Kernel Search for the Generalized Assignment Problem

The generalized assignment problem is a classical optimization problem. Its modifications and extensions occur frequently in practical problems in the area of industry, telecommunications, computer networks, logistics, and many others. It is a hard problem, thus an optimal solution to practical instances cannot be calculated in an acceptable amount of time. That is why approximate methods must be applied. The kernel search is one of the recently developed heuristics that exploit a mathematical programming formulation of the problem. It is a general and simple heuristic framework based on the idea of decomposition of the original problem into smaller sub-problems and using a mathematical programming solver as a black-box to solve them. In this paper we propose an implementation of the kernel search algorithm for the general assignment problem and evaluate the efficiency of the algorithm on benchmark instances.