5 June 2009 © David J Hill The Australian National University Massive Networks 112 October 2009 © David J Hill The Australian National University Adaptive Grids 1
Short Course on Future Trends for Power Systems, The University of Sydney, 12th October, 2009
Adaptive Power Grids:Responding to Generation Diversity
David J HillResearch School of Information Sciences
and EngineeringThe Australian National University
12 October 2009 © David J Hill The Australian National University Adaptive Grids 3
Outline
Future grids
Challenges
New control ideas
Example: Voltage control
Conclusions
5 June 2009 © David J Hill The Australian National University Massive Networks 412 October 2009 © David J Hill The Australian National University Adaptive Grids 4
Australian Transmission Network
12 October 2009 © David J Hill The Australian National University Adaptive Grids 5
Diverse Generation in Australia
• New– Wind – Solar – Bioenergy– Geothermal– Nuclear
• Old– Coal– Hydro
New Power Grids
• Diverse loads
• Diverse generation New
• Diverse storage New
• All– Distributed– Multi-level– Multi-scale– Multi-type– Volatile
5 June 2009 © David J Hill The Australian National University Massive Networks 712 October 2009 © David J Hill The Australian National University Adaptive Grids 7
Ref: J.Fan and S.Borlase, IEEE Power & Energy Magazine, Special Issue on the Next-Generation Grid, Vol.7, No.2, 2009
Big Changes
• Old model – variable load, adjust generation
• New models – variable load and generation
End-to-end control, i.e. generation, demand management, storage
12 October 2009 © David J Hill The Australian National University Adaptive Grids 9
Changes
• The existing grids typically do not have the right structure and capacities for large-scale renewables, e.g. will need wind and solar hubs quickly
• Generation is much more volatile, i.e. now on both sides of the generation = load equation
• Major new need is demand management
• New loads on horizon, e.g. plug-in (hybrid) electric vehicles (PHEV)
MUCH MORE UNCERTAINTY FOR THE GRID
Need ADAPTIVE end-to-end control
Uncertainty for Wind Generation
• Dependence on nature gives unpredictability
• Companies do not want to disclose their data, controls (IP for market)
• Manufacturers can disappear but their turbines keep operating
• Manufacturer models are very detailed, but need simpler models for grid studies
Challenges of Complexity
• Planning vs control
• Decision and control (performance, security)
• Massive amounts of data
• Optimizing (planning, control) on such a scale
• Validation
12 October 2009 © David J Hill The Australian National University Adaptive Grids 12
What is “Smart Grid”?
• Concept emerged in Europe; named in USA Energy Act 2007, Obama stimulus package
• Now a buzzword which captures other ideas: Intelligent Grid, EPRI; iGrid, Australia etc
• But Aus budget just gave A$100 million, US$4.6 billion in USA, so much anticipation
5 June 2009 © David J Hill The Australian National University Massive Networks 1312 October 2009 © David J Hill The Australian National University Adaptive Grids 13
Smart Grid Targets
• Meet environmental targets
• Accommodate greater emphasis on demand management
• Support new loads, e.g. PHEVs
• Support distributed generation and storage
• Maintain a level of availability, performance and security
12 October 2009 © David J Hill The Australian National University Adaptive Grids 14
More Monitoring, Computing and Control
Ref: A.Ipakchi and F.Albuyeh, IEEE Power & Energy Magazine, Special Issue on the Next-Generation Grid, Vol.7, No.2, 2009
“Smart Grid” as Control Engineering
• Large network of sensors
• Massive amounts of data, i.e. measurements, availability etc
• Distributed control operating at many levels, c.f. Internet
Thinking like the Internet
• Things just have to happen in time, e.g. the TV immediate, the toaster within 1 minute, but allow some scale
• A vision of a “plug and play” capability for the whole grid
• All controlled in (seven) layers
• Congestion handled by protocols, AQM, delays
• Major problems by re-routing
What looks useful
• Computer science– Machine learning– Planning and diagnosis, etc
• Automatic control
• Communications
• Mathematical algorithms
All working together have the tools to make major advances.
12 October 2009 © David J Hill The Australian National University Adaptive Grids 18
Outline
Future grids
Challenges
New control ideas
Example: Voltage control
Conclusions
12 October 2009 © David J Hill The Australian National University Adaptive Grids 19
Big Questions
Diverse generation makes planning, analysis and control all harder
• What level of renewables (or any given energy mix) can a given network support?
• How do we plan and control the power grid given all challenges?
