Rapid convergence to optimality through repeated collective learning and flying of swarms

We propose a new hierarchical collective learning (HCL) strategy for particle swarm optimization. The algorithm shows better performance than popular PSOs on standard benchmark functions. The algorithm has three components: a) initialization b) create exemplars c) fly under guidance of exemplars created in previous step. The HCL strategy is applied for the first two parts. In collective learning (CL), an exemplar is created by selecting the best combination of values for every dimension from the group members for the fittest one. In hierarchical part, this process is repeated for some levels, where in every level, the exemplars from the last level act as the member for the next higher level. During flying the initial population flies towards the final exemplar in proportion their fitness. These processes i.e. “create exemplars” and flying is repeated till a stop criterion is met. The process of flying reduces the common problems “oscillation” and “two steps forward one step backward” in PSO.