can complexity be captured with agent- based modeling … · can complexity be captured with agent-...

Copyright 2004 Argonne National Laboratory

Can Complexity Be Captured with AgentCan Complexity Be Captured with Agent--Based Modeling and Simulation?Based Modeling and Simulation?

Michael [email protected]

www.cas.anl.gov

2


The “Name Game”The The ““Name GameName Game””

• ABMS is known by many names:– ABM: “Agent-based modeling” or “anti-ballistic missile?”– ABS: Agent-based simulation or “anti-lock breaks?”– IBM: Individual-based modeling or “International Business

Machines Corporation?”• ABM, ABS, and IBM are all widely-used acronyms, but

“ABMS” will be used throughout the lectures to avoid confusion with the above mentioned terms

3


What is ABMS?What is ABMS?What is ABMS?

• ABMS seeks to create electronic laboratories (“e-laboratories”) that allow experimentation with simulated complex systems:– ABMS uses sets of agents and frameworks for simulating the agent’s

decisions and interactions – ABMS can show how a system could evolve through time in a way that

is difficult to predict from knowledge of the behaviors of the individual agents alone

• ABMS focuses on individual behavior with the agent rules are often based on theories of the individual such as Rational Individual Behavior, Bounded Rationality or Satisficing

• Based on these simple types of rules, ABMS can be used to study how patterns emerge

• ABMS may reveal behavioral patterns at a macro (system) level that are not obvious from an examination of the underlying agentrules alone – these patterns are called “emergent behavior”

4


ABMS is Often Used to ModelComplex Adaptive Systems

ABMS is Often Used to ModelABMS is Often Used to ModelComplex Adaptive SystemsComplex Adaptive Systems

• A Complex Adaptive System (CAS) is made up of agents that interact and reproduce while adapting to a changing environment

• Researchers such as John Holland are trying to isolate fundamental causes of adaptation and emergence of system-wide properties – in any CAS

• John Holland has identified the following properties and mechanisms that are common to all CAS:– Nonlinearity– Diversity– Aggregation– Flows– Tagging– Internal models– Building blocks

• ABMS incorporates some of the properties and mechanisms of CAS

5


There Are Many Examples of SystemsComprised of Interacting Individuals

There Are Many Examples of SystemsThere Are Many Examples of SystemsComprised of Interacting IndividualsComprised of Interacting Individuals

• Economic markets:– Producers– Distributors– Consumers

• Human immune system:– Antibodies– Bacteria– Viruses

• Social Systems:– People– Factions– Countries

• Ecosystems:– Species– Individuals– Hives– Flocks

6


Where Did ABMS Come From?Where Did ABMS Come From?Where Did ABMS Come From?

7


Repast is the One of the Most PopularAmong a Range of Available ABMS Toolkits

Repast is the One of the Most PopularRepast is the One of the Most PopularAmong a Range of Available ABMS ToolkitsAmong a Range of Available ABMS Toolkits

IMT flock.cbl.umces.edu/imt

Ease of Model Development

Mod

elin

g Po

wer

Easy Hard

Low

Hig

h

StarLogo www.media.mit.edu/starlogo

Participatory Simulation

Spreadsheets

Structured Languages (C, Pascal, etc.)

RePast repast.sourceforge.netAscape www.brook.edu/es/dynamics/models/ascape

Swarm www.swarm.org

Object Oriented Languages (Java, C++, etc.)

Mathematics Packages (Mathematica®, etc.)

Selected ExampleABMS Toolkits

NetLogo ccl.northwestern.edu/netlogo/

DIAS www.dis.anl.gov/DIAS/

8


What Useful InformationCan ABMS Provide?

What Useful InformationWhat Useful InformationCan ABMS Provide?Can ABMS Provide?

• ABMS can help to provide insight into and predictions of agent behaviors

• ABMS can help to anticipate system dynamics, structures, and possible evolutionary paths including suggesting answers to a variety of questions including the following:– What agent rules influence emergent behavior and how do

they do so?– Will a some types of agents tend to dominate?– Will changes come quickly or slowly?– Will some systems always be in a state of turbulence?

