Spanning the pareto front of a counter radar detection problem
Hans J.F. Moen, Harald Hovland · 2011
Radar system design and optimization are complex problems recently cast in the framework of multi-objective evolutionary algorithms. However, in the problem of counter radar detection and tracking, the state-of-the-art multi-objective optimization algorithm NSGA-II is unable to span the complete 2D Pareto front of the asymptotic and convex problem domain, leaving out vital information on the radar-jammer system dynamics. Common modifications to the domination principle employed will to some degree increase the span of the Pareto front, at the expense of slower convergence and a less dense front. In this paper, the new Surface Evolutionary Algorithm (SEA) is introduced to overcome these problems. The SEA characterizes all solutions by one single metric and uses interpolated attraction points along the boundary of the solution set as basis for selecting and evolving solutions in the optimizer. The SEA is proposed and analyzed in the context of the conflicting multi-objective optimization criteria of search efficiency, density distribution and span of the complete Pareto front of the counter radar detection problem. The SEA is shown to produce high performance solutions not easily obtained using the well-established optimization methods of NSGA-II and e-MOEA.