Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

The 6th International Conference on Renewable Power Generation (RPG)

19–20 October 2017

Robust optimisation for AC–DC power flow based on second-order cone


programming
Ye Zhou1, Yuan Tian1, Keyou Wang1, Mehrdad Ghandhari2
1
Department of Electrical Engineering, Shanghai Jiao Tong University, People’s Republic of China
2
Department of Electric Power Systems, KTH Royal Institute of Technology, Sweden
E-mail: wangkeyou@sjtu.edu.cn

Published in The Journal of Engineering; Received on 12th October 2017; Accepted on 2nd November 2017

Abstract: This study presents an adjustable robust optimisation method for AC–DC optimal power flow (OPF) considering the uncertainty of
renewable energy source (RESs). The optimal power flow of AC–DC system is modelled as a second-order cone (SOC) problem, and the
affinely adjustable robust OPF (AAROPF) formulation is proposed. To apply AAROPF, the base-point generation is calculated and deter-
mined to match the power with forecasted RES output before the realisation of the uncertainty. Also once the uncertainty is revealed, gen-
erators reschedule its output through participation factors responding to the uncertain fluctuation of RES output to ensure a feasible
solution for all realisations of RES output within a prescribed uncertainty set. Numerical results are obtained on a modified AC–DC IEEE
30-bus to minimise the expected operational cost. Results reveal a higher cost using AAROPF than the deterministic case, but it obtains a
more robust solution with higher successful rates.

θsr Vector of difference phase angle between sending and


Nomenclature
receiving ends at AC line (usrj ) with size of nlAC × 1
Consider the AC–DC system in Section 2.2 has nbAC AC buses and MPQ Incidence matrix associated with AC line active powers
nlAC AC lines, moreover, it has nbDC DC buses nlDC DC lines at DC with size of nbAC × nlAC
side. It operates with nG dispatchable generators and nR RESs: Ml Incidence matrix associated with AC line active power
losses with size of nbAC × nlAC
T
PG Vector of active power generation at each AC bus with MWAC MPQ
size of nbAC × 1 which has only nG non-zero elements PDC Injected/absorbed active power vector at DC buses
PR Vector of RES generation at each AC bus with size of Pr,DC Vector containing DC line flow active powers at
nbAC × 1 which has only nR non-zero elements receiving end of DC lines (Pr,DCj ) with size of nlDC × 1
PD Vector containing active power loads at each AC bus Ploss,DC Vector containing DC line flow active power losses
with size of nbAC × 1 (Ploss,DCj ) at AC lines with the size R of nlDC × 1.
PCONV Injected/absorbed active power vector of converters at RDC Diagonal matrix where Rj, j is resistance of DC lines
PCCs with size of nlDC × nlDC
Pr,AC Vector containing AC line flow active powers at α Vector of proportional factors of losses through
receiving end of AC lines (Pr,ACj ) with size of nlAC × 1 converters with respect to PCONV
Ploss,AC Vector containing AC line flow active power losses MPDC Incidence matrix associated with DC line active powers
(Ploss,ACj ) at AC lines with the size of nlAC × 1 with size of nbDC × nlDC
QG Vector of reactive power generation at each AC bus MlDC Incidence matrix associated with DC line active power
with the size of nbAC × 1 which has only nG non-zero losses with size of nbDC × nlDC
elements MWDC MPTDC
QD Vector containing reactive power loads at each AC bus KRES Vector of uncertainty factors of RESs which has only
with size of nbAC × 1 nRnon-zero elements kj with size of nbAC × 1
QCONV Injected/absorbed reactive power vector of converters
at PCCs
Qr Vector containing AC line flow reactive powers at 1 Introduction
receiving end of AC lines (Qrj ) with size of nlAC × 1 Due to the increasing penetration of intermittent renewable energy
Qloss Vector containing AC line flow reactive power losses sources (RESs), such as solar and wind power, have added addition-
(Qlossj ) at AC lines with the size of nlAC × 1 al uncertainties to AC–DC power system operation. Traditionally,
WAC Vector of square of AC bus voltage with size of the probabilistic and stochastic programming methods were
nbAC × 1 widely used to deal with uncertainties of RESs [1–4]. However,
R Diagonal matrix where Rj, j is resistance of AC lines these methods require more computational cost and also accurate
with size of nlAC × nlAC data of the probability distribution function of the uncertainties,
X Diagonal matrix where Xj, j is reactance of AC lines which is hard to acquire.
with size of nlAC × nlAC Thus, a promising approach to model the optimal power flow
B Diagonal matrix where Bi,i is shunt susceptance of AC (OPF) involving uncertainties is robust optimisation (RO) [5–8],
nodes with size of nbAC × nbAC which only requires knowledge of the range of variation of the

