Mean-Shift-Based Outliers-Robust Distributed Kalman Filter for Wireless Sensor Network Systems

Outliers can contaminate the communication and measurement processes of many wireless sensor networks (WSNs) systems, which may be induced by environmental disturbances, model uncertainties or mismatches, sensor faults or errors, subnetwork faults, or malicious cyberattacks. Once the distributed Kalman filter (DKF) is used in such WSN systems with outliers for distributed state estimation, the estimation performance can be seriously degraded in each sensor node. To address this problem, this article proposes mean-shift-based outliers screening and fuzzy-trust-based information fusion mechanisms to modify the DKF toward improving resilience and enhancing robustness against outliers. The proposed mechanism is capable of screening out the outliers-contaminated local measurements and estimates and adaptively narrowing the outliers-contaminated information pairs without modifying the original local state estimates to guarantee the fused estimation consistency and accuracy. This design leads to the development of mean-shift-based outliers-robust DKF (MSRDKF) for the state estimation over WSN systems in a distributed manner. It is proved that our proposed MSRDKF can generate mean-square-bounded-error estimation in each sensor node under bounded outlier disturbances, system collective observability, and network connectivity, and the upper bound of estimation error increases with the occurrence probability of outliers. Simulation and experimental results demonstrate the practical validity and superiority of the proposed MSRDKF in suppressing the effects of outliers, as evidenced by significantly lower root-mean-square-error (RMSE) and average-RMSE (ARMSE) compared to the DKF.