A new method of grouping variables for large-scale global optimization problems

Complexity and dimensionality of real-world optimization problems are rapidly increasing year by year. A lot of real-world optimization problems are complex, thus researchers consider these problems as ‘black box’ models due to the fact that the analysis of the problem is complicated or completely impossible, and partial information about the problem is rarely useful. Heuristic search algorithms have become an effective tool for solving such ‘black box’ optimi- zation problems. In recent decades, many researchers have designed a lot of heuristic algorithms for solving large- scale global optimization (LSGO) problems. In this paper, we proposed an innovative approach, which is called DECC-RAG. The approach is based on an original method of grouping variables (random adaptive grouping (RAG)) for cooperative cooperation framework. The RAG method uses the following idea: after a specified number of fitness evaluation in the cooperative coevolution with the SaNSDE algorithm, we choose a half of subcomponents with the worst fitness values and randomly mix indices of variables in these subcomponents. We have evaluated the DECC-RAG algorithm with 20 LSGO benchmark problems from the IEEE CEC’2010 and on 15 LSGO benchmark problems from the IEEE CEC’2013 competitions. The dimensionality of benchmark problems was equal to 1000. The experimental results have shown that the proposed method of optimization (DECC-RAG) outper- forms some well-known algorithms on the large-scale global optimization problems from LSGO CEC’2010 and LSGO CEC’2013.