解析のハイブリッドニューラルネットワークによる近似器の高精度化

抄録

In recent years, there has been a need for shorter analysis techniques in order to reduce the time required for design optimization. For this purpose, approximation by polynomial regression, which is a mathematical representation of the analytical structure, can be considered, but the difference from the real structure is large and the error is large. In addition, an approximator based on machine learning requires a lot of time to create training data, so a method that can be evaluated accurately even with a small amount of training data is essential. Therefore, by combining a neural network approximator that mimics polynomial regression and a neural network trained on errors with unknown governing equations, we have shown that the accuracy can be improved even with a small amount of training data.