Minimum entropy filtering for a single output non-Gaussian stochastic system using state transformation (Q6574429)

In this paper, the authors propose a filter design for the single-output stochastic non-linear systems described by the discrete-time equations \N\[\N \left\{ \begin{array}{ll} x_{k + 1} = Ax_{k} + \psi(y_{k}) + f\varphi(d^{\top}x_k)+ Gw_k, \\\Ny_{k} = c^{\top}x_{k} + v_{k}, \end{array}\right. \N\]\Nwhere \(x_k\in\mathbb R^{n}\) is the vector-valued system state to be estimated, \(y_k\in\mathbb R\) is the system output, the non-Gaussian random noise \(w_k\in\mathbb R^{m}\), zero mean-value Gaussian random measurement noise \(v_k\in\mathbb R\), \(\psi: \mathbb R\to \mathbb R^{n}\) denotes a known non-linear vector function with \(\psi (0) = 0\), \(\varphi: \mathbb R\to \mathbb R\) denotes a non-linear function with \(\varphi (0) = 0\), \( d, f , c \in\mathbb R\) stand for known constant vectors, and \(A, G\) are known constant matrices of appropriate dimensions. The non-linear term \(\psi(y_{k})\) can be considered as the generalized inverse dynamic (see [\textit{S.-J. Liu} et al., Automatica 43, No. 2, 238--251 (2007; Zbl 1115.93076)]) of the system which exists in many practical systems. For the described systems the authors propose an entropy-based filter so that the vector-valued unmeasurable states \(x_k\) of these non-Gaussian stochastic non-linear systems can be estimated. The convergence of the presented algorithm is analysed and a numerical example is given to verify the effectiveness of the presented filtering algorithm.