Professional Documents
Culture Documents
Exact Distribution Optimal Power Flow2019
Exact Distribution Optimal Power Flow2019
Exact Distribution Optimal Power Flow2019
Abstract—The distribution optimal power flow (D-OPF) mod- convexity either using convex relaxation techniques [3] or
els have gained attention in recent years to optimally operate a using linear approximation methods [4]. Note that a BIM OPF
centrally-managed distribution grid. On account of nonconvex formulation can be relaxed as semidefinite program (SDP) by
formulation that is difficult to solve, several relaxation methods
have been proposed; the exactness of the solutions obtained dropping the rank-1 constraint [5] and a BFM OPF model
from the relaxed models, however, remain a concern. In this can be relaxed as a second-order cone program (SOCP) by
paper, we identify one such problem related to radial distribution relaxing the quadratic equality constraints [6]. Although the
feeder where second-order cone program (SOCP) relaxation proposed SDP and SOCP relaxations result in a convex prob-
does not yield a solution that is feasible with respect to the lem thus reducing the complexity of nonlinear OPF model, the
original nonlinear OPF model. Specifically, we formulate an OPF
model for PV hosting capacity problem to obtain maximum exactness of the solution obtained from relaxed model is still
PV capacity that a feeder can integrate without violating the of concern. Consequently, several researchers have attempted
operating constraints. The SOCP relaxation for this problem to prove the exactness of relaxed OPF models. Sufficient
yields infeasible solutions. To address this concern, we propose a conditions were provided under which the relaxed SOCP and
convex iteration technique to simultaneously achieve optimal and SDP models are exact [7]–[11]. Further, it was proved in [12],
feasible OPF solution for the PV hosting (maximization) problem.
The proposed approach minimizes the feasibility gap with respect that SDP, chordal, and SCOP convex relaxation techniques are
to original nonlinear constraints at each SOCP iteration. The equivalent for the radial network while for mesh networks,
effectiveness of the proposed approach is validated using IEEE- SDP and chordal relaxation perform better.
13 node and IEEE-123 node test systems. The focus of this work is on D-OPF relaxation techniques
Index Terms—Distribution optimal power flow, convex relax- applied for radial distribution system. Therefore, we focus
ation, second-order cone programming (SOCP), convex iteration.
our attention to SOCP problems. It has been proved in the
I. I NTRODUCTION existing literature that the SOCP relaxation is exact for the
radial distribution feeders under certain conditions [9]. Authors
With the advancement in smart grid technology and in- also claim that even when these conditions are not exactly
creasing penetration of distributed energy systems (DERs), the satisfied, the relaxed SOCP formulation yields a solution that
electric power distribution system is rapidly transforming to is optimal and feasible with respect to the original problem.
an active and bidirectional network. In a centrally managed Further, in [9], authors have provided an excellent visualization
distribution grid, an optimal power flow (OPF) solver finds of conditions for the exactness of relaxed SOCP models. With
multiple applications related to effective management of all the help of a two-bus model, it is demonstrated that the SOCP
grid resources including but not limited to loss minimization, model leads to an optimal solution that lies at the boundary
volt-var optimization, and effective management of DERs [1], of the second-order power cone, thus achieving feasibility.
[2]. The related literature on OPF models from transmission Unfortunately, the exactness of the SOCP relaxation (for ra-
systems is not directly applicable to distribution grid on ac- dial feeder) is contingent upon the choice of objective function.
count of radial feeders, high R/X ratio, and large-variations in Although, SOCP relaxation is exact for most minimization
bus voltages. Consequently, several researchers have proposed problems (in power flow variables), a maximization problem
distribution OPF (D-OPF) formulations. D-OPF models have yields infeasible power flow solution. Notice that although
been largely based on two power flow methods: bus-injection most OPF problems do relate to minimizing a cost function
model (BIM) and branch-flow model (BFM). Although bus- of power flow variables, there are relevant cases when a
injection model applies to general radial/mesh feeders, branch maximization problem needs to be solved. One such case is
flow method is more suitable for modeling radial distribution identifying maximum photovoltaic (PV) penetration limits for
feeders. Both BIM and BFM based OPF models are non- the distribution feeder also known as PV hosting problem. This
convex and difficult to solve. requires maximizing the sum of nodal power injection (from
To address this concern, several relaxed models have been PVs) until system operating constraints (thermal limit, voltage,
proposed in literature that deal with the problem of non- reverse power flow) are not violated. Solving this optimization
This work was supported in part by the U.S. Department of Energy under problem using SOCP model yield infeasible results that lie
contract DE-AC05-76RL01830. inside the second-order cone and not at its boundary.
