Hao DENG a soutenu sa thèse de doctorat le vendredi 16 octobre 2026 : « Robust Control for Echo State Networks with Guaranteed Stability »

Due to a simplified training structure and reservoir computing technique, echo state networks (ESNs) have attracted increasing attention in the fields of sequential modeling and time series prediction. However, when implementing the echo state networks for sequential tasks, its robustness and stability remain a major concern. The main goal of this PhD thesis is to develop control synthesis design conditions for echo state networks with guaranteed stability as well as robustness toward uncertainties. This PhD thesis first proposes a fault-tolerant control method for systems with implicit structure, where the fault tolerance is based on the reconfiguration technique with an additional block constructed to reconfigure the faulty system such that it exhibits nominal system behavior. Furthermore, robustness with respect to polytopic uncertainties is also considered for implicit systems. The second part of this PhD thesis focuses on stability analysis for echo state network systems and its variant leaky-integrator ESNs. An incremental input-to-state stability criterion is developed for echo state networks, together with the corresponding control design conditions, through linear matrix inequalities. Besides, different control methods such as state feedback control and infinite horizon model predictive control are implemented for echo state networks. Additionally, robustness with respect to polytopic uncertainties is guaranteed for the considered echo state networks. The last part of the PhD thesis considers implicit echo state networks. Extensive stability analysis such as incremental input-to-state stability is conducted for implicit echo state networks, and polytopic uncertainties are further added. A comparative study between the implicit echo state network and the standard echo state network is developed, where the implicit structure illustrates an enhanced control performance.