An Intelligent Ranking Evaluation Method of Simulation Models Based on Graph Neural Network

To validate the alternative simulation models and select the most credible one when the models have multivariate and correlated outputs, an intelligent ranking evaluation method of simulation models based on Graph Neural Network (GNN) is proposed. The process of ranking evaluation is divided into three parts: graph structure conversion for evaluation data, feature extraction based on Graph Representation Learning (GRL) and ranking evaluation based on feature distance. A graph structure modeling method is presented to provide the pre-define graph structure for further GRL primarily. Next the interdependencies and dynamic evolutionary patterns among variables are captured by GNN so that the graph representations of evaluation data can be obtained. Then ranking evaluation is achieved by similarity measurement of the graph representations. In the end, the effectiveness of the proposed method on feature extraction of evaluation data and simulation models ranking is illustrated through an application example on a prediction model for aerodynamic parameters of a certain flight vehicle.