/ ^
X
ŝũ ŝũ
Maximize. pP
i
V
(5)
;Ɛ Ϳŝ
;Ɛ Ϳ
ũ
i∈N
ƵƐŬ
Subject to: (1)-(4), and
Ϯ
Authorized licensed use limited to: University of Gothenburg. Downloaded on July 26,2020 at 20:24:06 UTC from IEEE Xplore. Restrictions apply.
maximum PV panel rating that can be connected to bus i, γ k defines the ratio of error at current iteration and previous
Iij,rated is thermal rating of line {ij}, and Psub is the active iteration and is a tunable parameter (see [15] for details).
power flow out of the substation node.
e(k) ≥ γ k e(k−1) (13)
C. Convex Relaxation - SOCP Formulation (k) (k) (k) (k)
Also, by substituting for e(k) = (Pij )2 +(Qij )2 −vi lij
As detailed in several related literature, the quadratic equal- and using, γ where, γ k ≤ γ < 1, the error at k th
SOCP-
ity constraint (4) in power flow model makes the D-OPF iteration can be expressed using (14).
problem for hosting capacity in (5) nonconvex. As such this
(k−1) 2 (k−1) 2 (k−1) (k−1)
problem is a quadratically constrained problem and difficult e(k) ≥ γ × (Pij ) + (Qij ) − vi lij (14)
to solve. By relaxing (4) to an inequality as described in (11)
As constraint (14) is nonconvex, we linearize it using first-
a SOCP problem is obtained.
order Taylor series approximation in (15).
vi lij ≥ Pij2 + Q2ij ∀i ∈ N (11) 2Pij
(k−1) (k)
∆Pij + 2Qij
(k−1)
∆Qij − lij
(k)
∆vi − vi
(k−1)
∆lij ≥
(k) (k−1) (k)
The SOCP-relaxation for D-OPF based PV hosting capacity (k−1) 2 (k−1) 2 (k−1) (k−1)
(γ − 1) (Pij ) + (Qij ) − vi lij
problem is detailed as follows. (15)
X
Maximize. pP
i
V
(12) where, at k th iteration, Pij , Qij
(k−1)
, vi
(k−1)
, and lij
(k−1) (k−1)
i∈N th
in known from solving (k − 1) SOCP iteration of
Subject to: (1)-(3), (6)-(10), and (11). the problem. Therefore, (15) is linear in unknown,
(k) (k) (k) (k)
D. Exactness of the Relaxed D-OPF SOCP Problem ∆Pij , ∆Qij , ∆vi and ∆lij .
Note that in this iterative approach, the actual power flow
As described in [9], the SOCP relaxation includes all points solution at k th iteration is obtained by updating power flow
inside the second-order cone, however, minimizing a function variables using (16), where x defines set of problem variables.
of current (or power flow) for unconstrained problem yields
a feasible solution that lies at the boundary of the cone. x(k) = x(k−1) + α∆x(k) (16)
Unfortunately, for the case of PV maximization problem, the where, the acceleration factor, 0 < α < 1 and x(k−1) is the
maximum PV capacity a feeder can accommodate is bounded variable obtained at previous iteration and change in variables,
by upper limits on bus voltages. It has been demonstrated in ∆x(k) is determined at the current iteration.
