and some Retrospective Thoughts on Complex Systems Engineering

1
Intelligent Agents for Control of Distributed
Energy Resources

• (and some Retrospective Thoughts on Complex
  Systems Engineering)

Presented by Geoff James Friday 12 August 2005
2
The BIG Issues for DG in Australia

• What happens when Mr DG (times 5000) starts his
  unit in the morning and becomes a generator for
  the day?
• Network control and generation despatch is
  currently carried out top down gt 30 MW
• How can local generation be used to the advantage
  of the consumer and the network?
• Customer choice and network needs must be
  balanced
• Outcomes lower cost, improved reliability,
  reduced emissions
• How can demand side management options be
  deployed and controlled to reduce peak demand and
  prices and minimise GHG emissions?
• SCADA? Try autonomous smart distributed agents!

3
Distributed Energy Mgmt and Control
Project Goals

• A realistic solution to large-scale deployment of
  DE resources in the distribution network
• To impact the Australian network in 3 8 years
  time
• Adaptive, intelligent, distributed agents for
  various applications
• Local end-use optimisation
• Aggregation for network benefits
• A communications infrastructure
• A new set of features in the Australian NEM

4
Five Key Messages about DEMC

• One ICT infrastructure can have many applications
  that benefit the energy network
• Distributed agent technology gives consumer
  choice with cost effectiveness and scalability
• Key technology coordination of consumer loads
  and distributed generation
• Key technology scalable aggregation of
  distributed energy on a visionary scale
• This is the CSS bit
• We can begin now demonstration and industry
  trials as avenues to early deployments

5
But how did we get here?

• Ageless Aerospace Vehicles (2001-2005)
• Fantastic collaboration involving sensors, signal
  processing, distributed intelligence,
  communications, machine learning, and biology
• GREMLab (2002-2005)
• Tried to do engineering design for multi-agent CS
• Vision for self-assembly from macro to nano
  scales
• Hosted DAMAN and EDCCS projects for CSS
• Smart Spaces ESA (2002-2004)
• A collaborative vehicle and GREMLab participant
  that aimed to create a variety of self-organising
  smart spaces
• Attracted attention of Energy Transformed
  Flagship before being managed to death by
  oversight committee

6
Simulated tiled surface for an AAV
7
Concept demonstrator for NASA
8
Some absent friends
9
Inchwork response to a diagnosis
10
Towards complex systems engineering

• Weve done self-organising diagnosis
• Simulation and hardware
• Response is on the way
• Inchworm robot
• Critical damage reporting
• Prognosis is the big challenge now
• Existing project with CIP / CMIT / Boeing
• Fingers crossed for the Sentient Structures ESA
• Will carry forward our ideas and collaboration

11
GREMLAB
Leader Geoff Poulton
12
Herding cats
13
Idealised self-assembling mesoblock
/ - / 0
STATE MACHINE
/ - / 0
/ - / 0
/ - / 0
Sense
Change
14
Cool simulation of a sea of mesoblocks
15
An emergent behaviour
16
Another response mechanism for AAV
17
Two-layer hierarchy in nature
Design Goal Class of Proteins
Folding (emergent)
Enzymes
Ribosomes
mRNA messages
Amino acids, tRNA, etc.
Real evolution
emergent behaviour
Biochemical building-blocks
18
Two-layer hierarchy for self-assembly
Desired meso- or nano-structures - sensors,
actuators etc.
Construction
Constructor entities eg. Enzymes
Self-assembly
emergent behaviour
Meso- or nano- agents
19
Towards complex systems engineering

• Two-level hierarchy as framework methodology
• Avoids designing out the complexity
• Success in simulated environments
• DAMAN project for CSS
• Good publication record
• Yet to make a bridge with reality
• EDCCS project for CSS
• Fantastic model for collaboration
• Participating projects contribute resources
• Best times were when we had no budget!

