Subjects: Information Science and Systems Science >> Control science and technology submitted time 2024-01-05
Abstract: In the realm of soft sensing, missing data frequently occurs during the journey from data collection to application, significantly diminishing model accuracy. This paper introduces a filling model based on the Variational Autoencoder (VAE) and GRU neural network. Validation through industrial processes confirms the accuracy of the imputed data. Experimental results demonstrate that the VAE imputation model yields an RMSE and MAE of 3.396% and 2.458% for missing rates of 10%, and 3.549% and 3.078% for missing rates of 30%, respectively. Compared to alternative imputation algorithms like PCA and SVD, the VAE model exhibits significantly enhanced performance, affirming the feasibility of this model.
Peer Review Status:Awaiting Review