Particle swarm optimization using multi-level adaptation and purposeful detection operators

Particle swarm optimization (PSO) algorithm has shown favorable performance on global optimization problems. However, it is prone to premature convergence since a monotonic and static learning model is applied for all particles, which makes PSO unable to deal with different complex situations. Moreover, there is no efficient operator to help population detect some promising positions purposefully if the population has been trapped into potential local optima. To solve the shortcomings, a sophisticated PSO (SopPSO) algorithm based on multi-level adaptation and purposeful detection in this research. Relying on the multi-level adaptation, a particle not only updates its neighbors based on its current fitness landscape but also periodically re-selects target dimensions that the particle learns from its neighbors. The multi-level adaptive strategy applied in individual-level and dimension-level enables PSO to have a more accurate simulation on emergent collective behaviors. Furthermore, a purposeful detection operator based on some historical information is proposed to help the population to jump out of local optima. In addition, a simple local searching strategy is introduced to improve the accuracy of elitist particles. A set of experiment has verified the efficiency of each proposed component. At last, the extensive experimental study on CEC'13 test suites illustrates the effectiveness and efficiency of the modified PSO.