Generalized Correntropy Criterion-Based Performance Assessment for Non-Gaussian Stochastic Systems

Area of Science:

• Control Systems Engineering
• Statistical Signal Processing
• Industrial Automation

Background:

• Effective control loop performance assessment (CPA) is critical for industrial systems.
• Existing CPA methods and indicators have limitations, particularly for non-Gaussian systems.
• Accurate estimation of system parameters and disturbance noise probability density functions (PDFs) is challenging.

Purpose of the Study:

• To propose a novel evaluation method for CPA in non-Gaussian stochastic systems.
• To introduce generalized correntropy criterion (GCC) as a robust assessment index.
• To develop an efficient algorithm for estimating system parameters and noise PDFs.

Main Methods:

• Summarized shortcomings of existing CPA methods.
• Proposed a novel evaluation method based on generalized correntropy criterion (GCC).
• Developed a hybrid estimation of distribution algorithm (H-EDA) incorporating a 'wading across the stream' approach for parameter and PDF estimation.

Main Results:

• Generalized correntropy criterion (GCC) effectively characterizes non-Gaussian statistical properties.
• GCC can be directly used as an assessment index, even when system output is unknown.
• The hybrid-EDA algorithm accurately and quickly estimates system parameters and disturbance noise PDFs.
• Simulations validated the effectiveness of the proposed algorithm and GCC for a feedback control system.

Conclusions:

• The proposed GCC offers a more comprehensive approach to CPA for non-Gaussian systems.
• The H-EDA algorithm provides a faster and more accurate method for system identification.
• The novel approach enhances the reliability and efficiency of industrial control loop performance assessment.