Hunger relief networks consist of agencies that work as independent partners within a food bank network. For these networks to effectively and efficiently reduce food insecurity, strategic alliances between agencies are crucial. Agency preference for forming alliances with other agencies can impact network structure and network satisfaction. In this paper, we explore the compatibility and satisfaction achieved by alliances between different agencies. We introduce two agency norms: conservative and diversifying. We develop an agent-based simulation model to investigate alliance formation in a network. We evaluate network satisfaction, satisfaction among different types of agencies, and alliance heterogeneity. We test the statistical significance of satisfaction within a norm and between norms for different agencies. Findings reveal that the ‘diversifying’ norm in the network reduces gaps in satisfaction between strong and weak agencies, ensuring fairness for weaker agencies in the network, whereas the ‘conservative’ norm favors moderate agencies in the network.
Modern transportation research relies on seamlessly integrating traffic signal data with robust network representation and simulation tools. This study presents utdf2gmns, an open-source Python tool that automates conversion of the Universal Traffic Data Format, including network representation, signalized intersections, and turning volumes into the General Modeling Network Specification (GMNS) Standard. The resulting GMNS-compliant network can be converted for microsimulation in SUMO. By automatically extracting intersection control parameters and aligning them with GMNS conventions, utdf2gmns minimizes manual preprocessing and data loss. utdf2gmns also integrates with the Sigma-X engine to extract and visualize key traffic control metrics, such as phasing diagrams, turning volumes, volume-tocapacity ratios, and control delays. This streamlined workflow enables efficient scenario testing, accurate model building, and consistent data management. Validated through case studies, utdf2gmns reliably models complex urban corridors, promoting reproducibility and standardization. Documentation is available on GitHub and PyPI, supporting easy integration and community engagement.
Agent-based simulations for networked anagram games, often taking advantage of experimental data, are useful tools to investigate collaborative behaviors. To confidently incorporate statistical analysis of experimental data into agent-based simulations, it is crucial to conduct sufficient validation for the underlying statistical models. In this work, we propose a systematic approach to evaluate the validity of statistical methods of players' action sequence modeling for networked anagram experiments. The proposed method can appropriately quantify the effect and validity of expert-defined covariates for modeling the players' action sequence data. We further develop a Large Language Model (LLM)-guided method to augment the covariate set, employing iterative text summarization to overcome token limits. The performance of the proposed methods is evaluated under different metrics tailored for imbalanced data in networked anagram experiments. The results highlight the potential of LLM-driven feature discovery to refine the underlying statistical models used in agent-based simulations.
Although Large Language Models get a lot of attention, Generative Artificial Intelligence encompasses a variety of methods such as Generative Adversarial Networks, Variational Autoencoders or Diffusion Models, that all work very differently but are all capable of generating synthetic data. These methods have considerable potential to make simulation studies more efficient, especially through the creation of artificial data sets, automatic model parameterization and assisted result analysis. The aim of this study is to systematically classify generative methods and their applicability in the context of simulation studies. Based on a comprehensive literature review, applications, trends and challenges of generative methods that are used in combination with simulation are analyzed and structured. This is then summarized in a conceptual workflow that shows how and in which phase generative methods can be used advantageously in simulation studies.
Advances in Artificial Intelligence have enabled more accurate and scalable modeling of complex social systems, which depend on realistic, high-resolution population data. We introduce a novel methodology for generating hierarchical synthetic populations using differentiable programming, producing detailed demographic structures essential for simulation and analysis. Existing approaches struggle to model hierarchical population structures and optimize over discrete demographic attributes. Leveraging feedforward neural networks and Gumbel-Softmax encoding, our approach transforms aggregated census and survey data into continuous, differentiable forms, enabling gradient-based optimization to match target demographics with high fidelity. The framework captures multi-scale population structures, including household composition and socio-economic diversity, with verification via logical rules and validation against census cross tables. A UK case study shows our model closely replicates real-world distributions. This scalable approach provides simulation modelers and analysts with, high-fidelity synthetic populations as input for agent-based simulations of complex societal systems, enabling behavior simulation, intervention evaluation, and demographic analysis.
Agent-based modeling and simulation (ABMS) has become one of the most popular simulation methods for scientific research and real-world applications. This tutorial paper explores recent development in the use of artificial intelligence including large-language models and machine learning, and digital twin in ABMS research. Given the different perspectives on ABMS, this paper will start with ABMS basic concepts and their implementation using an online platform called AgentBlock.net.
