Modeling Concepts

Day 3 | System Dynamics vs Agent-Based Models

Sachiko Ozawa

Why Simulation?

  • Test policies before implementation
  • Understand complex system behavior
  • Explore “what-if” scenarios
  • Identify intervention leverage points

Two Paradigms

  1. System Dynamics (SD): Focus on flows and stocks
  2. Agent-Based Modeling (ABM): Focus on individual entities

System Dynamics

Think of a river system:

  • Stocks: Water levels at different points
  • Flows: Movement between locations
  • Feedback: How flows affect stocks

Good for understanding aggregate behavior and bottlenecks.

Agent-Based Modeling

Think of individual water molecules:

  • Agents: Individual entities with rules
  • Interactions: How agents affect each other
  • Emergence: System-level patterns from individual behavior

Good for understanding heterogeneity and emergence.

Comparison

Aspect System Dynamics Agent-Based
Unit of analysis Aggregate Individual
Data requirements Lower Higher
Computational cost Lower Higher
Heterogeneity Limited Full
Emergence Not captured Captured

Supply Chain Example: SD

[Manufacturer] → Flow → [Warehouse] → Flow → [Facility]
                  ↑                     ↑
             Orders ←←←←←←←←←←←←←← Demand

Tracks aggregate quantities flowing through the system.

Supply Chain Example: ABM

  • Each facility is an agent
  • Each medication lot is tracked
  • Agents make decisions based on rules
  • Can track specific medications, expiry dates

Why ABM for WISE?

  1. Heterogeneity: Facilities vary in size, location, capacity
  2. Tracking: Need to follow specific medications
  3. Targeted interventions: Different policies for different agents
  4. Emergence: System-wide patterns from local decisions

The Integration Challenge

How do we connect ML predictions to ABM?

  • ML: “Demand will be X next month”
  • ABM: Agent ordering behavior

Bridge: Use predictions to parameterize agent decision rules.

Integration Architecture

┌─────────────────┐     ┌─────────────────┐
│   ML Model      │────►│   Predictions   │
│ (Demand Forecast)│     │ (By region/time)│
└─────────────────┘     └────────┬────────┘
                                 │
                                 ▼
┌─────────────────┐     ┌─────────────────┐
│   ABM Simulation│◄────│  Agent Params   │
│ (Supply Chain)  │     │ (Order rules)   │
└─────────────────┘     └─────────────────┘

Hands-on Preview

In the next session, we’ll:

  1. Generate predictions from our ML model
  2. Format predictions for ABM input
  3. Explore a simplified ABM
  4. Connect the components

Questions?

Let’s discuss before moving to the hands-on session.