Sequential change detection/isolation with stochastic energy constraints

This paper deals with the performance analysis of sensors that are employed to detect and isolate changes in an observation sequence and are powered by renewable energy sources. The nature of energy arrival process is stochastic and the available energy needs to be utilized efficiently. The model we consider is an extension of the one by Geng and Lai [1] in which they study optimal power allocation policies for detecting a change in an observation sequence. For this we will first derive an asymptotic lower bound on worst mean detection delay for any detection/isolation algorithm and an arbitrary power allocation scheme. Then we will show that Matrix CUSUM algorithm combined with an immediate power allocation scheme achieves this lower bound and hence is asymptotically optimal. We also provide numerical simulations to augment the theoretical results.