Transitioning from narratives to formal system dynamics (SD) models is a complex task that involves identifying variables, their interconnections, feedback loops, and the dynamic behaviors they exhibit. This paper investigates how large language models (LLMs), specifically GPT-4o, can support this process by bridging narratives and formal SD structures. We compare zero-shot prompting with chain-of-thought (CoT) iterations using three case studies based on well-known system archetypes. We evaluate the LLM's ability to identify the systemic structures, variables, causal links, polarities, and feedback loop patterns. We present both quantitative and qualitative assessments of the results. Our study demonstrates the potential of guided reasoning to improve the transition from narratives to system archetypes. We also discuss the challenges of automating SD modeling, particularly in scaling to more complex systems, and propose future directions for advancing toward automated modeling and simulation in SD assisted by AI.
We study decentralized multi-agent multi-armed bandits in fully heavy-tailed settings, where clients communicate over sparse random graphs with heavy-tailed degree distributions and observe heavy-tailed (homogeneous or heterogeneous) reward distributions with potentially infinite variance. The objective is to maximize system performance by pulling the globally optimal arm with the highest global reward mean across all clients. We are the first to address such fully heavy-tailed scenarios, which capture the dynamics and challenges in communication and inference among multiple clients in real-world systems. In homogeneous settings, our algorithmic framework exploits hub-like structures unique to heavy-tailed graphs, allowing clients to aggregate rewards and reduce noises via hub estimators when constructing UCB indices; under M clients and degree distributions with power-law index α > 1, our algorithm attains a regret bound (almost) of order O(M^1 -1/αlogT). Under heterogeneous rewards, clients synchronize by communicating with neighbors, aggregating exchanged estimators in UCB indices; With our newly established information delay bounds on sparse random graphs, we prove a regret bound of O(M logT). Our results improve upon existing work, which only address time-invariant connected graphs, or light-tailed dynamics in dense graphs and rewards.
Many simulation projects are less than successful because "analysts" view simulation modeling as a complicated exercise in computer programming. This is probably caused by their education being limited to vendor training or an undergraduate simulation course that focuses on how to use a particular simulation-software package. Unsuccessful projects also result from lack of real-world experience in performing simulation studies. In this tutorial we discuss 23 critical pitfalls that can cause a simulation project to result in failure. These pitfalls fall into four categories: (1) modeling and validation, (2) simulation software, (3) modeling system randomness, and (4) design and analysis of simulation experiments.
This paper addresses challenges in creating ontologies for the semiconductor supply chain. Ontologies are crucial for seamless data exchange within the semantic web, enabling initiatives like GAIA-X and CatenA-X. Traditionally, ontology creation is complex. Here, we propose a novel AI-assisted method using large language models (LLMs) like ChatGPT 4 Turbo to support human experts. This collaboration aims to expedite ontology generation while maintaining quality. While initial tests show promise, refining the human-AI interface for clear content generation remains a focus. By improving this collaboration, we expect to create more accurate and complete ontologies, fostering efficient information sharing and strengthening the meaningfulness of standards within the semiconductor supply chain.
This tutorial serves as an introductory guide to Monte Carlo tree search (MCTS), a versatile methodology for sequential decision making under uncertainty through stochastic/Monte Carlo simulation. MCTS gained notoriety from its pivotal role in Google DeepMind's AlphaZero and AlphaGo, hailed as major breakthroughs in artificial intelligence (AI) due to AlphaGo defeating the reigning human world Go champion Lee Sedol in 2016 and the world's top-ranked Go player Ke Jie in 2017. AlphaZero, without requiring any domain-specific knowledge beyond the game rules (tabula rasa), achieved remarkable success by surpassing previous benchmarks in Go and outperforming leading AI opponents in chess (Stockfish) and shogi (Elmo) after just 24 hours of MCTS-driven reinforcement learning. We demonstrate the building blocks of MCTS and its performance through decision trees and the game of Othello, and provide an empirical simulation study for the latter.
Photovoltaic (PV) systems are pivotal in the global energy transition, where accurate solar power forecasting is critical. Traditional forecasting has leaned heavily on solar irradiance data, yet such reliance carries inherent uncertainties and measurement complexities, presenting significant forecasting challenges. This paper introduces a novel hybrid/ensemble model that reduces dependence on solar irradiance data, utilizing geographic, meteorological, and temporal data to predict solar power output. Combining the strengths of XGBoost and LightGBM algorithms through a linear regression meta-model, our approach demonstrates improved prediction accuracy, evidenced by a mean absolute error (MAE) of 0.033, and an R-squared value of 0.693. This study advances solar power forecasting, enhancing PV system efficiency, and reliability, and promoting sustainable energy investments.
This paper presents an introductory tutorial based on the Simulation Exploration Experience (SEE) 2024, highlighting a collaborative effort by NASA, SISO and international academic partners to model lunar facilities and habitats through the High-Level Architecture (HLA) for distributed simulations. Focused on federating simulations of lunar infrastructure, this paper outlines methodical steps for creating and executing models that incorporate communication systems and 3D visualizations to support educational and research initiatives in space exploration. Reflecting on SEE 2024’s advancements, the tutorial emphasizes significant progress in using simulation technology to promote innovation and collaboration across various scientific disciplines. This contribution, intended for discussion at the Winter Simulation Conference (WSC) 2024, showcases the role of HLA runtime infrastructure (RTI) in enabling realistic and interoperable simulation environments, enriching the discourse on simulation education.
