Documente Academic
Documente Profesional
Documente Cultură
P
I. INTRODUCTION
formulation successfully ensures voltage security, to
OWER flow studies play vital roles in power system overcome a well-known weakness of the current DC-based
planning and operation for ensuring both operational market clearing algorithm in modeling voltage security [3].
security and economic efficiency. Power flow solution Furthermore, the GDC-based formulation generates reactive
algorithms can be placed under two categories: the more power prices that represent market incentive for voltage
accurate nonlinear AC power flow algorithms and the support, an important ancillary service that has been regulated
simplified linear DC power flow algorithms. Although AC outside the electricity market.
power flow algorithms can achieve desirable accurate results, Both the proposed GDC power flow model and new
DC power flow algorithms have gained wide industrial market clearing algorithms enhanced with voltage security
acceptance and play important roles in many important constraints represent significant contributions to power
applications, such as the optimal power flow (OPF) studies system modeling and economic analysis.
for wholesale electricity market clearing. In these applications,
achieving robust solutions within acceptable timeframes is II. T HE T ANGENT-LINE L INEAR APPROXIMATIONS OF
among the most important objectives. In OPF applications, P OWER F LOW F UNCTIONS
specifically, the linearity of the DC power flow model also
helps to overcome model non-convexity and ensures model A. Linear approximation of power flow functions
solution. On the other hand, there are two well-known For an N-bus power system network model, the active
limitations in the DC power flow model (especially among the and reactive power injections at a given bus i relate to the
“cold start” DC model variants that solve without knowledge network states, i.e., bus voltage angles and magnitudes,
of an operation base state): 1) the model inaccuracy under through:
N
heavy load conditions and networks with high resistance-to- Pi ( x) = ∑ViV j [Gij cos(δ i − δ j ) + Bij sin(δ i − δ j )] (1)
reactance (R/X) ratios, and 2) the inability to solve for bus j =1
voltage magnitudes. These limitations are due to the key N
heuristic assumptions made in the DC power flow model, i.e., Qi ( x) = ∑ViV j [Gij sin(δ i − δ j ) − Bij cos(δ i − δ j )] (2)
j =1
zero network resistance and 1 per unit (p.u.) magnitude for all
bus voltages [1]. for i = 1, …, N, where the state variables = =
This work makes critical enhancements to the DC power [ … … ] ∈Ω and Ω = (δmin, δmax)×(δmin,
flow model to overcome its main model limitations. The δmax)× …×(δmin, δmax)×(Vmin, Vmax)×(Vmin, Vmax) ⊆ R2N;
proposed approach is to remove all dependencies on heuristic parameters δmin, δmax,Vmin and Vmax represents the lower and
assumptions and achieve linearity by applying the calculus upper bounds of voltage angles and magnitudes in power
concept of “tangent line approximation of nonlinear system steady-state operation, e.g., δmin = −π/2, δmax = π/2,
Vmin = 0.9 and Vmax=1.1; and parameters Gij and Bij are the real
1 and imaginary parts of the i-jth element of the Y-bus matrix.
Mingguo Hong, Zihan Ning, and Reza Jamalzadeh are with the
Electrical Engineering and Computer Science department at the
The tangent line approximations are now considered for
Case Western Reserve University in Cleveland, OH. Emails: (1) and (2):
mxh543@case.edu, zxn23@case.edu, rxj171@case.edu.
