Co-Evolutionary Sensor Deployment Optimization for Layered Low-Altitude Airspace Monitoring Under Worst-Case Traffic Composition

Layered sensor networks that monitor low-altitude airspace must detect heterogeneous aerial traffic, yet sensor deployment is usually optimized against a single fixed traffic case and can leave an entire target kind undetected when the composition changes. This paper formulates deployment against worst-case traffic composition through a high–low transition (HL) trajectory model, a library of 35 target templates in four kinds, and a per-kind fairness objective f4, the worst per-kind miss rate. An adaptive co-evolution (ACE) framework trains deployments against a co-evolved, kind-diverse population of traffic cases with a hall of fame and a robust final selection, and ACE-A6 instantiates it with a sensor-count dispatcher. Over four scenarios with five, four, three, and two sensors; 30 seeds; and two independent sets of 60 held-out cases, ACE-A6 lowers the worst per-kind miss rate of four state-of-the-art evolutionary optimizers on the four-, three- and two-sensor scenarios with large effect sizes (Cliff’s δ≥ 0.44, Holm-corrected p<0.05). Equal-budget experiments locate the gain in the training framework: each of five optimizers trained inside ACE breaks the f4,max=1 ceiling, and ACE-A6 performs on par with them. The advantage persists under probabilistic detection, masking, position error and moderate library shift.