Comparative Study of Decomposition and Merging Evolutionary Algorithms for Large-Scale Optimization Problems

Evolutionary Algorithms (EAs) have been shown to be effective when applied to various classes of optimization problems. However, EAs suffer performance degradation as the number of dimensions in a given Large Scale Optimization Problem (LSOP) rises. To improve the performance of EAs for LSOPs, the Cooperative Co-evolutionary (CC) framework has been introduced in the literature. In previous works, novel decomposing and merging methods were proposed for Cooperative Particle Swarm Optimization (CPSO). We propose decomposition and merging variants based on Differential Evolution (DE) and Artificial Bee Colony (ABC) and compare them with the CPSO variants. Two well-known large-scale optimization benchmark problems are used for comparison, i.e., CEC’2010 and CEC’2013 problems of increasing decision variables of up to 2000. The Decomposition Cooperative Differential evolution (DCDE) and Merging-CPSO (MCPSO) variants generally performed best. However, MCPSO saw a greater variance in its performance while DCDE’s performance was more stable. Additionally, Cooperative Artificial Bee Colony (CABC) performed particularly well for the separable functions, outright winning or tying every function. For the CEC’2013 benchmarks, which are designed to more closely resemble real-world applications, DCDE was the top performing algorithm on the partially-separable and nonseparable functions, though Merging-CCDE and CPSO algorithms were also competitive on the partially separable functions. The separable functions saw their bests results with CABC, while Merging-CPSO and Merging-CABC also performed well.