Conditionally Minimax Prediction in Nonlinear Stochastic Systems∗. Issue 11 (2015)

• Record Type: Journal Article
• Authors: Bosov, A.V.; Borisov, A.V.; Semenikhin, K.V.
• Is Part Of: IFAC-PapersOnLine. Volume 48:Issue 11(2015)
• Page Start: 802
• Page End: 807
• Publication Date: 2015
• Subjects: Adaptive Control and Signal Processing for Nonlinear Systems -- Modeling and Identification of Nonlinear Systems; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2015.09.288
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: This paper describes an approach of nonlinear prediction in discrete-time stochastic systems. The method consists of two steps: preliminary statistical estimation of second-order moment characteristics and minimax optimization of structural coeficients. The minimax prediction algorithms are based on semidefinite programming over several kinds of confidence regions for the uncertain mean and covariance matrix. The guaranteed level of the mean square error (MSE) is provided by the minimax theorem on normal correlation.