A Linearized OPF Model with

Reactive Power and Voltage

Magnitude: A Pathway to improve
MW-only DC OPF

SUBMITTED BY:
TANMAY TEWARI
17104078
 Motivation

 Optimal power Flow(OPF) calculation is crucial to facilitate the secure

and economic operation of power grids.
 Although numerous solving algorithms have been proposed, methods that
achieve the global optimum cannot guarantee the convergence
because of the nonlinearity and nonconvexity nature of the OPF problem.
 There are basically two methods to ensure computational convergence:
1) convex relaxation
2) network model linearization
 The network model linearization approach is widely used in power
industries.
 Most system operators use the DC OPF method due to the concerns
of computational robustness .
 However, the DC OPF method oversimplifies the network model,
which causes the following major drawbacks:
1) the reactive power (Q) and the voltage magnitude (v) are not
modelled, and
the voltage /VAR security cannot be considered.
2) the constraints of the active power flow are considered, but the
effect of the
reactive power flow on line current is ignored.
 In this study, a linear and convergence-guaranteed OPF method is
proposed.
 Key Contributions

 In this study, a linearly-constrained and convergence guaranteed OPF

method is proposed.
 The proposed method improves the performance of the MW only
DCOPF method without requiring any additional information.
 Compared with the DC OPF method, The proposed method explicitly
models the reactive power components such as Q and v whilst
improving the accuracy of the DCOPF solution.
 The proposed method has desirable computational performance and is
suitable for practical implementation of real-world market clearing.
 If the accuracy of the network model needs to be further improved , a
solution that is notably close to ACOPF optimum can be obtained by
taking one more iteration using the proposed method.
 Mathematical Modelling

 Branch Flow Linearization

 Exact power flow through branch (i,j)[here 𝑔ij and 𝑏ij are admittances
of branch (i,j)];
𝑃ij = 𝑔ij(vi2 –vivjcosij ) - 𝑏ijvivjsinij ;
𝑄ij = −𝑏ij(vi2 –vivjcosij ) - 𝑔ijvivjsinij ;
 To keep the linear and quadratic terms, the second-order Taylor series
expansions of the sine and cosine functions are used. Assuming that
the value of ij is normally a small number and the magnitude of v is
close to 1.0 p.u, the following equations are obtained:
𝑃ij = 𝑔ij(vi2 –vivj ) - 𝑏ijij + 𝑔ijij2/2 ;
𝑄ij = −𝑏ij(vi2 –vivj ) - 𝑔ijij - 𝑏ijij2/2 ;
 Regarding v2 as an independent variable, mathematical
transformation for the nonlinear voltage magnitude term is
done which gives a linear and a quadratic term
 By substituting this in reduced equations of line flows we get

𝑃ijA = 𝑔ij(vi2 – vj2)/2 - 𝑏ijij + 𝑃ijL ;

𝑄ijA = −𝑏ij(vi2 –vj2)/2 - 𝑔ijij + 𝑄ijL;

