Nonlinear editing method for test data

The invention relates to a nonlinear editing method of test data, which adopts a neural network model combining Transform and GAN to realize nonlinear fitting, reconstruction and automatic feature learning of the test data. Comprising the following steps: constructing a Transform-GAN model; a Transform-GAN training model is adopted, test data are input into a Transform encoder, the data are processed through a self-attention mechanism, position encoding and other technologies, and feature representation of the input data is obtained. Inputting the feature representation into a GAN generator, and optimizing a loss function of a discriminator and the generator to enable the generator to generate a sample with relatively high similarity with real data; and reconstructing data: reconstructing the original test data by using the trained generator. The reconstructed data can be used as new test data for subsequent analysis. The method has wide application prospect and value. 本发明涉及一种试验数据的非线性编辑方法，采用Transformer和GAN相结合