Particle Swarm Optimization inspired by starling flock behavior

• New Particle Swarm Optimization algorithm based on the collective response of starlings flock behavior called “Starling PSO.”
• Starling PSO introduces a new method to adjust the position and velocity of particles which will generate new feasible solutions to increase the diversity in the solution space.
• Starling PSO improves the performance of the original PSO and yields the optimal solution in many numerical benchmarking experiments.
• Starling PSO gives the best results in almost all clustering experiments.

Swarm intelligence is a meta-heuristic algorithm which is widely used nowadays for efficient solution of optimization problems. Particle Swarm Optimization (PSO) is one of the most popular types of swarm intelligence algorithm. This paper proposes a new Particle Swarm Optimization algorithm called Starling PSO based on the collective response of starlings. Although PSO performs well in many problems, algorithms in this category lack mechanisms which add diversity to exploration in the search process. Our proposed algorithm introduces a new mechanism into PSO to add diversity, a mechanism which is inspired by the collective response behavior of starlings. This mechanism consists of three major steps: initialization, which prepares alternative populations for the next steps; identifying seven nearest neighbors; and orientation change which adjusts velocity and position of particles based on those neighbors and selects the best alternative. Because of this collective response mechanism, the Starling PSO explores a wider area of the search space and thus avoids suboptimal solutions. We tested the algorithm with commonly used numerical benchmarking functions as well as applying it to a real world application involving data clustering. In these evaluations, we compared Starling PSO with a variety of state of the art algorithms. The results show that Starling PSO improves the performance of the original PSO and yields the optimal solution in many numerical benchmarking experiments. It also gives the best results in almost all clustering experiments.

Figure optionsDownload as PowerPoint slide