[9] that when upper bounds on node voltages are binding, The iterative SOCP model at the k th iteration after including
the SOCP relaxation is not exact. This is because the SOCP (15) is detailed in (17)-(23).
solution lies inside the cone and not at the boundary. Due to At k th SOCP-iteration,
this SOCP-relaxation of PV hosting capacity problem results X P V (k)
in optimal solutions that are not feasible with respect to Minimize. − ∆pi (17)
original quadratic equality constraint. i∈N
Subject to:
III. I TERATIVE SOCP F ORMULATION FOR E XACT X
(k−1) (k−1)
S OLUTION OF R ELAXED D-OPF M ODEL (Pij + ∆Pijk ) = (Pjk k
+ ∆Pjk )+
k:j→k (18)
In this section, we propose a convex iteration technique to (k−1) (k) P V,(k) P V,(k)
obtain an optimal and feasible D-OPF solution over multiple rij (lij + ∆lij ) + (pj + ∆pj ) − pload
j
Pij2 + Q2ij and vi lij is gradually reduced to zero. (k−1) (k) (k−1) (k)
2(rij (Pij + ∆Pij ) + xij (Qij + ∆Qij ))+ (20)
The proposed method is designed to specifically achieve this
2 (k−1) (k)
equality over successive iterations. We define an error term, (rij + x2ij )(lij + ∆lij )
e(k) , measuring the feasibility gap at k th iteration that is equal
(k) (k) (k) (k) (k) (k) (k) (k) (k−1) (k) (k−1) (k)
to (Pij )2 + (Qij )2 − vi lij , where, Pij , Qij , vi , lij (vi + ∆vi )(lij + ∆lij ) ≥
(21)
are power flow variables obtained by solving k th iteration of ((Pij
(k−1) 2 (k)
) + ∆(Pij )2 ) + ((Qij
(k−1) 2
) + ∆(Qij )2 )
(k)
Authorized licensed use limited to: University of Gothenburg. Downloaded on July 26,2020 at 20:24:06 UTC from IEEE Xplore. Restrictions apply.
0
P V,(k) P V,(k) P V,(k)
pP V
i,l − pi ≤ ∆pi ≤ pP V
i,u − pi (23) -50
maximum load
Authorized licensed use limited to: University of Gothenburg. Downloaded on July 26,2020 at 20:24:06 UTC from IEEE Xplore. Restrictions apply.
0.04 TABLE I
minimum load
maximum load
PV H OSTING CAPACITY FOR IEEE-13 AND 123 BUS SYSTEM ( IN K W)
0.03
Test Feeder loading condition Nonlinear OPF Iterative OPF
error
February 2014.
0.01 [2] M. Huneault and F. D. Galiana, “A survey of the optimal power flow
literature,” IEEE Transactions on Power Systems, vol. 6, no. 2, pp. 762–
0 770, May 1991.
1 2 3 4 5 6 7 8 9 10 11 12
[3] S. H. Low, “Convex relaxation of optimal power flowpart i: Formulations
iteration number and equivalence,” IEEE Transactions on Control of Network Systems,
vol. 1, no. 1, pp. 15–27, March 2014.
Fig. 4. IEEE-123 bus:Reduction in feasibility gap vs. number of iterations [4] Z. Yuan and M. R. Hesamzadeh, “Improving the accuracy of second-
order cone ac optimal power flow by convex approximations,” in 2018
IEEE Innovative Smart Grid Technologies - Asia (ISGT Asia), May 2018,
results obtained for the PV hosting capacity at the minimum pp. 172–177.
[5] X. Bai and H. Wei, “Semi-definite programming-based method for
and maximum loading conditions are shown in Table I. security-constrained unit commitment with operational and optimal
Next, we analyze the feasibility gap at each iteration of power flow constraints,” IET Generation, Transmission Distribution,
the proposed iterative SOCP formulation. Due to additional vol. 3, no. 2, pp. 182–197, February 2009.
[6] R. A. Jabr, “Radial distribution load flow using conic programming,”
constraints, the maximum error for the system at the minimum IEEE Transactions on Power Systems, vol. 21, no. 3, pp. 1458–1459,
load condition at first iteration is 0.0032 decreases over suc- Aug 2006.
cessive iterations. Similarly, for the maximum load condition, [7] D. K. Molzahn and I. A. Hiskens, “Convex relaxations of optimal power
flow problems: An illustrative example,” IEEE Transactions on Circuits
the error at first iteration is 0.0311 to 0.0032 at 12th iteration. and Systems I: Regular Papers, vol. 63, no. 5, pp. 650–660, May 2016.
