Multilayer Online Sequential Reduced Kernel Extreme Learning Machine-Based Modeling for Time-Varying Distributed...

Area of Science:

• Engineering
• Machine Learning
• Control Systems

Background:

• Industrial dynamic processes often exhibit time-varying and nonlinear characteristics.
• Accurate modeling of distributed parameter systems (DPSs) is crucial for control and optimization.
• Existing methods may struggle to capture both spatial and temporal dynamics effectively.

Purpose of the Study:

• To develop an online spatiotemporal modeling approach for time-varying DPSs.
• To address the challenges of nonlinearity and time-varying behavior in industrial processes.
• To propose a deep learning framework for efficient DPS modeling.

Main Methods:

• A multilayer online sequential reduced kernel extreme learning machine (ML-OSRKELM) was developed.
• The approach utilizes stacked online sequential reduced kernel extreme learning machine autoencoders (OSRKELM-AEs) for dimensionality reduction.
• An online sequential reduced kernel extreme learning machine (OS-RKELM) is employed for temporal modeling and spatiotemporal reconstruction.

Main Results:

• The ML-OSRKELM effectively translates spatiotemporal data into a low-dimensional temporal domain.
• The kernel trick and support vector selection optimize nonlinear learning and reduce redundant information.
• The sequential update scheme allows for real-time parameter adaptation.

Conclusions:

• The proposed ML-OSRKELM approach provides accurate and efficient online spatiotemporal modeling for time-varying DPSs.
• Experimental validation on a lithium-ion battery's thermal process demonstrates the model's excellent performance.
• This method offers a promising solution for capturing complex dynamics in real-world industrial applications.