Bayesian Discrete-Event Simulation — Posterior-informed stochastic process modeling
Also known as: Bayesian DES, BDES, Bayesian event-driven simulation, posterior-driven discrete-event simulation
Bayesian Discrete-Event Simulation (BDES) integrates Bayesian statistical inference with discrete-event simulation. Prior beliefs about system parameters — such as service rates, arrival times, or failure probabilities — are updated with observed data via Bayes' theorem, and the resulting posterior distributions directly drive the simulation engine. This coupling allows modelers to propagate both aleatory and epistemic uncertainty through event-driven process models.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use BDES when you need to model a sequential, event-driven process (queues, logistics chains, patient pathways, manufacturing lines) AND parameter uncertainty is substantial — especially when data are sparse, expensive to collect, or heterogeneous across sites. It is particularly valuable in health technology assessment, emergency services planning, and supply-chain risk analysis where decision-makers need calibrated uncertainty bounds rather than point estimates. Avoid BDES when event-level data are rich enough to estimate parameters with negligible uncertainty (standard DES suffices), when the process structure itself is poorly understood (system dynamics or ABM may be more appropriate), or when computational budget is too limited for MCMC plus repeated simulation runs.
Strengths & limitations
- Propagates both aleatory randomness (inherent process variability) and epistemic uncertainty (parameter estimation error) through the model, yielding calibrated credible intervals on KPIs.
- Formally incorporates expert prior knowledge, enabling useful predictions even when historical data are sparse or partial.
- Posterior distributions update coherently as new data arrive, supporting sequential learning without restarting the analysis from scratch.
- Produces full predictive distributions rather than single-point estimates, enabling probabilistic decision rules and value-of-information analyses.
- Compatible with standard discrete-event simulation software: the Bayesian layer governs parameter inputs; existing DES engines remain unchanged.
- Computationally intensive: MCMC or variational inference for posterior estimation, combined with hundreds or thousands of DES replications, can be prohibitive for large or complex models.
- Requires careful prior specification; poorly chosen priors can dominate results when data are limited, and sensitivity analyses add further workload.
- Model validation is more complex than standard DES because both the statistical and the process-simulation components must be separately and jointly validated.
- Interpretability decreases as model complexity grows; stakeholders unfamiliar with Bayesian reasoning may find posterior distributions difficult to communicate.
- Implementation demands cross-disciplinary expertise spanning Bayesian statistics, simulation modeling, and domain knowledge simultaneously.
Frequently asked
How is Bayesian DES different from standard stochastic DES?
Standard stochastic DES uses fixed input distributions (e.g., Exponential(lambda) with a pre-set lambda). Bayesian DES treats lambda itself as uncertain, placing a prior on it and updating it from data. Each simulation run samples from the posterior over lambda, so output uncertainty reflects both process randomness and parameter estimation error.
Do I need custom software to implement Bayesian DES?
Not necessarily. A practical workflow couples a Bayesian inference library (Stan, PyMC, JAGS) with an existing DES engine (SimPy, AnyLogic, Arena). The inference library produces posterior samples, which are passed as input parameters to the DES engine for each run.
How many simulation runs are needed?
A common minimum is 1,000 posterior samples times enough replications per sample to stabilize within-run randomness. The total required runs depend on the variance of outputs; convergence diagnostics (R-hat for MCMC, output variance across replications) should guide the choice rather than a fixed number.
When should I prefer BDES over a simpler Monte Carlo sensitivity analysis?
When parameter uncertainty is derived from real data and must be propagated formally. Simple Monte Carlo sensitivity assigns arbitrary ranges to parameters without grounding them in likelihood; Bayesian updating ensures the uncertainty envelope is data-consistent and internally coherent.
Can BDES handle non-stationary processes (e.g., time-varying arrival rates)?
Yes. Time-varying parameters (e.g., a Poisson arrival rate that changes by hour of day) can be modeled with Bayesian hierarchical or time-series priors, and the resulting posterior process drives the DES in the same pipeline framework.
Sources
- Onggo, B. S., & Kunc, M. (2016). Combining discrete-event simulation and Bayesian updating for incorporating evidence from real-world data. Journal of Simulation, 10(1), 1-12. link ↗
- Pidd, M. (2004). Computer Simulation in Management Science (5th ed.). Wiley. ISBN: 9780470092781
How to cite this page
ScholarGate. (2026, June 3). Bayesian Discrete-Event Simulation — Posterior-informed stochastic process modeling. ScholarGate. https://scholargate.app/en/simulation/bayesian-discrete-event-simulation
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Agent-based Discrete-Event SimulationSimulation↔ compare
- Bayesian Agent-Based ModelingSimulation↔ compare
- Bayesian Markov ModelSimulation↔ compare
- Discrete-Event SimulationSimulation↔ compare
- MONTE-CARLO-SIMULATIONDecision-making↔ compare
- Stochastic Discrete-Event SimulationSimulation↔ compare