Dynamic event-triggered distributed filtering design for sensor networks against spoofing attacks

This paper studies dynamic event-triggered distributed filtering for discrete-time sensor networks under spoofing attacks and measurement nonlinearities modeled by a saturation sector condition. First, spoofing on sensor and filter channels is modeled by independent Bernoulli sequences. A radial basis function neural network, trained offline using attack-free simulated trajectories, predicts nominal sensor/filter signals. The prediction residual is used to capture attack-induced deviations and compensate corrupted information. A nonlinear model is introduced to capture measurement saturation, and a boundedness condition is established. Moreover, based on this framework, a dynamic event-triggered scheme is designed to reduce redundant transmissions. The H ∞ performance of the filtering-error system is formulated in the ℓ 2 -gain sense and analyzed via a Lyapunov function, and filter gains are obtained by solving linear matrix inequalities. In the reported simulation, the proposed method reduces the overall root mean square error from 0.632169 to 0.273090 and the number of triggering events from 150 to 94 relative to the no-compensation baseline. Additional tests examine training mismatch, attack probabilities and attack-generation patterns. These finite-horizon results illustrate the potential of the approach within the scenarios considered.