Considerable number of studies confirm a remarkable performance for evolutionary algorithms (EAs); however, this performance deteriorates as EAs face large scale non-separable problems. This paper proposes a new collaborative approach based on a co-evolutionary framework for solving large scale nonseparable problems. Here, collaboration means using more interactive and intelligent particles in a search space for faster but not premature convergence. Proposed ideas for Collaborative Coevolutionary Particle Swarm Optimization (CLCPSO) are summarized in the following three items. First is adding “Skepticism parameter” to redistribute particles with Cauchy and Gaussian distributions, when the algorithm for two sequential runs shows same result. Second is adaptively tuning group diversity to overcome the problem of trapping in local optima, by adding a new particle called “Group energizer“ to call for random topology when the best fit particle has reached a certain age. Third is using a pseudo random number instead of random number for population initialization. Results show that these techniques improve the convergence issue. Application to several CEC2010 benchmarks and comparison against several state-of-the-art approaches such as cooperative co-evolutionary particle swarm optimization (CCPSO) confirm the merits of the approach

Keywords