Soft instrument training and sample supplementing method

Abstract

The invention relates to a soft instrument training and sample supplementing method, which comprises the following steps: firstly, using originally acquired data as a training sample of VAE to obtain a hidden layer variable z of the VAE; secondly, taking a decoding part of the VAE as a generator G of the WGAN, sampling from a hidden variable z, and taking as an input of the G to generate a new sample; and finally, inputting the real sample and the generated sample into a discriminator D, judging the difference between the real sample and the generated sample, and training the WGAN by optimizing an objective function to obtain a sample closest to real data. A new training sample is generated for the soft instrument by using the model, a data set is supplemented, the problem of insufficient original training data is solved, the training precision of the soft instrument is improved, the generated data sample has the highest similarity with a real sample, and the model has the highest convergence speed.

Description

technical field [0001] The invention belongs to the technical field of machine learning, and in particular relates to soft instrument training and sample supplementation methods, a deep generation model VA-WGAN based on variational auto-encoders (Variational auto-encoders, VAE) and generative adversarial networks (GAN) , a new structure and training method for generative models. And on this basis, apply this model to generate new training samples for the training of industrial soft instruments, supplementing the training data set. Background technique [0002] In the industrial process, there are many key variables closely related to product quality and production efficiency. However, because many industrial sites are in harsh environments such as high temperature, high dust, or corrosiveness, it is difficult to measure these key variables with hardware equipment. In addition, such The high manufacturing and maintenance costs of hardware measurement equipment also hinder th...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the deficiencies of the above-mentioned prior art, the purpose of the present invention is to provide soft instrument training and sample supplement method, generate new samples by learning the distribution of real data, to supplement the training data set of soft instrument, solve the problem of insufficient original training data , optimize the training process, improve the convergence speed, and improve the training accuracy of the soft instrument

Images