Adjusted load flow solution through a Mixed

Complementarity Problem framework

Lakshmi S P S Nagendra Rao
Dept. Electrical Engineering Dept. Electrical Engineering
Indian Institute of Science Indian Institute of Science
Bangalore -560012, India Bangalore -560012, India
[email protected] [email protected]

Abstract—Incorporating generator reactive power limits into the generator as a continuous function, it is possible to incorporate
load flow solution is one of the most important adjustments to be the adjustments automatically in the load flow problem
made in a practical load flow algorithm. Traditionally this formulation itself .The present work is also motivated by a
adjustment is carried out by bus type switching. There is similar thinking. However a new approach is proposed. It is
considerable divergence in incorporating this scheme across shown here that the Q limit constraints at a PV bus can be
implementation of the load flow algorithm. In this paper we effectively captured into a single equation if these constraints
propose a modified formulation for the load flow problem where are properly represented in a Mixed Complementarity Problem
in, this important adjustment is automatically included. It is (MCP) framework. Incorporating such an equation for each of
shown here that such a formulation for load flow problem is
the PV buses the load flow problem can be reformulated so as
possible if the PV bus equations are included in a Mixed
to get the adjusted solution directly.
Complementarity Problem framework. Effectiveness of the
scheme has been demonstrated by considering a number of The rest of the paper is organized as follows. First the MCP
standard test systems. problem is briefly introduced in section II. In section III the
load flow problem is formulated in the MCP framework by
Keywords-load flow soltions ,Newton Raphson method,Adjusted automatically incorporating the Q limit constraints, at PV
solutions,Generator Q limits,Mixed Complementarity Problem buses. In section IV simulation results is presented considering
I. INTRODUCTION several test systems. The effectiveness of the proposed method
is shown and is also compared with some of the existing
THE importance of the load flow problem in power system
analysis and optimization can hardly be over emphasized
methods. The paper is concluded in Section V.
.In view of its importance there has been a long history of
development .Load flow methods that are widely used in
industry are Newton-Raphson method [1],FLDF methods [2] II. MIXED COMPLEMENTARITY PROBLEM FRAMEWORK
etc. They have reached a very high level of maturity [3]. Even Two vectors x, y ϵRn are said to be complementary if each
though the basic unadjusted load flow algorithms are by and of their components is non-negative and the inner product is
large identical in various implementations, the same thing zero. i. e. Given x ≥ 0, y ≥ 0 and
cannot be said when it comes to adjusted solutions. Obtaining
adjusted load flow solutions is still more of an art than a xT y = 0 (1)
science and each implementation is to some extent ad hoc. (This implies for each i at least one of either xi , yi is zero)
Of the set of adjustments that are normally required in a Let F(x) → Rn be a continuous differentiable function.
practical load flow algorithm, handling generator Q limits is Mixed complementarity problem (MCP) consists of finding
the most important one. The suggested method for handling x ∗ ϵ [L, U] such that exactly one of the following conditions is
this adjustment either is bus type switching or sensitivity based true.
approach [1,2].In the former approach a generator bus is
considered as either PV or PQ in each iteration depending on xi∗ = Li and Fi xi∗ > 0 (2a)
whether its Q limits are satisfied or not .This results in the xi∗ = Ui and Fi xi∗ < 0 (2b)
system of equations considered for solution to change across xi∗ ϵ Li , Ui and Fi xi∗ = 0 (2c)
iterations .Also ,there is considerable divergence regarding at
what point in the iteration process the bus type switching
should be initiated and how to handle situations where buses
L and U are the upper and lower bound vectors such that
continuously alternate between PQ and PV types in
Li <Ui .In order to find the solution to this problem a nonlinear
consecutive iterations. The sensitivity method though appears
function called a merit function ρ x is defined .The MCP
attractive for constant matrix load flow methods, it is not in use
framework essentially involves defining this ρ(x) suitably
widely and there is not much recorded evidence of its
effectiveness. A novel method of handling generator Q limits such that the solution of ρ x = 0 turns out to be the solution
has been proposed in [4] where it is shown that by intelligently of the original MCP defined above in (2).
approximating the discontinuous QV characteristic of a

