simulation with system dynamics models ie 680 spring 2007 po-ching c. delaurentis april 19, 2007
TRANSCRIPT
Outline Systems Thinking System Dynamics (SD) Paradigm Comparison Basics of System Dynamics Quantification Challenges Simulation with System Dynamics- An Example Summary System Dynamics Resources
Systems Thinking
Systems Thinking Early 20th century physicists began to
challenge Newtonian precepts; Werner Heisenberg, Norbert Weiner, Von Bertalanffy
“An approach for developing models to promote our understanding of events, patterns of behavior resulting in the events, and even more importantly, the underlying structure responsible for the patterns of behavior”*
* http://www.systems-thinking.org
System Dynamics
System Dynamics Introduced by Jay Forrester of MIT in 1958
“The study of information-feedback characteristics of industry activity to show how organizational structure, amplification (in policies), and time delays (in decisions and actions) interact to influence the success of the enterprise” (Forrest 1958 & 1961)
Paradigm Comparison of System Dynamics, Discrete Event & Agent Based
Borshchev A , Filippov A. From System Dynamics and Discrete Event to Practical Agent Based Modeling: Reasons, Techniques, Tools. Proceedings of the 22nd International Conference, July 25-29, 2004, Oxford, England, UK.
High Abstraction Less Details Macro Level
Strategic Level
Middle Abstraction Medium Details
Meso Level Tactical Level
Low Abstraction More Details
Micro Level Operational Level Individual objects, exact sizes, distances, velocities, timings, …
Agent Based Active objects
Individual behavior rules
Direct or indirect interaction
Environment models
Discrete Event Entities (passive
objects)
Flowcharts and/or transport networks
Resources
Aggregates, Global Casual Dependencies, Feedback Dynamics, …
Mainly discrete
System Dynamics Levels (aggregates)
Stock-and-Flow Diagrams
Feedback loops
Mainly continuous
Basics of System Dynamics (cont’d)
Brownies_in_Stomach(t) = Brownies_in_Stomach (t - dt) + (eating - digesting) * dt
INIT Brownies_in_Stomach = 0
DOCUMENT: Initially Andy’s stomach is empty.
UNITS: brownies
eating = 1
DOCUMENT: Andy eats a brownie every hour.
UNITS: brownies/hour
digesting = 1/2
DOCUMENT: Andy digests 1 brownie every 2 hours. He therefore digests a half a brownie every hour.
UNITS: brownies/hour
Basics of System Dynamics (cont’d)
Wide Range of System Dynamics Applications Corporate planning and policy design Economic behavior Public management and policy Biological and medical modeling Energy and the environment Supply chain management
Quantification ChallengesQuantification Challenges
Quantification Challenges System dynamics is strategic in orientation and it
is often seen to have ‘soft’ variables
Example (Coyle 2000): Consumer Satisfaction as an influence on
New_Order_Inflow_Rate = Basic_Inflow * Satisfaction_Multiplier
A variable that may range from 0 to an upper limit, and have a nonlinear relationship with Consumer Satisfaction
Quantification Challenges (Cont’d) If it becomesNew_Order_Inflow_Rate = Basic_Inflow * Satisfaction_Multiplier * Quality_Multiplier * Price_Multimplier * etc.
The number of uncertainties becomes very large;
Strong assumption that multipliers are multiplicative.
EX: 0.5 * 0.5 * 0.5 = 0.125 only 25% of 0.5
0.510 = 0.000977 only 1/500 of 0.5
Simulation Example
Police & Driver– A System Dynamics Model for a Mixed Strategy Game (Kim & Kim, 1997) System dynamics: dynamic fluctuations of a system Game Theory
Players; preferences & strategies; payoff/utility functions Players change decisions in response to other players’
actions Finding equilibrium states in game situation Dominant vs. mixed strategies
Police & Driver Game “You are driving your car in a hurry…there are two states
of the world: either the police are nearby or they are not. There are two actions to choose from: either to violate the speed limit or to abide by the law.” (Tsebelis 1989)
Police
Patrol Not Patrol
DriverSpeed a1, a2 b1, b2
Not Speed c1, c2 d1, d2
Note: c1 > a1 b1 > d1
a2 > b2d2 > c2
p
1-p
q 1-q
Police & Driver Game (Cont’d) Mixed strategy equilibrium results:
p = prob. with which the driver chooses to speed q = prob. with which the policeman decides to patrol Solved for:
p* = (d2-c2)/(a2-b2+d2-c2)q* = (b1-d1)/(b1-d1+c1-a1)
Observation The probability of the driver’s law violation is not
determined by the payoffs for the driver Increase in penalty ( a1)
Argument A contradiction to common sense: increase in penalty is
conceived as one the most effective tools for policy implementation
Police & Driver Game with System Dynamics
Why does game theory produce theorems inconsistent with common sense? Game theory applicability Equilibrium applicability
How to model this game with SD? Probability of players’ behaviors
Two independent players in the game Population mixed-strategy game
Policemen in Office
Policemen in Patrolling
Drivers in Violation
Drivers in Conforming
Patrol to office
Quit Patrol
Go Patrol
Office to Patrol
Patrolling Time
Quitting Time
Difference eup eunp
euno
ppcnpppvnp
eupppvp
ppcp
prob v
Speed Down Time
Speed Down
Speed Up
Conform to ViolationSpeed Up
Time
Violation to Conform
Difference euv eunv
dpcnp
dpcp
eunv
prob p
euv
dpvp
dpvnp
System Dynamics Diagram of the Game (redrawn)
parameters
Simulation Results Oscillation Rather than Equilibrium
Tendency towards the equilibrium state
But it takes a long time!!
p*=0.25 (2:prob p)
q* = 0.5 (1:prob v)
Simulation Results (Cont’d) The Effectiveness of Penalty Increase
Increase in Penalty
p*=0.25 (2:prob p)
q* = 0.22 (1:prob v)
Tendency towards the equilibrium state
Prob. of violation reduced to < 0.25 for about 50 days
Simulation Results (Cont’d) Considering Information Delay Between the Police & the Driver
Larger amplitude
Not approaching steady state
10 days for Driver
5 days for Police
Simulation Results (Cont’d) The Effectiveness of Penalty Increase with Information Delay
Increase in Penalty
Prob. of violation is lightly reduced
Simulation Results (Cont’d) Effectiveness of Automatic Penalty Management (without info
delay) Police change the amount of penalty in line with probability of changes
in law violation
Equations
Equilibrium reached after a short period of fluctuation
Policy Implications
Simulation results suggest: Temporarily reduce the violation tendency of drivers by
changing the amount of penalty Penalty management can decrease the amplitude of
fluctuating behavior of drivers For policy makers
Temporary reductions of speed limit violation will be a sufficient incentive for introduction of penalty increase
Penalty management can reduce the amplitude of fluctuation violations
Max. level of violation probability of car accidents?
Insights
Information delay & policy interruptions exist in a dynamic system
Mixed-strategy equilibrium may be a poor guide Equilibrium vs. Steady State In the real world, players are usually myopic
System Dynamics: simulate evolutionary processes toward (non-)equilibrium states
Game Theory: a framework for modeling the world of competition and cooperation
Summary System Dynamics
Stock-and-flow Casual loops Higher, strategic level modeling Dynamic behaviors of a system Transient processes Challenges: quantification of influencing
factors; underlying effects
System Dynamics Resources
System Dynamics Society http://www.systemdynamics.org/
Sterman, Business Dynamics: Systems Thinking and Modeling for a Complex World, McGraw-Hill/Irwin, 2000
Software VenSim– http://www.vensim.com/ Stella– http://www.iseesystems.com/