Enhancing Observer Performance with Gaussian Processes: A Noise-Resilient High-Gain Approach

This paper introduces the Gaussian Process (GP) as a robust and versatile technique to address inherent data noise, a ubiquitous challenge in practical control applications. GP employs a probabilistic modeling approach to effectively handle data noise. When applied to mitigate high-frequency noise in a high-gain extended state observer (ESO), GP considers both underlying data trends and noise-induced uncertainty. This integration is realized through the GP's covariance structure, capturing process variability and noise characteristics. The Gaussian Process predictive distribution provides mean and variance estimations. For high-frequency noise scenarios, the distribution exhibits increased variance, effectively quantifying prediction uncertainty. This characteristic enhances comprehension of high-frequency noise's impact on predictions. The GP-based approach presented here shows considerable potential for managing high-frequency noise and enhancing control performance. The experimental results presented in this paper offer significant evidence of the efficacy of the proposed algorithm.