12 October 2009 © David J Hill The Australian National University Adaptive Grids 20
Many Challenges
• Protocols for access– cf. Internet “plug and play”
• Affect on system dynamics, collapse– Blackouts due to weak points
• Wide-area control architectures– How to coordinate 1000’s of controls at multiple levels– Lot more uncertainty
• Sensing technology and architectures
5 June 2009 © David J Hill The Australian National University Massive Networks 2112 October 2009 © David J Hill The Australian National University Adaptive Grids 21
Voltage Collapse
5 June 2009 © David J Hill The Australian National University Massive Networks 2212 October 2009 © David J Hill The Australian National University Adaptive Grids 22
Blackout 2003 USA-Canada
South Australia Wind Power Case*
1200MW wind scenario
Wind Generation Scenarios
South Australia Wind Power Case* --- continued
Long term voltage stability limits
* NEMMCO Report: Assessment of Potential Security Risks due to High Levels of Wind Generation in South Australia
Locations of Wind Turbines
0 20 40 60 80
0.20
0.25
0.30
0.35
0.40
Number of Turbines
Crit
ical
Cle
arin
g Ti
me
DFIG at G3Constant Speed at G3DFIG at G1Constant Speed at G1
G1
G2
G3
G4 WW
Ref: Bennett, Hill and Zhang, in prep
Comments and Conjectures
• All stability types affected
• Locations of generation types important
• Structure of network important
• More flexible (adaptive) control must be used
12 October 2009 © David J Hill The Australian National University Adaptive Grids 27
ARC LP Project 2009- 2012
1. Investigate proper, possibly generic, models for diverse generators and their local controls in different power networks and voltage levels;
2. Determine what kinds of stability or security issues could arise due to characteristics of renewable energy resources;
3. Check the limitations of available control mechanisms to guarantee power system quality of supply and stabilities;
4. Otherwise, design proper coordinated control schemes to maximize the stability margin of the power system;
5. Given the improved technology, implemented at some generic level, develop methods to assess what level of renewable generation could be supported at different sites;
6. Investigate whether available control in power system with diverse generation can guarantee levels of security and quality of supply for increasing levels of mandatory targets for certain technologies.
12 October 2009 © David J Hill The Australian National University Adaptive Grids 28
Outline
Future grids
Challenges
New control ideas
Example: Voltage control
Conclusions
12 October 2009 © David J Hill The Australian National University Adaptive Grids 29
Control Challenge
• A multi-level version of distributed adaptive control
• Attends to local and system control needs
• Reconfigurability plus tuning, i.e. can attack problems as they arise in staged response
Call it global control
12 October 2009 © David J Hill The Australian National University Adaptive Grids 30
From Ian Hiskens, Cornell Uni
We already do well but we can do more!• Currently SCADA has real-time data every 2 secs, state estimation,
optimal power flow, security analysis etc – that’s already “smart”
• But this is forty year old concept (following 1965 blackout etc)
• Also its confined to generation-transmission system level
• And tends to treat problems separately, e.g. angle stability, voltage stability
• We now have PMUs which can give data in millisecs
• And major advances in technology especially ICT, power electronics
• With whole ICT repertoire we can do control at all levels for distributed generation, load and storage
• And we can coordinate a lot better, e.g. use refined load control to help system stabilities in a cascading situation
5 June 2009 © David J Hill The Australian National University Massive Networks 3212 October 2009 © David J Hill The Australian National University Adaptive Grids 32
Global Control Framework (Leung, Hill and Zhang, 2009)
12 October 2009 © David J Hill The Australian National University Adaptive Grids 33
Other Ideas
• Our view of control “is autistic”; for massive systems get “cognitive overload”;
• Maybe just viewing the problem as computation reduction is inadequate;
• Will need more than just using structure better;
• In global control used ‘indicators’ and switching, c.f. economic control;
• Computer scientists have ad hoc techniques for ‘planning’ in large systems; we have systematic techniques for simple systems?
Comments by ANU Computer Scientist
• The machine learning area has learned a lot from the control area in the past
• We see adaptive control as a precise way to deal with simple systems
• Machine learning has a lot of tools and tricks, a bit ad hoc, but does deal with complex systems
• Maybe its time to see how machine learning can help control?
• Hewitt:• I've been training
extremely hard, putting in a lot of hours on the court …… (BBC Sports)
• An example of• “Learning by
doing”• Fast responses
needed
Learning-based Control
• Improves its performance based on past experiences (Fu, 1969; Farrell and Baker, 1993)
• Effectively recall and reuse the learned knowledge
• Use stability robustness to handle mismatch
• Can be used to reduce space for optimization
Pattern-based Control
• - used in large systems, e.g. Lissajous recordings of faults power systems
• - not developed in control area
• Dynamic pattern recognition• Switching/tuning control between different patterns
– patterns as local models– stability issues
Ref: Wang and Hill, Deterministic Learning Theory for Identification, Recognition and Control, CRC Press, 2009.