• ABMS can be used to help identify disequilibrium situations and their causes

• ABMS can be used to help identify sources of uncertainty in the underlying system

9


ABMS is Complementaryto Traditional TechniquesABMS is ComplementaryABMS is Complementaryto Traditional Techniquesto Traditional Techniques

• Analytics: Analytical modeling seeks to develop rigorous, provable statements about systems

• Statistical Methods: Statistical modeling specifies how outputs depend on inputs – systems are represented as a “black boxes”

• Optimization: Optimization modeling seeks to find optimal solutions relative to well-defined objectives and subject to specific constraints

• Discrete Event Simulation: Traditional discrete event simulation modeling represents the inner workings of dynamic processes and moves those representations forward through time at a system level

Out

put

Input-10 -5 0 5 10

xi1

-10-5

05

10

xk1

-4000

-3000

-2000

-1000

0

QP

-4000

-3000

-2000

-1000

0

QP

Wait for firstavailable server

CustomerCustomer Customer

Customerin Transit

Server 1Server 2Customer

Being Served

Customer

Customerin Transit

Customer

CustomerCustomer

Customer

Queue

10


Analytical Modeling Seeks to Develop Provable Statements About Systems

Analytical Modeling Seeks to Develop Provable Analytical Modeling Seeks to Develop Provable Statements About SystemsStatements About Systems

• An example is solving a well-posed problem in classical mechanics

• Difficulties:– Analytical models usually focus on global descriptions– Analytical models of complex systems can be extremely

unwieldy– Excessively “heroic” assumptions are required to create

analytically solvable models of many systems– Many systems cannot be analytically solved at all

11


Statistical Methods SeekRelationships Between Inputs and Outputs

Statistical Methods SeekStatistical Methods SeekRelationships Between Inputs and OutputsRelationships Between Inputs and Outputs

• Output = f(Input1, Input2, Input3, …., Inputn) where f is a statistically derived relationship

• Difficulties

SystemInput Output

– The derived relationships can be brittle– Not sensitive to many assumptions or

amenable to “what-if” scenarios– Insight into underlying causes is often

limited

Output

Input

12


Optimization Modeling Seeks “Best Values”Optimization Modeling Seeks Optimization Modeling Seeks ““Best ValuesBest Values””

• Max x1 and x2: (1 - e(x1+x2)) (Fitness)Subject to : 3 x1 + 5 x2 < 100 (Size constraint)

• Difficulties -10 -5 0 5 10

xi1

-10-5

05

10

xk1

-4000

-3000

-2000

-1000

0

QP

-4000

-3000

-2000

-1000

0

QP

– Optimization models usually focus on global descriptions

– Even for well defined problems, finding optimal solutions can be extremely difficult

– Brittle formulations and solution points can result

13


Wait for firstavailable server

CustomerCustomer Customer

Customerin Transit

Server 1Server 2Customer

Being Served

Customer

Customerin Transit

CustomerBeing Served

CustomerCustomer

Customer

Customer

Queue

1. Customer Arrives to System,Customer Enters Queue (ifservers busy)

2. Customer Begins Transitto Server

3. Customer ArrivesServer,Customer BeginsService,Server BecomesBusy

4. CustomerCompletes Serviceand DepartsSystem,Server BecomesFree

The Queueing Simulation Has Four Types of Eventsand Two Activities

Discrete Event Simulation Modeling Represents the Detailed Steps in a Process As It Unfolds Over TimeDiscrete Event Simulation Modeling Represents the Discrete Event Simulation Modeling Represents the Detailed Steps in a Process As It Unfolds Over TimeDetailed Steps in a Process As It Unfolds Over Time

• Difficulties– The emphasis is on fixed processes instead of adaptive

actors– Defining the process representation can be difficult since

there is no a clear delineation between too little and too much detail

14


ABMS is Being Applied to a Wide Range of CASABMS is Being Applied to a Wide Range of CASABMS is Being Applied to a Wide Range of CAS

• Social systems:– Political systems– Small groups– Businesses

• Markets:– Financial markets– Energy markets

• Industrial supply chains• Infrastructure systems• Ecosystems• Immune systems• Electrical power markets

