Variable Neighbourhood Search (VNS) is a metaheuristic for solving (combinatorial) optimization problems. It iteratively explores growing neighbourhoods to find better local optima in each iteration.
I'm trying to apply VNS to a problem where a/the solution is represented by a finite vector in which each element is a finite strictly positive scalar (e.g. [1, 2, 6, 22, 6])
What are general guidelines for the design of neighbourhoods? How should neighbourhoods relate to each other? Can the quality of neighbourhoods be somehow measured? What are good references on this topic?
Maybe this could be a reference question.