Improving the accuracy of auto service simulation models through three-level vehicle arrival generation

The article discusses the task of creating a vehicle arrival generator for discrete-event simulation models of an auto service enterprise. This generator aims to improve the accuracy of auto service simulation models by utilizing machine learning technologies and neural networks. The paper presents a three-component generator for simulation models within the ANYLOGIC environment. The work justifies the choice of a specific ML model and the DeepLearning4J neural network. Increased efficiency is achieved by using three complementary methods for generating vehicle arrivals into the model during simulation. Thus, the generator accounts for the random nature of vehicle arrivals, seasonal fluctuations in arrivals (including dependence on weather conditions), as well as dependencies on time of day, calendar weekends, and holidays. The generator is implemented by loading relevant libraries from the internet and program code written in Java, embedded into the Anylogic program. A weighted assessment is used for the intelligent selection of the best forecasting method. An experimental reduction in forecasting error for flow intensity is proven.