Towards development of a human-like learning and control methodology
Ref: Yusheng Xue, PSCC 2005
Aim: Maintain steady voltages at all buses.
Control devices: Tap changers, capacitors, load shedding
Voltage controlG G
GG
G
GG GG G
30
39
1
2
2537
29
17
26
9
338
16
5
4
18
27
28
3624
35
22
21
20
34
2319
3310
11
13
14
15
8 31
126
32
7
The New England 39-bus Power System
Coordinated Voltage Control
Coordinated Voltage Control (CVC)CVC is a scheme relies upon the simulated performance of a power system, coordinatedscheduling and switching voltage control devices
• sequencing: decide the order of control actions• timing: decide the switching time of each control
action• tuning: decide the values of the adjustable
parameters of each control action
CVC include three aspects of system design:
Coordinated Voltage Control
On-line Multi-Objective CVC System
MCVC System:
• Off-line global search
• On-line flexible control
• On-line learning
irefiti t
vi vvJ −=∑ ∑∑min
cact nJ min=
∑=k
loadload knJ min
Mid-term
On-line Learning
Objective functions:
Power System
Output
Global Search: Get non-dominated solutions
Data Base: 1.faults, 2.order of effective controllers 3.objective values of non-dominated solutions
Short term
Local Search: 1.Get available controllers, 2.Searching neighborhood
Control
Get
Ava
ilabl
e C
ontr
olle
rs
Get from Database
Off-line SearchingOn-line Adaptive Control
Evaluation
Some Possible Faults
Multiple Criteria Decision Making
Learning Mid-term
Short term
Case Study
Generator32 tripped at 15sG G
GG
G
GG GG G
30
39
1
2
2537
29
17
26
9
338
16
5
4
18
27
28
3624
35
22
21
20
34
2319
3310
11
13
14
15
8 31
126
32
7
Case1: Tripping Generator 32
Case1: Tripping Generator36 and Line2-3
Case Study
No. 1 2 3 4 5 6 7 8 9
Ctrl Ltc31 Ltc30 Ltc35 Ltc11 Ltc12 Ltc33 C13 C13 Ltc37
move +1 +1 +1 +1 +1 +1 +0.15 +0.30 +1
No. 10 11 12 13 14 15 16 17 18
Ctrl C7 C7 C8 C8 C4 C4 Ltc38 Ltc36 Ltc34
move +0.15 +0.30 +0.15 +0.30 +0.15 +0.30 +1 +1 +1
No. 19 20 21 22 23 24 25 26 27
Ctrl C15 C15 C3 C3 C18 C18 C16 C16 Ltc39
move +0.15 +0.30 +0.15 +0.30 +0.15 +0.30 +0.15 +0.30 +1
No. 28 29 30 31 32 33 34 35 36
Ctrl C24 C24 C27 C27 C21 C21 C26 C26 C25
move +0.15 +0.30 +0.15 +0.30 +0.15 +0.30 +0.15 +0.30 +0.15
No. 37 38 39 40 41 42 43 44 45
Ctrl C25 C23 C23 C28 C28 C29 C29 C20 C20
move +0.3 +0.15 +0.30 +0.15 +0.30 +0.15 +0.30 +0.15 +0.30
Case1: Tripping Generator 32
Order of effective controllers:
Control preferences:
System performance:
(1) the solution which can recover bus voltages very fast is the most desirable one. Totally 28 controllers, 39 movements of control are used.
(2) a solution uses less control actions is the best one. Totally 26 controllers, 29 movements of control are used.
Case Study
Control Scenario
Time Event
30s Line3-2 tripping
60s G36 tripping
180s Line3-2 and G36 reconnection
540s Line3-2 and G36 tripping together
660s Line3-2 and G36 reconnection
1140s Line3-2 and G36 tripping together
Case2: Tripping Generator 36 and Line 2-3
System performance:
12 October 2009 © David J Hill The Australian National University Adaptive Grids 45
Outline
Future grids
Challenges
New control ideas
Example: Voltage control
Conclusions
Future Work
• Combine– Computer science for learning, planning,
diagnosis, visualization, data structures etc– Networks for structure – Control for dynamics
to give algorithms which scale
• Link to other levels: power electronics, economics