The main challenges of adaptability of swarm intelligence algorithms

Abstract

Analyzed three swarm intelligence algorithms, namely Ant Colony Optimization (ACO), Bee Colony Optimization (BCO), Particle Swarm Optimization (PSO) and the adaptability of these algorithms to a dynamic environment. Firstly, the ACO algorithm was analyzed, the behavior of ants in nature, the purpose of the algorithm, and its shortcomings in a dynamic environment. Then the existing modifications of this algorithm to changing environments were investigated, namely AСO with dynamic pheromone updating (AACO), ACO with adaptive memory (ACO-AP), ACO with multi-agent system (MAS-ACO), ACO with machine learning algorithms (MLACO). The advantages and disadvantages of these modifications are also discussed in detail. The software tools that implement the functionality of this algorithm, such as AntTweakBar, AntOpt, EasyAnt have been mentioned. These software tools provide an opportunity to develop new modifications of the ACO algorithms and to study existing ones. Furthermore, the capabilities of the BCO algorithm were clarified and the behavior and parameters of this algorithm were described, its pros and cons in a dynamic environment were investigated. The following BCO modifications were considered: Group Bee Algorithm (GBA), Artificial Bee Colony (ABC), and open source software: PySwarms, PyABC. The third part of the article investigates the work of the PSO algorithm, its advantages and disadvantages of adaptation to dynamic environments. Dynamic Particle Swarm Optimization with Permutation (DPSO-P), Dynamic Multi-swarm Particle Swarm Optimization Based on Elite Learning (DMS-P50-EL) are considered as modifications of PSO to adapt to dynamic environments. The libraries for work such as SciPy, DEAP, PyGAD, Particleswarm, JSwarm (has a wide API and well-written documentation), Dlib have been mentioned. Finally, a comparative table with the most important properties (resistance to environmental changes, complexity of implementation, the possibility of using for a UAV swarm, etc.) for all three algorithms was created, a brief description of similar articles comparing algorithms of swarm intelligence was also made, and the conclusions of the study were drawn.

Keywords

• Ant Colony Optimization (ACO)
• Bee Colony Optimization (BCO)
• Particle Swarm Optimization (PSO)
• Adaptation to dynamic environments