High accuracy sensor fault detection for energy management applications

Wireless sensor networks (WSNs) have found many applications in building environment monitoring and energy management applications. However, WSN data is not completely reliable and hence its use in energy management has to be carefully considered. Experiments conducted on a WSN test bed monitoring a campus classroom highlight how sensor data is prone to errors, underlining the need to detect faulty measurements from the data. This paper describes the experiments showcasing faulty measurements as well as methods to detect these faulty measurements. A key contribution of this paper is highly accurate fault detection methods developed using two machine learning approaches: neural networks and support vector machine. Experimental results showcasing the successful application of the proposed methods for fault detection are discussed. The paper also provides insights on implementation of the proposed methods as well as its integration into an energy management platform.