are an example…

Consumers

Transmission Link

Transmission Node Generators

ISO/RTO/ITP DISPATCH

FUNCTIONS

TRAN

SMIS

SION

and

DIST

RIBU

TION

COMPA

NIES

DistributionCompanies

TransmissionCompanies

DistributionService Territory

BUSINESS LAYERS

PHYSICAL LAYER

REGULATORY LAYER

Consumer Contracts and Tariffs

Demand Agents

Generation Companies Generators

Generator Ownership

Bilateral Contracts

GENE

RATI

ON C

OMPA

NIES

and

DEM

AND

AGEN

TS

Consumers

POOL

MARKETS

ENERGY MARKET ANCILLARY SERVICES MARKET

ISO/RTO/ITP MARKET

OPERATION FUNCTIONS

Market Information System

REGULATOR

Consumers

Transmission Link



FUNCTIONS

Consumers

Transmission Link



FUNCTIONS

TRAN

SMIS

SION

and

DIST

RIBU

TION

COMPA

NIES




TRAN

SMIS

SION

and

DIST

RIBU

TION

COMPA

NIES




BUSINESSLAYERS

PHYSICAL LAYER

REGULATORY LAYER


Demand Companies


Generator Ownership

Bilateral Contracts

GENE

RATI

ON C

OMPA

NIES

and

DEM

AND

COM

PANI

ESConsumers

POOL

MARKETS


ISO/RTO/ITP MARKET

OPERATION FUNCTIONS


Consumers

POOL

MARKETS


ISO/RTO/ITP MARKET

OPERATION FUNCTIONS

ISO/RTO/ITP MARKET

OPERATION FUNCTIONS

Market Information SystemMarket Information System

REGULATORREGULATORREGULATORSpecial Event

Generator

Special Event

Generator

Special Event

Generator

15


Electric Utility Systems Are EvolvingElectric Utility Systems Are EvolvingElectric Utility Systems Are Evolving

• Until recently, most electric power systems were managed by regulated, vertically integrated monopolies

• Several systems, such as those in California and the UK, have implemented open electricity markets that seek to:– Promote competition among suppliers – Provide consumers with a choice of services

• The results have been, at best, mixed• Many places throughout the nation are planning such changes

despite the initial outcomes found in places such California• In the old systems, decision-making was centralized within the

managing monopolies (constrained system)• However, in deregulated systems decision-making is distributed

among many competing organizations (agents)

16


The Electricity Market CAS Model (EMCAS) Applies ABMS to Model Decentralized Electricity Markets

The Electricity Market CAS Model (EMCAS) Applies ABMS to The Electricity Market CAS Model (EMCAS) Applies ABMS to Model Decentralized Electricity MarketsModel Decentralized Electricity Markets

• EMCAS is an agent-based electricity market model

• EMCAS agents take on the roles of individual market participants (generators, distributors, transmission system operators, demand aggregators, customers, regulators)

• The agents operate in multiple layers within nested time scales (hourly, daily, weekly, monthly, yearly, multi-year)

Consumers

Transmission Link


ISO/RTO/ITP DISPATCH FUNCTIONS

TRAN

SMIS

SION

and

DIST

RIBU

TION

COMPA

NIES




BUSINESS LAYERS

PHYSICAL LAYER

REGULATORY LAYER


Demand Agents


Generator Ownership

Bilateral Contracts

GENE

RATI

ON C

OMPA

NIES

and

DEM

AND

AGEN

TS

Consumers

POOL

MARKETS


ISO/RTO/ITP MARKET

OPERATION FUNCTIONS


REGULATOR

Consumers

Transmission Link



Consumers

Transmission Link



TRAN

SMIS

SION

and

DIST

RIBU

TION

COMPA

NIES




TRAN

SMIS

SION

and

DIST

RIBU

TION

COMPA

NIES




BUSINESSLAYERS

PHYSICAL LAYER

REGULATORY LAYER


Demand Companies


Generator Ownership

Bilateral Contracts

GENE

RATI

ON C

OMPA

NIES

and

DEM

AND

COM

PANI

ESConsumers

POOL

MARKETS


ISO/RTO/ITP MARKET

OPERATION FUNCTIONS


Consumers

POOL

MARKETS


ISO/RTO/ITP MARKET

OPERATION FUNCTIONS

ISO/RTO/ITP MARKET

OPERATION FUNCTIONS



Generator

Special Event

Generator

Special Event

Generator

17


EMCAS Is Designed to Explore System PossibilitiesEMCAS Is Designed to Explore System PossibilitiesEMCAS Is Designed to Explore System Possibilities

• “Prediction” is not a goal• ECMAS is intended to provide ranges of possibilities rather

than “point answers:”– The ranges of possibilities are created through multiple

simulations– The ranges are intended to discover potential weaknesses in

electricity markets rather than say whether or not a given agent (company) will actually exploit a given weakness

• These results can be used by decision makers to form better market policies (market rules) and make better market decisions

