Prof.Dr.-­‐Ing.  Luigi  Vanfretti

iTesla  Workshop  – Brussels,  BelgiumNov.  4-­‐5,  2015

iTesla  RaPId -­‐ Rapid  Parameter  IdentificationA  Toolbox  for  Model  Validation  and  Identification

luigiv@kth.seAssociate  Professor,  Docent

Electric  Power  Systems  Dept.KTH

Stockholm,  Sweden

Luigi.Vanfretti@statnett.noSpecial  Advisor

Research  and  Development  Division  Statnett SFOslo,  Norway

Outline

• Recap– The  iTesla  Power  Systems  Modelica  Library  (iPSL)– Where  is  Modelica  used  in  iTesla?  And  where  does  RaPId  fit?

• Why  Model  Validation– Motivation– Model  validation  loop– Validation  levels– Requirements  for  a  model  validation  SW  tool

• iTesla  RaPId– Model  validation  software  architecture  based  using  Modelica  tools  and  FMI  

Technologies– RaPId  Intro  and  Demos

• Conclussions and  Recommendations

RECAPFirst,  you  need  a  good  model  to  validate!

Modeling,  Simulation  Tools  and  Model  Validation

Current/new  tools  will  need  to  perform  simulations:• Of  complex  hybrid  model  components  

and  networks  with  very  large  number  of  continuous   and  discrete  states.

• Models  need  to  be  shared,  and  simulation  results  need  to  be  consistentacross  different  tools  and  simulation  platforms…

Assume:• If  models  could  be  “systematically  shared  

at  the  equation   level”,  and  simulations  are  “consistent  across  different  SW  platforms”  – we  would  still  need  to  validate  each  new  model  (new  components)  and  calibrate  the  model  to  match  reality.

iPSL:iTesla  Power  Systems  Modelica   Library

• The  iTesla  Power  Systems  library  developed  using  as  reference  domain  specific  software  tools  (e.g.  PSS/E,    Eurostag,  PSAT  and  others)

• The  library  has  been  used  with  several  Modelica  compliant  tools  software:  OpenModelica,  Dymola,  SystemModeler,  Jmodelica,  etc.

• Components  and  systems  are  validated  against  proprietary  tools  and  one  OSS  tool  used  in  power  systems  (domain  specific)

Model  Editing  in  OpenModelica

Model  Editing  inDymola

iPSL! Now  Available  as  OSS!

• Download  at:• https://github.com/itesla/ipsl

Get  it  while  it’s  hot!

Reminder:                                  models  are  used  as  a  key  enabler  of  the  iTesla  Toolbox!

Sampling  of  stochastic variables

Elaboration  of  starting network  

states

Impact  Analysis(time  domainsimulations)

Data  mining on  the  results of  simulation

Data  acquisition  and  storage

Merging module

Contingency screening  (several stages)

Time  domainsimulations

Computation   of  security rules

Synthesis of  recommendationsfor  the  operator

External data  (forecasts and  snapshots)

Improvements of  defence and  

restoration plans

Offline  validation  of  dynamic models

Data  management

Data  mining  services

Dynamic  simulation Optimizers Graphical  

interfaces

Modelica  use  fortime-­‐domain  simulation

WHY  POWER  SYSTEM  MODEL  VALIDATION?

Assume  that  you  have  a  “good  enough”  model,  then  what?

Why  “Model  Validation”?• iTesla  tools  aim  to  perform  

“security  assessment”• The  quality  of  the  models  

used  by  off-­‐line  and  on-­‐line  tools  will  affect  the  result  of  any  SA  computations– Good  model:  approximates  

the  simulated  response  as  “close”  to  the  “measured  response”  as  possible

• Validating  models  helps  in  having  a  model  with  “good  sanity”  and  “reasonable  accuracy”:  – Increasing  the  capability  of  

reproducing  actual  power  system  behavior  (better  predictions)

2 3 4 5 6 7 8 9-2

-1.5

-1

-0.5

0

0.5

1

Δ P

(pu)

Time (sec)

Measured ResponseModel Response

US  WECC  Break-­‐up  in  1996

BAD  Model   for  Dynamic  Security  Assessment!!!

The  Model  Validation Loop• The  major  ingredients  of  the  model  validation  loop  below  have  been  

incorporated  into  the  proposed  general  framework  for  validation:

• A  model  can  NEVER be  accepted  as  a  final  and  true  description  of  the  actual  power  system.– It  can  only  be  accepted  as  a  suitable  (“Good  Enough”)  description  of  the  system  for  specific  

aspects  of  interest  (e.g.  small  signal  stability  (oscillations),  etc.)– Model  validation  can  provide  confidence on  the  fidelity  and  quality  of  the  models  for  

replicating  system  dynamics  when  performing  simulations.– Calibrated  models  will  be  needed  for  systems  with  more  and  more  dynamic  interactions

(1)Prior  Knowledge(Experience)

Parameter  Tuning

Choice  of  Model  Structure  (2)

Criterion  fit:-­‐  fitting  the  data?-­‐  purpose  of  the  

model?  (3)

OK!Model  meets  the  

performance  criteria/indicator.

