This study is concerned with the problem of ℋ2 state-feedback control design for discrete-time Markov jump linear systems (MJLS), assuming that the transition probability matrix is not precisely known, but belongs to a polytopic domain, or contains unknown or bounded elements. As a first contribution, the uncertainties of the transition probability matrix are modelled in terms of the Cartesian product of simplexes, called multi-simplex. Thanks to this representation, the problem of robust mean square stability analysis with an ℋ2 norm bound can be solved through convergent linear matrix inequality (LMI) relaxations constructed in terms of polynomial solutions. The proposed conditions yield a better trade-off between precision and computational effort when compared with other methods. As a second contribution, new conditions in terms of LMIs with a scalar parameter lying in the interval (− 1, 1) are proposed for ℋ2 state-feedback control with complete, partial or no observation of the Markov chain. Owing to the presence of the scalar parameter, less conservative results when compared with other conditions available in the literature can be obtained, at the price of increasing the associated computational effort. Numerical examples illustrate the advantages of the proposed methodology.

Many practical non-linear systems can be described by non-linear auto-regressive moving average (NARMA) system models, whose stabilisation problem is challenging in the presence of large parametric uncertainties and non-parametric uncertainties. In this work, to address this challenging problem for a wide class of discrete-time NARMA systems, in which there are uncertain periodic parameters as well as uncertain non-linear part with unknown periodic time delays, we develop adaptive predictive control laws using the key ideas of ‘future outputs prediction’ and ‘nearest-neighbour compensation’, among which the former is carried out to overcome the non-causalness problem and the latter novel idea is proposed to completely compensate for the effect of non-linear uncertainties as well as unknown time delays. To achieve the desired asymptotic tracking performance in the presence of semi-parametric uncertainties with time delays, an ‘n-step parameter update law’ is first designed, based on which an ‘one-step update law’ is then elaborately constructed to obtain smoother closed-loop signals. This study in general develops a systematic adaptive control framework for periodic NARMA systems with guaranteed boundedness stability and asymptotic tracking performance, which are established by rigorous theoretic proof and verified by simulation studies.

This study is concerned with model predictive control (MPC) for discrete-time Markovian jump linear systems subject to polytopic uncertainties both in system matrices and in transition probabilities between modes. The multi-step mode-dependent state-feedback control law is utilised to minimise an upper bound on the expected worst-case infinite horizon cost function. MPC designs for three cases: unconstrained case, constrained case and constrained case with low online computational burden (LOCB) are developed, respectively. All of them are proved to guarantee mean-square stability. In the constrained case, the minimisation of the expected worst-case infinite horizon cost function and constraints handling are dealt with in a separate way. The corresponding algorithm is proved to guarantee both the mean-square stability and the satisfaction of the hard mode-dependent constraints on inputs and states. To reduce the computational complexity, an algorithm with LOCB is developed by making use of the affine property of the solution to linear matrix inequalities. Finally, a numerical example is given to illustrate the proposed results.

Stabilisation of sampled-data systems with time-varying and uncertain sampling periods via dynamic output-feedback controllers is considered. Extending the existing discrete-time approaches to the more general setup of output-feedback control, sufficient linear matrix inequalities conditions are developed for the design of linear, constant-parameter controllers. The applicability of the proposed design method is demonstrated through numerical examples, which also indicate improvement with regards to the other approaches.