WK,(((3(6$VLD3DFLILF3RZHUDQG(QHUJ\(QJLQHHULQJ&RQIHUHQFH$33((&

An Interval Power Flow Method based on

Linearized DistFlow Equations for Radial
Distribution Systems
Jinqiao Du Jie Tian Zhi Wu
Shenzhen Power Supply Bureau Co., Shenzhen Power Supply Bureau Co., School of Electrical Engineering
Ltd. Ltd. Southeast University
Shenzhen, China Shenzhen, China Nanjing, China
[email protected] [email protected] [email protected]

Ao Li Ghulam Abbas Qirun Sun

School of Electrical Engineering School of Electrical Engineering School of Electrical Engineering
Southeast University Southeast University Southeast University
Nanjing, China Nanjing, China Nanjing, China
[email protected] [email protected] [email protected]

Abstract—With consideration of the radial structure of most interval arithmetic method is proposed for power flow
distribution systems, this paper proposes an interval power flow calculation of a balanced radial distribution system. With
method based on linearized Distflow equations. In the proposed consideration of the radial structure and wide ranging
method, the calculation of interval power flow solution has been
modeled as an optimization problem, and the uncertainties resistance and reactance values of distribution systems, this
derived from the randomness of distributed renewable method is based on a backward sweep/forward sweep
resources as well as electric vehicle loads are properly involved. procedure. Furthermore, on the basis of the method proposed
The proposed method is more high-efficiency than the existing in [6], a modified method is proposed in [7] to accommodate
methods since its linear property, and the 33-bus distribution three-phase unbalanced distribution systems, and the
system as well as a practical urban distribution system is uncertain sources include both load demands and the feeder
employed to verify its effectiveness.
parameters. To reduce the overestimation of bounds yielded
Keywords—distribution system, distributed renewable by correlation of variables in interval arithmetic, an affine
resources, electric vehicle, interval power flow, linearized arithmetic-based algorithm is proposed in [8]. Compared with
Distflow equations, optimization model the method proposed in [6], the affine arithmetic-based
algorithm can obtain narrower power flow solution ranges. In
I. INTRODUCTION
[9], the affine arithmetic-based algorithm has been extended
With the increasing scale of distributed generation and to three-phase unbalanced distribution systems. And
electric vehicles integrated into distribution systems, the meanwhile an index of relative influence of uncertain
problems of uncertainties in power systems have become variables on outcomes is proposed for quantifying the impacts
significant, due to the randomness and intermittency of of individual uncertain factors on power flow distributions. In
renewable energy (e.g. wind power and solar power) as well [10], regional control strategies are considered to smooth the
as electric vehicle loads [1,2]. Power injections at some buses unscheduled power fluctuation in distribution systems, and
have changed from deterministic values to fluctuating random then an affine arithmetic-based modeling method is proposed
values, which make it difficult for traditional deterministic to describe the unscheduled power tracking characteristic of
methods to represent the operating state of power systems and dispatchable generation resources, which results in more
calculate the power flow. Based on this background, how to accurate power flow solutions. In [11], the interval power flow
calculate uncertain power flow solutions becomes a critical problems of distribution systems are creatively modeled as
problem. optimization models, and the second-order cone programming
To describe the influence of uncertainties on power flow method is used to manage the non-convex characteristic.
distribution, many researchers have carried out substantial Based on this method, distribution ranges of power flow can
studies from diverse perspectives. These studies can be be accurately achieved.
broadly divided into three categories according to the To the best of the authors’ knowledge, none of the existing
representation methods of uncertainties. To be detailed, researches in interval power flow fields have taken into
stochastic power flow, fuzzy power flow and interval power consideration the DistFlow equations [12], which is
flow respectively correspond to probability theory, fuzzy customized for radial distribution systems and significantly
theory and interval theory [3-5]. Among these researches, reduces the complexity of optimization problems. Based on
interval power flow uses a restricted range representing each this background, this paper constructs an optimization model
uncertain parameter or variable, and thereby achieves the to solve interval power flow problems. With the linearized
upper bounds and lower bounds of the operation state of procedure on DistFlow equations, the optimization model is
power systems, which is simple and intuitive. In [6], an represented as an linear programing (LP) problem, and can be
easily solved through existing optimization algorithms.
This work was supported by the Science and Technology Project of
China Southern Power Grid Co., Ltd. (090000KK52170127).

