZIP MODEL AND THE LEAST SQUARES OPTIMIZATION

K qz = 1 − ( K qi + K qc + K q1 + K q 2 ) (8)
A. ZIP Model
Where Qfrac is the fraction of the bus load represented by the As mentioned before, generally, the constant active &
static model. reactive power loads is used in the general load flow
In the above equations, P0 and Q0 are the initial active and programs, but it should be considered that current, voltage and
reactive of power consumptions from the power flow base impedance of the load in all the times are not constant and
case. They are the nominal load powers at nominal voltage varied dependent on the load, so the results of constant power
and frequency. P and Q are the consumed load powers that are consumption based calculations can not be correct.
related with voltage and frequency. The constant current, constant power and constant
The equations (5) , (7) are the main and key models. Other impedance model that is known as “ZIP” model is one of the
models of loads are derived from the fifth and seventh most important and appropriate models.
equations. Equations (9) , (10) are the important derived This model is presented and discussed about in this paper.
equations. This model is formulated in different shapes that is shown
B.1. Exponential Model in equations (17) ,(18) [4]:
As mentioned above, this model is a derivation of the last
model [10]. P ⎡ V 2 V ⎤ (17)
= ⎢ P1 ( ) + P2 ( ) + P3 ⎥ (1 + L DP Δ f )
⎛ V V ⎞ P0 ⎣ V 0 V0
P = P0 ⎜⎜ Pa1 ( ) KPV 1 (1 + KPF 1 × Δf ) + Pa 2 ( ) KPV 2 ⎟⎟ (9) ⎦
⎝ V0 V0 ⎠
⎡ V V ⎤
P = P0 ⎢ Z p ( ) 2 + I p ( ) + Pp (1 + K p f ) ⎥ (18)
⎛ V V ⎞ ⎣ V 0 V 0 ⎦
Q = Q0 ⎜⎜ Qa1 ( ) KQV 1 (1 + KQF 1 × Δf ) + Qa 2 ( ) KQV 2 (1 + KQF 2 × Δf ) ⎟⎟
⎝ V0 V0 ⎠ (10)
These models are other derivations of the general model in
Pa 1 + Pa 2 = 1 (11) the fifth and seventh equations.
This paper ignored from the frequency variations, so the
Q 0 discussed model is in equations (19) , (20) [13]:
Q a1 + Q a 2 = (12)
P0
⎡ V V ⎤
P = P0 ⎢ Z p ( ) 2 + I p ( ) + Pp ⎥
B.2. Residential, Commercial, Industrial Load. ⎣ V0 V0 ⎦ (19)
This model is a derivation of the main model in equations
(5) , (7) [10]:
⎡ V V ⎤ (20)
α (13)
Q = Q 0 ⎢ Z q ( ) 2 + I q ( ) + Pq ⎥
P = P0 .V ⎣ V0 V0 ⎦
Q = Q0 .V β (14)
Z p + I p + Pp = 1 (21)
B.3. Residential, Commercial, Industrial and Building Loads
This model is a derivation of the model in equations (5) ,
(7) too [10].
Z q + I q + Pq = 1 (22)

P = K .V np (15)
P0 , Q0 , V0 are the nominal values in these equations.
B. Least Squares Method
Q = K 0 + K n .V nq (16)
ZIP parameters can be determined by some methods. The
optimization method that is used in this paper is Least Squares
There are other models that can be derived from the key method.
and the main load model in the equations (5) , (7) and we do Least Squares regression is the most basic form of LS
not want to write all of them in this paper. There are a lot of optimization problem. Suppose you have a set of
nonlinear equations about load models that is presented in measurements yn gathered for different parameter values, xn.
many papers but a collection of these models is presented in the LS regression problem is to find the relation (23).
Ref [10].

N
min
P
∑
n =1
( y n − px n )2 (23)
 4

IV. CALCULATION OF PARAMETERS

The objective function that should be minimized is as
equation (24): ⎡ n n n
⎤ ⎡ n 2⎤
⎢ ∑ 2Vi ∑ 2V ∑ 2V ⎢ ∑ 2 Pi .Vi ⎥
4 3 2
2 i ⎥
i
n
⎛V ⎞ ⎛V ⎞ ⎛P ⎞ ⎢ i =1 i =1 i =1
⎥ ⎡Z P ⎤ ⎢ i =1
⎥ (32)
λ= ∑ ( Z P ⎜⎜ i ⎟⎟ + I P ⎜⎜ i ⎟⎟ + PP − ⎜⎜ i ⎟⎟ ) 2 (24) ⎢ n
⎢ ∑ 2Vi
3
n