 Where 𝑃ijA , 𝑄ijA are linear approximation of active Reactive

Power Flow and 𝑃ijL , 𝑄ijL are active reactive losses on branch
(i,j) given by:
𝑃ijL = 𝑔ij(vij2 + ij2)/2 ;
𝑄ijL = −𝑏ij(vij2 + ij2)/2 ;
 Nodal Power Balance Equation
 Nodal active and reactive power injection at bus i are given by:
Pi = 𝑗=1 𝐺ijvivjcosij
𝑁
+ Bijvivjsinij ;
Qi = - 𝑗=1 Bijvivjcosij
𝑁
− Gijvivjsinij ;
Here G and B are elements of Ybus matrix.
 When compared with Power flow equations they can be written as
follows
Pi = ( 𝑗=1 𝐺ij)
𝑁
vi2 + (𝑖,𝑗)∈𝑘 Pij ;
Qi =( 𝑗=1 −Bij)
𝑁
vi2 + (𝑖,𝑗)∈𝑘 Qij ;
 Loss Linearization
 In the power flow equation only the loss terms are quadratic
𝑃ijL = 𝑔ij(vij2 + ij2)/2 = 𝑃ij,vL + 𝑃ij,L ;
𝑄ijL = −𝑏ij(vij2 + ij2)/2 = 𝑄ij,vL + 𝑄ij,L ;
 Based on the first-order Taylor series expansion, term 𝑃ij,L can be
linearized as follows:
𝑃ij,L=𝑔ij ij2/2 ≈ 𝑔ijij,0 (ij - ij,0) ;
 As shown above , 𝑃ij,vL is converted into a function of v2 and then
linearized:
2 2 − v2 2
Step 1: v +
𝑃ij,vL = 𝑔ijvij2 /2 ≈ 𝑔ij v i + vjv v
vij2 /2 =𝑔ij v i + vj /2
i,0 j,0 i,0 j,0

Step 2: 𝑃ij,vL = 𝑔ij vvi,0 +− vvj,0 (vi2 − vj2) - 𝑔ij(vi,0 − vj,0 )2 /2;
i,0 j,0
 Based on the base case operating condition the losses can be
linearized as:
PijL ≈LFP,T + LFP,vTv2 + offsetP,i,j;
QijL ≈LFQ,T + LFQ,vTv2 + offsetQ,i,j;
 According to above equations , the expressions for the sensitivities
LFP,T and LFP,vT (where v is in fact v2 ) and offsetP,i,j is given by:

LFP,T = gijij,0 Mij and LFP,vT = gij vvi,0 +− vvj,0 Mij

i,0 j,0
offsetP,i,j = - gij [ ij,02 +(vi,0 − vj,0 )2 ]/2
 Similarly LFQ,T , LFQ,vT and offsetQ,i,j can be derived.
 PROPOSED OPF SOLUTION
 A general formulation of the objective function of the proposed linear
OPF model is described as follows:
min f(Pg , Qg , v2 , ) =
𝑔∈𝐺 (Cg PP 2 + Cg PP + Cg P )+(Cg QQ 2 + Cg QQ + Cg Q )
,1 g ,1 g ,0 ,2 g ,1 g ,0
 Power Flow Equations
 Because of the quasi-linear P-θ relationship, the values of 𝑃ij,L are
usually restricted by the optimization objective of minimizing the
operational costs. To avoid the negative values of 𝑃ij,vL (since losses can
only be positive), the following constraint is added:
𝑃ij,vL + ε+i,j ≥0, (i,j)∈ K
The penalty factor for ε+i,j is added to the objective function.
 Constraints
 Nodal Power Balance Equations