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

k,(((
 WK,(((3(6$VLD3DFLILF3RZHUDQG(QHUJ\(QJLQHHULQJ&RQIHUHQFH$33((&

The remaining parts of this paper are organized as follows. where Usub corresponds to the voltage magnitude of substation
Section 2 describes the interval models of renewable energy bus and is always valued 1 p.u..
resources and electric vehicle loads. Section 3 minutely
represents the constructed optimization model for interval In the linearization process, some tolerable assumption
power flow calculation. Section 4 explains the concrete and simplification have been adopted, as follows:
procedures of the proposed method. Section 5 illustrates and 1) The nonlinear items represent the losses in (1) and (2)
analyzes the numerical results on test systems. And finally, are ignored. As verified in [12], these nonlinear items are
conclusion is summarized in Section 6. much smaller than the branch power items Pj and Qj.
II. INTERVAL MODELS OF RENEWABLE ENERGY RESOURCES 2) It is assumed that (Ui-Usub)Ĭ0 and thereby U2i ĬU2sub
AND ELECTRIC VEHICLE LOADS
+2Usub(Ui-Usub).
In the operation of distribution systems, the outputs of
renewable energy (e.g. wind power and solar power), as well Based on the presented linearized Distflow equations, the
as the electric vehicle loads are hard to be accurately predicted. optimization model for interval power flow calculation can be
They are always fluctuated in certain ranges, and thereby can established, as shown in (7). For a certain bus i or branch ij,
be modeled as intervals with lower and upper bounds. The through adjusting the objective function in (7) and solving this
interval models of renewable energy resources and electric corresponding optimization model can obtain the required
vehicles have been deeply studied in [11] and [13], readers boundaries of power flow variables.
may refer to these literatures for more detailed information.
­U i
III. OPTIMIZAITON MODEL °
Min or Max ® Pij
Taking the radial structure into consideration, Distflow °
¯Qij
equations derived from [12] are shown in (1)-(3).
­ ¦ Pij ¦ Pjl  Pj j  : N
( Pij ) 2  (Qij ) 2 °iv ( j ) lu ( j )
¦ [P
iv ( j )
ij 
(U i ) 2
˜ Rij ] ¦P
lu ( j )
jl  Pj j  : N (1) °
° ¦ ij ¦ Q jl  Q j j  : N
Q
(7)
°iv ( j ) lu ( j )
( Pij ) 2  (Qij ) 2 °° Pij ˜ Rij  Qij ˜ X ij
¦ [Q
iv ( j )
ij 
(U i ) 2
˜ X ij ] ¦Q
lu ( j )
jl  Qj j  : N (2) s.t. ®U j U i  ij  : L
° U sub
(U j ) 2 (U i ) 2  2( Rij Pij  X ij Qij )  ° P - d P d P + j  :
° j j j N
( Pij ) 2  (Qij ) 2 (3) °Q j d Q j d Q j j  : N
- +
[( Rij ) 2  ( X ij ) 2 ] ij  : L
(U i ) 2 °
°̄U sub =1
where
where
u(j) and v(j) respectively represent the set of downstream and
P-j and P+j are respectively the lower bound and upper bound
upstream buses connected with bus j;
of the equivalent active power demand at bus j;
Pij and Qij are the active and reactive power in branch from i
Q-j and Q+j are respectively the lower bound and upper bound
to j;
of the equivalent reactive power demand at bus j.
Ui stands for the voltage magnitude at bus i;
The established optimization model belongs to linear
Rij and Xij indicate the resistance and reactance of branch from programing (LP), and can be easily solved through existing
i to j; optimization algorithms.
Pj and Qj are the equivalent active and reactive power IV. CONCRETE PROCEDURES
demands at bus j;
The concrete steps of the proposed method are shown as
ΩN and ΩL are, respectively, the sets of buses and branches in fol1ows.
the distribution system.
1) Go through the distribution system and number all the
According to [14], when voltage at each bus not deviate buses.
per unit value too much, Distflow equations can be linearized
2) Taking into consideration of the cumulative effect of
with acceptable errors. The linearized Distflow equations are
distributed generation and original load, calculate the
shown in (4)-(6).
equivalent power demands and their tolerances at each bus in
¦P ¦P
iv ( j )
ij
lu ( j )
jl  Pj j : N (4)
the distribution system.
3) Regarding the target power flow variable (e.g. the lower
¦Q ¦Q
iv ( j )
ij
lu ( j )
jl  Q j j : N (5) or upper bound of voltage magnitude, branch active power and
Pij ˜ Rij  Qij ˜ X ij so on) to be calculated, construct the optimization model
U j Ui  ij : L (6) according to (7).
U sub
 WK,(((3(6$VLD3DFLILF3RZHUDQG(QHUJ\(QJLQHHULQJ&RQIHUHQFH$33((&

