Multi-grid cellular genetic algorithm for optimizing variable ordering of ROBDDs

This paper presents a cellular genetic algorithm for optimizing the variable order in Reduced Ordered Binary Decision Diagrams. The evolution process is inspired by a basic genetic algorithm. The population evolves on a bidimensional grid and is implicitly organized in geographical clusters that present a form of structural similarity between individuals. Two feature functions are used to measure the similarity between chromosomes. The approach considers multiple parallel evolving grids. A similarity based communication protocol between clusters of individuals from parallel grids is defined. The exchange of genetic material proves to considerably boost the quality of the solution. The extensive experimental evaluation uses difficult classical benchmarks and proves the efficiency and the stability of the algorithm. The approach systematically produces better results than the used basic genetic algorithm and better or similar results with other heuristic methods.