Efficient event detection using self-learning threshold for wireless sensor networks

Abstract

Event detection is an essential issue for wireless sensor networks research. Energy saving and reliable detection are major challenges for the resource constraints on sensor nodes. To give attention to both of them, this paper presents a distributed event detection approach using self-learning threshold to fully exploit the energy-reliability tradeoff in wireless sensor networks. In the proposed approach, a stream of real-valued sensor readings is mapped into symbol sequences in order to reduce data dimensionality and simplify event description. A dynamic conversion granularity is adopted to improve the effectiveness of symbolic representation. Then the anomaly probabilities of symbol sequences are estimated through Markov model, and sensor nodes participating in the event detection make local decisions in a distributed manner based on the learned anomaly detection threshold. A timer-based node sleep scheduling is developed to prolong network lifetime during the detection process. Subsequently, the final detection decision is made by a bitwise voting based on the local decisions. A comprehensive set of simulations demonstrate that the proposed approach achieves considerable energy conservation while maintaining fast and accurate event detection.