An abstract interface for large-scale continuous optimization decomposition methods

ABSTRACTDecomposition methods are valuable approaches to support the development of divide-and-conquer metaheuristics. When the problem structure is unknown, such as in black-box problems, this structure can be inferred through several decomposition mechanisms. In the context of continuous optimization, the most efficient meta-heuristics to deal with a large number of decision variables involve decomposition methods. However, choosing a suitable decomposition method is not a trivial task since each strategy requires an appropriate set of parameters. In this context, this paper proposes a C++ library called Continuous Optimization Problem Decomposition (COPD) that provides the most recent decomposition methods, interfaces for new methods, and integration with solvers. Furthermore, the proposed library can aid related studies since the decomposition for continuous optimization problems can be easily applied with different methods. The experimental results demonstrate a high grouping accuracy for most methods on large-scale problems.