Hybrid tuning of an evolutionary algorithm for sensor allocation

The application of evolutionary algorithms to the optimization of sensor allocation given different target configurations requires the tuning of parameters affecting the robustness and run time of the algorithm. In this context, parameter settings in evolutionary algorithms are usually set through empirical testing or rules of thumb that do not always provide optimal results within time constraints. Design of experiments (DOE) is a methodology that provides some principled guidance on parameter settings in a constrained experiment environment but relies itself on a final inductive step for optimization. This paper describes a sensor allocation tool developed for intelligence, surveillance and reconnaissance (ISR) in the maritime domain and introduces a hybrid methodology based on DOE and machine learning techniques that enables the tuning of an embedded particle swarm optimization (PSO) algorithm for different scenarios.