Transcript 2008 Multi-Party Multi-Rate Real

____________________
..
____________________
Computational Challenges in the
Simulation of Modern Electrical Power
Systems
Roy Crosbie
California State University, Chico
CICSyN 2010
Liverpool
28 July 2010
Acknowledgements
____________________
..
____________________
The research described in this presentation is based on the work of a
research team at the McLeod Institute of Simulation Sciences at
California State University, Chico, USA.
Team Members
Richard Bednar, Professor Emeritus
Roy Crosbie, Professor Emeritus and Institute Director
Nari Hingorani, Visiting Research Professor
Dale Word, Associate Professor, Electrical & Computer Engineering
John Zenor, Professor Emeritus
Financial support by the US Office of Naval Research is gratefully
acknowledged
CICSyN, Liverpool, 28 July 2010
2
Conference Themes
____________________
..
____________________
• Computational Intelligence > System Modeling & Simulation
• Communication Systems> Real-time Simulation & Control
• Networks> Distributed Power System Control
CICSyN, Liverpool, 28 July 2010
3
Traditional Approach to
Simulation of Power Systems
____________________
..
____________________
A. Steady State Load Flow Studies
B. Dynamic Simulation of Transient Behavior
–
Seminal Analysis by Dommel
–
Nodal Circuit Analysis + Implicit Trapezoidal Integration
–
Non-linearities require iterative procedures
–
Electromagnetic Transients Program (EMTP)
–
50 microsecond maximum integration steps
CICSyN, Liverpool, 28 July 2010
4
Modern Power Systems
____________________
..
____________________
• Much greater use of power converters (ac to dc & dc to ac)
• High-voltage d.c. transmission
• Renewable energy generation (solar, wind etc.)
• Independent power systems for ships etc.
CICSyN, Liverpool, 28 July 2010
5
6-pulse Back-to-Back
Converter System
____________________
..
____________________
23 ODEs, 12 switches, 2 PWM controllers with sine/triangle comparison PI control plus power calculations
6
CICSyN, Liverpool, 28 July 2010
Distributed Energy System
(Adel Ghandakly)
____________________
..
____________________
WTPEC
Wind Turbine Unit
Integration System
Monitoring & Control
Photo Voltaic Unit
Booster Rectifier
Unit
Inverter
Rectifier
Unit
PowerGrid
DSPEC
PVPEC
Battery Storage
Unit
BSPEC
Load
Power System for Electric Ship
____________________
..
____________________
Questions?
8
CICSyN, Liverpool, 28 July 2010
High-Speed Real-Time Simulation
____________________
..
____________________
Why Real-Time?
Simulation running at true speed allows connection to real hardware
Hardware can be tested in absence of real system
Plant operators, pilots etc. can be trained under realistic conditions
Why High-Speed?
For many systems frame times can be tens of milliseconds or longer
Systems with fast dynamics or rapid switching need shorter frames
Power electronic systems often need microsecond frame times
CICSyN, Liverpool, 28 July 2010
9
Choice of Technology
____________________
..
____________________
• Many real-time simulations use a real-time version of
Linux running on a high-performance PC
• Operating system jitter (of the order of 10 μS) limits
minimum frame time
• Higher-performance is possible from systems with
Pentium or PowerPC based processors but only with
custom designs
• Initial solution: arrays of digital signal processors
inserted in PCI bus of conventional PC with Windows
OS running on host – off-the-shelf components; no
problems with OS jitter
CICSyN, Liverpool, 28 July 2010
10
TS201 Board Architecture
____________________
..
____________________
CICSyN, Liverpool, 28 July 2010
11
DSP Issues
____________________
..
____________________
• Scheduling Processor Tasks
– Equalizing processor execution times
– Minimise inter-processor data transfers
• Internal Data Transfer
– Common memory vs. link ports
• External Data Transfer
– Digital and analog outputs and inputs
• Code efficiency
– Hand-coding vs compiler efficiency
– Identify efficient HLL code sequences
CICSyN, Liverpool, 28 July 2010
12
Software Issues
____________________
..
