Two-fidelity Data Fusion Modeling with Particle Swarm Sampling

Abstract

Computer simulation and experimental testing are commonly used means of data acquisition in industrial design, so the combination of low-precision data represented by cheap calculation data and high-precision data represented by expensive test data for fusion modeling is a hot topic in practical applications question. In order to improve the quality of the model and control the usage of high-precision data, this paper proposes a double-precision fusion modeling method based on particle swarm sampling. The algorithm uses the Kriging model to model high- and low-precision data, and uses PSO (Particle swarm optimization) to obtain low-precision model information for high-precision modeling. Combined with the characteristics of sufficient amount of low-precision data, high-precision data is expensive and difficult to obtain, the low-precision model modeled after sufficient sampling is used to guide the initial sampling of the high-precision model, and the fusion model is obtained by training with two high-precision and low-precision models, and two The point-adding criterion is used for sequential modeling, and the fusion model is continuously optimized. The algorithm is tested on a standard function and an application problem. The experimental results show that the algorithm improves the quality of the modeling without increasing the amount of high-precision data, which verifies the effectiveness of the proposed algorithm.