31247 Chap.14 Scenarios, Calibration and Agent-Based Models
Scenarios, Calibration and Agent-Based Models
Using a model to compare futures
Scenario development runs a model under different assumptions to prepare for an uncertain future.
The Week 9 lab varies the contact rate in the epidemic model, and an optional lab calibrates unknown parameters against observed data.
Behaviour from individuals
Week 10 adds agent-based modelling, in which a population of agents with their own states and rules produces the system’s behaviour.
The lecture closes with the challenge of choosing a method and the obstacles every participatory modelling project meets.
Choosing the method that fits the question
The Week 10 lecture warns that the variety of tools makes choosing hard, and that in practice the modellers’ prior experience often decides the method more than the problem does.
Matching the question to the tool, rather than the tool to the team’s habits, is the last skill the subject teaches.
What this chapter covers
- 01
Scenario development
- 02
The parameter variation experiment
- 03
Calibration against historical data
- 04
Agents, statecharts and transitions
- 05
Emergent behaviour
- 06
Choosing the right modelling method
- 07
Challenges in participatory modelling
Worked example · free
Read the contact-rate scenarios from the epidemic model
- 1Timing: as the contact rate rises from 0.5 to 2.0, the peak moves earlier, from about day 113 to about day 45.
- 1Size: the peak rises from about 2,550 to about 4,050 infectious people.
- 1Shape of the effect: going from 0.5 to 1.25 adds about 1,190 people at the peak, but going from 1.25 to 2.0 adds only about 310, so the effect weakens at higher rates.
- 1Conclusion for a report: reducing contacts both delays and lowers the peak, with the largest gain at the lower end of the range.
Key terms
- Scenario Development
- A systematic way of exploring possible futures by running a model under different assumptions.
- Calibration
- Tuning a model’s parameters so its output matches an observed pattern.
- Agent-Based Model
- A model in which individual agents with their own behaviours produce the system’s behaviour.
- Statechart
- A diagram of the states an agent can be in and the transitions between them.
- Parameter Variation
- A series of model runs that differ in one or more parameters, compared on one chart.
- Emergence
- System behaviour that arises from many individual behaviours rather than being specified directly.
Scenarios, Calibration and Agent-Based Models FAQ
What does the parameter variation lab show?
It runs the epidemic model with the contact rate varied from 0.3 to 2 in steps of 0.1, eighteen runs on one chart. The lecture concludes that a higher contact rate makes the infection spread faster.
Do I have to complete the calibration lab?
The Week 9 tutorial marks it as optional, because the calibration experiment needs AnyLogic’s Professional edition, available on a 30-day free licence. The parameter variation experiment in the same week uses the ordinary edition and AnyLogic Cloud.
When should I use an agent-based model instead of system dynamics?
When individual differences, interactions or locations drive the result, agents fit better; when the question is about totals and rates, stocks and flows are simpler. The lecture also warns that modellers often pick the method they already know rather than the one the problem needs.
What are the main challenges of participatory modelling?
The Week 10 lecture lists selecting stakeholders, time and cost, complexity, facilitation, biases, beliefs, heuristics and values, change, uncertainty, and gaining trust while staying neutral, and it advises interpreting results together with the stakeholders.
What does calibration change in a model?
It changes the values of parameters that cannot be measured directly, such as infectivity, so that the model’s output matches an observed history. The software repeats runs, compares output with the data and adjusts the values until the difference is as small as possible.
What is an agent in an agent-based model?
Any object whose behaviour you model individually: people, vehicles, units of equipment, projects, products or organisations. In the Week 10 lab each of 5,000 consumers is an agent with a statechart that moves it from potential user to user.
Assessment move
Define three scenarios for your own stock and flow model, change one parameter in each, and write a sentence on what each result means for an improvement criterion. List which of your parameters are measured and which are assumptions. Then sketch a two-state statechart for the main actor in your problem to see whether an agent-based view adds anything. Ask Sia to question your weakest assumption.
Note in the report which results depend on assumed rather than calibrated values.
Working through Scenarios, Calibration and Agent-Based Models in 31247? Sia is AskSia’s AI Information Technology tutor — ask any 31247 Scenarios, Calibration and Agent-Based Models question and get a clear, step-by-step explanation grounded in how 31247 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.