A Hybrid Genetic Algorithm for Solving the Unsplittable Multicommodity Flow Problem: The Maritime Surveillance Case

Abstract

Large volume surveillance missions are characterized by the employment and collaboration of several agents processing diverse information sources’ inputs in order to ensure a surveillance task. Given the time dependant relevance of the shared information, an efficient global routing policy needs to be set up to optimize information exchange in the backbone of the surveillance network. We propose to model this problem as a single path multicommodity flow problem, where several commodities are to be shared in a capacitated network. The considered objective function is to minimize the overall network congestion. As the problem is NP-Hard, a hybrid genetic approach is proposed as a solution approach. A greedy search procedure based on the nearest neighbor method is transplanted into the genetic algorithm. The empirical validation is done using a simulation environment called Inform Lab. A comparison to a state-of-the-art ant colony system approach is performed based on a real case of maritime surveillance application and some randomly generated instances. The analysis of the results obtained in the two sets was supported by statistical nonparametric Wilcoxon signed-rank tests. The experimental results show that the hybrid genetic algorithm performs consistently well for large sized problems.