Predefined-time decentralized dynamic surface control for interconnected system.

A decentralized dynamic surface control approach with predefined-time convergence guarantees is developed for high-order interconnected systems subject to model uncertainties and external disturbances. This strategy ensures that subsystem tracking errors reach a vicinity of the equilibrium point within a user-defined time interval. Specifically, neural networks approximate the unknown interconnection effects using local subsystem states and the reference states of coupled subsystems. This method removes the conventional requirement that interconnections must satisfy a matching condition and possess a known upper bound. To circumvent the repeated differentiation of virtual control signals inherent in traditional backstepping, a predefined-time nonlinear filter is introduced. This modification not only prevents potential control-input chattering but also establishes an explicit bound on the first derivative of each virtual control signal. Furthermore, leveraging predefined-time stability theory, a compensation mechanism mitigates performance degradation caused by filtering errors during the transient phase. A novel neural network disturbance observer, grounded in the universal approximation principle, is proposed to handle both plant model uncertainties and unknown disturbances concurrently. Numerical simulations and a practical application case study ultimately validate the effectiveness of the presented approach.