⎡ N ⎛ N ⎞ ⎤
⎡ N ⎤ ⎢ ∑ B1 j − B12 ... − B1N ⎜ G11 + ∑ G1 j ⎟ G12 ... G1N ⎥
⎜ ⎟
⎢ − ∑ G1 j ⎥ ⎢ j =2 ⎝ j =1 ⎠ ⎥
⎢ j =1 ⎥ ⎢ ... J1 ... ... J2... ⎥ ⎡ δ1 ⎤
⎡ P1( x) ⎤ ⎢ N... ⎥ ⎢ ⎢δ ⎥
N −1
⎛ N ⎞ ⎥ 2
⎢ ... ⎥ ⎢− ∑ G ⎥ ⎢ − BN1 − BN 2 ... ∑ BNj GN1 GN 2 ... ⎜⎜ GNN + ∑ GNj ⎟⎟ ⎥ ⎢⎢ ... ⎥⎥
⎢ PN ( x) ⎥ ⎢ j =1 Nj ⎥ ⎢ j =1
⎝ j =1 ⎠ ⎥ ⎢δ N ⎥ + O 2 (Δδ , ΔV ) (4)
⎢ Q ( x) ⎥ = ⎢ N ⎥+⎢ ⎥ V
⎢ 1 ⎥ ⎢ ∑ B1 j ⎥ ⎢ N ⎛ N ⎞ ⎥⎢ 1 ⎥
⎜ − B11 − ∑ B1 j ⎟ − B12 ... −
⎥ ⎢ ∑ G1 j − G12 ... − G1N
B
⎢ ... ⎥ ⎢ j =1 ⎜
⎝ j =1
⎟
⎠
1N
⎥ ⎢V2 ⎥
⎢⎣QN ( x)⎥⎦ ⎢ ... ⎥ ⎢ j = 2 ⎥ ⎢ ... ⎥
⎢ N ⎥ ⎢ ... J3 ... ... J4 ... ⎥ ⎢⎣VN ⎥⎦
⎢ ∑ BNj ⎥ ⎢ − G N −1
− GN 2 ... ∑ GNj ⎛ N ⎞⎥
⎢⎣ j =1 ⎥⎦ ⎢ N1 − BN1 − BN 2 ... ⎜⎜ − BNN − ∑ BNj ⎟⎟⎥
j =1
⎣ ⎝ j =1 ⎠⎦
⎡∂P ∂P ⎤ associated with the P-Q buses. Vectors g , g , and of
⎡P(x)⎤ ⎡P(x)⎤ ⎢ ∂δ ∂ V ⎥ ⎡Δδ ⎤ 2 appropriate dimensions correspond to the first term on the
⎢Q(x)⎥ = ⎢Q(x)⎥ + ⎢∂Q ⋅ +O (Δδ , ΔV ) (3)
⎣ ⎦ ⎣ ⎦ x0 ⎢ ∂Q ⎥ ⎢⎣ΔV ⎥⎦ RHS of eq. (4).
⎣ ∂δ ⎥
∂V ⎦ x0 Following routine procedure [4], the bus voltage
magnitudes representing P-V bus voltage setpoints are
The choice for the point of tangency is arbitrary, but a
given; the slack bus voltage angle (first element of ) is set to
reasonable one is the “flat voltage profile”, i.e., δi = 0 and Vi =
zero; P-V and P-Q buses active power injections:
1 for i = 1, …, N. Let x0 = [0 … 0 1…1]T denote this choice of
(excluding the 1st element) and , are known, as well as
point of tangency. Also, substituting Δδi =δi − 0 and ΔVi =Vi −
1, one obtains eq. (4). reactive power injections at all P-Q buses . Eliminating
equations of the first row of vector ( ), all rows ( ), and
In eq. (4), the first term on the right-hand-side (RHS) is a the slack bus angle from , one has:
constant vector with elements − ∑ j =N1 Gij and ∑ j =N1 Bij . The ⎡ P ' ⎤ ⎡− g '⎤ ⎡ J1' J1'12 J 2'12 ⎤ ⎡δ '1 ⎤ ⎡ J 2'11 ⎤
1 1 11
Jacobian matrix is partitioned into four N×N sub-matrix ⎢ 2 ⎥ ⎢ 2
⎢ P ⎥ = ⎢− g ⎥ + ⎢ J1' J122 J 222 ⎥ ⋅ ⎢ δ 2 ⎥ + ⎢ J 221 ⎥ ⋅ V1
⎥ ⎢ 21 ⎥ ⎢ ⎥ ⎢ ⎥
blocks: ⎢⎣Q2 ⎥⎦ ⎣ b2 ⎦ ⎣ J 3'21 J 322 J 422 ⎦ ⎢⎣V2 ⎥⎦ ⎣ J 421 ⎦
N
a. In J1, the ith diagonal elements are ∑ j =1 Bij and i-jth off- + O 2 ( Δδ , ΔV ) (6)
diagonal elements − Bij . In fact, the well-known DC power flow formulation can
be derived from (4) by ignoring all network conductance
N
b. In J2, the ith diagonal elements are Gii + ∑ and i- j =1 Gij
(setting all Gij = 0). Eq. (6) is the GDC power flow
jth off-diagonal elements Gij . formulation. The approximated power flow solution can be
N directly solved from eq. (6):
c. In J3, the ith diagonal elements are ∑ j =1 Gij and i-jth off-
⎡δ '1⎤ ⎡ J1'11 J1'12 J2'12⎤−1 ⎧⎡P1'⎤ ⎡− g1'⎤ ⎡J2'11⎤ ⎫
diagonal elements −Gij . ⎢δ ⎥ ≈ ⎢J1' J1 J2 ⎥ ⋅ ⎪⎢P ⎥ − ⎢− g ⎥ − ⎢J2 ⎥ ⋅V ⎪ (7)
⎢ 2 ⎥ ⎢ 21 ⎨ 2 21 1⎬
⎥ ⎪⎢ ⎥ ⎢ ⎥ ⎢J4 ⎥ ⎪
22 22 2
d. N V
