A Lightweight Alternating Direction Method of Multipliers for Decentralized Event Detection

In this paper we address the problem of decentralized event detection in a large-scale wireless sensor network (WSN). Comparing with existing centralized solutions, decentralized algorithms are superior in energy efficiency and network scalability, and thus fit for the distributed nature of a WSN. We formulate the event detection problem as a linear program, and solve it with the alternating direction method of multipliers (ADMM). Under mild conditions, this iterative algorithm is shown to be fully decentralized. Further, in view of the fact that the communication burden per iteration directly decides the energy consumption of sensor nodes, we simplify the classic ADMM to a lightweight one, which requires much lower communication burden while keeps the global convergence of the classic ADMM. Effectiveness of the proposed algorithm is validated with simulation results.