This is an open access article published by the IET under the Creative Commons J. Eng., 2017, Vol. 2017, Iss. 13, pp. 2164–2167
Attribution License (http://creativecommons.org/licenses/by/3.0/) doi: 10.1049/joe.2017.0713
uncertain parameters. A robust OPF formulation is proposed in [5] 2.2 AC and DC line flow based equations
that accounts for load and renewable generation uncertainties by
In this section, the line-flow-based equations [9] are referred to fully
considering several scenarios. However, it does not ensure full
represent AC and DC parts as well as converters of an AC–DC
coverage of the range of uncertain parameters, thus resulting in
network.
solutions which are not immunised against all realisations of the
Consider a hybrid AC–DC system with nG dispatchable genera-
RES output variability. In [6], robust optimisation-based OPF
tors and nR RESs at AC side. In the formulation, AC and DC parts
(ROPF) of an AC network considering wind and solar power uncer-
have their separate line flow based equations and let the variables of
tainty is proposed to determine optimal power dispatch and loca-
power through the converters at PCCs connect the two parts to-
tional marginal prices in a day-ahead market. In [7], an adjustable
gether as an AC–DC system. The equality constraints of the AC–
robust OPF (AROPF) of a DC network is studied and participation
DC OPF problem are as follows:
factors are introduced to realise the adjustable part of generators.
Moreover, because the RO may lead to an extreme conservative
PG + PR − PD + PCONV − MPQ Pr,AC − Ml Ploss,AC = 0 (2)
optimal solution compared to the traditional deterministic optimisa-
tion, this paper takes the price of robustness into account to solve
QG − QD + QCONV − MPQ Qr − Ml Qloss + BWAC = 0 (3)
the conservative problem of RO. An affinely adjustable robust
OPF (AAROPF) model was proposed in [8] for AC power networks
2RPr,AC + 2XQr + RPloss,AC + XQloss − MWAC WAC = 0 (4)
with RESs. It defines the adjustable part of generators an affine
function of uncertain variables in which the participation factors usr − XPr,AC + RQr = 0 (5)
are employed to ensure feasible re-dispatch for all possible realisa-
tions and then minimises the total cost of the conventional genera- XPloss,AC − RQloss = 0 (6)
tors to balance the load and RES output variations. However, most
ROPF literatures [6–8] consider only AC or DC networks but PDC − (1 + a)PCONV = 0 (7)
without hybrid AC–DC networks. In addition also, the DC power
flow model is usually considered to simplify the AC networks in PDC + MPDC Pr,DC + MlDC Ploss,DC = 0 (8)
order to avoid non-linear and non-convex constraints, which
neglects transmission losses and may lead to an infeasible OPF 2RDC Pr,DC + RDC Ploss,DC − MWDC WDC = 0 (9)
solution.
This paper applied an affinely adjustable robust OPF method for where (2)–(6) formulate the AC grid constraints and (7)–(9) are
a hybrid AC–DC network with RESs. The proposed method refers constraints of DC grid. Equations (2) and (3) represent the active
to an AC–DC second-order cone (SOC) OPF model which is dis- and reactive power balance constraints at AC buses, respectively,
cussed and derived in [9] as a basic mathematical model. (4) is associated with voltage drop equation constraints for AC
Furthermore, considering the uncertainty of RESs in the AC–DC lines and (5) represents the phase angle difference on AC lines,
system, an affinely adjustable strategy modified from [8] is (6) represents the relation between active and reactive power
applied to the OPF model. It firstly optimises base-point generations losses at AC lines. At DC grid, (7) represents the relation
to balance the forecasted or expected RES outputs and then employ- between injected DC power at DC network and converter power
ing participation factors to control generations acting as automatic at PCCs, (8) is associated with power balance constraints at DC
generation control (AGC) systems and re-dispatch when RES buses and (9) represents the voltage drop constraint on DC lines.
outputs vary in a defined uncertainty set.
The rest of this paper is organised as follows: Section 2 briefly 2.3 SOC relaxation method
introduces an AC–DC OPF model in SOCP format which is
derived in [9]. Section 3 applies the affinely adjustable strategy In (1)–(8) above, losses at AC and DC lines Ploss,AC, Ploss,DC can be
modified from [8] to the formulated OPF model taking the uncer- formulated as
tainty into account to minimise the expected generation cost.  
Furthermore, it presents appropriate and critical techniques to 2
Ploss,ACj Wr,ACj = Pr,AC + Q 2
R j,j (10)
j r j
solve the AC–DC AAROPF problem. In Section 4, numerical
results are obtained on a modified AC–DC IEEE 30-bus to minim- 2
Ploss,DCj Wr,DCj = Pr,DC R
j DC j,j
(11)
ise the expected operational cost. Results reveal a higher cost using
AAROPF than the deterministic case, but it obtains a more robust
solution with higher successful rates. Equations (10) and (11) are non-linear equality constraints which
can be transformed to inequality with a form of SOC and hence
can be solved efficiently. Two sets of new variables P̃loss,ACj and
P̃loss,DCj are introduced, and for an arbitrary positive and sufficiently
2 SOCP OPF for AC–DC network small constant ɛ ≥ 0, one can obtain

