Process Operational State Assessment Based on Stacked Enhanced Denoising Autoencoder

This paper presents a process operational state assessment method based on a deep neural network-specifically, the Stacked Enhanced Denoising Autoencoder (SEDAE)-tailored for complex industrial production data characterized by strong coupling, nonlinearity, and significant noise interference. The approach initially employs SEDAE to extract deep features closely associated with the operational states from process data. Subsequently, the extracted features are processed using a SoftMax classifier to accomplish the operational state assessment. Finally, the proposed method is validated through simulation using actual production data from the cyanide leaching process. The results demonstrate the superiority of the SEDAE model compared to others, maintaining a leading position under varying noise ratios. Even at a noise ratio of 60%, the SEDAE model sustains an accuracy level of around 90%.