Simulation modeling

Simulation modeling is a research method in which a real-world system is replaced by its model, which reproduces the system's structure and behavior with sufficient accuracy. In general, a simulation model is understood as a software system (a computer model) used to conduct computational experiments: "running" the model on a computer with different sets of input parameters and analyzing the resulting outcomes.

The goal of such modeling is to understand the properties of the system under study or to evaluate the consequences of various operational strategies without conducting direct experiments on the real-world object. The method is considered a special case of mathematical modeling, assuming that an analytical (formula-based) solution to the problem is either unknown or too complex to obtain. A simulation model reproduces the progression of processes over time, often involving random factors, and allows for obtaining numerical estimates of system characteristics through statistical processing of the results from multiple model runs.

Key Properties and Theorems

Unlike analytical modeling, simulation modeling produces results as a set of numerical realizations, rather than an explicit formula. Therefore, to obtain reliable information about the system under study, it is necessary to conduct a series of experiments with the model and process the results statistically.

Statistical Foundations

• Law of Large Numbers: A fundamental property underlying the method. According to this law, as the number of independent model runs (the number of random realizations) increases, the average values of the output characteristics converge to their theoretical expected values.
• Central Limit Theorem: This result allows for estimating the accuracy of simulation results. It states that the deviation of the average value obtained from $N$ independent runs from the true mean value is approximately proportional to $1/\sqrt{N}$. This makes it possible to construct confidence intervals for the obtained estimates.
• Little's Law: In queueing theory, this law ($\bar{L}=\lambda \bar{W}$) relates the average number of customers in a system ($\bar{L}$), their arrival rate ($\lambda$), and the average time they spend in the system ($\bar{W}$). It serves as an important tool for the verification (correctness checking) of simulation models of queues.

Main Modeling Paradigms

Several approaches are distinguished in simulation modeling, differing in their level of abstraction and the type of process being considered.

• Discrete-Event Simulation (DES): Focuses on events that change the system's state at discrete points in time (e.g., a customer's arrival or the completion of a service). This is the dominant paradigm in operations research for analyzing queueing systems, supply chains, and production lines.
• System dynamics (System Dynamics, SD): Operates with aggregate variables and models the continuous dynamics of systems by solving systems of differential equations. It is suitable for analyzing complex socio-economic systems.
• Agent-Based Modeling (ABM): Considers a system as a collection of autonomous objects (agents), each acting according to a set of predefined rules. The global behavior of the system emerges from the interactions of many agents. It is used for modeling social systems, epidemics, and markets.

In modern practice, multi-paradigm tools (e.g., AnyLogic) are often used, allowing these approaches to be combined within a single model.

Examples

Example 1: M/M/1 Queueing System

Let's consider a system with a single service channel, a Poisson arrival process with rate $\lambda$, and an exponential service time with rate $\mu$ ($\lambda <\mu$). It is analytically known that the average queue length in a steady state is $L={\rho }^2/(1-\rho )$, where $\rho =\lambda /\mu$. A simulation model of such a system, after running a sufficient number of arrivals, will show an empirical average queue length close to the theoretical value. For example, with $\lambda =2$ and $\mu =3$, the theoretical value is $L=4$. The simulation result might yield $\bar{L}\approx 3.98$, which confirms the model's correctness.

Example 2: Monte Carlo Method for Calculating π

The Monte Carlo method is a special case of simulation modeling. To estimate the number π, one can inscribe a quarter-circle of radius 1 within a unit square. Then, a large number $N$ of points are randomly "thrown" into this square. The ratio of the number of points that land inside the quarter-circle ($K$) to the total number of points $N$ will be approximately equal to the ratio of their areas: $K/N\approx (\pi \cdot 1^2/4)/1^2=\pi /4$. From this, $\pi \approx 4K/N$. This approach is widely used for solving multidimensional integral problems and in probabilistic analysis.

Application in Operations Research

Simulation modeling is one of the key tools in operations research, especially when a system is too complex for an analytical solution. Traditionally, it was considered a "method of last resort", but the advancement of computing technology has made it a standard and powerful tool for solving problems such as:

• Manufacturing and logistics: analysis of production lines, inventory management, design of warehouses and transportation networks.
• Service industry: optimization of call centers, hospitals (modeling patient flows), and banks.
• Economics and finance: analysis of business processes, risk assessment, modeling of financial asset price movements.
• Digital Twin: a modern approach where a continuously updated simulation model of a specific physical object (a machine tool, a car, a building) is created, which is synchronized with data from the real-world object. This allows for real-time behavior prediction and control optimization.

See also

• Mathematical modeling
• Self-organizing systems
• System dynamics
• Stochastic programming
• Operations research
• Modeling process