FundamentalTheoryforScalablePowerSystemsControl
TakayukiIshizaki(TokyoInstituteofTechnology)
JST-NSF-RCNWSonDistributedEMS
Collaborative Works Overview
Wind-IntegratedPowerSystem
Prof.Chakrabortty
Prof.Imura
RetConIntervalOptimization
Prof.Ramdani
ConAn
Sadamotovisit
Multiperiod ElectricityMarketDesign
?ConAn
Prof.Ilić
intelligentBAs
RetConNMR
Sasahara
visit
Profs.Johansson&Sandberg
NetworkControlTheory
RetConNMRIEEE TRANSACTION ON AUTOMATIC CONTROL, VOL. X, NO. X, ... 20XX 6
In addition to the preexistent controller, we suppose that aset of additional decentralized controllers is designed such thateach of them stabilizes the corresponding disjoint subsystem,namely !
˙xi = Aixi +Biu′i
y′i = Cixi.
In particular, we suppose that an additional dynamical mapKi(·) is designed such that
˙xi = Aixi +BiKi(Cixi) (27)
is stable. In the following, we denote the index set of sub-systems to which the additional decentralized controllers areimplemented by L. Unless otherwise stated, we denote the ithcomponent of a stacked symbol by that with the subscript ofi, whereas the stacked version of a symbol is denoted by thatwithout the subscript of i, e.g., xi and x.
On the premise of the definition above, we consider theinput signal in the form of
u = w + w, (28)
for which w and w are to be constructed by cooperative use ofK(·) and Ki(·) for i ∈ L. To realize a practical control system,it is desirable that the local subsystem control is individuallymanaged by each of Ki(·), while the entire system stability isto be ensured by K(·). The simplest way to use both K(·) andKi(·) for i ∈ L would be implementing them as
w = K(y), wi = Ki(yi),
where wi is assumed to be zero for i ∈ L. However, this simpleimplementation does not necessarily guarantee the stability ofthe feedback system; thereby possibly inducing the instabilityof the entire feedback system.
To prevent the induction of instability, let us considergiving a compensation signal to each additional decentralizedcontroller. This is performed in a manner such that u in (28)is constructed by
K : w = K(y) (29)
in conjunction with the combination of
Ki : wi = Ki(yi − yi),
Σi : yi = Fi
"wi, {γj}j∈Ni
# (30)
where wi is assumed to be zero for i ∈ L, and Fi(·)denotes the dynamical map of a compensator Σi. Note thateach compensator measures the interconnection output signalsfrom the neighborhood subsystems and the input signal fromthe preexistent controller, i.e., {γj}j∈Ni and wi. Thus, thecombination of Ki and Σi in (30) can be regarded as adistributed controller using the output signals of neighborhoodsubsystems. In this formulation, we address the followingcompensator design problem.
Problem 4.1: Consider an interconnected system Σ in (23),each of whose subsystem is given by Σi in (22). Let a feedbackcontroller K in (29) be given such that (26) is stable, and letKi in (30) be given such that (27) is stable for every i ∈ L.Then, design a set of compensators Σi for i ∈ L in the formof (30) satisfying the following specifications.
Preexistent controller
Additional controllers
Compensators
connection
Fig. 4. Signal-flow diagram of hierarchical distributed control systems.
– The entire closed-loop system is stable under the feed-back control of (28).
– The design scheme of each compensator Σi is reliant onlyon the corresponding subsystem model of Σi, and not onthe controller models of K and Ki.
These specifications are satisfied when the stability of theentire feedback system is guaranteed for any combination of Kand Ki for i ∈ L such that (26) and (27) are stable. This typeof distributed control systems, referred to as a hierarchicaldistributed control system, is reasonable in the sense thateach retrofit controller, which corresponds to the combinationof Ki and Σi in (30), can be designed and implementedindependently of the other controllers. The entire signal-flowdiagram is depicted as in Fig. 4.