2.1 Objective of AC–DC OPF 2


2P̃loss,ACj Wr,ACj ≥ Pr,ACj
+ Q2rj
The traditional OPF problem of a system is to minimise the oper- (12)
2
ational cost of generators and meanwhile constrained by different 2P̃loss,DCj Wr,DCj ≥ Pr,DCj
operating constraints such as described in Section 2.2. And the ob-
jective of OPF can be formulated as a quadratic cost function 2R j,j
Ploss,ACj = P̃
(1 + 1) loss,AC j
(13)
2RDC j,j
nG nG
  2
Ploss,DCj = P̃loss,DC j
(1 + 1)
   
min Ci PGi = c2i PGi + c1i PGi + c0i (1)
i=1 i=1
Therefore, non-linear quadratic equality constraints (10) and are
relaxed and can be formulated by SOC inequalities and linear
where Ci (PGi) denotes the quadratic cost function of the ith gener- equalities (13) and AC–DC OPF model is transformed into a
ator. PGi is the output and c2i, c1i, c0i are the pre-defined factors. convex problem.

J. Eng., 2017, Vol. 2017, Iss. 13, pp. 2164–2167 This is an open access article published by the IET under the Creative Commons
doi: 10.1049/joe.2017.0713 Attribution License (http://creativecommons.org/licenses/by/3.0/)
3 Affinely adjustable robust OPF factor a
3.1 Uncertainty of RES
PCONVi = PCONVi(base) − ai ksum z0 (18)
Even with better understanding and modelling of forecasted RES
output, an inherent and irreducible uncertainty will always occur
in every forecast which is the key factor to the robustness of the
system. To take uncertainties of RESs into account for traditional 3.3 Solution for AAROPF
OPF, the output of RES j ( j∈nR) can be defined as
Due to the linearity of constraints (2)–(9), some other intermediate
PRj = PR0j + zj , ∀zj [ GU (14) variables can also be expressed by non-adjustable and adjustable
variables with different participation factors, take (2) as an
where PRj is the actual output of RES j, PR0j denotes the expected example in the vector form
value of the forecasted output, ζj is the uncertain parameter denot-
ing the unexpected variation or forecasted error of the output and ΓU PG = PG base − ksum z0 b
defines the corresponding uncertainty set. PR = PR0 + z = PR0 + z0 KRES
Accordingly, the power generations should provide responses to PCONV = PCONV base − ksum z0 a
the uncertain fluctuation of RESs to ensure the feasibility of the (19)
problem under uncertain conditions. The power generation can be Pr,AC = Pr,AC base − ksum z0 b
expressed according to the uncertain variables ζj by different Ploss,AC = Ploss,AC base − ksum z0 c
strategies.
In addition substitute (2) by (19) and the constraint can be written as
3.2 Formulation of AAROPF