20
SmartLands (CLW, CLI, CTIP, CMIT)
21
The electricity network
22
Towards complex systems engineering

• Smart Spaces ESA had an evolving purpose
• Sharpened goal and reduced collaboration
• Distributed Energy MC grew out of it
• Among other things
• Demo focus means less room for complexity
• Top-down coordination of loads and generators
• But using optimisation tools from GREMLab
• Large-scale aggregation is essentially complex
• Collaboration with VUA is warming up nicely

23
1. One Infrastructure, Many Applications
24
Example Application DSM
NEM
Aggregated demand response can also be used to
defer capital expenditure
This quantity is tradable
Retailer
Aggregated response gt 30 MW
Rewards
SME
SME
PDA agent
Fleck agent
SME
SME
PC agent
SME
Mote agent
PDA agent
1000s of these
25
2. The Agent Mindset
Agents run on local devices and measure, make
decisions, and act in the real world

• Local control is good for
• Robustness
• Scalability
• Consumer acceptance
• Contrast with SCADA
• Prohibitively expensive to extend to consumer
  level
• Top-down control is not scalable and sometimes
  not desirable
• Opportunity agents can be a last-mile solution

26
A Multi-Agent System (Domestic Case)
27
Distributed Software Agents

• Natural model for distributed energy management
• Agents run on local hardware and represent
  consumers interests keeping data local as much
  as possible
• Agents are intelligent and can model the
  resources theyre responsible for learning as
  they go
• Agents can optimise locally and interact to
  achieve system benefits
• They can provide desirable properties
• Cheap, no single point of failure, safe fail
• Easy to add and remove agents and services
• No additional infrastructure (agents chat on the
  internet)
• Provision for intuitive consumer user interface

28
3. Coordination of Loads and Generators

• Coordinating a set of loads and generators to
  achieve both local and system goals
• Expressing as an optimisation problem
• Goals vary with application typically local cost
  effectiveness and participation in an aggregated
  system response
• Local modelling of capabilities and constraints
  of loads and generators
• Machine learning to
• Improve models based on measured performance
• Predict generating capacity for wind and
  photovoltaics
• Adapt price sensitivity for agent goal setting

29
4. Aggregation on a Visionary Scale

• Scalable and timely aggregation of distributed
  capacity across 104, 105, 106, consumers
• System response gt 30 MW in order of minutes with
  communication delays in order of seconds
• BREAKTHROUGH WE AIM AT demonstrating emergent
  behaviour to a desired outcome
• Complex systems techniques decentralised
  clustering, dynamic hierarchies, scale-free
  networks,

30
Two Levels of Aggregation
(May use similar or different methods)
Customer agent
Customer agent
Customer agent
Aggregation WITHIN Customers
Customer agent
Customer agent
Customer agent
Customer agent
Aggregation BETWEEN Customers
Grid / Market Interaction
31
5. Begin Now

• Writing an agent-based software platform
• Collaboration with Infotility (Boulder / San
  Francisco)
• Alpha release under test from April
• Creating a uniform agent environment and a
  reliable platform across a diverse set of devices
• Developing multi-agent coordination algorithms
• Focus coordination in 04/05 and scalability in
  05/06
• Demonstrating in hardware at Newcastle
• Cooperating loads and generators by June
• Embarking on a trial with an industry partner
• We wont do front-end deployment ourselves
• Looking for commercial partners in 05/06

32
Recap Five Key Messages for Today

• One ICT infrastructure can have many applications
  that benefit the energy network
• Distributed agent technology gives consumer
  choice with cost effectiveness and scalability
• Key technology coordination of consumer loads
  and distributed generation
• Key technology scalable aggregation of
  distributed energy on a visionary scale
• We can begin now demonstration and industry
  trials as avenues to early deployments

33
Energy Transformed Flagship
The Energy Transformed Flagship is aimed halving
greenhouse gas emissions and doubling the
efficiency of the nations new energy generation,
supply and end use, and to position Australia for
a future hydrogen economy. Theme 4
Distributed Energy this technology Theme
directly targets a step change in energy
efficiency and reduction in GHG emissions by
accelerating the uptake of distributed energy
systems that provide local power, heat and
cooling to industrial and commercial sites. The
Centre for Distributed Energy and Power CenDEP
is an alliance of organisations, joining with
CSIRO to help put distributed energy on the map
in Australia.
34
Contacts
For more information, see www.csiro.au or
contact
Thank you