Electrical Power and Energy Systems 30 (2008) 547–552

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Characteristics of the dynamics of distribution electrical networks

Bin Huang a,*, E. Handschin b
a
China Electric Power Research Institute, No. 15, Xiaoying East Road, Beijing 100085, China
b
Institute of Power Systems and Power Economics, University of Dortmund, 44227 Dortmund, Germany

a r t i c l e i n f o a b s t r a c t

Article history: This paper quantifies the characteristics of the distribution network dynamics by use of eigenvalue anal-
Received 17 August 2007 ysis. The study reveals that: (1) the parameters of the lines and cables, rather than the network topology,
Received in revised form 18 July 2008 are the predominant factors to determine the distribution of the network eigenvalues; (2) the range of
Accepted 15 August 2008
them is bounded by the relevant values of the lines and cables; (3) in a real distribution system in Europe,
the real parts of the eigenvalues should be within 69.3 to 33,055, namely, the dynamic procedure of
the network itself dies down to 5% of its initial value within less than 0.043 s and (4) thus after the inte-
Keywords:
gration of DG, adopting a steady state model to represent the distribution network is acceptable.
Electrical network dynamics
Electrical network model
Ó 2008 Elsevier Ltd. All rights reserved.
Distributed generation (DG)

1. Introduction after the integration of DG, though the results at the end may still
be right. To meet the need, eigenvalue analysis is employed in this
The growing trend of distributed generation (DG) is easily rec- paper to achieve some quantified conclusions.
ognized because of various political, economic and environmental
reasons. The DG, on the one hand, usually has a relatively large
capacity up to several megawatts when compared to the tradi- 2. Modelling of distribution electrical networks
tional electricity sources/loads connected to the distribution net-
work; on the other hand, usually operates more or less in a 2.1. Relation of network eigenvalues under abc and dq0 frames
stochastic manner. These lead to a concern whether the appar-
ently continuous and large amplitude stochastic inputs from DG The electrical network is modelled under dq0 reference frame.
will induce harmful responses in the distribution electrical net- The reason is if the analysis demonstrates the negligence of the
work (as in Fig. 1, and the DG is considered as a stochastic network dynamics is inappropriate, the developed model can be
source, namely no detailed DG type is referred, to represent the incorporated to other system component dynamic models for the
problem in a more general sense). A key point for the investiga- consequent analysis. Kron transformation [5] is adopted for the
tion is how to treat the electrical network dynamics. If they are modelling and the reference frame is designated as: the q-axis
ignored, one can expect to adopt Monte–Carlo method through coincides with the system reference phasor and leads the d-axis
power flow calculation [1] or use frequency scanning techniques by 90°.
[2] to achieve some results. If not, the problem will be more The theoretical precondition for investigating the properties of
complicate. the network dynamics under abc frame by evaluating the network
Though it is well known the electrical network dynamic proce- eigenvalues under dq0 frame, is the eigenvalues under abc frame
dure dies out quite fast so that it is often neglected [3,4], in distri- can be reconstructed from those under dq0 frame. This is feasible
bution systems, since the parameters of the lines/cables as well as based on the following observation: a symmetric network is as-
the topology differ a lot from those of a transmission network, it is sumed with 3N state variables, namely 3N eigenvalues. Under
still a question if the ignoring of the network dynamics is correct. dq0 reference frame, these 3N eigenvalues consist of N pairs of
To the authors’ knowledge, no documents discuss in detail to quan- eigenvalues which are complex-conjugate/double-real ones and
tify the characteristics of the electrical network dynamics, espe- relate only to the dq components (lately referred to as kdq), and N
cially those of the distribution systems. Thus, it is not rigorous to eigenvalues which relate only to the 0 components (lately referred
directly adopt the steady state model to represent the network to as k0). Correspondingly, under abc frame the 3N eigenvalues can
be divided into two groups, one contains N pairs of eigenvalues (la-
tely referred to as kabc2N), the other includes the rest (lately referred
* Corresponding author. to as kabcN). Then the eigenvalues under abc and dq0 reference
E-mail address: [email protected] (B. Huang). frame have the following relations:

0142-0615/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijepes.2008.08.001
548 B. Huang, E. Handschin / Electrical Power and Energy Systems 30 (2008) 547–552

model has nine eigenvalues under abc frame as given in Fig. 2b.
Under a balance operation, only the dq circuit components are of
interests. Thus in Fig. 2c only kabc2N is kept (the counterpart of
kdq) while kabcN is dropped. The CE of the given system is deliber-
ately selected as in Fig. 2d, which preserves all the necessary infor-
mation of kabc2N. Compared to Fig. 2c, the pair of zero eigenvalues is
dropped in Fig. 2d, because they have trivial meaning and their
occurrence as well as the quantity can be expected by investigating
the network topology. A further discussion is given in Appendix B.