____________________
• Choice of numerical integration algorithm
–
–
–
–
Euler vs Runge-Kutta vs implicit trapezoidal vs state-transition methods
Analyse and monitor accuracy and stability of numerical integration
Combine differential equations with integration algorithm before coding
Minimize total mathematical operations
• Hand coding vs optimizing compiler
– Hand coding may be needed if compiler can’t exploit processor architecture
– Use HLL constructs that produce more efficient code
CICSyN, Liverpool, 28 July 2010
13
Real-Time Simulation with FPGA
____________________
..
____________________
• FPGA offers competitive alternative to DSP; shorter frame times
• Can be programmed using Simulink blockset, VHDL, M-code
• Full 6-pulse model ported to larger FPGA
• Soft processor used for slow Ethernet interface
• Direct programmed high-speed Ethernet interface
CICSyN, Liverpool, 28 July 2010
14
ML506 Board
____________________
..
____________________
15
CICSyN, Liverpool, 28 July 2010
FPGA Performance vs DSP
____________________
..
____________________
Model/Platform
Minimum Frame
Time
Processor
Clock Rate
6 Pulse BTB - Hammerhead
Board, 23 ODEs
16 µs
AD 21160 DSP
80Mhz
6 Pulse BTB - TigerSharc
Board, 23 ODEs
3.85µs
AD TS101 DSP
250Mhz
6 Pulse BTB - TigerSharc
Board, 23 ODEs
2.02µs
AD TS201 DSP
500Mhz
12 Pulse BTB - TigerSharc
Board, 39 ODEs
4.5µs
AD TS201 DSP
500Mhz
6 Pulse BTB - Xilinx ML506
Board, Virtex 5, 23 ODEs
450nS
Virtex 5 FPGA
100Mhz
CICSyN, Liverpool, 28 July 2010
16
FPGA Based Performance vs DSP
____________________
..
____________________
Step Size
2.02us
Step Time
Begins
Main
Communicator
Loop Begins
Communicator
Ends
Interrupt Handler
Communications
0.398
1.622 Main Communicator Loop
Start Signals Sent by Communicator
End Signals Sent to Communicator
Main
Controller
Loop Begins
Controller
1.630 Main Controller Loop
0.118
0.239
1.869 1.990
Main
Converter
Right Loop
Begins
Converter Right
End Signal
Received by
Communicator
1.506 Main Converter Right Loop
0.130
0.251
1.757
1.878
Main
Converter
Left Loop
Begins
Converter Left
1.406 Main Converter Left Loop
0.142
FPGA
Setup
Time
TBD
0.263
1.669
1.790
.230 Converter
Left Loop
0.121 is used to send and receive handshaking variables
between that processor and the communicator.
This is the delay between when a new simulation frame begins
and when the processor is sent handshaking variables.
CICSyN, Liverpool, 28 July 2010
17
The Need for Multi-Rate
Real-Time Simulation
____________________
..
____________________
• CSU, Chico developed HSRT simulations with frame rates up to
2 MHz (500 nS frame times)
• These frame rates are needed for power electronic components
but not for slower system components such as motors,
mechanical components, thermal effects etc.
• Multi-rate real-time simulations simulate different subsystems at
different frame-rates on different simulation platforms.
• The slower components are simulated in real-time using a
commercial RTOS, often with Simulink support, for faster,
cheaper model development.
• Multi-rate also improves performance of non real-time
simulations.
• Multi-rate raises questions of stability and accuracy.
CICSyN, Liverpool, 28 July 2010
18
Multi-Rate Example:
Unmanned Underwater Vehicle
____________________
VTB Battery
Model
Controller/Converter Model
(CSU Chico)
..
____________________
UUV Physical Model
(Glasgow)
VTB Synchronous,
Permanent Magnet Motor
Model
Vehicle Control Inputs
Controller
Converter
Low Rate
High Rate
Medium Rate
Low rate
VTB Multi-Rate Solver (USC)
CICSyN, Liverpool, 28 July 2010
19
Multi-Rate Results
____________________
..
