We consider the problem of decentralized sequential change detection, in which K sensors monitor a system in real time, and at some unknown time there is an anomaly in the environment that changes the distribution of the observations in all sensors. The sensors communicate with a fusion center that is responsible for quickly detecting the change, while controlling the false alarm rate. We focus on two families of decentralized detection rules with minimal communication requirements. First, we assume that each sensor runs a local CUSUM algorithm and communicates with the fusion center only once, when it detects the change. The fusion center then declares that a change has occurred when m of the K sensors have raised an alarm. Assuming that all sensors have the same signal strength, we show that the asymptotic performance of these one-shot schemes is free of m to a first order, but decreases with m to a second-order, suggesting that the best strategy for the fusion center is to detect the change with the first alarm. Second, we consider schemes that detect the change when m of the K sensors agree simultaneously that the change has occurred. While a first-order asymptotic analysis suggests that it is optimal for the fusion center to wait for all sensors to agree simultaneously, a second-order analysis reveals that it can be better to wait fewer (but more than half) of the sensors to agree. The insights from these asymptotic results are supported by a simulation study.
[1]
J. Kiefer,et al.
Asymptotically Optimum Sequential Inference and Design
,
1963
.
[2]
Yongguo Mei,et al.
Information bounds and quickest change detection in decentralized decision systems
,
2005,
IEEE Transactions on Information Theory.
[3]
G. Lorden.
PROCEDURES FOR REACTING TO A CHANGE IN DISTRIBUTION
,
1971
.
[4]
Steven A. Orszag,et al.
CBMS-NSF REGIONAL CONFERENCE SERIES IN APPLIED MATHEMATICS
,
1978
.
[5]
E. S. Page.
CONTINUOUS INSPECTION SCHEMES
,
1954
.
[6]
Alexander G. Tartakovsky,et al.
Nearly Optimal Change-Point Detection with an Application to Cybersecurity
,
2012,
1202.2849.
[7]
G. Moustakides.
Optimal stopping times for detecting changes in distributions
,
1986
.
[8]
A. Gut.
Stopped Random Walks: Limit Theorems and Applications
,
1987
.
[9]
M. Woodroofe.
Nonlinear Renewal Theory in Sequential Analysis
,
1987
.
[10]
Howard S. Burkom,et al.
Statistical Challenges Facing Early Outbreak Detection in Biosurveillance
,
2010,
Technometrics.
[11]
George V. Moustakides,et al.
Decentralized CUSUM Change Detection
,
2006,
2006 9th International Conference on Information Fusion.
[12]
V. Veeravalli,et al.
Asymptotically Optimal Quickest Change Detection in Distributed Sensor Systems
,
2008
.