Distributed Fault Detection in Sensor Networks using a Recurrent Neural Network

摘要

In long-term deployments of sensor networks, monitoring the quality ofgathered data is a critical issue. Over the time of deployment, sensors areexposed to harsh conditions, causing some of them to fail or to deliver lessaccurate data. If such a degradation remains undetected, the usefulness of asensor network can be greatly reduced. We present an approach that learnsspatio-temporal correlations between different sensors, and makes use of thelearned model to detect misbehaving sensors by using distributed computationand only local communication between nodes. We introduce SODESN, a distributedrecurrent neural network architecture, and a learning method to train SODESNfor fault detection in a distributed scenario. Our approach is evaluated usingdata from different types of sensors and is able to work well even withless-than-perfect link qualities and more than 50% of failed nodes.