It is to be noted that as the number of iteration increases the [8] M. Farivar and S. H. Low, “Branch flow model: Relaxations and
feasibility gap is driven towards zero. We also calculate the PV convexification: Part I,” IEEE Transactions on Power Systems, vol. 28,
no. 3, pp. 2554–2564, Aug 2013.
hosting capacity using original nonlinear-OPF model for both [9] S. H. Low, “Convex relaxation of optimal power flow Part II: Exactness,”
loading conditions (see Table 1). The results show that the IEEE Transactions on Control of Network Systems, vol. 1, no. 2, pp.
proposed iterative approach leads to same solution and hence 177–189, June 2014.
[10] M. Farivar, R. Neal, C. Clarke, and S. Low, “Optimal inverter var control
is successful in achieving an optimal and feasible solution for in distribution systems with high pv penetration,” in 2012 IEEE Power
the PV hosting problem. and Energy Society General Meeting, July 2012, pp. 1–7.
[11] L. Gan, N. Li, U. Topcu, and S. H. Low, “Exact convex relaxation of
optimal power flow in radial networks,” IEEE Transactions on Automatic
V. C ONCLUSIONS Control, vol. 60, no. 1, pp. 72–87, Jan 2015.
[12] S. Bose, S. H. Low, T. Teeraratkul, and B. Hassibi, “Equivalent relax-
The optimal power flow problems are nonconvex and diffi- ations of optimal power flow,” IEEE Transactions on Automatic Control,
vol. 60, no. 3, pp. 729–742, March 2015.
cult to solve. Although several relaxation methods have been [13] E. Grover-Silva, R. Girard, and G. Kariniotakis, “Multi-temporal optimal
proposed, the exactness of the solutions obtained from the power flow for assessing the renewable generation hosting capacity of
relaxed models remain a concern. In this paper, we identify an active distribution system,” in 2016 IEEE/PES Transmission and
Distribution Conference and Exposition (T D), May 2016, pp. 1–5.
one such problem related to second-order cone relaxation of [14] S. Y. Abdelouadoud, R. Girard, F. Neirac, and T. Guiot, “Iterative linear
branch-power flow based OPF model when attempting to solve cuts strenghtening the second-order cone relaxation of the distribution
a maximization problem. The problem of maximizing the system optimal power flow problem,” in 2014 IEEE PES T D Conference
and Exposition, April 2014, pp. 1–4.
feeder’s PV hosting capacity is modeled as an OPF problem [15] A. Jiang and H. K. Kwan, “Minimax design of iir digital filters using
and relaxed using SOCP relaxation. On solving the relaxed iterative socp,” IEEE Transactions on Circuits and Systems I: Regular
model, it is demonstrated that the results are not feasible with Papers, vol. 57, no. 6, pp. 1326–1337, June 2010.
[16] M. Farivar and S. H. Low, “Branch flow model: Relaxations and
respect to the original nonlinear problem. In order to solve this convexificationpart i,” IEEE Transactions on Power Systems, vol. 28,
problem, we have proposed a convex iteration technique that no. 3, pp. 2554–2564, Aug 2013.
by solving multiple iterations of the relaxed-SOCP problem [17] K. P. Schneider, B. A. Mather, B. C. Pal, C. . Ten, G. J. Shirek, H. Zhu,
J. C. Fuller, J. L. R. Pereira, L. F. Ochoa, L. R. de Araujo, R. C. Dugan,
results in an exact solution for the original nonlinear-OPF S. Matthias, S. Paudyal, T. E. McDermott, and W. Kersting, “Analytic
problem. The proposed approach is validated on IEEE-13 and considerations and design basis for the ieee distribution test feeders,”
IEEE-123 bus test feeders. The results demonstrate that the IEEE Transactions on Power Systems, vol. 33, no. 3, pp. 3181–3188,
May 2018.
proposed iterative method reduces the feasibility gap observed
Authorized licensed use limited to: University of Gothenburg. Downloaded on July 26,2020 at 20:24:06 UTC from IEEE Xplore. Restrictions apply.