Chapter 6. Automatic Determination of Parameters
6.5 Normalization of Dynamic Reactive Load Models
times the load time constant, we will also obtain an accurate value for the steady state behavior and for the time constant.
Figure 6.11: Exponential recovery of the load.
The common value Qo used in the normalization of the reactive load models may be close to zero due to the effect of reactive compensation, and therefore the reactive load parameters that describe the load response will deviate to very large values. The measured data is collected with different amounts of reactive compensation. This makes it possible to describe how the reactive load parameters change with the value of Qo.
6.5.1 Determination of parameters in reactive load models
Based on the described nonlinear model in Chapter 5, equations (5.1) and (5.2), the corresponding reactive load representation is given by (6.1), (6.2), and characterized by three parameters, reactive steady state load-voltage dependence, reactive transient load-voltage dependence and a load-recovery time constant.
W V
8R 4R 8 8R
4R 8 GW 4
7T G4U U β β
−
=
+ (6.1)
W
8R 4R 8 4
4O U β
+
= (6.2)
Uo and Qo are the voltage and power consumption before a voltage step change. Qr is the reactive power recovery, Ql is the total reactive power response, Tq is the reactive load recovery time constant, βt is the transient reactive load-voltage dependence, and βs is the steady state reactive load-voltage dependence. In analogy with Chapter 5 the non-linear model is simplified to a linear identification problem by linearizing around an operating point (U*). New quantities have been introduced, and the state space representation is given by (6.3) and (6.4):
0 1
0 1
0
*
*
U u U U
Bq U U
Aq U
s t
s t
=∆
⋅
=
⋅
=
−
− β
β
β β
U U
T U O
4 X 4
$T X 4 GW %T 7 4
X 4
$T 4 4
∆
−
⋅
⋅
−
⋅
⋅
∆ =
⋅
⋅
⋅ +
∆
=
∆
By introducing Tq1=Tq⋅Aq, the reactive load variation ∆Ql is given by (6.5). A similar equation can be derived for the active load variation.
8
8 7T V
7T V
%T 4
4O ⋅∆
⋅ +
⋅
= +
∆
The transfer function (6.5) represents the load response when a voltage change is occurring in the system. It is characterized by three parameters, reactive steady state voltage dependency Bq, reactive transient voltage dependency Aq, and reactive time constant Tq. ∆Ql and ∆U represent deviations from the steady state values Qo and Uo. The reactive load equation has been normalized by using Qo. The aim is then, to use the described linearised model (6.5) and to identify its parameters with measured data. The least squares optimization algorithm is used in the identification.
6.5.2 Normalization in dynamic load models
The identification process proposed above has been applied to different sets of data, as exemplified by test No.6.1 and 6.2.
Test No.6.1
The test, check Section 4.2, is based on previous measurements in the power system in the South of Sweden, and they represent an extension of previous studies [Le Dous, 1999], [Karlsson, 1992]. The data originates from six different experiments in the 400-130 kV-transmission system during June 1996. The voltage changes were made by simultaneous manual operation of tap changers on the 400/130 kV transformer. Six different sets of data have been used in the simulation.
No ∆V/V0 [%]
Tp
[s]
αt αs
1 -1.8 135 1.36 0.25
2 +1.9 40 1.70 -0.10
3 -3.7 61 1.31 -0.16
4 +3.7 74 1.35 -0.54
5 -5.3 70 1.65 -0.32
6 +5.4 78 1.60 -0.08
Table 6.5: Identified parameters for active load response under six different voltage steps.
No ∆V/V0
[%]
Tq
[s]
βt βs Qo
[Mvar]
1 -1.8 89 -181.80 7.90 -0.974
2 +1.9 256 -687.20 975.60 -0.046
3 -3.7 88 87.67 31.86 -2.850
4 +3.7 105 104.50 -148.56 0.867
5 -5.3 78 77.89 19.35 -2.840
6 +5.4 94 94.18 -49.75 1.254
Table 6.6: Identified parameters for reactive load response under six different voltage steps. Normalizing factor of the reactive load Qo.