4) Solve the constructed optimization model through 19 20 21 22

existing optimization algorithms, and obtain the target

boundary. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

5) Back to step 3) and calculate power flow variables at

other buses, until all the required boundaries are obtained.
26 27 28 29 30 31 32 33

6) Conclude all the boundaries and exhibit the interval

power flow distribution.
23 24 25
To describe the concrete steps more clearly, a flowchart is
Fig. 2 the 33-bus distribution system
shown in Fig. 1.

Start
11 12
15 13
Go through the distribution system and number
all the buses.
1 2 3 4 5 6 7 8 9
Calculate the equivalent power demands and their
tolerances at each bus in the distribution system.
10
16 17 14
Select a target power flow variable.
Fig. 3 A practical urban distribution system in Shenzhen, China

Construct the optimization models corresponding to the

upper and lower bounds of the target power flow variable. TABLE I. Parameters of the employed distribution system in Shenzhen

Branch Start End Pj Qj

Rij (Ω) Xij (Ω)
Solve the constructed optimization models No. Bus i Bus j (kW) (kVar)
through existing optimization algorithms. 1 1 2 0.0052 0.0069 0 0
2 2 3 0.0385 0.0514 0 0
3 3 4 0.0881 0.1174 0 0
4 4 5 0.1080 0.1440 0 0
Obtain the target boundaries. 5 5 6 0.0223 0.0298 0 0
6 6 7 0.0003 0.0004 0 0
7 7 8 0.0065 0.0087 217 5
Conclude all the boundaries and exhibit the interval 8 8 9 0.0041 0.0054 73 11
power flow distribution. 9 8 10 0.0128 0.0170 1680 511
10 8 11 0.0101 0.0135 98 12
11 8 12 0.0212 0.0282 0 0
End 12 7 13 0.0102 0.0136 665 90
13 7 14 0.0066 0.0088 40 5
Fig. 1 The flowchart of the proposed method 14 5 15 0.0022 0.0029 1680 511
15 4 16 0.0814 0.1086 0 0
16 16 17 0.0067 0.0089 0 0
V. CASE STUDIES
To demonstrate the feasibility and effectiveness of the
proposed method, the 33-bus distribution system [12] as well B. Power Flow Solutions and Comparison
as a practical urban distribution system in Shenzhen, China, is To make the comparison more evident, active and reactive
employed in this paper. The presented optimization models loads at all the buses in Case 1 and Case 2 are assumed to be
corresponding to various power flow variables are established uncertain sources. And the fluctuation range is set to be ±20%.
through YALMIP toolbox in MATLAB environment, and are The power flow solutions as well as the comparison are shown
solved by commercial solver GUROBI. as follows.
A. Case Overview and parameters Case 1: the power flow solutions of Case 1 are shown in
Case 1: the 33-bus distribution system is minutely Table II and III, as well as Fig. 4-6. To make the verification
described in [12], and its topology is shown in Fig. 2. The 33- more convincing, comparison between the proposed method
bus distribution system consists of 3.715 MW active loads and and Monte Carlo method are illustrated in Fig. 4, where
2.3 MVar reactive loads, and readers may refer to [12] for method A represents the proposed method and method B
more detailed information. represents Monte Carlo method.
Case 2: this test system is collected from Shenzhen, China, TABLE II. Voltage magnitudes in Case 1
with 4.45 MW active loads and 1.14 MVar reactive loads, and Lower Upper Lower Upper
the voltage level is 10 kV. The topology and parameters of Bus No. bound bound Bus No. bound bound
Case 2 are respectively shown in Fig. 3 and Table I. (p.u.) (p.u.) (p.u.) (p.u.)
1 1.0000 1.0000 18 0.9034 0.9356
2 0.9966 0.9978 19 0.9960 0.9973
3 0.9807 0.9871 20 0.9917 0.9945
 WK,(((3(6$VLD3DFLILF3RZHUDQG(QHUJ\(QJLQHHULQJ&RQIHUHQFH$33((&

