Enhanced comprehensive learning particle swarm optimization

• ECLPSO overcomes the low solution accuracy weakness of CLPSO.
• ECLPSO performs better than OLPSO-L.
• ECLPSO is easier to implement than OLPSO-L.
• ECLPSO requires much less memory space than OLPSO-L for many real-world large-scale problems.

Comprehensive learning particle swarm optimization (CLPSO) is a state-of-the-art metaheuristic that encourages a particle to learn from different exemplars on different dimensions. It is able to locate the global optimum region for many complex multimodal problems as it is excellent in preserving the particles’ diversity and thus preventing premature convergence. However, CLPSO has been noted for low solution accuracy. This paper proposes two enhancements to CLPSO. First, a perturbation term is added into each particle’s velocity update procedure to achieve high performance exploitation. Normative knowledge about dimensional bounds of personal best positions is used to appropriately activate the perturbation based exploitation. Second, the particles’ learning probabilities are determined adaptively based on not only rankings of personal best fitness values but also the particles’ exploitation progress to facilitate convergence. Experiments conducted on various benchmark functions demonstrate that the two enhancements successfully overcome the low solution accuracy weakness of CLPSO.