SIMULATION: Transactions of The Society for Modeling and Simulation International

Anderson, Taylor; Shepherd, Patrick

doi: 10.1177/00375497251343097pmid: N/A

Liu, Huasheng; Deng, Haoran; Li, Jin; Yang, Sha; Dong, Kui; Zhao, Yuqi

doi: 10.1177/00375497241268740pmid: N/A

Lane-level differences in traffic conditions on urban roads are becoming increasingly significant. To remedy this problem, this study proposes a method for the microscopic traffic simulation calibration problem that considers the complexity of traffic conditions on-road sections and the differences in operating states between lanes. A simulation model was established by collecting actual data. Calibration parameters were determined using sensitivity analysis. A calibration model was built to minimize the relative errors of the roadway efficiency and lane differential indicators. The values of these parameters were obtained using a genetic algorithm (GA). The calibration processes were automated using programming. To assess the reliability of the proposed method, we conducted five sets of comparative experiments focusing on two aspects: calibration methods and algorithm utilization. Results indicate that the proposed method significantly enhances simulation accuracy, particularly in lane-level traffic simulations. In comparison to approaches considering only section-level traffic characteristics and default application software parameters, the proposed method yielded reductions in errors by 3.7%, 5.8%, 6.6%, and 3.2% for simulating lane occupancy rate and cross-section flow. The proposed method demonstrated a simulation error of approximately 5%, while the artificial neural network method was about 7%, validating the effectiveness of the algorithms employed. It can play a crucial role in multilane traffic flow, intelligent driving tests, vehicle–road cooperation, and other related study areas.

Clemen, Thomas; Tolk, Andreas; Clemen, Ulfia A; Glake, Daniel; Günther, Gerrit

doi: 10.1177/00375497241295765pmid: N/A

The enormous complexity of political decisions, especially with regard to crisis situations, requires innovative concepts for decision support. The focus here is always on people’s well-being. Artificial societies based on agent-based simulation models are a fairly new, forward-looking paradigm for this. Digital twins, on the contrary, are one of the most promising enabling technologies for realizing a seamless integration between the virtual and the physical or biological world. In this paper, we propose an architecture that combines these two concepts to develop a research agenda to address research topics that will allow us to provide more effective decision support systems. We reflect on the current state of development in this area and formulate possible future research directions. A particular focus here is on integrating generative artificial intelligence (AI) methods and supporting cross-disciplinary collaboration, as decision support in the political environment is a highly cross-domain task. We conclude this article with a call for action to gain experience with the proposed architecture. We hope that this will encourage greater cross-disciplinary exchange for this important task.

Ramesh, Yenda; Panduranga Rao, MV

doi: 10.1177/00375497241264815pmid: N/A

Statistical model checking (SMC) for the analysis of multi-agent systems has been studied in the recent past. A feature peculiar to multi-agent systems in the context of statistical model checking is that of aggregate queries–temporal logic formula that involves a large number of agents. To answer such queries through Monte Carlo sampling, the statistical approach to model checking simulates the entire agent population and evaluates the query. This makes the simulation overhead significantly higher than the query evaluation overhead. This problem becomes particularly challenging when the model checking queries involve multiple attributes of the agents. To alleviate this problem, we propose a population sampling algorithm that simulates only a subset of all the agents and scales to multiple attributes, thus making the solution generic. The population sampling approach results in increased efficiency (a gain in running time of 50%–100%) for a marginal loss in accuracy (between 1% and 5%) when compared with the exhaustive approach (which simulates the entire agent population to evaluate the query), especially for queries that involve limited time horizons. Finally, we report parallel versions of the above algorithms. We explore different strategies of core allocation, both for exhaustive simulations of all agents and the sampling approach. This yields further gains in running time, as expected. The parallel approach, when combined with the sampling idea, results in improving the efficiency (a gain in running time of 100%–150%) with a minor loss when compared with the exhaustive approach in accuracy (between 1% and 5%).

Banihashemi, Sepideh; Veksler, Keren; Abhari, Abdolreza

doi: 10.1177/00375497241298962pmid: 40458739

Analyzing social media networks is crucial for understanding and uncovering common interests and characteristics among users within human societies. In this context, we simulated a simple application of human interaction in social networks, which involves users following others based on text similarity. We then investigated the effects of various machine learning (ML) algorithms employed in the applications to be used as recommendations to decision-making users. A novel agent-based social network simulator called distributed system and multinode processing is developed to assess the parallelization of the ML algorithms (i.e., K-means clustering, cosine similarity, support vector machine, multilayer perceptron) using bag of words (BoW) term frequency-inverse document frequency vectorization by evaluating their performance when executed in parallel across distributed heterogeneous resources. In addition, this simulator compares the effects of BoW with the Doc2Vec model on network structure by observing the differences in detected communities and resulting network graphs when a selected user follows the recommendations produced by an employed algorithm. Three real datasets were used in the experiments: Twitter, Scientific Research Papers, and Retail. This work’s contribution is a unique in-house agent-based simulator developed to analyze the impact of common ML algorithms, including supervised and unsupervised learning, on social networks.