3. Simulations and results

3.1. A brief view of the study networks

Fig. 1. Potential problems caused by the stochastic input of DG. In this work, two networks are available for the investigation:
Study network I (SNI)—which is based on the IEEE 34 node test
feeder (Fig. 3) [6]. Some modifications are made as follows: (1) the
 The eigenvalues of kabc2N are assumed as r1 ± jx1. After the dq0 two voltage regulators sited in line segment 814–850 and 852–832
transformation, the corresponding eigenvalues are r1 ± are omitted. The transformer sited in line segment 832–888 is re-
j(x1  x0) and r1 ± j(x1 + x0). placed by a line with a 1 km length. (2) All lines are assumed to be
 The eigenvalues of kabcN hold unchanged after the dq0 symmetric with the parameters given in Section 2.2.
transformation. Study network II (SNII)—the structure and data are from a real
distribution system in Germany (Fig. 4). Compared to SNI, this net-
Detailed discussions are given in Appendix A. work connects to the transmission system through node 1 as well
as node 2. More information is available in Appendix C.
2.2. Concept of the characteristic eigenvalues Later on, the analysis focuses in detail on the SNI. Then the con-
clusions are verified by studying the SNII.
Before going further, the term ‘characteristic eigenvalues’ (CE) is
introduced in advance. Fig. 2a presents a simple system: a load is 3.2. Properties of the network eigenvalues
fed by a generator via a transmission line. The length of the line
is 1km and the parameters are as: R = 0.128 X/km, X = 0.122 X/ 3.2.1. Distribution of the characteristic eigenvalues of the study
km, B = 116.239 lS/km (R the resistance, X the reactance and B network I and its variants
the susceptance. These parameters are from a cable made in Ger- Both the network topology and the parameters are considered
many with the type No. 20 kV NEKEBA 3  150). Its p equivalent to be relevant for the distribution of the CE of the network. In order

Fig. 2. Explanation for the selection of the characteristic eigenvalues.

Fig. 3. Topology of the SNI.

 B. Huang, E. Handschin / Electrical Power and Energy Systems 30 (2008) 547–552 549

to find the key factor, the topology as well as the parameters are the altered susceptance. For a 1 km length altered line, its CE is
altered hence some variants are developed from the SNI. For topol- 659.22 + j235,960. (3) P3, besides the alteration given in P2, lines
ogy, there are three options as: (1) T1, which refers to the original 830–854, 854–856 and 854–852 are changed as: Rn = R/2,
topology (Fig. 3); (2) T2, in which the line between node 824 and Xn = 2  X, Bn = B. For a 1 km long altered line, its CE is
828 is removed, and a new line between node 826 and 834 is 41.2 + j117,980.
added; (3) T3, in which the line between node 824 and 828 is re- The distribution of the CE of the SNI and its variants is plotted in
moved, and a new line between node 814 and 858 is added. For Fig. 5. From Fig. 5a–c, it can be seen that though the network topol-
line parameters, there are also three options as: (1) P1, which refers ogy changes, the real parts of the CE remain unchanged, which
the parameters given in Section 2.2. As shown in Fig. 2, the CE of match the one given in Fig. 2d. This indicates that, compared to
the line with 1 km length is 164.8 + j166,850. (2) P2, in which network topology, line parameters predominantly determine the
the parameters of the lines 808–812, 816–824 and 860–836 are al- real parts of the CE. This observation can be verified by investigat-
tered as: Rn = 2  R, Xn = X/2, Bn = B, where R, X, B are the same as in ing Fig. 5d–i. Taking Fig. 5d–f as an example, in which the distribu-
P1 and Rn is the altered resistance, Xn the altered reactance and Bn tion of the CE are similar; while the comparison of Fig. 5a–c to Fig.

Fig. 4. Topology of the SNII.

Fig. 5. Distributions of the characteristic eigenvalues of the SNI and Its variants.
550 B. Huang, E. Handschin / Electrical Power and Energy Systems 30 (2008) 547–552

Fig. 6. Distributions of the characteristic eigenvalues of the SNII.

