Distributed distribution-based optimization for sensor fault detection

Faulty sensor data undermine the usability and reliability of sensor networks. Distributed automatous faulty sensor detection is a critical component of a self-managed and sustainable sensor network. In this paper, we present a new distribution-based optimization approach for distributed faulty data detection. We propose a new statistical measurement mutual divergence as the optimization objective function. Based on correlations between related sensor nodes, mutual divergence measures the deviation of a sensor data from its expected value. It captures the behaviors of a broad range of common sensor faults. The distribution-based method is based on the probabilistic collectives theory. It searches for solutions in the form of probability distributions and is robust to noise and uncertainty. Experimental results show that the distribution-based method performs better than other methods with significantly higher detection accuracy.