Research on quality prediction of industrial processes based on deep learning with missing data

A quality prediction method based on data interpolation and mixed gate units is proposed to address the quality prediction problem caused by data loss in industrial processes. Firstly, use Generative Adversarial Interpolation Network (GAIN) for data interpolation of time series; Secondly, redesign the loss functions of the GAN generator and discriminator to enable them to learn the original distribution of missing data and effectively fill in the missing parts; Finally, a mixed gate unit (MIXGU) quality prediction model is constructed to predict the quality of the interpolated samples. Through predictive experiments, it can be concluded that the problem of insufficient generalization in different data can be solved, further improving the accuracy of quality prediction.