The affinely adjustable strategy defines that the power which is gen- PGbase + PR0 − PD + PCONVbase − MPQ Pr,ACbase
eration consists of the base-point value PGi(base), which is the output
corresponding to nominal conditions, and the change in generation   
−Ml Ploss,ACbase + −b − a + MPQ b + Ml c (20)
corresponding to fluctuation in RESs outputs [8]

nR
∗ksum + KRES ∗z0 = 0

PGi = PGi(base) − bi zj (15)
j=1 Apply the adjustable robust optimisation method and (20) can be
separated as two constraints
where βi is the participation factor denoting the rate of change of the
generator output corresponding to the total fluctuation of RESs, and PGbase + PR0 − PD + PCONVbase − MPQ Pr,ACbase
the change in generation which is expressed as the latter part of (21)
− Ml Ploss,ACbase = 0
(15), appears an affine relationship with the uncertainties.
With the affine policy formulated above, the adjustable robust  
OPF method then optimises the base-point generations before the −b − a + MPQ b + Ml c ∗ksum + kRES = 0 (22)
realisation of the uncertain RESs outputs (non-adjustable variables),
and reschedules the generations change through participation where (21) represents the ‘here and now’ process to schedule the
factors once the uncertainty is revealed (adjustable variables), i.e. base-point value which should be constrained to balance the
it considers the whole decision process as two steps [8]: (i) the non- expected or forecasted RES output before the uncertainty is rea-
adjustable variables, corresponding to the ‘here-and-now’ deci- lised. Furthermore, after the uncertainty is revealed, (22) denotes
sions, are used to schedule a base-point generations and cannot that adjustable variables should be rescheduled by employing par-
be adjusted after the uncertainty is revealed, (ii) the adjustable vari- ticipation factors to balance the fluctuation of RESs outputs thus
ables represent the ‘wait-and-see’ decisions because they can adjust to ensure the feasibility of the problem. Also constraints (3)–(9)
themselves to balance the actual RESs outputs by applying partici- and (12) and (13) can be written similarly with (21) and (22).
pation factors. AAROPF in this paper is to minimise the expected Accordingly, the objective of AAROPF is to minimise the
operational cost and meanwhile yield adjustable variables which expected operational cost
are affine functions of RES uncertainty to ensure feasibility for
all possible realisations of the uncertainty set. nG
  
To simplify the formulation of the problem, denote that uncertain min Ci PGi(base) (23)
variables of different RESs can be expressed as i=1

zj = k j z0 (16) By applying certain appropriate transformations discussed above,


AAROPF of AC–DC system considering the uncertainty of RES
where kj is the corresponding uncertainty factor which is a constant is formulated as an SOCP problem and can be efficiently solved.
defined according to the output property of RES. By applying (16),
uncertain variables are set to correlate linearly with each other and 4 Numerical results
(15) can be expressed as The AAROPF formulations of AC–DC system considering uncer-
tainties are programmed in MATLAB and the SOCP is solved
PGi = PGi(base) − bi ksum z0 (17) using Gurobi Optimizer 6.0 [10]. The simulations are carried out
on an Intel Core i3-4160 CPU running at 3.60 GHz with 8 GB of
n
where ksum = S j=1
R
kj , thus variable PGi can be defined by only one RAM.
uncertain variable ζ0. The proposed method is tested on a modified AC–DC
Moreover, active and reactive power through converters at PCCs IEEE 30-bus network [9] with three RESs of 100, 150, 150 MW
PCONV and QCONV, which represent voltage source converters forecasted power at bus 12, 21, 30, respectively (Fig. 1.).
(VSC), should also be formulated as (17) with another participation The uncertainty level of the RES output is given by

This is an open access article published by the IET under the Creative Commons J. Eng., 2017, Vol. 2017, Iss. 13, pp. 2164–2167
Attribution License (http://creativecommons.org/licenses/by/3.0/) doi: 10.1049/joe.2017.0713
forecast and dispatches more power to the conventional generators
as reserve rather than RESs.
Table 2 shows the comparison of successful rates between
AAROPF and deterministic scenarios stochastically generated by
Latin hypercube sampling (LHS) method [11] with different
values of r. Note that successful rates of sampled scenario are
lower as the uncertainty level increases thus the problem may
become infeasible. Whereas AAROPF can obtain a 100% success-
ful rate under different levels of uncertainty. Despite the conserva-
tism of robust optimisation compared to the traditional deterministic
optimisation, AARO can provide a more robust programming
process against all realisations of the RES output variability

5 Conclusion
An affinely adjustable robust OPF method based on an AC–DC
system considering the uncertainty of renewable energy sources is
proposed in this paper. It formulated the OPF model into a SOC
problem, and an affinely adjustable strategy is introduced to the
robust optimisation to optimise the base-point generations before
the realisation of uncertainty and reschedule the generations
through participation factors once the uncertainty is revealed to
ensure a feasible solution for all realisations of uncertainty.
Numerical results are obtained on a modified AC–DC IEEE 30-bus
to minimise the expected operational cost. Results reveal that the
cost using AAROPF is higher than the deterministic case, whereas
it obtains a more robust solution with higher successful rates than sto-
chastic scenarios generated by Latin hypercube sampling method.

