An Improved Locally Weighted PLS Based on Particle Swarm Optimization for Industrial Soft Sensor Modeling

Area of Science:

• Chemical Engineering
• Process Control
• Data Science

Background:

• Soft sensors are crucial for industrial quality and safety.
• Traditional global models degrade over time due to process changes, requiring frequent maintenance.
• Locally Weighted Partial Least Squares (LWPLS) offers a just-in-time learning approach for soft sensors.

Purpose of the Study:

• To develop an optimized LWPLS method for industrial soft sensor modeling.
• To address the critical impact of the bandwidth parameter (h) in LWPLS.
• To enhance prediction performance and model robustness in dynamic industrial environments.

Main Methods:

• Proposed a two-phase bandwidth optimization strategy combining Particle Swarm Optimization (PSO) with LWPLS.
• Evaluated the PSO-LWPLS method using a numerical simulation and an industrial application case.
• Compared the performance against traditional global methods and LWPLS with fixed bandwidth.

Main Results:

• The proposed PSO-LWPLS method demonstrated superior prediction performance compared to traditional methods.
• PSO-LWPLS outperformed LWPLS with a fixed bandwidth.
• The method showed significant advantages in scenarios with changing data density.

Conclusions:

• The PSO-LWPLS strategy effectively optimizes soft sensor models for industrial applications.
• This approach provides a more robust and accurate solution for real-time process monitoring.
• The optimized LWPLS method is particularly beneficial for handling dynamic industrial process variations.