Knowledge based recursive non-linear partial least squares (RNPLS).

Area of Science:

• Chemical Engineering
• Data Science
• Process Control

Background:

• Soft sensors are crucial for indirect measurement in industrial plants, but face challenges like data collinearity, non-linearity, and time-varying dynamics.
• Partial Least Squares (PLS) regression is widely used for linear modeling with correlated data, yet struggles with non-linear and dynamic process characteristics.

Purpose of the Study:

• To develop a novel knowledge-based methodology for a Recursive Non-linear Partial Least Squares (RNPLS) algorithm.
• To address the limitations of traditional PLS in handling non-linear and time-varying industrial processes.

Main Methods:

• Implemented a non-linear PLS algorithm using an augmented input matrix with knowledge-based non-linear transformations.
• Employed Recursive Exponentially Weighted PLS to adapt the model to process changes.
• Validated the RNPLS algorithm on a sugar industry evaporation station and a simulated wastewater treatment plant.

Main Results:

• The RNPLS methodology successfully adapted to process changes, demonstrating robustness.
• Accurate predictions of critical variables were achieved by integrating process-specific knowledge.
• The approach effectively managed non-linear and time-varying process dynamics.

Conclusions:

• Knowledge-based RNPLS offers a powerful solution for enhancing soft sensor performance in complex industrial environments.
• The algorithm's adaptability and accuracy make it suitable for real-world applications with evolving process conditions.
• Integrating expert or emulated knowledge is key to successful soft sensor implementation for difficult-to-measure variables.