∑ 2V 2
∑
n
⎥ ⎢ ⎥ ⎢ n ⎥
2Vi ⎥ × ⎢ I P ⎥ = ⎢ ∑ 2 Pi .Vi ⎥
i =1 ⎝ V0 ⎠ ⎝ V0 ⎠ ⎝ P0 ⎠ i
⎢ i =1 i =1 i =1 ⎥ ⎢ P ⎥ ⎢ i =1 ⎥
⎢ n n
⎥ ⎣ P⎦ ⎢ n ⎥
⎢ ∑ 2Vi ∑ 2V ⎢ ∑ 2 Pi
2
The Vi and Pi are the input values. These are the voltage i 2n ⎥ ⎥
⎣ i =1 i =1 ⎦ ⎣ i =1 ⎦
and the power consumption at the correspond voltage of a bus.
To simplify the above relation, we assume that the Vi and Pi
values are in pu base. So the above equation (24) is simplified V. TEST SYSTEM
to the equation (25): In order to show the applicability of the presented
algorithm for calculating of the load model, the method is
n applied on a 2 bus test system. A computer program has
∑ (Z ) 2
λ = P .V i 2 + I P .V i + PP − Pi (25) written in matlab7 to calculate the algorithm on this system,
i =1 but the algorithm is applicable on any test system with more
than 2 buses. Fig.2 shows the test system.
The procedure of solving the problem is to derive from the
above equation to each parameters(Zp , Ip , Pp ). After
derivation, each equation should be equal to zero. At last there
will be 3 equations in 3 variables that should be solved.
The solving process is as follows:

dλ n

dZ P
= ∑ 2V (Z
i =1
i
2
P .V i 2 + I P .V i + PP − Pi = 0 ) (26)

dλ n

dI P
= ∑ 2V (Z
i =1
i P .V i 2 + I P .V i + PP − Pi = 0 ) (27) Fig.2 The 2 Buses Test System

The test system has 2 buses. The first bus is the reference
dλ n

dP P
= ∑ ( 2 Z P .V i 2 + I P .V i + PP − Pi = 0 ) (28) bus and the second one is the load bus. The load is consist of a
i =1 constant consumption with 50MW active and 30MVAr
reactive powers. a constant impedance Z=1+2i is connected to
this bus. A line with a negligible impedance connects the
These equations are simplified to equations (29) , (30) , reference bus to the load bus. So the load consist of (1)
(31) : constant power , (2) constant impedance. The input values are
in the table (1).
n n n n
The inputs are resulted from the load flow program with
Z P .∑ 2V i 4 + I P .∑ 2V i 3 + PP ∑ 2V i 2 = ∑ 2V i
2
. Pi (29) variation in voltage of the first bus.
i =1 i =1 i =1 i =1

n n n n
Z P ∑ 2V i 3 + I P ∑ 2V i 2 + PP ∑ 2V i = ∑ 2V P i i
(30) VI. SIMULATION RESULTS
i =1 i =1 i =1 i =1
The load model (33) , (34) is obtained from the algorithm:

n n n
Z P ∑i =1
2V i 2 + I P ∑
i =1
2 V i + PP . 2 n = ∑
i =1
2 Pi (31) P = 0 / 2857 × V 2
+ 0 / 7143 (33)

The matrix equation (32) is resulted from the above

2
equations. Q = 0 . 5714 × V + 0 / 4286 (34)
So the matrix of variables is calculated from the reverse of
the matrix (3*3) multiplied in the matrix (3*1). This models show that there is (35) and (36) percentages in
the load.
 5

