Sequential Detection of an Abrupt Change in a Random Sequence with Unknown Initial State

TL;DR

This paper develops a Bayesian sequential detection method for abrupt changes in IID sequences with unknown initial states, ensuring asymptotic correctness and computational efficiency, and compares favorably to CUSUM in simulations.

Contribution

It introduces a recursive Bayesian approach for change detection with unknown initial distributions, improving accuracy and computational simplicity over previous methods.

Findings

Asymptotic vanishing of incorrect detection probability.

Average delay approaches optimal CUSUM delay after transient.

Method maintains constant per-unit-time computation.

Abstract

The problem of sequentially detecting an abrupt change in a sequence of independent and identically distributed (IID) random variables is addressed. Whereas previous approaches assume a known probability density function (PDF) at the start of the sequence, the problem addressed is the detection of a single change in distribution among a finite number of known 'equal-energy' PDFs, but where the initial and final distributions are not known a priori. A Bayesian multiple hypothesis approach is proposed where (i) unlike previous threshold policies, the minimum cost hypothesis is tracked through time, (ii) under an exponential delay-cost function that satisfies an upper bound determined by the distances between hypotheses, the probability of detecting a change from an incorrect initial distribution asymptotically vanishes with time, (iii) computation is recursive and constant per unit time, and (iv) the unknown initial state gives rise to unavoidable incorrect detections that be made to vanish with a constant test threshold with negligibly small effect on correct detection delay for change times beyond a lower bound. Simulations illustrate the analysis and reveal that average delay approaches that of the optimal CUSUM test after an initial transient period determined by an incorrect detection probability constraint.