Distributed Machine Learning for Cyber-Physical Systems

Wireless sensor networks (WSN) are increasingly used for environmental monitoring over extended periods of time. To facilitate deployments in remote areas, sensor nodes are typically small, solar-powered devices with limited computational capabilities. Over the duration of the deployment, harsh weather conditions can lead to problems like mis-calibration or build-up of dust on sensors and solar panels, leading to incorrect readings or shorter duty-cycles and thus less data. Our goal is to automatically learn the normal system behaviour and to use this model to detect anomalies. In our approach, sensor nodes participate in a distributed recurrent neural network, where each of the sensor nodes hosts a few units and communicates only with its local neighbours. Our online learning is a variant of backpropagation-decorrelation (BPDC) learning [1] with intrinsic plasticity [2] (IP). In a similar setting, we have proposed a distributed fault detection [3] based on echo state learning [4], but the offline learning approach is computationally too demanding to be directly executed on sensor nodes. Our new Spatially Organised and Distributed Backpropagation-Decorrelation (SODBPDC) architecture and learning algorithm (Section II) is suited for directly learning on sensor nodes because of a smaller memory footprint than echo state learning during training. In Sect. III, we present results of an application of SODBPDC to fault detection in sensor network data.