4 0.9724 0.9816 21 0.9909 0.9939

5 0.9641 0.9761 22 0.9901 0.9934
6 0.9437 0.9625 23 0.9765 0.9843
7 0.9398 0.9599 24 0.9687 0.9792
8 0.9344 0.9563 25 0.9649 0.9766
9 0.9275 0.9517 26 0.9416 0.9610
10 0.9211 0.9474 27 0.9387 0.9592
11 0.9202 0.9468 28 0.9261 0.9507
12 0.9185 0.9457 29 0.9170 0.9447
13 0.9118 0.9412 30 0.9131 0.9421
14 0.9093 0.9395 31 0.9085 0.9390
15 0.9077 0.9385 32 0.9075 0.9383
16 0.9062 0.9375 33 0.9072 0.9381
17 0.9040 0.9360

TABLE III. Active and reactive power in branches in Case 1 Fig. 5 Active power in branches in Case 1
Lower Upper Lower Upper
Branch Start End bound bound bound bound
No. Bus i Bus j of Pij of Pij of Qij of Qij
(kW) (kW) (kVar) (kVar)
1 1 2 2972 4458 1840 2760
2 2 3 2604 3906 1664 2496
3 3 4 1788 2682 1272 1908
4 4 5 1692 2538 1208 1812
5 5 6 1644 2466 1184 1776
6 6 7 860 1290 408 612
7 7 8 700 1050 328 492
8 8 9 540 810 248 372
9 9 10 492 738 232 348
10 10 11 444 666 216 324
11 11 12 408 612 192 288
12 12 13 360 540 164 246
13 13 14 312 468 136 204 Fig. 6 Reactive power in branches in Case 1
14 14 15 216 324 72 108
15 15 16 168 252 64 96 Case 2: the power flow solutions of Case 2 are shown in
16 16 17 120 180 48 72 Table IV and V, as well as Fig. 7-9.
17 17 18 72 108 32 48
18 2 19 288 432 128 192 TABLE IV. Voltage magnitudes in Case 2
19 19 20 216 324 96 144
Lower Upper Lower Upper
20 20 21 144 216 64 96
Bus No. bound bound Bus No. bound bound
21 21 22 72 108 32 48
(p.u.) (p.u.) (p.u.) (p.u.)
22 3 23 744 1116 360 540
23 23 24 672 1008 320 480 1 1.0000 1.0000 10 0.9812 0.9875
24 24 25 336 504 160 240 2 0.9996 0.9998 11 0.9816 0.9877
25 6 26 736 1104 760 1140 3 0.9969 0.9979 12 0.9816 0.9877
26 26 27 688 1032 740 1110 4 0.9905 0.9937 13 0.9817 0.9878
27 27 28 640 960 720 1080 5 0.9828 0.9885 14 0.9818 0.9879
6 0.9818 0.9879 15 0.9827 0.9885
28 28 29 592 888 704 1056
29 29 30 496 744 648 972 7 0.9818 0.9879 16 0.9905 0.9937
30 30 31 336 504 168 252 8 0.9816 0.9877 17 0.9905 0.9937
9 0.9816 0.9877
31 31 32 216 324 112 168
32 32 33 48 72 32 48
TABLE V. Active and reactive power in branches in Case 2