B. SolutionOn the basis of hierarchical state-space expansion, we give
a solution to Problem 4.1. The key to solving the probleminvolves selecting the parameters in Lemma 3.2 as
A = diag(Ai), P = I,
which lead to Γ in (18) whose (i, j)-block is given by
Γi,j =
!αi,jLi, j ∈ Ni
0, otherwise.(31)
Then, we have the following result.Theorem 4.1: With the notation in Section IV, consider the
set of compensators each of whose dynamics is given by
Σi :
!˙xi = Aixi + Li
$j∈Ni
αi,jγj +Biwi
yi = Cixi(32)
for i ∈ L. Then, the closed-loop system under the feedbackcontrol of (28) is stable for any choice of an index set L andany combination of feedback controllers K and Ki such that(26) and (27) are stable.
Proof: See the proof of Theorem 4.2, which includesTheorem 4.1 as a special case.
In Theorem 4.1, the dynamics of Σi can be regarded as anobserver for the interconnection output signals {γj}j∈Ni , towhich the input signal wi from the preexistent controller Kis injected. The compensator Σi plays the role cancelling outthe effect of interferences with neighborhood subsystems and
RetrofitControlNetworkModelReductionConvexAnalysis
Me
Koike
Large-Scale Power Systems ControlIEEJEAST30Model (stablesystem composedof30generators)
�����
�����
����
�����������
�������
����
�����
����
Tokyoarea
Northeastarea
Controllerdesignbylocalmodel?Stability?Betterperformance?
parameters?
parameters?
parameters?
parameters?
parameters?
Completeglobalmodelingisunrealistic!!
retrofitAreaofinterest
localdisturbance
4/13
Problem Formulation: Retrofit Control
Subsystemofinterest (modelavailable)
Othersubsystem(s) (modelunavailable)
possiblylargescale
"Problem# Findaretrofitcontroller suchthat(a)thewholesystemiskeptstable and(b)ismade small forany
Assumption: (i)aremeasurable(ii)thepreexistingsystemwithoutisstable
Hierarchical State-Space ExpansionCoupledstateequationofanda
stable (assumption)
state-spaceexpansiontocascaderealization
Hierarchicalrealization -dim
stable stabilizedby
decoupled!
"Lemma#
and forany
If andthen
6/13
Localized Controller Design
7/13
Hierarchicalrealization modelavailable!
Howtoimplement??
"Lemma# Designacontrollersuchthat
isstableand
Thentheclosed-loopsystemisstableand
Generalizationtodynamicalcontrollerdesignisstraightforward
constant
Controller ImplementationHowtoimplement??
"Theorem#
Localizingcompensator
Theclosed-loopsystemwiththeretrofitcontroller
isinternallystableanditsatisfies
with
with
Demonstration by Swing Equation Model
Localperf.
Glob
alperf.
High
Time
:designbasedoninfinitebusmodel:retrofitcontrol
unstable!
Freq
. dev
iatio
n
highgainLQR
lowgainLQR
retrofit
Scalabledevelopmentoflarge-scalestablenetworksystemsbasedondistributeddesignandimplementation ofmultipleretrofitcontrollers
6 MVA
w/ controllerw/o controller
Enhanced Damping of Wind Power Systems
20 MVA 40 MVA
Time (s)
rela
tive
gen.
angl
es (d
eg)
retrofitretrofit
Retrofitcontrolofwindpowerplantcanenhancedampingperformance
Poorlydampedaswindpenetrationlevelincreases
Time (s)rela
tive
gen.
angl
es (d
eg)
Freq
. diff
. (H
z)
byT.Sadamoto,A.Chakrabortty
RetrofitcontrolLocalizationofcontrollerdesignandimplementationStabilityguaranteeandcontrolperformanceimprovement
Concluding Remarks
Hierarchicalstate-spaceexpansionRedundantrealizationwithcascadestructureSystematicanalysisforstabilityandcontrolperformance
Thankyouforyourattention!
GeneralTheory: T.Ishizaki,T.Sadamoto,J.Imura,H.Sandberg,K.H.Johansson:RetrofitControl:LocalizationofControllerDesignandImplementation.arXiv
PowerSystemsApplication: T.Sadamoto,A.Chakrabortty,T.Ishizaki,J.Imura:ARetrofitting-BasedSupplementaryControllerDesignforEnhancingDampingPerformanceofWindPowerSystems.ACC2017