Discrete Event Simulation (DES)

The system state only changes at discrete, countable points in time when an event occurs. Between events, the state is assumed to remain constant. This contrasts with continuous simulation where state variables change smoothly over time.

• An event is an instantaneous occurrence that alters system state
• Examples: order arrival, machine breakdown, shipment departure
• The simulation clock jumps from one event timestamp to the next
• No computational resources are wasted on periods of inactivity

The Future Event List is the central data structure that drives DES execution. It is a priority queue of all scheduled events, ordered by their occurrence time. The simulation engine repeatedly removes the most imminent event and processes it.

• Events are scheduled by other events during execution
• Example: An 'arrival' event schedules the next 'arrival' event
• Conditional events wait for a specific state to become true
• Efficient FEL implementation is critical for simulation performance

DES relies on probability distributions to model real-world uncertainty. Rather than using fixed averages, stochastic inputs capture the variability inherent in supply chain processes.

• Inter-arrival times often follow an exponential distribution
• Service times may use gamma, lognormal, or Weibull distributions
• Demand quantities frequently modeled with Poisson or negative binomial
• Goodness-of-fit tests validate distribution choices against historical data

Because DES models are stochastic, output metrics are random variables, not single-point estimates. Rigorous statistical analysis is required to draw valid conclusions from simulation runs.

• Multiple replications with different random seeds are essential
• Confidence intervals quantify the precision of performance estimates
• Warm-up period detection removes initial transient bias
• Batch means method handles autocorrelation in steady-state simulations

DES models are structured around entities that flow through a system of resources and queues. This architecture naturally maps to supply chain objects like orders, pallets, and trucks.

• Entities: Dynamic objects that move through the system (orders, parts)
• Resources: Static objects that provide service (machines, docks, workers)
• Queues: Waiting lines that form when demand exceeds capacity
• Attributes: Properties attached to entities (priority, due date, size)
• Resource seizure and release logic governs contention

DES models are classified by their run duration characteristics, which determines the appropriate statistical analysis method.

• Terminating: Has a natural end event (e.g., a warehouse shift ends)
  • Initial conditions matter; multiple replications required
• Non-terminating: Runs continuously without a natural end (e.g., a 24/7 port)
  • Requires steady-state analysis with warm-up period removal
  • Steady-state detection algorithms identify when equilibrium is reached

A bottom-up simulation paradigm where autonomous decision-making entities (agents) interact according to defined behavioral rules. Unlike DES's process-flow focus, ABM excels at modeling emergent phenomena like market dynamics, crowd behavior, and decentralized logistics networks.

• Agents possess internal states, goals, and heterogeneous attributes
• System-level patterns emerge from local interactions without centralized control
• Ideal for modeling supplier negotiation behaviors and competitive marketplaces

A stochastic technique that executes thousands of randomized trials to map the probability distribution of possible outcomes. While DES sequences events chronologically, Monte Carlo methods focus on static uncertainty propagation—perfect for quantifying financial risk in inventory investments or lead time variability.

• Uses repeated random sampling from defined probability distributions
• Generates confidence intervals and Value-at-Risk metrics
• Computationally intensive but embarrassingly parallelizable

A mathematical framework for sequential decision-making under uncertainty, defined by states, actions, transition probabilities, and rewards. MDPs provide the theoretical backbone for reinforcement learning agents that optimize warehouse routing and dynamic pricing within DES environments.

• Satisfies the Markov property: future depends only on current state
• Solved via value iteration, policy iteration, or Q-learning
• Foundation for autonomous agent behavior in supply chain control towers

A high-performance computing technique that partitions a DES model across multiple processors to execute independent events concurrently. Synchronization protocols—either conservative (Chandy-Misra) or optimistic (Time Warp)—prevent causality violations while dramatically accelerating large-scale supply chain simulations.

• Requires careful model partitioning to minimize cross-processor dependencies
• Global Virtual Time (GVT) bounds memory usage in optimistic protocols
• Enables real-time digital twin simulation of global logistics networks

A systematic methodology for planning simulation runs to efficiently identify which input factors most significantly impact key performance indicators. DOE replaces ad-hoc parameter sweeping with statistically rigorous designs like factorial experiments and Latin hypercube sampling.

• Minimizes computational cost while maximizing information gain
• Identifies interaction effects between multiple variables
• Essential for sensitivity analysis and simulation model calibration

The rigorous three-phase quality assurance process ensuring simulation credibility. Verification confirms the model is built correctly (no coding errors). Validation confirms the model represents reality accurately (output matches historical data). Accreditation is the official certification for a specific use case.

• Face validity, Turing tests, and statistical hypothesis testing for validation
• Essential for defense logistics and FDA-regulated pharmaceutical supply chains
• Without VV&A, simulation-based decisions carry unquantified risk