5d–f reveals that the CE subjects to a sharp change. Hence, though

the topology has an influence on the CE, the parameters of the line
are of much greater influence.
A more instructive principle about the relation between the line
parameters and the CE of the network can be formed as: a network
is assumed to consist of M different types of lines, then the maxi-
Fig. 8. Examples for investigating the properties of the characteristic eigenvalues.
mum and minimum values of the real part of the lines’ CE span a
range, which covers that of the network CE. Take Fig. 5i as an
example, in which, the network has three different types of lines.
from 1 kV to 20 kV in Europe. Its databank contains the parameters
The real parts of their CE are 41.2, 164.8 and 659.22. They
of 361 types of lines/cables. These data are fully studied and the re-
span a range from 41.2 to 659.22, which totally covers the
sults reveal that, the maximum real part value of the CE of these
one spanned by that of the network CE (44.9 to 526.2).
lines/cables is 69.3, while the minimum one of that is 33,055.
Thus, it can be expected that, the eigenvalues of a real distribution
3.2.2. Distribution of the characteristic eigenvalues of the study
network in Europe will have real parts sited within the range from
network II
69.3 to 33,055. This means that the dynamic procedure of the
The CE of the SNII is plotted in Fig. 6. Meanwhile, the CE of the
network itself caused by a certain disturbance will die down to
cables in the SNII is given in Table 1. It is worth noting when cal-
5% of its initial value within 0.0001–0.0433 s.
culating the CE, all the cables are assumed with a 1 km length.
The real part of the lines’ CE spans a range from 160.01 to
3.4. Proper electrical network model after the integration of DG
673.63, which totally covers the one spanned by that of the net-
work CE (160.01 to 513.27).
Though there are no enough statistics as proofs, it is sensible to
judge that normally the time intervals between two disturbances
3.3. Characteristics of the distribution network dynamics in Europe
caused by DG will be much longer than 0.0433 s (here the altera-
tions of the DG status, for example, the switching on and off, are
Neplan [7] provides comprehensive information about the
also deemed as disturbances). This demonstrates that after the
lines/cables for the distribution systems with the voltage levels
integration of DG, the steady state model of the distribution net-
work is still acceptable.
Table 1
Characteristic eigenvalues of the cables in the SNII
4. Conclusions and comments
Type of cable CE Type of cable CE
NA2XS(F)2Y 3  1  185 160.01 + j138820 NKBA 3  70 409.31 + j191350 By studying the eigenvalues of the electrical network, the fol-
NAKBA 3  240 229.44 + j161700 NAKBA 3  95 507.37 + j184290
NAKBA 3  150 350.92 + j172360 NAKBA 3  70 673.63 + j191350
lowing conclusions are drawn:

Fig. 7. System under abc and dq0 frames.