Contemporary parcel delivery companies face a significant surge in demand, along with increased customer expectations for flawless and timely delivery. They must meet these expectations within a shrinking window of opportunity in an increasingly competitive world while dealing with various micro and macro level uncertainties. Current industry practice relying on localized analysis to meet these expectations has turned out ineffective. This paper argues that imparting adaptiveness and resilience to parcel delivery network is the key to a pragmatic solution. It presents a holistic approach based on simulatable digital twins and composable agents to enable “in silico” business experimentation wherein a set of what-if scenarios are simulated to help evaluate efficacy of current strategy and identify suitable modifications to the strategy if necessary. The paper illustrates the proposed approach on a case study from the parcel industry and demonstrates its utility and efficacy on a set of real-life scenarios.
In recent times, eco-industrial parks (EIP) have taken on a significant role in addressing environmental challenges and supporting sustainable practices. To enhance the efficient utilization of energy within the park and ensure the achievement of production targets among its members, the implementation of energy and production scheduling is imperative. This paper addresses this optimization challenge by formulating it into a constraint programming (CP) model. The optimized scheduling solutions generated by the CP model will be directly used to guide the operations within the simulation environment of the EIP. The paper compares the CP results with the results of another method which we developed in our previous research stage. The outcomes of CP solver demonstrate its effectiveness in optimizing energy utilization and meeting production targets. This research contributes to developing a practical decision support system for resolving real-life scheduling problems in industry parks.
For the purpose of computing the expected value of a stochastic optimization problem via simulation, we propose a method for efficiently constructing importance sampling distributions using surrogate modeling. A software implementation of the methods called SMAIS is available on github. We use this software in experiments to demonstrate that our method can outperform Monte Carlo simulation. We also show good parallel efficiency for up to 16 processors allowing a speed up of more than 10. Our method uses adaptive sample sizes so it is not very sensitive to sample size parameters.
This paper introduces methodologies aimed at predicting the cycle time of robotic arm spot welding operations, which are essential for vehicle body assembly process plans. Predicting the cycle time of robot arm is crucial for process plan, as it is closely linked to overall production efficiency and safety considerations. However, it is common for companies in the vehicle body assembly industry to rely on rough estimates for cycle time prediction. We propose methodologies that ensure ease of use and accuracy based on simulation data to cope with this problem. This paper provides an overview of the overall process of each methodology, including data collection and model construction. Additionally, experiments are conducted to compare the performance of each methodology, with results indicating that our proposed approach outperforms conventional methods. Through this research, we found the potential for the development of advanced methods applicable in the industry.
Epidemic models have become increasingly useful, especially in the wake of the recent COVID-19 pandemic, emphasizing the crucial role of human behavior in the spread of disease. There has been a recent rise in the usage and popularity of generative artificial intelligence (GenAI), such as ChatGPT especially with its ability to mimic human behavior. In this study, we demonstrate a novel application of GenAI for epidemic modeling. We employed GenAI for creating agents living in a hypothetical town in simulations and simulating their behavior within the context of an ongoing pandemic. We performed a series of simulations to quantify the impact of agent traits and the availability of information for health condition, virus, and government guidelines on the disease spread patterns in terms of peak time and epidemic duration. We also characterized the most influential factors in agents' decision-making using random forest model.
Multistage stochastic optimization problems appear commonly in various disciplines in operations management and system control. While computational methods have been studied, they are mostly restricted to two-stage problems or otherwise require strong assumptions. In this paper, we propose a simple data-driven solution technique and uncertainty quantification method for these problems, based on natural multisample generalizations of the well-known sample average approximation and the so-called single replication procedure. Under the assumptions of stagewise independence and the use of parameterized policies, we justify statistical consistency and a coverage guarantee on bounding the optimality gap using our approaches. Our developments entail the establishment of several new statistical properties of the so-called multisample $U$-process that closely connect to multistage stochastic optimization.
This paper presents an algorithmic framework for solving unconstrained stochastic optimization problems using only stochastic function evaluations. We employ central finite-difference based gradient estimation methods to approximate the gradients and dynamically control the accuracy of these approximations by adjusting the sample sizes used in stochastic realizations. We analyze the theoretical properties of the proposed framework on nonconvex functions. Our analysis yields sublinear convergence results to the neighborhood of the solution, and establishes the optimal worst-case iteration complexity (O(epsilon(-1))) and sample complexity (O(epsilon(-2))) for each gradient estimation method to achieve an epsilon-accurate solution. Finally, we demonstrate the performance of the proposed framework and the quality of the gradient estimation methods through numerical experiments on nonlinear least squares problems.
