Simulation Optimization Techniques for Stochastic Systems

Simulation optimisation for stochastic systems combines the power of computer-based experiments with mathematical search strategies to identify optimal or near-optimal solutions when uncertainty rules out closed-form analysis. Such systems often involve randomness in inputs or dynamic interactions over time, as found in queueing networks, inventory management, supply chains and complex service systems. Core techniques address the dual challenge of sampling variability and large decision spaces. Ranking and selection procedures allocate simulation effort to discriminate among a finite set of designs; metamodel-based approaches construct surrogate models to approximate performance surfaces and guide exploration; multi-armed bandit adaptations borrow ideas from reinforcement learning to balance exploration and exploitation; and multi-fidelity frameworks exploit low-cost approximations to accelerate convergence. Recent advances emphasise adaptive allocation of samples based on statistical measures of uncertainty, hybridisation of evolutionary heuristics with memory components to preserve historic observations, and sequential configurations that embed simulation runs within optimisation loops for real-time decision-support. Collectively, these methods enhance reliability and efficiency across diverse applications—from digital twins and healthcare systems to emergency response and infrastructure planning—by marrying statistical rigour with computational ingenuity.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

A novel genetic multi-armed bandit algorithm integrates reinforcement-learning inspired memory into evolutionary search, retaining all past simulation observations rather than discarding those outside the current population. This approach systematically refines sample-mean estimates and guides selection with a global convergence guarantee, outperforming standard benchmarks across a suite of discrete optimisation test problems while reducing computational effort. A multi-fidelity simulation modelling method formulates the construction of discrete event models as a bi-objective optimisation problem, seeking the best trade-off between speed and accuracy. An efficient multi-objective algorithm based on hypervolume selection identifies high-quality model fidelities and demonstrates superior performance in a digital-twin emergency department case study, highlighting the value of automated model generation in Industry 4.0. Finally, sequential simulation–optimisation configurations have been applied to fugitive interception on road networks, embedding rich behavioural simulation as constraints within optimisation formulations. This configuration reduces computation times by an order of magnitude compared with traditional simulation model optimisation, enabling timely support for real-time law enforcement decision-making under uncertainty.

Simulation Optimization Techniques for Stochastic Systems publication trend

The graph below shows the total number of articles in simulation optimization techniques for stochastic systems across all publications each year (not limited to Nature Index journals).

Technical terms

Stochastic simulation: A computational experiment that replicates the random behaviour of a system to estimate performance measures.

Simulation optimisation: The process of finding the best decision variables for a simulation model under uncertainty, subject to performance objectives.

Optimal computing budget allocation (OCBA): A statistical strategy that allocates limited simulation runs across alternatives to maximise the probability of correct selection.

Multi-fidelity simulation: An approach that combines models of varying detail and computational cost to accelerate optimisation without sacrificing accuracy.

Genetic multi-armed bandit: An algorithmic framework that merges genetic search with bandit-style sampling to balance exploration and exploitation in stochastic environments.

References

1. Genetic Multi-Armed Bandits: A Reinforcement Learning Inspired Approach for Simulation Optimization. IEEE Transactions on Evolutionary Computation (2024).
2. Review on ranking and selection: A new perspective. Frontiers of Engineering Management (2021).
3. Simulation optimization: a review of algorithms and applications. Annals of Operations Research (2015).
4. Multi-Fidelity Simulation Modeling for Discrete Event Simulation: An Optimization Perspective. IEEE Transactions on Automation Science and Engineering (2022).
5. An efficient simulation procedure for the expected opportunity cost using metamodels. Automatica (2023).
6. Simulation–optimization configurations for real-time decision-making in fugitive interception. Simulation Modelling Practice and Theory (2024).