• The focus is on supporting decisions by exploring potential consequences of different market conditions (initial and changing)

18


EMCAS Operates atSix Time Scales or Decision Levels

EMCAS Operates atEMCAS Operates atSix Time Scales or Decision LevelsSix Time Scales or Decision Levels

19


Live Simulations Were Used toPrototype EMCAS

Live Simulations Were Used toLive Simulations Were Used toPrototype EMCASPrototype EMCAS

• To better understand the requirements of decentralized electricity market modeling, a live electricity market simulation was created

• The live simulation that was developed used individuals to play the role of generation companies:– Each generation company in the market simulation game had three

generators– Players submitted bids electronically based on publicly posted:

Prices

Demands

Supplies

Weather• One additional person played the role of the system operator

20


The Live Simulation includedExogenous Variability

The Live Simulation includedThe Live Simulation includedExogenous VariabilityExogenous Variability

• The system operator collected the players’ bids on a periodic basis and used to them to simulate the operation of an electricity spot market:– The simulation calculated market prices and player profits

based on internally derived demands, supplies, and weather– The actual simulation demands, supply, and weather differed

from the publicly posted projections by small random amounts– Generating units also suffered from unannounced random

outages• Several versions of the live simulation where run• Results from the live simulation contributed to the design

of EMCAS

21


Several Types of Agents Illustrate the ABMS Approach Resulting from the Live Simulation

Several Types of Agents Illustrate the ABMS Approach Several Types of Agents Illustrate the ABMS Approach Resulting from the Live Simulation Resulting from the Live Simulation

• Generation company agents illustrate ECMAS’approach to modeling competitive decision-making

• Independent System Operator/Regional Transmission Organization (ISO/RTO) agents illustrate ECMAS’approach to modeling coordinated behavior

Consumers

Transmission Link



TRAN

SMIS

SION

and

DIST

RIBU

TION

COMPA

NIES




BUSINESS LAYERS

PHYSICAL LAYER

REGULATORY LAYER


Demand Agents


Generator Ownership

Bilateral Contracts

GENE

RATI

ON C

OMPA

NIES

and

DEM

AND

AGEN

TS

Consumers

POOL

MARKETS


ISO/RTO/ITP MARKET

OPERATION FUNCTIONS


REGULATOR

Consumers

Transmission Link



Consumers

Transmission Link



TRAN

SMIS

SION

and

DIST

RIBU

TION

COMPA

NIES




TRAN

SMIS

SION

and

DIST

RIBU

TION

COMPA

NIES




BUSINESSLAYERS

PHYSICAL LAYER

REGULATORY LAYER


Demand Companies


Generator Ownership

Bilateral Contracts

GENE

RATI

ON C

OMPA

NIES

and

DEM

AND

COM

PANI

ESConsumers

POOL

MARKETS


ISO/RTO/ITP MARKET

OPERATION FUNCTIONS


Consumers

POOL

MARKETS


ISO/RTO/ITP MARKET

OPERATION FUNCTIONS

ISO/RTO/ITP MARKET

OPERATION FUNCTIONS



Generator

Special Event

Generator

Special Event

Generator

22


Generation Company Agents Facea Difficult Situation

Generation Company Agents FaceGeneration Company Agents Facea Difficult Situationa Difficult Situation

• Generation company agents sell generation into each of several markets, with different rules in each market:– There is a bilateral contract market that is privately negotiated

between individual buyers and sellers– There is a “spot” energy futures market that is centrally cleared– There are four backup generation options markets that are each

centrally cleared• The decision-making process for generation company agents is

difficult:– The commodity they produce (electric power) cannot typically be

stored– The power “transportation” system (the electric grid) follows well

understood, but highly complicated, rules– Generation company agents have limited knowledge about the other

players in the market

23


The Power “Transportation” System FollowsHighly Complicated Rules

The Power The Power ““TransportationTransportation”” System FollowsSystem FollowsHighly Complicated RulesHighly Complicated Rules

GeneratorCustomer 2

Customer 1

Customer 3

3

1

1

0.5

0.5

1

233

•All power in tens of MW•Notional example

$10/MWh

$25/MWh

$15/MWh

24


Generation Company Agent Decisionsare Based On Several Factors (1 of 2)

Generation Company Agent DecisionsGeneration Company Agent Decisionsare Based On Several Factors (1 of 2)are Based On Several Factors (1 of 2)