𝑔∈𝑖 Pg -Pi,d= ( 𝑗=1 𝐺ij) vi2 +

𝑁
P A ;
(𝑖,𝑗)∈𝑘 ij

𝑔∈𝑖 Qg -Qi,d=( 𝑗=1 −Bij) vi 2 +

𝑁
Q A ;
(𝑖,𝑗)∈𝑘 ij
 Branch Flow Limits
(𝑃ijA )2 + (𝑄ijA)2  Sij,max2
 This constraint is quadratic which can be easily linearized. In this paper,
the piecewise linearization method is used to facilitate the linear
formulation.
 This method is illustrated with the help of given figure(in next slide) where x
and y axis represent the reactive and active power flow in branch (i,j).
Constraint responds to area within the circle.
Here 1u = 2u = 1d = 2d =300 and M =N=20
 The M+N linear constraints characterized by obtained lines are used to
approximate constraints:
Li,ju,n ≥ 0 , n=1,…..,N
л(𝑃ijA ,𝑄ijA)= ,(𝑖, 𝑗) ∈ 𝑘
Li,ju,m ≥ 0 , m=1,…..,M
 Operational Constraints
Pgmin ≤ Pg ≤ Pgmax ; Qgmin ≤ Qg ≤ Qgmax ; g∈N
vi,min2 ≤ vi2 ≤ vi,max2 ; i∈N
 Further Improving Accuracy
 This section provides a method to obtain a solution that is notably close to
the optimum of the AC OPF model by taking only one more iteration.
 Formulation of the Warm-Start OPF Model
 Based on the values of (v1 , 1) obtained from the solution of the
proposed linear OPF model, a warm-start network model can be
formulated.
 Approximation of sin and cos are improved as:
sinij ≈ ij cosij,1- ij,1 cosij,1 + sinij,1 = sij1ij – sij0 ;
cosij ≈ ij (-sinij,1)- ij,1 (-sinij,1)+ cosij,1 = cij1ij – cij0
 An improved approximation for the nonlinear term vivjθij is used:
vivjθij ≈ vi,1vj,1θij + (vivj- vi,1vj,1)θij,1
 The improved warm start model is given below (including linearized
losses) which is then substituted in the OPF problem:
𝑃ijA = 𝑔ijvi2 – 𝑔ijP (vi2 +vj2)/2 - 𝑏ijP (ij - ij,1)+ 𝑃ij,vL ;
𝑄ijA = -𝑏ijvi2 + 𝑏ijQ (vi2 +vj2)/2 - 𝑔ijQ (ij - ij,1)+ + 𝑄ij,vL;
where 𝑔ijP = (𝑔ij cij0 + 𝑏ij sij0 )+ (𝑔ij cij1 + 𝑔ij sij1 ) ij,1 ;
𝑏ijP = (𝑔ij cij1 + 𝑔ij sij1 ) vi,1vj,1 ;
𝑏ijQ = (−𝑔ij sij0 + 𝑏ij cij0 )- (𝑔ij sij1 - 𝑏ij cij1 ) ij,1 ;
𝑔ijQ = (𝑔ij sij1 - 𝑏ij cij1 ) vi,1vj,1 ;
𝑃ij,vL = 𝑔ijP vvi,1 +− vvj,1 (vi2 − vj2) - 𝑔ijP (vi,1 − vj,1 )2 /2;
i,1 j,1
𝑄ij,vL = −𝑏ijQ vvi,1 +− vvj,1 (vi2 − vj2) + 𝑏ijQ(vi,1 − vj,1 )2 /2;
i,1 j,1
 Key Results
 The proposed method is simulated along with DC OPF model and exact AC
OPF model for IEEE 30-bus system(shown here) and Polish 2383-bus system to
compare the performances.

Here M_I is ACOPF, M_II is DCOPF, M_III is proposed method,

M_IV is warm start OPF using M_III solution
 Error in objective Maximum
function Error in Qij
P Q M_II M_III M_IV M_III M_IV
(in %) (in %) (in %) (in %) (in %) (in%) (in %)

20 20 0.56 0.47 0.025 0.27 0.024

30 40 1.21 0.31 0.050 0.22 0.022
40 20 0.96 0.078 0.046 0.26 0.025
40 40 1.44 0.10 0.046 0.27 0.024
50 50 1.77 0.30 0.044 0.26 0.024
60 60 2.39 0.57 0.045 0.27 0.028
 Conclusions
 The technique presented in this paper significantly improves the DC
OPF solution without requiring any additional effort.
 Also the added iterations of checking AC feasibility of a DC OPF
solution are not required here since warm start technique solves this
problem.
 In existing SLP approach the step size of voltage angles and
magnitudes must be controlled to reduce chances of divergence.
This control increases the no of iterations required whereas here only
one iteration is required of warm start OPF.
 The accuracy of loss linearization via proposed method is
acceptable when power flow pattern on branches does not
dramatically change between base case operating condition for
loss linearization and the obtained OPF solution. If that is not the
case solution is different from optimum.
Thank You