More Vensim and “Stuff” Fall 2010

1. More Vensim and “Stuff” Fall 2010

2. TODAY • Recitation • Lecture • Hands-on

3. Recitation • Table lookups use _____ ______ between data points, by default

4. Comment about Table Lookups • To find the ordinate value corresponding to a particular abscissa value that is given, let b = ordinate desired, a given abscissa. Then b = bi + (a – ai)*(bi+1 – bi) / (ai+1 – ai) Where a has been determined to lie between abscissas ai and ai+1

5. In what two types of places can delays occur?

6. What two types of delays are there?

7. Comment about delays The modeler must ask… How do physical processes in the actual system create lagged behavior? How much disaggregation is necessary to represent the delay accurately?

8. Simulation Time Step • Should be between .5 and .25 of the shortest time constant (delay) in the model • Look at all of the time constants • Adjustment time • Perception time • Delivery delay time • Construction time • Lifetime • Find smallest • Set simulation time step appropriately

9. Integration Method • Euler for models with discrete events • RK4 for models with oscillation

10. Equilibrium • Transient • Steady-state

11. Dynamic Test Inputs • Purpose • Reveal inherent behavior • Create extreme conditions • Examples • Pulse • Step • Ramp—one we didn’t look at • Exponential growth • Noise—randomness

12. Generating surprises • Test for asymmetric responses to positive and negative disturbances • Test small and large amplitude inputs • Test policies at multiple points in system • Test multiple patterns of behavior

13. Extreme Conditions • Purpose • Reveal weaknesses • Generate insight • Methods • Remove contents of stock with PULSE function • Cut off inflows or outflows • Artificially force variables to 0 or to infinity

14. Reality Check • Purpose • Automate model quality checks • Format • Test input • THE CONDITION: Staff = 0 • Consequence • IMPLIES: Production = 0

15. Partial Model Testing • Purpose • Divide and Conquer • Develop understanding of subsystems • Test response of subsystems to driving data • Methods • Cut & paste structures into a new model • Use data variables or test inputs to drive behavior

16. Feedback Elimination • Purpose • Identify feedback loops that are causing behavior • Methods • Sever flow connections • Replace variables with constants or test inputs • Insert 0*… in equations • Flatten lookups

17. Parameter Sensitivity Analysis • Purpose • Link behavior to feedback loop structure • Identify leverage points • Search for equilibria • Methods • Vary parameters and initial conditions • Stretch and shift lookup table shapes

18. Types of Sensitivity • Insensitive • Pendulum always comes to rest at bottom • Numerical • Numerical values change, but behavior “looks” the same • Behavior mode • Shift from s-shaped growth to oscillation • Policy • Policy conclusions change

19. Policy Evaluation • Purpose • Develop effective policies • Identify conditions for effectiveness • Identify weakness in formulation of existing policies • Tools • Sensitivity Analysis • Optimization • Gaming

20. Automation • Sensitivity analysis • Calibration • Optimization • Command files