Dynamic and Evolutionary Multi-objective Optimization for Target Tracking in Sensor Networks

When large sensor networks are applied to the task of target tracking, it is necessary to successively identify subsets of sensors that are most useful at each time instant. Such a task involves simultaneously maximizing target detection accuracy and minimizing querying cost, addressed in this paper by the application of multi-objective evolutionary algorithms (MOEAs). The objective of maximizing detection accuracy is quantified as the minimization of the posterior Cramer-Rao Lower Bound. NSGA-II, a well-known MOEA, is demonstrated to be successful in obtaining multiple solutions (trade-off points), when compared to a ”weighted sum” approach that combines both objectives into a single cost function. We also explore an improvement, LS-DNSGA, which incorporates periodic local search into the algorithm, and outperforms standard NSGA-II on the sensor selection problem.