Enhancing Kernel Search with Pattern Recognition: the Single-Source Capacitated Facility Location Problem
We introduce Pattern-based Kernel Search (PaKS), a two-phase matheuristic for the solution of the Single-Source Capacitated Facility Location Problem (SSCFLP). In the first phase, PaKS employs a pattern recognition technique to identify an implicit spatial separation of potential locations and customers into subsets, called regions, within which location and assignment decisions are strongly interdependent. In the second phase, PaKS employs an enhanced Kernel Search (KS) heuristic that leverages the interdependencies among the decision variables identified in the first phase. On a set of 112 benchmark instances, consisting of up to 1,000 locations and 1,000 customers, computational results show that PaKS consistently outperforms both a standard KS implementation and the current state-of-the-art heuristic for solving the SSCFLP, as well as CPLEX when run with a time limit. For these instances, PaKS achieved an average gap compared to the best known solution of 0.02%. Experimental results conducted on a large set of new very large test problems, comprising up to 2,000 locations and 2,000 customers, demonstrate that PaKS outperforms both the standard KS heuristic and CPLEX in terms of quality of the solution found, finding the largest number of best solutions, and achieving the smallest average gap.