A Dynamic Optimization Approach to the Design of Cooperative Co-Evolutionary Algorithms


Cooperative co-evolutionary algorithm (CCEA) decomposes a problem into several subcomponents and optimizes them separately. This divide-and-conquer feature endows CCEAs with the capability of distributed and high-efficiency problem solving. However, traditional CCEAs trend to converge to Nash equilibrium rather than the global optimum due to information loss accompanied with problem decomposition. Moreover, the interactive nature makes the subcomponents' landscapes dynamic, which increases the challenge to conduct global optimization. To address these problems, a multi-population mechanism based CCEA (mCCEA) was proposed to compensate information in dynamic landscapes. The mCCEA is decentralized for each subcomponent since it doesn't need centralized archive or information sharing. It focuses on both the global and the local optima of each subcomponent by maintaining multiple populations and conducting local search in dynamic landscapes. These optima are seen as the current representatives of the subcomponents and used by the other subcomponents to construct their complete solutions for fitness evaluation. Experimental study was conducted based on a wide range of benchmark functions. The performance of the proposed algorithm was compared with several peer algorithms from the literature. The experimental results show effectiveness and advantage of the proposed algorithm.