The identified parameters for the active load, Table 6.5, exhibit low variability and correspond to acceptable values. The reactive parameters, however, are less reliable. Even though the reactive transient (βt) and steady state (βs) parameters fit the proposed model in each case, they deviate from the expected values. An important factor contributing to this may be that the value Qo used in the normalization in equations (6.2) may be close to zero because of the effect of reactive compensation. By checking the factor Qo
in Table 6.6 it is observed that for those values closer to zero, the deviation in the identified parameters is larger. Figure 6.12 shows a simple representation of the total load as the sum of the effect of the active and reactive power.
Figure 6.12: Total load So, as the sum of the active and reactive effect, Po and Qo, when reactive compensation is and is not needed, figure b) and a) respectively.
The load response in figure 6.12.a is not affected by reactive compensation while it is in figure 6.12.b. The value Qo is reduced to a value closer to zero due to the effect of capacitors, but Po remains constant. Since theoretically, when Qo tends to zero, the model parameters will go to infinity, it is necessary to normalize the reactive load model by some other factor that is not affected by this phenomenon.
Suggested candidates are Po or So. By normalizing by So the reactive load representation is then given by equations (6.6) and (6.7):
t s
Uo So U Uo
So U dt Q
Tq dQr r
χ χ
−
=
+ (6.6)
t
Uo So U Q Ql r
χ
+
= (6.7)
Uo and So are the voltage and apparent power levels before a voltage step change. Qr is the reactive power recovery, Ql is the total reactive power response, Tq is the reactive load recovery time constant, χt is the transient reactive load-voltage dependence, and χs is the steady state reactive
load-6R 3R
4R
6R 4R
3R
Figure 6.12.a)
Figure 6.12.b)
voltage dependence. Table 6.7 shows the results obtained by normalizing the reactive dynamic load in equation (6.2) by So. The quality of the reactive power parameter estimates is now similar to those for the active power.
No ∆V/V0
[%]
Tq
[s]
χt χs
1 -1.8 88 2.43 0.12
2 +1.9 165 2.02 -0.64
3 -3.7 88 2.08 -0.77
4 +3.7 89 2.42 -0.98
5 -5.3 78 2.22 -0.48
6 +5.4 92 2.62 -0.49
Table 6.7: Identified parameters for reactive load response under six different voltage steps. Normalizing factor of the reactive load So.
Test No.6.2
The data originates from a continuous acquisition from normal operation data in the 130-50-20 kV distribution system in the South of Sweden, see Section 4.3. Due to the fact that the continuous acquisition of data provides large amount of information that corresponds to non-stationary load data sequences, i.e. every data case differs from the others, it has been possible to study the impact of reactive compensation during daily operations on the representation of the reactive load.
Measurements at the 20 kV-level during February 2002 have been selected for the analysis. The identification procedure described in this thesis has been applied to the mentioned data.
Figure 6.13: Effect of the reactive compensation in the identification of the parameter χt, using the normalization factor Qo and So respectively.
Figure 6.13 shows the result of the identification for the reactive transient part of the load, plotted against the power level Qo, when the reactive load equation has been normalized by Qo, asterisk ‘*’, and by So, circle ‘o’, respectively. By normalizing with Qo, when the value Qo goes to zero, the transient parameters, βt, tend to plus or minus infinity, and they apparently deviate significantly from their normal values. On the other hand, by normalizing with So, the transient parameter χt, exhibits small variability.
When the value Qo is in the neighborhood of 0, the parameters do not deviate from the physically expected values , since So is hardly affected by the effect of reactive compensation.
Figure 6.14 presents the result of the identified parameter χt, when the normalization factor is So. The identified parameters remain in a limited interval.
Qo aa
-3 0 3
-150 0 150
Qo βt(Qo)
χt(So)
βt (Qo) χt(So)
Figure 6.14: Effect of the reactive compensation in the identification of the parameter χt using the normalization factor So.
6.5.3 Conclusions
The effect of normalization on dynamic load models has been studied. The factor Qo previously used to normalize the reactive load model is inappropriate since it may be equal to zero due to the effect of switching capacitor operations. The identification of parameters for the reactive load model when normalizing by Qo shows that these parameters tend to infinity when Qo goes to zero. On the other hand, an accurate representation of the reactive load is achieved by normalizing with the factor So, since this factor is hardly affected by reactive compensation.
15 20 25 30 35
0 0.5 1 1.5 2 2.5
So χt
103