Abstract
Congestion controller design for distributed sensor networks deployed within highly dynamic environments constitutes a complex problem due mainly to inherent network-induced time-varying delays. Conventional linear time-invariant models and controller design techniques can produce unacceptable behavior in such scenarios; the same holds true when the problem is treated as slowly time-varying and/or piecewise time-invariant. In this paper, a novel genetic algorithm that enables one to design buffer level controllers that are robust against such time-varying delays is proposed. Its features include a high mutation rate for increased exploratory behavior, a very aggressive selection criterion for fast convergence and a simulation-based fitness evaluator. The effectiveness of the proposed algorithm is demonstrated via several simulation experiments.