Not  OK!Revise  from  (1),  (2),  or  (3)

Experiment  Design(Staged  Tests)

Data(System  Measured  

Response)

Assumed  Model  Structure(s)(Model  of  the  System  During  

the  Test) Simulate  the  Model

(System  Simulated  Response)Chosen  Criterion  of  Fit

(Performance  Indicator)

Validate  the  Model

(Model  of  the  System  During  the  Test)

?

Different  Validation  Levels

• Component  level  response– e.g.  generator  such  as  wind  

turbine  or  PV  panel

• Cluster  level  behavior– e.g.  gen.  cluster  such  as  wind  

or  PV  farm

• System  level  behavior– e.g.  power  system  small-­‐signal  

dynamics  (oscillations)

G1

G2

G3

G4

G5

G7

G6

G8

G9

What  is  required  from  a  SW  architecture  for  model  validation?

Models

Static Model

Standard Models

Custom Models

Manufacturer Models

System LevelModel Validation

Measurements

Static Measurements

Dynamic Measurements

PMU Measurements

DFR Measurements

Other

Measurement, Model and Scenario

Harmonization

Dynamic Model

SCADA MeasurementsOther EMS Measurements

Static Values:- Time Stamp- Average Measurement Values of P, Q and V- Sampled every 5-10 sec

Time Series:- GPS Time Stamped Measurements- Time-stamped voltage and current phasor meas.

Time Series with single time stamp:- Time-stamp in the initial sample, use of sampling frequency to determine the time-stamp of other points- Three phase (ABC), voltage and current measurements- Other measurements available: frequency, harmonics, THD, etc.

Time Series from other devices (FNET FDRs or Similar):- GPS Time Stamped Measurements- Single phase voltage phasor measurement, frequency, etc.

Scenario

Initialization

State Estimator Snap-shop

DynamicSimulation

Limited visibility of custom or manufacturer models will by itself put a limitation on the methodologies used for model validation

• Support  “harmonized”  dynamic  models

• Process  measurements  using  different  DSP  techniques

• Perform  simulation  of  the  model

• Provide  optimization  facilities  for  estimating  and  calibrating  model  parameters

• Provide  user  interaction

THE  RAPID  TOOLBOXA  model  validation  and  parameter  identification  SW

Coupling  Models  with  Simulation  &  Optimization:  FMI  and  FMUs

• FMI  stands  for  flexible  mock-­‐up  interface:– FMI  is  a  tool  independent  standard to  support  both  model  exchange  and  co-­‐simulation  

of  dynamic  models  using  a  combination  of  xml-­‐files  and  C-­‐code,  originating  from  the  automotive   industry  

• FMU  stands  for  flexible  mock-­‐up  unit– An  FMU  is  a  model  which  has  been  compiled  using  the  FMI  standard  definition

• What  are  FMUs  used  for?– Model  Exchange

• Generate  C-­‐Code  of  a  model  as  an  input/output   block  that  can  be  utilized  by  other  modeling  and  simulation  environments

– FMUs  of  a  complete  model  can  be  generated  in  one  environment  and  then  shared  to  another  environment.• The  key  idea  to  understand  here  is  that  the  model  is  not  locked  into  a  specific  

simulation  environment!• We  use  FMI  technologies  to  build  RaPId

The  FMI  Standard  is  now  supported   by  40  different  simulation  tools.

User  Target(server/pc)

Model  Validation  Software

iTesla  WP2  Inputs  to  WP3:  Measurements  &  Models

Mockup  SW  ArchitectureProof  of  concept  of  using  MATLAB+FMI

EMTP-­‐RV  and/or  other  HB  model  simulation  traces  and  simulation  configuration

PMU  and  other  available  HB  measurements

SCADA/EMS  Snapshots  +  Operator  Actions

MAT

LAB

MATLAB/Simulink  (used  for  simulation  of  the  Modelica  Modelin  FMU  format)

FMI  Toolbox  for  MATLAB(with  Modelica  model)

Model  Validation   Tasks:

Parameter  tuning,  model  optimization,  etc.

User  Interaction

.mat  and  .xml  files

HARMONIZED  MODELICA  MODEL:Modelica  Dynamic  Model  Definition  for  Phasor Time  Domain  Simulation

Data  Conditioning

iTesla  Data  Manager

Internet  or  LAN.mo files

.mat  and  .xml  files

FMU  compiled  by  another  tool

FMU

Proof-­‐of-­‐Concept  ImplementationThe  RaPId  Mock-­‐Up  Software  Implementation

• RaPId is our proof of conceptimplementation (prototype) of a softwaretool for model estimation and validation.The tool provides a framework for modelidentification/validation, mainlyparameter identification.

• RaPId is based on Modelica and FMI –applicable to other systems, not onlypower systems!

• A Modelica model is fed through anFlexibleMock-­‐Unit (i.e. FMU) to Simulink.

• The model is simulated and its outputs arecompared againstmeasurements.

• RaPId tunes the parameters of the modelwhile minimizing a fitness criterionbetween the outputs of the simulationand the experimental measurements ofthe same outputs provided by the user.

