SYSTEM DYNAMICS 101
FAMES
Sgouris Sgouridis
MIT
Cambridge, MA
May, 2nd 2006
1
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
System Dynamics Overview
• System dynamics:
– a way of thinking
– a modeling tool
• Basic SD premise:
– “Policy resistance” can be anticipated
through systems thinking.
– SD facilitates systems thinking by model
construction and visualization of the
modeled system behavior.
2
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Policy Resistance
• A definition: unanticipated effects of wellintentioned solutions.
– Related to behavioral complexity.
• Some examples:
– Antibiotics and drug-resistant strains.
– Fire suppression and forest fires.
– Proliferation of taxis in San Diego.
– California’s deregulation of electricity
markets.
• Can you think of others?
3
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
SD as a way of thinking: Exposing mindsets
The open-loop, event-oriented view of the world
The feedback view of the world
Goals
Decisions
Problem
Decision
Results
Goals
Situation
Environment
Our decisions alter our environment, leading to new decisions,
Decisions
Goals
Side effects
Environment
Goals of other
agents
Figures by MIT OpenCourseWare, adapted from Sterman.
Actions of
others
but also triggering side effects, delayed reactions, changes in
goals and interventions by others. These feedbacks may lead
to unanticipated results and ineffective policies.
Source: Sterman 2002
4
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Some drivers of policy resistance:
• Time-delays
• Non-linearities
• Feedback
5
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Shower: a metaphor and the origin of SD
• Achieving the right temperature is a control
problem: there are delays in the system even
if uncertainty is removed.
• Other control problems: driving, flying a plane.
Humans and machines can learn to anticipate
and adjust to short term feedback delays.
• The modeling part of SD originates from
control theory and servomechanism design.
• Jay Forrester in the 1950s at MIT was the first
to adapt those ideas to policy applications.
6
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Folding a paper
• Fold a paper of 0.1mm thickness
• How thick does it become if you fold
it ~45 times?
– <0.5m
– 1m
– 100 m
– >100 m
Human brains are not adapted to deal
with non-linearities.
7
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Beer games
Wholesaler
Beer cases
Retailer
Orders
Feedback: (coupling
of action and
reaction – usually
several stakeholders
are involved).
What is the plausible
reference mode?
Demand
Distributor
Producer
8
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Casual loop diagrams
• A way to capture feedback
Inventory
Balancing Loops
+
-
Gap
Orders
+
+
Target
Inventory
+
Users
Rabbits
+
+
+
Available
Females
Stores
Informed
Non-users
+
Reinforcing Loops
Number of
Births
+
9
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Causal Loops recap
• Unambiguous
– Direction
– Link polarity
• Tracing the loop to identify loop polarity
• Example:
– price and revenue
• Example:
–
–
–
–
Attractiveness of market
Number of competitors
Price
Profits
10
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Beyond causal loops: the mechanics
• Stocks
• Flows
• Integration
Flows (liters/minute)
100
75
Inflow
Outflow
Inflow
Bathtub
Outflow
50
25
0
0
Figures by MIT OpenCourseWare, adapted from Sterman.
2
4
6
8
10
12
14
16
Time (minutes)
Source: Sterman 2002
Assuming the initial quantity is 100lit. How does the “stock” (quantity of water) behave?
11
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Fundamental Reference Modes
Exponential Growth
Goal Seeking / Exponential Smooth
st
Linear
Fundamental
Modes
(1 Order)
R
B
B
State
of the
System
Corrective
Action
B
Carrying capacity
Discrepancy
B
time
Carrying
Capacity
B
Resource
Adequacy
Source: Piependbrock 2004, Sterman 2000
State
of the
System
Goal
B
Corrective Discrepancy
Action
Carrying
capacity
R
R
time
Overshoot & Collapse
Carrying
capacity
B
Net
Increase
Rate
Goal
S-shaped growth with
Overshoot & Oscillation
S-shaped Growth
pendulum
time
Net
State
Increase R of the
Rate
System
Fractional
Net Increase
Rate
(2nd Order, delay = stock)
(1 Order)
time
Nonlinear
(Mode-Switching)
Oscillation
st
time
B1
R
B2
time
Net
Increase
Rate
Fractional
Net Increase
Rate
State
R of the
System
B
Net
State
Increase R of the
Rate
System
Carrying
Capacity
Fractional
Resource
Net
Increase
Adequacy
Rate
of Carrying
Capacity
Figure by MIT OpenCourseWare, adapted from Piependbrock and Sterman.
12
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Brief Return to the Beer Game model
• Examining various behaviors
13
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Traditional Model Testing: Validation,
Verification, Accreditation
• Validation: the model structure reflects the real
world accurately enough
• Verification: the model operates according to
how it was designed
• Accreditation: the model is considered credible
by the decision-makers
Methods:
Inspection
Historical data and forecasting / backcasting
Statistical testing
14
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
SD Modeling Process review
• Problem articulation
– Purpose (why?)
– Boundary selection (space + time)
– Reference modes (expected behaviors)
• Formulating dynamic hypotheses
– Endogenous explanation
• Testing and validation
• Policy Design and Evaluation
15
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
SD and the CLIOS Process
• SD is considered a tool in the CLIOS Process
toolkit.
• SD as a way of thinking shares a lot of
similarities with the CLIOS Process but it is not
explicit with regard to the nested complexity.
• CLIOS Diagrams can be a good starting point
for creating SD models.
• Preliminary SD models can be a good starting
point for engaging stakeholders and making
explicit their tacit knowledge.
16
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Developing an SD model for SNF
target fraction of
nuclear
Application
process
Reactor
Construction
Rate of
Construction
Existing
Reactors
Rate of
Decommisioning
Reactor
Decommisioning
Energy demand
Alternative energy
sources
fraction of
nuclear
Cost/MW of
alternatives
Max MW
Capacity
Uranium
extraction
Cost/MW
Rate of use
Reactor
type
Uranium in
usage
Useful life
Unstored
SNF
Stored SNF
Storage rate
Reprocessing
17
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Resources
Software:
Vensim, Stella, iThink.
Anylogic!
Books:
J. Sterman Business Dynamics
P. Senge The Fifth Discipline
Attached Papers
18
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Back-up
19
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Agenda
•
•
•
•
•
Purpose of SD
Origin of SD
Mind-sets vs. SD
SD Notation and terminology
Applying SD thinking in SNF
20
Cite as: Joseph Sussman, and Sgouris Sgouridis, course materials for ESD.04J Frameworks and Models in Engineering Systems,
Spring 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].