6 Acknowledgements
Fig. 1 Modified AC–DC IEEE-30 system with RES
This work was partially supported by National Natural Science
Foundation of China (51407115), National Science and
Table 1 Comparison of total cost in two cases Technology Support Plan of China (2015BAA01B02) and
Research Project on Planning and Optimisation of AC–DC Power
Deterministic case (r = 0) AAROPF (r = 50%) System Based on Global Energy Internet of State Grid China.

Total cost, $/hour 18114.92 18210.96 7 References

[1] Madaeni S.H., Ramteen S.: ‘The impacts of stochastic programming


and demand response on wind integration’, Energy Syst., 2013, 4, (2),
z0 [ [ − 100 MW, 100 MW] × (r/100), where r, percentage of un- pp. 109–124
certainty relative to the forecasted or expected output value, denotes [2] Papavasiliou A., Oren S.S., O’Neill R.P.: ‘Reserve requirements for
the uncertainty level, i.e. the forecasted error of the RES output. wind power integration: a scenario-based stochastic programming
Also the corresponding uncertainty factors of 3 RESs are set to framework’, IEEE Trans. Power Syst., 2011, 26, (4), pp. 2197–2206
1.0, 1.5, and 1.5, respectively. [3] Yuan Y., Li Q., Wang W.: ‘Optimal operation strategy of energy
storage unit in wind power integration based on stochastic program-
Table 1 compares the expected operational cost of AAROPF with ming’, IET Renew. Power Gener., 2011, 5, (2), pp. 194–201
the nominal deterministic case (r = 0). It is evident that the expected [4] Pappala V.S., Erlich I., Rohrig K.: ‘A stochastic model for the optimal
operational cost of the deterministic case is lower than the uncertain operation of a wind-thermal power system’, IEEE Trans. Power Syst.,
case for the reason that the higher the uncertainty level is, the less 2009, 24, (2), pp. 940–950
the operator would be confident about the accuracy of the power [5] Yu H., Rosehart W.D.: ‘An optimal power flow algorithm to achieve
robust operation considering load and renewable generation uncer-
tainties’, IEEE Trans. Power Syst., 2012, 27, (4), pp. 1808–1817
[6] Samakpong T., Ongsakul W., Polprasert J.: ‘Robust optimisation-
Table 2 Comparison of successful rates as the uncertainty level increases
based AC optimal power flow considering wind and solar power un-
certainty’. Int. Conf. Utility Exhibition on Green Energy for
r, % Deterministic cases, % AAROPF, % Sustainable Development (ICUE) 2014, 19–21 March 2014,
Pattaya City, Thailand, 2014
0 100 100 [7] Ding T., Bie Z.H., Bai L.Q.: ‘Adjustable robust optimal power flow
10 100 100 with the price of robustness for large-scale power systems’, IET
20 92.1 100 Gener. Trans. Distrib., 2016, 10, (1), pp. 164–174
30 89.4 100 [8] Jabr R.A.: ‘Adjustable robust OPF with renewable energy sources’,
40 85.0 100 IEEE Trans. Power Syst., 2013, 28, (4), pp. 4742–4751
[9] Baradar M., Hesamzadeh M.R., Ghandhari M.: ‘Second-order cone
50 81.2 100
programming for optimal power flow in VSC-Type AC–DC grids’,
60 79.5 100 IEEE Trans. Power Syst., 2013, 28, (4), pp. 4282–4291
70 77.7 100 [10] Gurobi Optimizer 6.0: February 2015 Available: http://www.gurobi.
80 74.9 100 com
90 72.2 100 [11] Chen Y., Wen J., Cheng S.: ‘Probabilistic load flow method based on
100 70.1 100 Nataf transformation and Latin hypercube sampling’, IEEE Trans.
Sust Energy, 2013, 4, (2), pp. 294–301

J. Eng., 2017, Vol. 2017, Iss. 13, pp. 2164–2167 This is an open access article published by the IET under the Creative Commons
doi: 10.1049/joe.2017.0713 Attribution License (http://creativecommons.org/licenses/by/3.0/)

You might also like