An Intelligent Simulation Result Validation Method Based on Variational Autoencoder

Validation for complex system simulation faces the challenges of a large amount of validation data, which comes in the form of extensive samples and prolonged time series. To address this issue, this paper proposes an intelligent simulation result validation method based on Variational Autoencoder (VAE), achieving model validation through intelligent feature extraction and feature consistency metrics. Initially, we present the overall validation framework, which consists of feature extraction and simulation result validation, and detail the steps of the validation process. Subsequently, the principles and procedures of the feature extraction and simulation result validation methods are thoroughly discussed. The feature extraction is achieved by learning the distribution parameters of the latent space using a VAE model. Feature differences across various aspects are measured by three distance metrics and are further converted into simulation credibility. Finally, the effectiveness and superiority of this method are demonstrated through a case study on validating a drone delivery model.