Adaptive quantile control for stochastic system. (April 2022)

• Record Type: Journal Article
• Authors: Ma, Xuehui; Qian, Fucai; Zhang, Shiliang; Wu, Li
• Is Part Of: ISA transactions. Volume 123(2022)
• Page Start: 110
• Page End: 121
• Publication Date: 2022-04
• Subjects: Adaptive quantile control -- Asymmetric Laplace Distribution -- Bayesian quantile sum estimator -- Certainty equivalence principle; Engineering instruments -- Periodicals; Engineering instruments; Periodicals; Electronic journals; 629.805
• DOI: 10.1016/j.isatra.2021.05.032
• Languages: English
• ISSNs: 0019-0578

Abstract

Abstract: Adaptive control has been successfully developed in deriving control law for stochastic systems with unknown parameters. The generation of reasonable control law depends on accurate parameter estimation. Recursive least square is widely used to estimate unknown parameters for stochastic systems; however, this approach only fits systems with Gaussian noises. In this paper, the adaptive quantile control is first proposed to cover the case where stochastic system noise follows sharp and thick tail distribution rather than Gaussian distribution. In the proposed approach, the system noise is modeled by the Asymmetric Laplace Distribution, and the unknown parameter is online estimated by our developed Bayesian quantile sum estimator, which combines recursive quantile estimations weighted by Bayesian posterior probabilities. With the real-time estimated parameter, the adaptive quantile control law is constructed based on the certainty equivalence principle. Our proposed estimator and controller are not computationally consuming and can be easily conducted in the Micro Controller Unit to fit practical applications. The comparison with some dominant controllers for the unknown stochastic system is conducted to verify the effectiveness of the adaptive quantile control. Highlights: Adaptive quantile control for the stochastic system with sharp and thick tail noise is proposed. The sharp and thick tail noise is modeled by Asymmetric Laplace Distribution. Bayesian quantile sum estimator combines different quantile estimated parameters weighted by Bayesian posterior probability. The control law is based on the certainty equivalence principle.