• RaPId was  developed  in  MATLAB.– The  MATLAB  code  acts  as  wrapper to  

provide  interaction  with  several  other  programs  (which  may  not  need  to  be  coded  in  MATLAB).

• Advanced  users  can  simply  use  MATLAB  scripts  instead  of  the  graphical  interface.

• Plug-­‐in  Architecture:– Completely   extensible  and  open  

architecture  allows  advanced  users  to  add:• Identification  methods• Optimization  methods• Specific  objective   functions• Solvers   (numerical  integration  

routines)

Options  and  

Settings

Algorithm  Choice

Results  and  Plots

Simulink  Container

Output  measurement  data

Input  measurement  data

What  does  RaPId  do?

Output   (and  optionally   input)  measurements  are  provided   to  RaPId  by  the  user.

At  initialization,  a  set  of  parameters  is  pre-­‐configured   (or  generated  randomly  by  RaPId)The  model   is  simulated  with  the  parameter  values  given  by  RaPId.

The  outputs   of  the  model  are  recorded  and  compared  to  the  user-­‐provided  measurementsA  fitness  function   is  computed   to  judge  how  close  the  measured  data  and  simulated  data  are  to  each  otherUsing  results  from  (5)  a  new  set  of  parameters  is  computed  by  RaPId.

1

2

3

4

5

2’

ymeas

t

ymeas , ysim

tSimulink  ContainerWith  Modelica  FMU  Model

Simulations  continue  until  a  min.  fitness  or  max  no.  of  iterations  (simulation  runs)  are  reached.

1

2

3

4

5

RaPId! Now  Available  as  OSS!

• Download  at:• https://github.com/SmarTS-­‐Lab/iTesla_RaPId

Get  it  while  it’s  hot!

Video  Demo!

Validating  the  Excitation  System  Model  of  the  Mostar  Power  Plant!

Stay  for  the  workshop  to  work  with  this  example!

More  On-­‐line  Video  Demos!GUI  example

https://www.youtube.com/watch?v=e7OkVEtcz6ACLI  example:

https://www.youtube.com/watch?v=4qrPASIWdiY

Stay  for  the  workshop  to  work  with  RaPId!

TAKE AWAYS!Opportunities  and  Conclussions

Modelica  and  FMI  Modern  computer  technologies  opening  new  opportunities

• Validating  power  system  models  requires  to  develop  new  methods  and  new  tools  itself:– The  tools  for  model  validation  can  be  built   independent   from  a  specific  power  system  

simulator,  thanks  to  the  development   of  the  Modelica  library  which  allows  to  run  the  models  with  different   tools  and  using  FMUs.  

– Model  validation  tools  developed   in  this  approach  will  provide  additional   flexibility   to  couple  in  a  modular  fashion:  simulation,  optimization  and  signal  processing  tools.

iTesla Dynamic  Database

Power  System  Model  Definition   in  

Modelica

Model  Validation  Tools

(RaPId)

Model  validation   results,  including   validation  metrics  and  parameters  are  sent  back  to  the  dynamic  database  

which  updates  these  specifications

Simulation

Signal  Processing

Optimization

New  Method  ImplementedTo  Deal  with  Difficult  Validation  Problem

Conclusions  andLooking  Forward

• Modeling  power  system  components  with  Modelica  (as  compared  with  domain  specific  tools)  is  very  attractive:– Formal  mathematical  description  of  the  model  (equations)– Allows  model  exchange  between  Modelica  tools,  with  consistent  (unambiguous)  

simulation  results• The  FMI  Standard  allows  to  take  advantage  of  Modelica  models  for:

– Using  Modelica  models  in  different  simulation  environments– Coupling  general  purpose  tools  to  the  model/simulation  (case  of  RaPId)

• There  are  several  challenges  for  modeling  and  validating  “large  scale”  power  systems  using  Modelica-­‐based  tools:– A  well  populated  library  of  typical  components   (and  for  different  time-­‐scales)– Support/linkage  with  industry  specific  data  exchange  paradigm  (Common  Information  

Model  -­‐ CIM)• Developing  a  Modelica-­‐driven  model  validation  for  large  scale  power  systems  is  more  

complex  challenge  than  the  case  of  RaPId.  • However,  the  results  obtained  so  far  by  Tractebel are  encouraging,  and  are  a  good  starting  

point  to  build  CIM-­‐compliant  tools  for  model  validation.• We  have  released  RaPId  as  a  Free  and  Open  Source  Software,  and  the  iTesla  Power  Systems  

Modelica  library  will  be  released  shortly.

Get  it  while  it’s  hot!

Questions?

luigiv@kth.seluigi.vanfretti@statnett.no

Thank  you!

25

RaPId:  Now  Available  as  OSS!:  https://github.com/SmarTS-­‐Lab/iTesla_RaPId

• A  view  of  the  future:– What  new  modelingand  simulation  technologies  can  allow  users  to  do  

with  their  models  when  they  are  free  from  a  specific  tool.– Collaboration  with  Michael  Tiller,  Xogeny:  http://www.xogeny.com

Modelica  and  FMI  Modern  power  system  computing  using  xengen