Datasets:
tensorkelechi
/
arxiv-cstext

Dataset card Data Studio Files Community
1
text
stringlengths
0
89.3k
5 Solve the graphbased problem with updated
weights wk1 w1 w2 w Nusing
Plasmojl
6end
7Return wwτ where τ U1 K
As shown in Algorithm 1 the inputs of this algorithm
include the initial values of weights w0 step size β and
welldefined multiobjective functions eg f1 f2 f n
andg1 g2 g nN 2 The weighedsum of the given
objective functions can be written into nnodes graphbased
formulation as in 1 During the iteration from k 1K
the weights ware updated by the step size βinto the per
objective stochastic gradient estimates hk
w wkfwkg
then the new weights will be its projection to the N
simplex defined by NwRNwi0i
NP
iNwi 1 For each iteration the problem will be
solved as a single objective graph in Plasmojl and after Kit
erations the output of the algorithm will be uniformly chosen
from the generated Pareto frontier weights w1 wK
It is imperative to mention that while numerous con
ventional methods are available to tackle multiobjective
optimization problems the reason to choose the proposed
gradientbased algorithm is threefold Firstly computa
tional efficiency when solving largescale graphbased multi
objective optimization problem is still a critical issue espe
cially when binary variables exist in the problem The graph
based formulation can be easily integrated in our proposed
algorithm for computing its projected gradient without losing
its convexity Secondly the proposed algorithm supports
vectorvalued nonlinear objectives and constraints which cannot be directly solved in existing Plasmojl orJuMP multi
objective solvers yet Finally this approach is more scalable
to graphbased formulation when combined with Plasmojl
to application in largescale graph network
III C ASE STUDY OFFSHORE WINDFARM
INTERCONNECT
We apply our methodologies on the wellknown WSCC
9bus system Two OWFs are connected to buses 4 and
6 through a fourterminal MMCbased MTDC network to
transmit the wind power 29 30 Two BESS are attached
to buses 4 and 6 to help reduce total cost and improve system
efficiency The system structure is shown in Figure 2
Fig 2 IEEE9 bus system interconnected with OWFs using MMCMTDC
grid
We treat OWFs with converters as active power source
and energy storage devices with converters as active power
load or sources depending on their chargingdischarging
operations MMCs connecting AC and MTDC grids are
simply modeled with fixed coefficients 31 32
The original nonconvex formulation of our MTDC and AC
power flow CCD problem is relaxed to a secondorder conic
program SOCP 33 34 It usually gives more accurate
solutions than linear models and does not have the local
optimality issue in solving nonconvex problems SOCP can
either be solved by many optimization solvers or calculated
through decomposition algorithms
Control Codesign Optimization Formulation
The following two objectives are considered in our model
We can easily extend it to contain additional objetives
minX
iSfEI
iBSi X
iBACX
tTfG
iPG
it QG
it
X
iSX
tTfEO
iPch
it Pdis
it 2
minX
iCX
tTPLoss
itX
ijLACX
tTgijciitcjjt2cijt
X
ijLDCX
tTgDCijviitvjjt2vijt 3
The objective function eq 2 is to minimize total cost
including BESS installation fEI regular generator fuel
cost fGand BESS operation cost fEO We use S
BAC andTto represent the index sets of batteries busesin AC grid and time intervals respectively In the cost
functions BSiis the size of BESS iPch
itandPdis
itare
their charged and discharged power at time t respectively
andPG
itandQG
itare active and reactive power output of
generator iat time t respectively
The objective function eq 3 is to minimize total power
loss including those in MMCs AC grid transmission and
DC grid transmission In this objective function CLAC and
LDCare index sets of MMCs branches of AC and MTDC
grids respectively variable PLoss