 B. Huang, E. Handschin / Electrical Power and Energy Systems 30 (2008) 547–552 551

Table 2 too complex to achieve. Here the discussion is based more on a

Eigenvalues of the systems given in Fig. 8 physical meaning rather than on a mathematical reasoning.
Case of kabc2N kabc2N is assumed to have eigenvalues of r1 ± jx1. Firstly, the in-
Fig. 6a 80.331, 80.331, 49.716, 49.716, 25.133, 25.133 put signal Iabc in Fig. 7 is assumed to be positive sequential sinusoid
Fig. 6b 37.667 ± j8115.9, 37.667 ± j8115.9, 24.764 ± j6205, with a frequency of x1. It is well known the output signal Oabc can
24.764 ± j6205,15.159 ± j16598, 15.159 ± j16598, 0, 0 be considered as a combination of a transient signal as well as a
Fig. 6c 34.048 ± j7866.8, 34.048 ± j7866.8, steady signal. For simplifying the discussion, the steady output sig-
23.51 ± j3972.7,23.51 ± j3972.7,
20.032 ± j13372, 20.032 ± j13372
nal is assumed to have a prominently large amplitude, namely,
Fig. 6d 36.927 ± j7995.1, 36.927 ± j7995.1, 28.096 ± j4706.2, there exists a resonance in the response. After the dq0 transforma-
28.096 ± j4706.2, 12.566 ± j11107, 12.566 ± j11107 tion, Idq0 consists of only dq components with a frequency of
Fig. 6e 37.762 ± j7639.4, 37.762 ± j7639.4, 24.64 ± j3298.6, x1  x0, while the 0 component constantly equals to zero [8].
24.64 ± j3298.6, 15.188 ± j16584, 15.188 ± j16584
Applying the dq0 transformation to Oabc is equivalent to apply
the transformation separately to the transient signal as well as
the steady signal in Oabc. For the transient signal, the decaying ratio
 There is a close relation for the network eigenvalues under abc keeps unchanged, which means the real parts of the network
and dq0 reference frame. eigenvalues hold unchanged after the transformation. For the stea-
 The parameters of the lines/cables are the predominant factors dy signal, the frequency is shifted to x1  x0, which means under
to determine the network eigenvalues. the dq0 frame the resonance frequency is shifted to x1  x0. Hence
 The distribution of the network eigenvalues is bounded by the by applying the dq0 transformation the network has eigenvalues of
relevant values of the lines/cables. r1 ± j(x1  x0). It is worth noting the transformation also alters
 In a real distribution network in Europe, the eigenvalues will the frequency of the transient signal. In order to simplify the anal-
have the real parts within 69.3 to 33,055. ysis the above discussions focus only on the impact on the decay-
 It is acceptable to employ the steady state model to represent ing ratio.
the distribution network after the integration of DG. Then, Iabc is assumed to be negative sequential sinusoid with a
frequency of x1. Since the negative sequence signal has nearly
So far, some conclusions are achieved, though some limitations the same impact to the network as that of the positive sequence
still exist, such as: signal, a similar reasoning as above reveals the network under
the dq0 frame has eigenvalues of r1 ± j(x1 + x0).
 Though the eigenvalues of kabcN are not included in the above Finally, kabcN is assumed to have eigenvalues of r2 ± jx2 and Iabc
discussion, it is expected that the maximum and minimum val- is assumed to be three phase identical (zero sequence) with a fre-
ues of the real parts of the lines’ kabcN span a range, which covers quency of x2. A similar deduction as above reveals the eigenvalues
the interval of the network’s kabcN. hold unchanged under abc and dq0 frames (applying the dq0 trans-
 Frequency dependence of line parameters is not considered formation to the zero sequence signals leads to zero in dq -axis
when calculating the eigenvalues. A possibly improved approach while a sinusoidal signal the same as the input one in 0-axis).
is: (1) assuming the system is stepwise frequency independent;
(2) forming the state matrix within this frequency band; (3) cal- Appendix B
culating the eigenvalues, and those whose imaginary parts con-
form to the frequency band belong to the set of the eigenvalues Three kinds of node of the distribution system are important for
of the real system and (4) steps 1–3 can be executed iteratively the analysis as: (1) starting node (SN) – which is usually the com-
until all the eigenvalues have been found. During this procedure, mon coupling point of the transmission network and the distribu-
it is important to note that as the frequency keeps rising, the p tion network; (2) ending node (EN) – which is the node where only
equivalent model for lines/cables will not be proper even for the one transmission line is connected and (3) connecting node (CN) –
distribution networks. In this case distributed parameter models which connects at least two transmission lines as well as other en-
can be adopted. With the improved approach, though the net- ergy sources or sinks (such as DG, shunt capacitor and load). If a
work eigenvalues subject to a change, the achieved conclusions node connects more than one transmission lines but without any
are expected to be still applicable. energy sources or sinks, it is not a CN.
 The studied networks in this paper are based on radial topology. It is observed that the kabc2N of the network consists of two
It is open for further study if the results are also applicable to eigenvalues at the origin, if the SN, EN and CN are connected with
meshed networks. However, a positive answer can be reason- an equivalent shunt capacitor.
ably expected. Firstly, the following will demonstrate if the requirements are
met, the system has eigenvalues at the origin. The definition re-
veals that SN, EN and CN are the interface of the network to other
Acknowledgment system components. Assuming the state space model of the net-
work is x_ ¼ Ax þ Bu , then if each SN, EN and CN connects an
Bin Huang acknowledges the financial support from the Gradu- equivalent shunt capacitance, the input vector u only consists of
ate School of Production Engineering and Logistics, University of currents ðu ¼ ½ i1 ; i2 ; :::; in T Þ. Here all the currents are set to zero
Dortmund. except i1 ðu ¼ ½ i1 ; 0 ::: 0 T ), i1 is a constant value) to simplify the
discussion. Since all the nodes are either isolated from the ground
Appendix A or connected to the ground via a shunt capacitance, theoretically
the constant current source i1 will drive all the nodal voltages
Padiyar [5] describes the relations of the network eigenvalues increasingly to infinity. This matches the observation that the net-
under abc and dq0 frames. But the discussion is confined to the ba- work has eigenvalues at the origin. However, if some of the SN, EN
sic components such as a series impedance and a shunt capaci- and CN do not have an equivalent shunt capacitance, then u con-
tance. It is not known if the principles are applicable to a more sists of both currents and voltages (u ¼ ½ i1 ::: v1 ::: 0 T ). If u is still
general network. A mathematically strict proof for them may be specifically chosen as ½ i1 ; 0 ::: 0 T , this physically means node i is
552 B. Huang, E. Handschin / Electrical Power and Energy Systems 30 (2008) 547–552