1
 The merit functions that solve MCP have to be smooth and Considering the modified load flow equations with these two
continuous, as most of the algorithms seeking to solve sets augmented, the linear set of equations to be solved in each
ρ x = 0 use the gradient of the function. One such widely iteration in the proposed method is given by (9).In (9) for the
used merit function is given by: sake of simplicity of representation, the additional equations
ρ x = φ xi − Li , φ Ui − xi , −Fi (xi ) (3) corresponding to the reactive power at PV buses (8) are
augmented after the PQ buses and are denoted by ∆QPV .The
Where, if a and b are two functions,φ a, b is of the form last set of equations correspond to (7).
φ a, b = a2 + b 2 − a − b (4)
∆P H N N′ 0 ∆δ
This function is known as Fischer–Burmeister function. ∆QPQ J L L′′ 0 ∆V PQ /Vi
This function has been analyzed for the MCP application in PV = J′ L′ (9)
∆Q L′′′ K ∆V PV /Vi
[5].It is easy to see that if (3) is satisfied then the conditions ∆𝛒 0 0 M S ∆Q
given in (2) are satisfied.
III. LOAD FLOW PROBLEM AS MCP The H, N, J, L are the same as in (5) of the conventional
Newton’s method. N', J' are similar to N and J respectively
In conventional Newton-Raphson load flow, the active and and L', L'', L''' have the same form as L. It is easy to see that
reactive power balance are considered at the PQ buses and the leading 3 by 3 matrix in (9) corresponds to the
only active power balance at the PV buses. Hence, the linear conventional jacobian, if all the buses are considered as PQ
system of equations to be solved in each of the iteration is of buses. The new terms introduced in (9) are K, M, and S and
the form [2]: they correspond to:
∆P H N ∆δ
= (5)
∆QPQ J L ∆V PQ /V
∂ρi
Mii = V PV i (10a)
∂ V PV i
In the method proposed, it is intended to include the Q
equations at the PV buses into the load flow model. However,
at the PV buses there are constraints to be considered. It is
necessary that at these buses the specified voltage magnitude ∂ρi
Mij = V PV j = 0 ,∀i ≠ j (10b)
∂ V PV
𝑉 sp must be maintained as long as the reactive power j

generation required at the PV bus is within the specified limits

i.e. Q min < Q < Q max . ∂ρi
Sii = ∂Qi (10c)
∂ρi
Sij = ∂Qj = 0 ∀i ≠ j (10d)
For the reactive power balance at PV buses, we can restate the
above condition as follows:
K ii = −1, ∀i ≠ j K ij = 0 (10e)
sp
if , Q i= Qmin
i then (Vi − Vi ) > 0 (6a)
sp
if , Q i= Qmax
i then (Vi − Vi ) < 0 (6b) The detailed expressions for the elements are given in the
min max sp appendix A.
if , Q i < Q i < Q i then (Vi − Vi ) = 0 (6c)
IV. SIMULATION RESULTS
We can clearly see the equivalence between (6) and (2). The
Qmin and Qmax correspond to the lower bound (L) and upper The effectiveness of the algorithm is demonstrated by
bound (U) vectors respectively. In view of the similarity, the considering a number of test systems. The data for these
constraints at PV buses represented by (6) can be captured by systems is obtained from [8]. The solutions obtained by the
proposed scheme are verified by comparing them with MAT
defining a merit function 𝝆𝒊 of the type (3). For the ith PV bus
POWER [7] solutions. In all the cases the solutions obtained by
𝝆𝑖 is given by (7) the two approaches are seen to be identical. The number of
iterations required to obtain the solution for both the methods
𝝆i = φ Q i − Qmin
i , φ Qmax
i − Q i , −(Vi − ViSP ) (7) considering a convergence tolerance of 10-6 p.u. is given in
Table I.
Augmenting this set of equations to the original set of the load
flow equations (P at all buses and QPQ at all PQ buses), we see From Table I it is seen that for these test cases, the performance
that we need another set of equations (one for each of the PV of the new algorithm and MATPOWER are quite comparable.
bus) since (6) introduces two new variables for each PV bus, The two numbers given in Table I for some cases (for the
normally not considered in the Load Flow Formulation. This MATPOWER based solution) represent the number of
additional set of equations is obtained by including (8) for the iterations required at each stage of the solution since this
ith PV bus. approach makes use of the basic N-R algorithm in stages. At
the end of each stage Q limit violations are checked. If
Q i − Qcal
i =0 (8)
violations are seen, bus type switching is adopted and the next

