A Supervised Bidirectional Long Short-Term Memory Network for Data-Driven Dynamic Soft Sensor Modeling

Data-driven soft sensors have been widely adopted in industrial processes to learn hidden knowledge automatically from process data, then monitor difficult-to-measure quality variables. However, extract and utilize useful dynamic latent features accurately for efficient estimations remains one of the most important research issues sensor modeling. In this article, a supervised bidirectional long short-term memory (SBiLSTM) is proposed data-driven The SBiLSTM incorporates extended information with moving window up $k$ time steps enhances learning efficiency by architecture. With novel structure, can nonlinear both variables variables, further improve prediction performance significantly. effectiveness network-based model demonstrated through two case studies on debutanizer column an wastewater treatment process. Results show that outperforms state-of-the-art traditional deep learning-based models.