Agent-Based Discrete-Event Simulation — Hybrid Simulation Combining Autonomous Agents and Event Queues
- Introduced by
- Hybridization formalized by multiple authors; Siebers & Aickelin, Lagergren & Buckley among key contributors
- Year
- 2000s
- Type
- Hybrid simulation paradigm
- Family
- Simulation / optimization
What does it do?
Agent-based discrete-event simulation (AB-DES) is a hybrid modeling paradigm that couples autonomous agent behavior with an event-driven execution engine.
It captures the decision-making heterogeneity of individual entities while maintaining the precise, time-stamped flow control of discrete-event simulation, making it suitable for complex systems where both individual agency and process sequencing matter.
The intuition behind the method
Some systems are too complex for either agent-based modeling or discrete-event simulation alone. A hospital, for example, has patients and staff who make autonomous choices (agent behavior), but also a strict sequence of service events — triage, treatment, discharge — that must be tracked precisely (discrete events). AB-DES lets modelers represent both: agents carry personal attributes and decision rules, while the event queue schedules and processes state transitions in chronological order. This hybrid approach captures emergent collective behavior and process bottlenecks simultaneously, providing richer insights than either paradigm in isolation.
History
Discrete-event simulation originated in the 1960s with Tocher and Gordon. Agent-based modeling emerged in the 1990s through Schelling, Axelrod, and Epstein. Hybrid AB-DES approaches were formally explored in the 2000s as researchers recognized that complex sociotechnical systems required both paradigms. Healthcare and logistics applications drove early adoption, and the approach is now supported by major simulation platforms including AnyLogic.
When to use it
Use AB-DES when the system involves both heterogeneous autonomous actors making individual decisions and well-defined sequential processes with precise timing. It is especially appropriate for healthcare pathways, supply chain logistics, emergency response, and manufacturing systems where worker or customer behavior varies but process stages are fixed. Do not use AB-DES when the system is homogeneous and process-driven with no meaningful agent heterogeneity (standard DES suffices), or when interactions are purely emergent with no event-driven sequencing (pure ABM suffices). Avoid it when data on both agent behavior and event timing are unavailable, as the model will be under-constrained.
Limitations and considerations
- Model design and implementation complexity is significantly higher than pure ABM or pure DES.
- Requires calibration data for both agent behavior parameters and event timing distributions, which may be difficult to obtain.
- Computational cost can be high for large agent populations with fine-grained event granularity.
- Verification and validation requires expertise in both simulation paradigms.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Captures both individual heterogeneity (agent behavior) and process sequencing (event timing) in a single coherent model.
- Enables richer what-if analysis than either ABM or DES alone by varying both behavioral rules and process parameters.
- Naturally represents systems where human decision-making interacts with institutional processes.
- Supports parallel and distributed execution for large-scale scenarios.
- Validated in real-world domains including healthcare, logistics, and emergency management.
Applications
- Hospital patient flow modeling where staff behavior and treatment queues interact
- Supply chain disruption analysis combining supplier decision-making with shipment event schedules
- Emergency evacuation planning with heterogeneous pedestrian agents and event-triggered exits
- Manufacturing cell simulation with worker skill heterogeneity and machine event queues
- Urban transportation systems with driver behavior and signal event control
- Policy evaluation in social services where case worker decisions interact with scheduled appointments
How it works
How it works
Define the system boundary: identify which entities will be modeled as agents (with attributes, state, and decision rules) and which processes will be modeled as discrete events (instantaneous state changes scheduled on a global event calendar). Clarify what questions the simulation must answer.
Design the dual-layer architecture. The agent layer specifies agent types, attributes, behaviors, and interaction protocols. The event layer specifies event types, triggering conditions, and the global event calendar structure. Define how agents generate, consume, and react to events.
Specify agent decision logic: rule sets, utility functions, or behavioral heuristics that govern how each agent selects actions when an event occurs or at scheduled intervals. Agent heterogeneity is encoded here, allowing different agents of the same type to behave differently based on their internal state.
Initialize the global event calendar with seed events. At each simulation step, the engine extracts the earliest event, advances simulation time to that event's timestamp, identifies the affected agent(s), and invokes the relevant agent decision rules. Agents may schedule new events as a result.
Run the simulation for the desired number of replications or until a stopping criterion is met. Collect trajectory data: event histories, agent state timelines, and aggregate statistics. Use multiple independent replications to obtain statistically stable output estimates.
Analyze output data using statistical methods: compute means and confidence intervals across replications, perform sensitivity analysis on key parameters, visualize agent flow through the event network, and compare alternative policy scenarios. Validate the model against real-world data before drawing conclusions.
Common pitfalls
- Conflating simulation time with real time: event timestamps must be managed rigorously to avoid logical inconsistencies.
- Over-engineering agent rules when simpler behavioral assumptions would suffice, leading to opaque and uncalibrated models.
- Neglecting warm-up periods: early simulation output reflects initialization artifacts rather than steady-state behavior.
- Running too few replications and mistaking a single run's output for a representative result.
Frequently asked questions
How does AB-DES differ from pure agent-based modeling?
Pure ABM advances time in fixed steps and focuses on emergent collective behavior. AB-DES uses a global event calendar to advance time only when events occur, which is more efficient for sparse processes and preserves exact timing of state changes — critical when sequencing matters.
What software tools support AB-DES?
AnyLogic is the most widely used commercial platform that natively supports all three paradigms (ABM, DES, SD) in a single model. Repast Simphony and MASON can be extended to incorporate event queues, though they require more custom development.
How many replications are needed?
There is no universal answer, but common practice is to run a pilot study of 10–30 replications, estimate output variance, and calculate the required sample size for the desired confidence interval width. Stochastic models typically require at least 30 replications for reliable inference.
When should I choose pure DES over AB-DES?
Choose pure DES when entities can be treated as homogeneous (all patients, all customers, all parts behave identically) and individual decision logic is not needed. AB-DES adds complexity that is only justified when behavioral heterogeneity meaningfully affects system-level outcomes.
Is AB-DES suitable for real-time decision support?
Generally no — AB-DES models are calibrated and validated offline. Some fast, simplified AB-DES models can run quickly enough for near-real-time scenario testing, but full-scale models are typically used for planning and policy analysis rather than live operational control.
Related methods
Sources
Lagergren, J. H., & Buckley, E. (2010). A hybrid approach to simulation: Combining agent-based and discrete event simulation. Proceedings of the 2010 Winter Simulation Conference, pp. 170–181. IEEE.
Open publication referenceCopied.Agent-based model. Wikipedia.
Open publication referenceCopied.