Lower Upper Lower Upper

Branch Start End bound bound bound bound
No. Bus i Bus j of Pij of Pij of Qij of Qij
(kW) (kW) (kVar) (kVar)
1 1 2 3562 5344 916 1374
2 2 3 3562 5344 916 1374
3 3 4 3562 5344 916 1374
4 4 5 3562 5344 916 1374
5 5 6 2218 3328 507 761
6 6 7 2218 3328 507 761
7 7 8 1654 2482 431 647
8 8 9 58 88 9 13
9 8 10 1344 2016 409 613
10 8 11 78 118 10 14
11 8 12 0 0 0 0
12 7 13 532 798 72 108
Fig. 4 Comparison between the proposed method and Monte Carlo method 13 7 14 32 48 4 6
in Case 1 14 5 15 1344 2016 409 613
15 4 16 0 0 0 0
16 16 17 0 0 0 0
 WK,(((3(6$VLD3DFLILF3RZHUDQG(QHUJ\(QJLQHHULQJ&RQIHUHQFH$33((&

solutions on different distribution systems verify the

generality of the proposed method.
VI. CONCLUSION
With consideration of the radial structure of distribution
systems, this paper proposes an interval power flow method
based on linearized Distflow equations. Due to the linear
characteristic of the constructed optimization model, the
proposed method is more high-efficiency than the existing
methods and can effectively solve the interval power flow
problems caused by the randomness and intermittency of
renewable energy resources as well as electric vehicle loads.

Fig. 7 Voltage magnitudes in Case 2 REFERENCES

[1] Liu B, Huang Q, Zhao J, et al. A Computational Attractive Interval
Power Flow Approach with Correlated Uncertain Power Injections.
IEEE Transactions on Power Systems, 2020; 35(1): 825-828.
[2] Mazza A, Chicco G, Bakirtzis E, et al. Power flow calculations for
small distribution networks under time-dependent and uncertain input
data. In: IEEE PES T&D Conference and Exposition; 2014.
[3] Liao H. Fast deterministic sampling for mean and covariance
estimation in stochastic load flow. In: IEEE PES General Meeting;
2007.
[4] Sun H, Yu D, Xie Y. Application of fuzzy set theory to power flow
analysis with uncertain power injection. In: IEEE PES Winter Meeting;
2000.
[5] Liao X, Liu K, Zhang Y, et al. Interval method for uncertain power
flow analysis based on Taylor inclusion function. IET Generation,
Transmission & Distribution, 2017; 11(5): 1270-1278.
[6] Das B. Radial distribution system power flow using interval arithmetic.
International Journal of Electrical Power & Energy Systems, 2002;
Fig. 8 Active power in branches in Case 2 24(10): 827-836.
[7] Das B. Consideration of input parameter uncertainties in load flow
solution of three-phase unbalanced radial distribution system. IEEE
Transactions on Power Systems, 2006; 21(3): 1088-1095.
[8] Gu W, Luo L, Ding T, et al. An affine arithmetic-based algorithm for
radial distribution system power flow with uncertainties. International
Journal of Electrical Power & Energy Systems, 2014; 58: 242-245.
[9] Wang S, Han L, Wu L. Uncertainty tracing of distributed generations
via complex affine arithmetic based unbalanced three-phase power
flow. IEEE Transactions on Power Systems, 2014; 30(6): 3053-3062.
[10] Luo L, Gu W, Wang Y, et al. An affine arithmetic-based power flow
algorithm considering the regional control of unscheduled power
fluctuation. Energies, 2017; 10(11): 1794.
[11] Guo X, Gong R, Bao H, et al. The second-order cone programming
method for interval power flow of active distribution network with
distributed generation. IEEJ Transactions on Electrical and Electronic
Engineering, 2018; 13(5): 696-703.
[12] Baran M E, Wu F F. Network reconfiguration in distribution systems
for loss reduction and load balancing. IEEE Transactions on Power
Fig. 9 Reactive power in branches in Case 2 Delivery, 1989; 4(2): 1401-1407.
[13] Luo L, Zhu J, Yang S, et al. Interval arithmetic based influence analysis
Through comparison in Fig. 4, it is obvious that the ranges on power flow caused by integration of wind power and electric
of power flow solutions derived from the proposed method vehicles. In: 19th IFAC World Congress; 2014.
completely contain those of Monte Carlo method. This [14] Yeh H G, Gayme D F, Low S H. Adaptive VAR control for distribution
phenomenon effectively demonstrates the feasibility and circuits with photovoltaic generators. IEEE Transactions on Power
Systems, 2012; 27(3): 1656-1663.
effectiveness of the proposed method. And moreover, the test