____________________
• Multi-Rate Configuration
–
–
–
–
–
Converter, Switch Controller
Feedback Controller
Motor/Propeller
Battery, Ship
Graphics
2 µsec
800 µsec
50-100 µsec
.1 sec
.1 sec
• Multi-Rate Performance on 2.16 GHz Mac Running Windows XP
– All components at 2 µsec:
– Multi-rate, Motor/Propeller 50 µsec
– Multi-rate, Motor/Propeller 100µsec
CICSyN, Liverpool, 28 July 2010
.001x real time
1.2x real-time
2.0x real-time
20
UUV Effects of Multirate
Ship at .1sec vs .001 sec (Identical Plots)
____________________
..
____________________
CICSyN, Liverpool, 28 July 2010
21
UUV VTB 3D Model Output
____________________
..
____________________
CICSyN, Liverpool, 28 July 2010
22
Power System Control
____________________
..
____________________
Hierarchical control combines local controllers at stations
and system wide control at control centers
As more and more raw data is being sent from stations to
control centers communication channels are overloaded
On-line real-time simulators at stations can reduce data
volume through processing of raw data
This can facilitate more rapid detection of critical behavior
and more rapid action to minimize its effect
CICSyN, Liverpool, 28 July 2010
23
Power System Communication
____________________
..
____________________
Regional
Control
Center
Local Station
Local Station
CICSyN, Liverpool, 28 July 2010
Local Station
24
Power System ControlNetwork
____________________
..
____________________
25
CICSyN, Liverpool, 28 July 2010
Acknowledgement
____________________
..
____________________
The following material is based on:
Power System Stability: New Opportunities for Control
By Anjan Bose
Chapter in
Stability and Control of Dynamical Systems and
Applications,
Derong Liu and Panos J. Antsaklis eds
http://gridstat.eecs.wsu.edu/Bose-GridCommsOverview-Chapter.pdf
CICSyN, Liverpool, 28 July 2010
26
Power System Networks:
Stability
____________________
..
____________________
• Power system networks in North America & Europe are
the world’s’ largest man-made interconnected networks
• All the rotating generators in one network rotate
synchronously
• Any large disturbance (e.g. equipment short circuit) can
make the power system unstable.
CICSyN, Liverpool, 28 July 2010
27
Power System Networks:
Control
____________________
..
____________________
• Control uses a combination of isolating switches,
continuous control of voltage and power, and powerelectronic switch-based control.
• These controls are all local (equipment/control in same
substation)
• Regional and system-wide control is mainly limited to
adjusting generation levels to adjust to slowly changing
power loads
CICSyN, Liverpool, 28 July 2010
28
Power System Networks:
Communication
____________________
..
____________________
• System-wide control needs communication between
contol centre and substations (microwave, telephone
lines, increasing use of optical fibre)
• Lower costs, increasing bandwidth, GPS time
synchronization, improved power electronics offer
opportunities for fast distributed controls
• Increasing amount of data gathered at substations at
mS rates is too voluminous for real-time transmission
and control. OK for later study.
CICSyN, Liverpool, 28 July 2010
29
Power System Networks:
New Technologies
____________________
..
____________________
• Faster, cheaper computers
– Embedded in equipment
– Provide intelligence in the control loops
• Low-cost broadband communications
– Greater volume of real-time data
– Possibilities for decentralizing control
• Better power electronic controls
– FACTS – Flexible AC Transmission Systems
CICSyN, Liverpool, 28 July 2010
30
Future Research
____________________
..
____________________
The Goal
Automatic global control for system-wide transient stability.
The Need
Computation to analyze the situation and compute necessary control
actions, has to match the time-frame of current protection schemes
(milliseconds).
“Whether this is possible with today’s technology is unknown. However,
the goal is to determine what kind of communication-computation
structure is needed to make this feasible.” (Bose)
CICSyN, Liverpool, 28 July 2010
31
Conclusion
____________________
..
____________________
Modern electric power systems provide research
opportunities that synthesize the conference
themes: computational intelligence,
communication systems and networks
CICSyN, Liverpool, 28 July 2010
32