• The success of generation company agent decisions are not guaranteed

• Agents “weigh” the relative rewards of success against the costs and risks of failure

• The anticipated success or failure rate is based on experience:– Each generation company agent keeps an ongoing private

record of historical events (i.e., private memory) including a history of decisions made in the past and these results of those decisions under various supply and demand conditions

– Information such as system outages, loads, location-based market prices are posted by the ISO on publicly available bulletin board

25


Generation Company Agent Decisionsare Based On Several Factors (2 of 2)

Generation Company Agent DecisionsGeneration Company Agent Decisionsare Based On Several Factors (2 of 2)are Based On Several Factors (2 of 2)

• The level of risk that an agent is willing to take is an integral part of its decision-making:– More conservative agents that have a lower tolerance for risk

may have lower profits but have a steady stream of income– More aggressive agents may have the potential for higher

profits but experience financial failure if anticipated market behaviors do not come into fruition

• Some business choices that the generation company agent can consider– Bid on contracts or bid into the pool market– Bid into the energy market and/or the ancillary services market– Adjust/change bid price strategy (production cost, low bid to

ensure acceptance, bid high on last portion of capacity, withhold capacity)

26


Generation Company Agents Use a Sophisticated Decision-Making Process

Generation Company Agents Use a Sophisticated Generation Company Agents Use a Sophisticated DecisionDecision--Making ProcessMaking Process

27


The Decision-Making Processes ofGeneration Company Agents Allows Learning

The DecisionThe Decision--Making Processes ofMaking Processes ofGeneration Company Agents Allows LearningGeneration Company Agents Allows Learning

28


ISO/RTO Agents Act as Coordinators (1 of 2)ISO/RTO Agents Act as Coordinators (1 of 2)ISO/RTO Agents Act as Coordinators (1 of 2)

29


ISO/RTO Agents Act as Coordinators (2 of 2)ISO/RTO Agents Act as Coordinators (2 of 2)ISO/RTO Agents Act as Coordinators (2 of 2)

• ISO/RTO agents match buyers and sellers in each of the five public markets

• ISO/RTO agents approve bilateral contracts to insure physical stability

• ISO/RTO agents manage payments for the public markets• ISO/RTO agents post information on a public bulletin board:

– Historical generation, outages, weather, loads, and location-based prices are posted

– Projected outages, weather, and loads are posted

30


An EMCAS Case has been CreatedBased on the Live Simulation

An EMCAS Case has been CreatedAn EMCAS Case has been CreatedBased on the Live SimulationBased on the Live Simulation

• Specific agents representing individual live simulation players were implemented by using EMCAS’ agent architecture:– The strategies of the individual players were determined by asking

them to write short descriptions of their approaches after the completion of the live simulation and then following up the writing with a series of focused interviews

– Once the strategies were determined, agents implementing each of the strategies were programmed

• The individual agents developed to emulate the live simulation players were run using the same dataoriginally used for the live simulation

• EMCAS closely matched the resultsof the six player live simulation

31


EMCAS Is Undergoing Thorough Verification and Validation Using a Variety of Techniques

EMCAS Is Undergoing Thorough Verification and EMCAS Is Undergoing Thorough Verification and Validation Using a Variety of TechniquesValidation Using a Variety of Techniques

• Unit testing has and is being preformed for the main components currently in use

• EMCAS’ internal algorithms have been and are being reviewed in detail by domain experts

• EMCAS output has and is being compared to analytically solvable special cases

• EMCAS output has and is being compared to some historical cases

• EMCAS is being expanded so the verification and validation is continuing!

32


Complexity Can Be Captured with Agent-Based Modeling and Simulation!

Complexity Can Be Captured with AgentComplexity Can Be Captured with Agent--Based Based Modeling and Simulation!Modeling and Simulation!

• CAS are structures composed of many components that interact and reproduce while adapting to a changing environment

• CAS often have numerous nested levels of interaction that span many scales of measurement

• ABMS can be used to build models for both single-scale and multi-scale CAS

33


Can Complexity Be Captured with Agent-Based Modeling and Simulation?

Can Complexity Be Captured with AgentCan Complexity Be Captured with Agent--Based Based Modeling and Simulation?Modeling and Simulation?

Are there additional questions?

Michael [email protected]

www.cas.anl.gov

can complexity be captured with agent- based modeling … · can complexity be captured with agent-...

Documents