中国科学院自动化研究所机构知识库

The modified H-infinity filtering problem for a class of Markovian jump systems with unknown nonlinear dynamics is investigated in the work by developing the neural event-triggered filter co-design method. Moreover, the true system modes are assumed to be inaccessible such that the estimated jumping modes are utilized for the mode-dependent filters. In particular, a novel event-triggered mechanism is introduced to improve filtering communication efficiency, where the unknown nonlinearity approximation is conducted by a neural network. By virtue of employing Lyapunov-Krasovskii method, sufficient filtering conditions are constructed to ensure the optimal H-infinity performance under the mean-square framework, based on which desired mode-dependent filter gains, event-triggering, and neural network parameters are co-designed with an aid of matrix techniques. Illustrative simulations with two practical examples are finally carried out to validate the usefulness and advantages of our developed approach.