Multi-fidelity agent model modeling method based on canonical correlation analysis

Abstract

The invention provides a multi-fidelity agent model modeling method based on canonical correlation analysis. Incidence relation between different fidelity models is evaluated by utilizing a canonical correlation analysis method, so that the method can fully fuse and utilize information of different fidelity models in the modeling process, and the model precision and the prediction robustness are ensured to a certain extent. The modeling process comprises the following steps: estimating overall correlation between a high-fidelity model and a low-fidelity model; constructing a difference function between the high-fidelity model and the low-fidelity model; minimizing the error function, and carrying out hyper-parameter optimization; and establishing a prediction expression of the multi-fidelity agent model. The incidence relation between different fidelity models can be utilized in the modeling stage of the variable fidelity model, information of different fidelity models can be fully utilized, and the prediction precision is good.

Description

technical field [0001] The invention belongs to the technical field of agent models, and relates to a multi-fidelity agent model modeling method based on canonical correlation analysis. Background technique [0002] Traditional mechanical design optimization obtains experimental data by building an experimental platform, or relies on a large amount of process knowledge and engineering experience to explore the response law of the mechanical system and improve the design of the existing mechanical system. Traditional design methods based on empirical knowledge have extremely high technical requirements for workers, and the design reliability and optimization degree are low; while traditional design methods based on experiments usually have high experimental costs, difficulty in extracting experimental data, and long design cycles. With the development of mathematical calculation methods, computer simulation technology has shown its great advantages in the field of mechanical ...

AI Technical Summary

Problems solved by technology

Usually, the single-fidelity proxy model method is constructed based on a single high-fidelity sample information. For complex problems with high dimensionality and high nonlinearity, it is difficult to obtain measured data, that is, high-fidelity sample points are still It is difficult to obtain, and it is necessary to establish a variety of different low-fidelity simulation models, so a variable-fidelity proxy model technology that fuses multi-fidelity model information is proposed

However, the existing multi-fidelity proxy models do not consider the overall correlation between different fidelity models in the modeling stage, and cannot make full use of the limited number of high-fidelity and low-fidelity training point information.

Images