2
stage is initiated. From the Table I it can also be seen that there APPENDIX A
are no limit violations in the 57 and 14 bus examples (only one Here we obtain the expressions to the new elements
stage of the MATPOWER solutions).In the other systems introduced in the jacobian in (9).We consider the two equations
violations are noticed. The number of violating buses are 1(30 (7) and (8) rewrite them as (A.1) and (A.2) respectively.
bus), 6(118 bus) and 10 (300 bus).
Q i − Qcal
i = 0 (A.1)
In order to assess the robustness of the proposed algorithm,
number of test cases were generated by increasing the loads sp 2
ρ= Q i − Qmin
i
2 + ∅ Qmax
i – Q i , −(Vi − Vi ) − Q i − Qmin
i
and generation at all buses. In addition, the Q capabilities at
the generator buses were reduced to various degrees. It is sp
− ∅ Qmax
i − Q i , −(Vi − Vi ) (A.2)
observed that the performance of the proposed algorithm is
where,
uniformly good for all such test cases. As a representative
sp
example a summary of this investigation considering the ∅ Qmax
i – Q i , −(Vi − Vi ) =
IEEE-118 bus test system is given in Table II. sp sp
From the results in Table II it is seen that the number of = (Qmax
i – Q i )2 + (−(Vi − Vi ))2 − (Qmax
i − Q i ) − (−(Vi − Vi )) (A.3)
violating buses increase with load. However the performance
of the proposed approach is seen to remain more or less ∂∅ −(Q max −Q i )
∂Q I = ∅q = +1
i
(A.4)
unchanged. It is interesting to note that the performance of (Q max
i −Q i )2 +(−(V i −V i ))2
sp

MAT POWER algorithm seems to deteriorate as the number sp

∂∅ (V i −V i )
of violations increases. ∂Vi = ∅v = (Q max −Q i )2 +(−(V
sp 2
+1 (A.5)
i i −V i ))

TABLE I. NUMBER OF ITERATIONS FOR CONVERGENCE ∂ρi −(Q min

i –Q i )+∅∗∅q
∂Q i = 2 sp 2
− 1 − ∅q (A.6)
SYSTEM 118 57 30 14 Q i −Q min
i +∅ Q max
i –Q i ,−(V i −V i )
SIZE 300 BUS BUS BUS BUS BUS
∂ρi
MAT Sii = ∂Q i (A.7)
POWER 4,4 3,2 3 3,2 2
NEW ∂ρi
It can easily be seen that, Sij = ∂Q j = 0 ∀i ≠ j
METHOD 7 7 4 4 4
∂ρ ∅∗∅𝐯
TABLE II. PERFORMANCE OF THE ALGORITHM FOR THE STRESSED ∂Vi = (Q i −Q min )2 +∅ Q max −Q i ,−(V i −V i )
sp 2
− ∅𝐯 (A.8)
i i
SYSTEMS
∂ρi
Load increase factor for Mii = V PV (A.9)
∂ V PV i
i
reactive power 1 1.1 1.3 1.5 1.9
Load increase factor for ∂ρi
It can easily be seen that, Mij = ∂Vj = 0, ∀i ≠ j
real power 1 1.1 1.3 1.6 1.5
Q limit reduction factor 1 1.2 1.6 2 2 Differentiating (A1) with respect to Qi we get,
MAT POWER
Iterations 3,2 3,3 3,3,3 3,3,4,4 3,4,4,4 K ii = −1 and ∀i ≠ j K ij = 0 (A.10)
Violations 6 6 20,4 27,5,2 28,7,3
NEW METHOD REFERENCES
Iterations 7 8 9 10 10 [1] Tinney W.F. and Hart C.E. (1967), “Power flow solution by Newton’s
Violations 6 6 24 34 38 method”. IEEE PAS-88,1449-60,Nov 1967
[2] B. Stott and O. Alsaq, “Fast Decoupled Load Flow”IEEE Transactions
V. CONCLUSION on Power Apparatus and Systems,. Vol. PAS-93, No.3, May/June 1974,
This paper presents a modified load flow scheme which pp. 859-867.
implicitly includes the reactive power generation constraints [3] B.Stott, “Review of load flow calculation methods”. Proceedings of
IEEE, VOL. 62, NO. 7, JULY 1974
into the formulation. Using the Mixed Complementarity
Problem framework, the discontinuity in the Q characteristics [4] Bijuna Kunju K. Nagendra Rao P.S., “A novel method to handle Q
generation limits at PV buses in load flow problem”.NPSC (2002)
of the generator has been handled. The algorithm has been
[5] Michael C.Ferris,Christian Kanzow ,Todd S.Munson “Feasible descent
extensively tested on a large number of systems. The algorithms for Mixed complementarity Problems“.March 1998.
performance of the algorithm has been compared with the [6] T. De luca,F.Fachinei and C kanzow “A semismooth equation approach
MATPOWER software. The results are quite promising. Also, to the solution of nonlinear complementarity problems”.Mathematical
it is envisaged that further investigations could help in Programming 75,1996 pp 407-439
improving the proposed scheme further. In addition it appears [7] MATPOWER SOFTWARE by Ray D. Zimmerman, CarloMurillo-
that the MCP approach can be of use in the solution of many Sánchez & Deqiang(David) Gan s E.
other power system problems http://www.pserc.cornell.edu/matpower/
[8] Power system test case archive:
http://www.ee.washington.edu/research/pstca/index.html