Adaptive control of nonlinear system using online error minimum neural networks.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Control Systems Engineering

Background:

• Extreme Learning Machine (ELM) algorithms are widely used for various machine learning tasks.
• Existing ELM variants like OS-ELM and EM-ELM have limitations in network redundancy and performance evaluation.
• Adaptive control systems require robust algorithms capable of handling environmental changes.

Purpose of the Study:

• To propose a novel learning algorithm, Online Error Minimized-ELM (OEM-ELM), to improve upon existing ELM methods.
• To enhance the identification capability and avoid network redundancy in ELM-based systems.
• To develop an adaptive control system with improved environmental adaptability using the OEM-ELM algorithm.

Main Methods:

• The OEM-ELM algorithm is developed by integrating online learning, network performance evaluation, and dynamic adjustment of hidden nodes.
• The algorithm combines advantages from OS-ELM and EM-ELM to optimize network structure and learning.
• An adaptive control system is designed and implemented using the OEM-ELM algorithm.

Main Results:

• The proposed OEM-ELM algorithm effectively avoids network redundancy compared to traditional ELM.
• The algorithm demonstrates improved identification capabilities and network performance.
• Adaptive control based on OEM-ELM shows enhanced adaptability to environmental changes.

Conclusions:

• OEM-ELM offers a significant improvement over traditional ELM by optimizing network structure and learning.
• The algorithm provides superior control performance and adaptability, particularly for complex systems like the Continuous Stirred Tank Reactor (CSTR).
• OEM-ELM presents a promising approach for advanced adaptive control applications in chemical processes and beyond.