Modern simulation model design involves a wide range of specialists from various fields. Additional resources are also required for the development and debugging of software code. This study is aimed at demonstrating the capabilities of large language models (LLM) applied at all stages of creating and using simulation models, starting from the formalization of dynamic systems models, and assessing the contribution of these technologies to speeding up the creation of simulation models and reducing their complexity.The model development methodology includes stages of formalization, verification, and the creation of a mathematical model based on dialogues with LLMs. Experiments were conducted using the example of creating a multi-agent community of robots using hybrid automata. The results of the experiments showed that the model created with the help of LLMs demonstrates identical outcomes compared to the model developed in a specialized simulation environment. Based on the analysis of the experimental results, it can be concluded that there is significant potential for the use of LLMs to accelerate and simplify the process of creating complex simulation models.
Keywords: Simulation modeling, large language model, neural network, GPT-4, simulation environment, mathematical model
The modern cycle of creating simulation models is not complete without analysts, modelers, developers, and specialists from various fields. There are numerous well-known tools available to simplify simulation modeling, and in addition, it is proposed to use large language models (LLMs), consisting of neural networks. The article considered the GPT-4 model as an example. Such models have the potential to reduce costs, whether financial or time-related, in the creation of simulation models. Examples of using GPT-4 were presented, leading to the hypothesis that LLMs can replace or significantly reduce the labor intensity of employing a large number of specialists and even skip the formalization stage. Work has been conducted comparing the processes of creating models and conducting experiments using different simulation modeling tools, and the results have been formatted into a comparative table. The comparison was conducted based on the main simulation modeling criteria. Experiments with GPT-4 have successfully demonstrated that the creation of simulation models using LLMs is significantly accelerated and has great perspective in this field.
Keywords: Simulation modeling, large language model, neural network, GPT-4, simulation environment, mathematical model