The errors of the algorithm resulted from the software are

⎧ 28 / 57 %
⎪0 8/5203e-22 for the active load model and 6/5395e-21 for the
⎪ % (35) reactive load model.
⎨ Activ
⎪ 71 / 43 % Table.1 The Input Values
⎪⎩ Eror = 8 . 5203 e − 22
Q(MV Q(MV
V(pu) P(MW) V(pu) P(MW)
Ar) Ar)
0/90 66/200 62/400 1/01 70/402 70/804
⎧57 / 14 % 0/91 66/562 63/124 1/02 70/808 71/616
⎪0 % Reactive
⎪ (36) 0/92 66/928 63/856 1/03 71/218 72/436
⎨ 0/93 67/298 64/596 1/04 71/632 73/264
⎪ 42 / 86 %
⎪⎩ Eror = 6 / 5395 e − 21 0/94 67/672 65/344 1/05 72/050 74/100
0/95 68/050 66/100 1/06 72/472 74/944
0/96 68/432 66/864 1/07 72/898 75/796
In order to check the correctness of the above relations, we 0/97 68/818 67/636 1/08 73/328 76/656
calculate these parameters with another formulations. It is 0/98 69/208 68/416 1/09 73/762 77/524
considered from the Fig.1 that there is a constant active power 0/99 69/602 69/204 1/10 74/200 78/400
50MW (0/71429pu) and 30MVAr (0/42857pu) with initial 1/00 70/000 70/000
value (base case) 70 MVA. The constant current load is equal
to zero as seen in Fig.1, so the residue load is for constant
impedance(0/28571 pu for active load and 0/57143 pu for VII. CONCLUSION
reactive load).
As considered, all calculations from the methods have the Some important load models are shown in this paper and
same results. the ZIP model is selected as an appropriate load model to
Fig.3,4 show the modeled obtained from the algorithm(P- study. This model is consist of 3 terms:
V, Q-V curves). • constant power
• constant current
• constant impedance
The least squares optimization method is selected to
determine the parameters of the load model. The results
obtained from the test system with the negligible errors show
the applicability and the correctness of the method. So, this
algorithm can be applied on any systems to obtain the load
models.
It is noted that this model is independent of frequency
variations.

VIII. REFERENCES
Fig.3 Load Model with the Input Values(P-V) [1] Y.Liang, C.O.Nwankpa, R.Fischl, A.DeVito and S.C.Readinger,
"Dynamic Reactive Load Model," IEEE Trans. Power Systems, vol. 13,
No. 4, November 1998.
[2] Ebrahim Vaahedi, H.Magdy Zein El-Din, William W.Price, "Dynamic
Load Modeling in Large Scale Stability Studies," IEEE Trans. On
Power Systems, Vol. 3,No.3, August 1988.
[3] M.Sedghizadeh, A.Rezazadeh, “Load Modeling for Power Flow and
Transient Stability Computer Studies at BAKHTAR Network,”
Proceeding of World Academy of Science Engineering and Technology
, Vol. 26, December 2007.
[4] IEEE Task Force, “Standard Load Models for Power Flow and
Dynamic Performance Simulations,” IEEE Trans. Power Systems, Vol.
10, No. 3, August 1995.
[5] ML.Coker, H.Kgasoane, “ Load Modeling,” IEEE Conference, Vol.2,
pp.663-668, September 1999.
[6] Dias, L.G. , M.E.El-Hawary, “Nonlinear Parameter Estimation
Experiments for Static Load Modeling in Electrical Power Systems,:
Fig.4 Load Model with the Input Values(Q-V) IEE Proceeding, Vol. 136, No.2, pp. 68-77, Mar. 1989.
[7] Hajagos, L.B.Danai, “ Laboratory Measurements and Models of Modern
Loads and Their Effect on Voltage Stability Studies,” IEEE Trans. Vol.
13, No. 2, pp.584-591, May 1998.
[8] Beghzouz, Y.C.Quist, “ Composite Load Model Derivation from
Recorded Field Data,” pp. 713-718, IEEE1998.
 6

[9] Ohyama, T.A.Watanabe, K.Nishimura, S.Tsuruta, “Voltage Dependence

of Composite Loads in Power Systems,” IEEE Trans. PAS-104, No. 11,
pp 3604-3073, Nov. 1985.
[10] IEEE Task Force, “ Bibliography on Load Models for Power Flow and
Dynamic Performance Simulation,” IEEE Trans. Vol. 10, No. 1, pp.
523-538, Feb. 1995.

[11] Khaled Hamed Al Ghamedi, "Environmental Impact on Residental Load

Modeling," MSC. dissertation, Dept. Electrical Eng. , Univ. King Fahad
University of Petroleum and Minerals Dhahran, Saudi Arabia, May
2003.
[12] Ines Romero Navaro, “Dynamic Load Models for Power Systems,”
BSC. Dissertation, Dept. Industrial Electrical Eng., Univ. Lund ,
Sweden, September 2002.
[13] M.Haghifam, K.E.Jome, B.Khaki, “Analyzing The Effect of Different
Load Models on Load Flow of the Distribution Power Systems,” The
Ninth Students Electrical and Electronic Eng. Conference, Tehran, Iran.