Niche search: an application in vehicle routing

The authors describe a hybrid strategy for solving combinatorial optimisation problems, obtained by coupling a local search method to an evolutionary algorithm, and they provide an application to a particular variant of the vehicle routing problem. The local search method has been devised specifically for this class of problems. It is based on a composite neighbourhood, which is searched iteratively up to the point where no further improvements are made. The evolutionary structure is the niche search, an algorithm based on the evolution of several independent niches. Niches whose individuals' fitness is good remain, and the others tend to be replaced. The separation of the population into niches allows for a good compromise between intensive search (inside each niche) and diversification (through the separation between the niches). They also describe how they integrate specific problem knowledge into an evolutionary structure, in order to achieve a high performance optimisation algorithm. All the steps that they consider necessary are described in detail: finding an appropriate representation, determining what is a relevant neighbourhood, setting up a local search method and finally integrating the local search into an evolutionary algorithm.