Fault detection filtering for memristive neural networks in the presence of communication constraints (Q6051488)

The authors deal with the memristive neural networks (MNNs) described by the equations \[ \left\{ \begin{array}{ll} x(k+1) = A(x(k))x(k) + W(x(k))g(x(k)) + Cd(k)+Ev(k), \\ y(k) = Bx(k) + Dd(k) + Fv(k), \end{array}\right. \] where \(x(k)\in\mathbb R^{n}\) and \(y(k)\in\mathbb R^{m}\) denote the neural state and the measurement output, respectively, \(v(k)\in\mathbb R^{n_l}\) refers to a fault to be detected, the unknown disturbance \(d(k)\in\mathbb{R}^{n_w}\) belongs to \(\ell_2[0,\infty)\), the nonlinear activation function \(g(x(k))\), \(A(x(k))\) and \(W(x(k))\) are state-dependent interconnection weight matrices, \(B,C,D,E,F\) are matrices subject to suitable dimensions. In contrast to traditional neural networks, MNNs exhibit complex dynamic characteristics, including state-dependent interconnection matrices. With their memory function, MNNs have superior synapse approximation abilities, leading to widespread adoption. Significant progress has been made in developing MNNs in both continuous- and discrete-time domains. The authors of this study investigate the fault detection filtering problem for the discrete-time memristive neural networks in the presence of the dynamic event-triggered mechanism (DETM) and a stochastic communication protocol (SCP). A parameter-dependent DETM is developed to enhance communication efficiency by considering the interaction between the triggering mechanism and dynamic quantization, which is essentially different from the existing results with state-dependent DETM. A unified framework is proposed to regulate whether to broadcast measurements and what measurements to broadcast simultaneously based on the advantages of parameter-based DETM and SCP scheduling, with the principle of reducing communication frequency while avoiding packet collisions. To address the mode mismatch between SCP and filter caused by practical factors, an asynchronous FDF based on hidden Markov models is developed. Based on the Lyapunov theory, sufficient conditions are attained for guaranteeing the stochastic stability of the filtering error system A numerical example is presented to illustrate the effectiveness of the theoretical results.