In J4, the ith diagonal elements are − Bii − ∑ j =1 Bij and i-jth ⎣⎢ 2 ⎦⎥ ⎣ J3'21 J322 J 422 Q b
⎦ ⎩⎣⎢ 2 ⎦⎥ ⎣ 2 ⎦ ⎣ 21⎦ ⎭
off-diagonal elements − Bij . With approximation error:
−1
When the network conductance is ignored, i.e., Gij = 0 for ⎡ J1'11 J1'12 J 2'12 ⎤
i, j = 1, …N, the first N equations P1(x) through PN(x) reduce e (Δδ , ΔV ) = ⎢ J1'21 J122 J 222 ⎥ ⋅ O 2 (Δδ , ΔV ) (8)
to the well-known DC power flow equations. For this reason, ⎢J 3' J 3 J 4 ⎥
⎣ 21 22 22 ⎦
eq. (4) is also called the Generalized DC or GDC Power Flow
functions in this study. C. Case Study Validation of GDC model
To validate the accuracy of the GDC power flow solution,
B. The GDC power flow formulation eq. (12) is solved on both the IEEE 14 and 118 Bus Test
Eq. (5) can then be rewritten in a concise format as: Systems. The GDC Power flow solutions are compared
⎡ P1 ( x) ⎤ ⎡ − g1 ⎤ ⎡ ⎡ J 111 J112 ⎤ ⎡ J 211 J 212 ⎤ ⎤ ⎡ δ1 ⎤ against 1) the full AC power flow solution solved with the
⎢ P ( x ) ⎥ ⎢− g 2 ⎥ ⎢ ⎢ J 1
⎣ ⎥ ⎢⎣ J 2 21 J 2 22 ⎥⎦ ⎥ ⎢δ ⎥
21 J 122 ⎦ Newton Raphson method, and 2) the traditional DC power
⎢ 2 ⎥=⎢ ⎥+⎢ ⎥⋅⎢ ⎥
2
⎢ Q1 ( x) ⎥ ⎢ b1 ⎥ ⎢ ⎡ J 311 J 312 ⎤ ⎡ J 411 J 412 ⎤ ⎥ ⎢V1 ⎥ (5) solution. The solution results are shown in Table 1 and Fig. 2.
⎢⎣Q2 ( x)⎥⎦ ⎢⎣ b2 ⎥⎦ ⎢⎣ ⎢⎣ J 321 J 322 ⎥⎦ ⎢⎣ J 4 21 J 4 22 ⎥⎦ ⎥⎦ ⎢⎣V2 ⎥⎦
Table 1: Comparison of solutions using the IEEE 14 bus model
+ O 2 ( Δδ , ΔV ) Bus GDC Full AC DC
where ( ), ( ), , are M×1 vectors of active power, No. V (pu) δ (°) V (pu) δ(°) V δ(°)
reactive power, bus voltage angles and voltage magnitudes 1 1.060 0 1.060 0 1 0
2 1.045 -5.23 1.045 -4.97 1 -5.01
associated with the slack bus (bus 1) and the P-V buses; 3 1.010 -13.23 1.010 -12.68 1 -12.96
( ), ( ), , are (N−M) ×1 vectors of similar quantities 4 1.038 -15.47 1.038 -14.62 1 -15.17
5 1.085 -14.49 1.085 -13.66 1 -14.07 approximation leads to significant benefit of the GDC method
6 1.029 -11.00 1.024 -10.38 1 -10.59 for 1) more accurate modeling of networks with high R/X
7 1.034 -9.42 1.028 -8.86 1 -9.09
8 1.050 -14.49 1.046 -13.66 1 -14.07
ratios, and 2) the accurate estimates of bus voltage
9 1.039 -16.32 1.035 -15.36 1 -15.89 magnitudes. Our case studies also confirm that the GDC
10 1.031 -16.50 1.028 -15.53 1 -16.19 solutions significantly improve over the first iteration of NR
11 1.031 -16.14 1.030 -15.22 1 -15.88 solutions. The simplicity (no heuristic assumptions) and
12 1.024 -16.42 1.024 -15.51 1 -16.27 solution features of the GDC model makes it an ideal choice
13 1.020 -16.53 1.020 -15.61 1 -16.44
for application in electricity market clearing algorithms,
14 1.012 -17.51 1.010 -16.52 1 -17.43
which for long have been searching for an accurate and
computationally tractable approach to model voltage security
constraints [3, 5].
(a) Bus voltage angles