Table 3
Line segment data of the SNII

Node Type of the cable Length (km) Node Type of the cable Length (km)
1–3 NAKBA 3  150 0.01 19–20 NA2XS(F)2Y 3  1  185 0.9086
3–4 NAKBA 3  150 0.2409 20 – 21 NAKBA 3  240 0.2885
4–5 NAKBA 3  150 0.226 21–22 NAKBA 3  240 0.8286
1–6 NAKBA 3  150 0.217 22–23 NA2XS(F)2Y 3  1  185 0.3374
6–7 NKBA 3  70 0.142 15–24 NA2XS(F)2Y 3  1  185 0.7855
7–8 NKBA 3  70 0.1334 24–25 NA2XS(F)2Y 3  1  185 0.9508
8–9 NA2XS(F)2Y 3  1  185 0.0416 25–26 NAKBA 3  150 0.4218
9–10 NKBA 3  70 0.2549 26–27 NAKBA 3  150 0.0982
10–11 NKBA 3  70 0.4249 2–28 NA2XS(F)2Y 3  1  185 0.4994
11–12 NAKBA 3  70 0.3664 28–19 NA2XS(F)2Y 3  1  185 0.9216
12–13 NAKBA 3  70 0.6168 19–29 NA2XS(F)2Y 3  1  185 0.4956
1–14 NA2XS(F)2Y 3  1  185 0.2619 29–30 NA2XS(F)2Y 3  1  185 0.2
14–15 NA2XS(F)2Y 3  1  185 0.4069 30–31 NA2XS(F)2Y 3  1  185 0.2
15–16 NAKBA 3  150 0.685 20–32 NAKBA 3  150 0.0661
16–17 NAKBA 3  150 1.6537 20–33 NAKBA 3  150 0.2606
17–18 NAKBA 3  150 0.71 33–34 NAKBA 3  150 0.2034
18–19 NAKBA 3  95 0.3076

Appendix C
Table 4
Parameters of the cables given in Table 3
See Tables 3 and 4.
Type of the cable R (X/km) X (X/km) B (lS/km)
NAKBA 3  70 0.446 0.104 103.673 References
NAKBA 3  95 0.323 0.1 116.239
NAKBA 3  150 0.21 0.094 141.372 [1] Papaefthymiou G, Schavemaker PH, et al. Integration of stochastic generation in
NAKBA 3  240 0.13 0.089 169.646 power systems. Electr Power Energy Syst 2006;28:655–67.
NA2XS(F)2Y 3  1  185 0.164 0.161 127.235 [2] Thiringer T. Frequency scanning for power system property determination-
NKBA 3  70 0.271 0.104 103.673 applied to a wind power grid. IEEE Trans Power Syst 2006;21(2):702–8.
[3] Cate EG, Hemmaplardh K, Manke JW, Gelopulos DP. Time frame notion and time
response of the models in transient, mid-term and long-term stability
programs. IEEE Trans Power Appl Syst 1984;103(1):143–51.
grounded. Thus i1 results in only a limited system response. In this [4] Sauer PW, LaGesse DJ, Ahmed-Zaid S, Pai MA. Reduced Order Modeling of
Interconnected Multimachine Power Systems Using Time-Scale Decomposition.
case there is no eigenvalues at the origin. IEEE Trans Power Syst 1987;PWRS-2(No.2):310–20.
Then, the following will verify the quantity of the zero eigen- [5] Padiyar KR. Analysis of subsynchronous resonance in power systems. Kluwer
values is two. It is difficult to directly prove the conclusion thus Academic Publishers; 1999.
[6] Kersting WH. Radial distribution test feeders. In: IEEE power engineering
some examples are studied and reported instead. The line param- society winter meeting; 2001.
eters in Fig. 8 the same as those of SNI. The symbols of the gener- [7] http://www.abb.com/global/abbzh/abbzh251.nsf!OpenDatabase&db=/global/
ators and loads in Fig. 8b–e are omitted to make the plot clear. In seitp/seitp408.nsf&v=9AAC910040&e=us&c=514877510D363EF7C1257018000
7CEE3, accessed October 2005.
Fig. 8a, all nodes do not have a shunt capacitance. On the contrary,
[8] Krause PC. Analysis of electric machinery. McGraw-Hill; 1987.
all the SN, EN and CN in Fig. 8b have a shunt capacitance. While in
Fig. 8c–e, one of the SN, EN and CN does not have a shunt capaci-
tance. The kabc2N of them are reported in Table 2. It is seen only in
the case of Fig. 8b, two eigenvalues appear at the origin.