Dragonfly Algorithm: Swarm Search Rules, Adaptation, and Benchmark Limits

Summary

The article describes the Dragonfly Algorithm, a population-based optimizer that represents candidate solutions as agents in a swarm. Each agent’s movement combines separation from neighbors, alignment with their movement, cohesion toward the group, attraction to the best solution, and repulsion from the worst. Inertia carries forward some prior movement. When an agent is isolated and the food source is out of reach, the method uses a Lévy flight, allowing occasional long jumps alongside smaller steps.

The algorithm adapts its neighborhood radius and movement weights over time to shift from broad exploration toward focused search. The article tests a modified implementation on optimization functions and reports that its performance fell short of expectations; it says the method can get stuck and recommends higher-ranked alternatives when accuracy is important. These benchmarks are not trading-strategy tests. The author also cautions that the implementation may differ from canonical versions, so the results apply to the tested variant and functions rather than every use of the algorithm.

Key ideas

• Each candidate solution moves according to swarm separation, alignment, cohesion, attraction to the best solution, and avoidance of the worst.
• An isolated agent can use a Lévy flight to combine small search steps with occasional larger jumps.
• Adaptive neighborhood radius and weights are intended to move the search from exploration toward exploitation.
• The tested implementation underperformed on the article’s benchmark functions and may become stuck.
• The results concern a modified optimizer and test functions, not trading strategies or all canonical versions of the algorithm.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.