opal-rt rt14 conference: state-space nodal

13
The 7th International Conference on Real-Time Simulation Technologies Montreal | 9-12 June, 2014 1 Christian Dufour, Ph.D. OPAL-RT TECHNOLOGIES [email protected] State-Space Nodal (SSN) New challenges of 2014

Upload: opal-rt-technologies

Post on 03-Jul-2015

160 views

Category:

Software


2 download

DESCRIPTION

State-Space Nodal (SSN): New challenges of 2014, by OPAL-RT

TRANSCRIPT

Page 1: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

1

Christian Dufour, Ph.D.OPAL-RT TECHNOLOGIES

[email protected]

State-Space Nodal (SSN)

New challenges of 2014

Page 2: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

2

• SSN solver in brief

• SSN usage in distribution grid applications

• SSN usage in More Electric Aircrafts.

• SSN usage in super-large fusion reactor converters.

• Iterative methods in SSN.

Outline

Page 3: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

3

• SimPowerSystems (SPS) is the main blockset of Simulink to simulate electric circuits and power systems

• ARTEMiS is a real-time enabler for SPS, its adds adapted real-time solvers like SSN (State-Space Nodal)

Simulink

SimPowerSystems

Rea

l-Ti

me

ARTEMiS

eMEGAsim

ARTEMiS, SSN and SimPowerSystems

Page 4: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

4

State-space vs. ‘Nodal method’

• Directly find the state-space equations (‘ABCD’ equations)

• not so easy to do in general, many special cases exists (ex: state dependency)

• 1st , make small pieces (or groups!) of the network connected to nodes

• Find the equation of each one• Ex: Capa: Vc=1/C*INT{i*dt}

• Solve the common voltages/currents with nodal admittance equation VY=I

Page 5: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

5

State-Space Nodal Solver (SSN)• Reducing the node number is critical because the LU factorisation of Y matrix time is

proportional to the cube (O3) of the size of matrix.

• SSN allow user to select the node location and limit the number of nodes.

• Below is an example of this node reduction effect: • In EMTP/RTDS/Hypersim: using the standard set of {R, RL, RLC, etc…} branches, we end up with 30 nodes.

• In SSN, the user choose the node location: in the case below, it produces only 1 node!

Comparison of node number for standard nodal admittance method and SSN

Using EMTP/RTDS/Hypersim: 10 nodes Using SSN: 1 node!

Page 6: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

6

Simulation challenges of large power grids• Grids represent a very large simulation problem

• Simulation can be parallelized by using the line propagation delays (-> maps into inter-core delays!)

6-core eMEGAsimGrid model

1 2

3 4

5 6

Page 7: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

7

Simulation challenges of power systems• So what do we do to parallelize the simulation when no transmission lines?

• Ex: distribution systems, on-board power systems…

Example of renewable integration into a distribution gridAircraft on-board power system

Page 8: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

8

Parallel State-Space Nodal Solver (SSN)• SSN allow the parallelisation of the network equations without delay.

• How is this possible? Reducing the node number has the effect of creating much bigger branch equations. These SSN group equations are computed on many CPUs in parallel!

• Note that this would be possible in theory for standard EMTP algo but in would be practically inefficient because of the inter-core communication costs for a large number of branches.

SSN groups computed mostly in parallel

LU solution is madeon a different core

Page 9: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

9

Active distribution grid: Actual client cases• These 2 results below are from actual clients.

• They used parallel SSN with 4 cores of 3.3 GHz i7 Intel PC, with absolutely no delay or stubline in the network.

• Lilles configuration is published but not the ‘Italy client’.

• Both are radial networks like the image here:

Client Network nodes(EMTP equivalent)

SSN nodes # of LC states Simulation time step

Number of RTDS racks required

Lilles L2EP 1 650 18 387 105µs np3

‘Italy client’ 2 740 9 981 70 µs np3

Lilles distribution grid configuration

3 np: Not Possible (even with a large number of RTDS racks)Because no decoupling if possible and RTDS is limited to 72 nodesper nodal solution

60µs (expected)

Page 10: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

10

SSN in More-Electric-Aircraft (Bombardier Global Express)

• GLEX stats:• 25 SSN groups

• 109 switches

• Y size: 48x48

• Time step 39 µs

0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.514

16

18

20

22

24

26

28

30

Time (s)

Vol

tage

(V

DC)

SPS 1us

SPS 50us

SSN 50us

L. Montealegre Lobo, C. Dufour, J. Mahseredjian, “Real-time Simulation of More-Electric Aircraft Power Systems”, Proceedings of the 15th European Conference on Power Electronics and Applications (EPE’13 ECCE Europe), Lille, France, Sept. 3-5, 2013

Page 11: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

11

• Undisclosed client.

• Application: fusion reactor coupled coil drives.

• Complained about very slow simulation in SPS (1 day typ.)

• Stats: 121 SSN groups, 468 switches and 291 nodes (Y matrix 95% sparse)• Also used some Rte-Drive buck converters.

• Results in 33-100X accelerationdepending on size of system.

Super-Large Converter applications (fusion reactor)

Coil current drive similar to the one discussed

Page 12: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

12

• Iterative MOV and Iterative Switches under development• Available in off-line mode since Q1/2014

• Q3/2014 expected date for RT.

• Available in HYPERSIM in RT-mode since 2012.

Iterative methods in SSN (MOV and Switches)

C. Dufour, O. Tremblay, “Iterative Algorithms of Surge Arrester for Real-Time Simulators”, 18th Power Systems Computation Conference (PSCC 2014), August 18-22, 2014, Wroclaw, Poland.

Page 13: OPAL-RT RT14 Conference: State-Space Nodal

The 7th International Conferenceon Real-Time Simulation TechnologiesMontreal | 9-12 June, 2014

13

• New challenges shows the existing and surprising capability of SSN

• New challenges required the coding of new features like iterations.

Conclusion