Soft Sensing of Key Variables Based on Dual-Enhanced Parallel Autoencoder Model

Recently, data-driven soft sensing technology has been widely used in industrial processes, where autoencoder are used to extract deep features, which play an important role in predicting key variables that are difficult to measure directly. However, existing autoencoder models would take a long time in feature extraction, and the extracted features are difficult to consider both key variable and process variables. To address these issues, this paper proposes a dual-enhanced parallel deep autoencoder (DE-PDAE) model. Different from the standard deep autoencoder model, the proposed model improves the feature extraction performance of deep autoencoder by embedding input variables into hidden layers through supervised learning in the pre-training stage while embedding output variable into decoder layer. Additionally, this model adopts a data parallel training strategy, which can reduce the training time. The experimental results on the sulfur recovery unit process verify the effectiveness and feasibility of the proposed model.