A novel class of non-Gaussian system performance assessment and controller parameter tuning methods.

Area of Science:

• Control Systems Engineering
• Statistical Signal Processing
• Non-Gaussian Data Analysis

Background:

• Traditional control performance assessment (CPA) and controller parameter tuning (CPT) methods often overlook non-Gaussian external disturbances.
• This oversight limits their effectiveness in real-world systems subjected to complex noise environments.

Purpose of the Study:

• To develop novel CPA and CPT methods specifically designed for single-input single-output (SISO) systems affected by non-Gaussian disturbances.
• To introduce a unified framework for robust control performance evaluation and tuning under non-Gaussian conditions.

Main Methods:

• Proposed data Gaussianization (inverse) transformation methods to convert non-Gaussian data into virtual Gaussian data by maximizing mutual information via quantile transformation.
• Utilized CARMA model-based recursive extended least squares and least absolute deviation algorithms for identifying virtual Gaussian and non-Gaussian system models.
• Developed a unified framework integrating CPA and CPT for non-Gaussian control systems.

Main Results:

• The proposed data Gaussianization methods effectively transform non-Gaussian data into a virtual Gaussian domain for analysis.
• The CARMA-based algorithms successfully identified system models in both virtual Gaussian and non-Gaussian domains.
• The unified framework provided a consistent benchmark criterion for evaluating performance across different non-Gaussian noises.

Conclusions:

• The novel data Gaussianization (inverse) transformation methods offer a significant advancement for CPA and CPT in non-Gaussian systems.
• The proposed strategy ensures robust controller parameter tuning and provides reliable performance assessment, outperforming traditional methods.
• The developed unified framework is effective for handling non-Gaussian control challenges.