Source-linked AI summary
A comprehensive review of firefly algorithms
Iztok Fister, Iztok Fister, Xin-She Yang, Janez Brest
TL;DR
The paper addresses how the firefly algorithm can be applied across diverse optimization and engineering problems despite problem-dependent performance. It comprehensively reviews FA applications and variants, finding broad practical applicability while identifying limited theoretical convergence analysis as an open boundary.
Problem
The original FA may require modification or hybridization to address diverse optimization problems, while its theoretical convergence behavior remains insufficiently analyzed.
Method
The paper conducts a comprehensive review of FA applications, algorithm characteristics, variants, and development directions.
Results
FA is reported as suitable for multi-modal, function, and combinatorial optimization, with applications spanning practically every problem domain.
Takeaways & Limitations
FA is presented as a simple, flexible, and versatile general problem solver and local search heuristic for diverse real-world problems.
Takeaways & Limitations
Theoretical analyses of FA convergence and stability remain very limited, despite rapid convergence often observed in real applications.
Abstract
from arXiv · showhide
The firefly algorithm has become an increasingly important tool of Swarm Intelligence that has been applied in almost all areas of optimization, as well as engineering practice. Many problems from various areas have been successfully solved using the firefly algorithm and its variants. In order to use the algorithm to solve diverse problems, the original firefly algorithm needs to be modified or hybridized. This paper carries out a comprehensive review of this living and evolving discipline of Swarm Intelligence, in order to show that the firefly algorithm could be applied to every problem arising in practice. On the other hand, it encourages new researchers and algorithm developers to use this simple and yet very efficient algorithm for problem solving. It often guarantees that the obtained results will meet the expectations.
1. Introduction
Swarm intelligence studies self-organized, decentralized collective behavior and informs optimization methods such as the firefly algorithm. This review presents FA applications, characteristics, and directions for extending its use across engineering and real-world problems.
- Swarm intelligence: Swarm intelligence models coordinated behavior emerging from simple individuals interacting through local information.It is concerned with self-organized and decentralized systems.
- Firefly algorithm: FA is a stochastic, nature-inspired meta-heuristic that searches for solutions through randomization and local improvement.Randomization helps avoid local optima, while local search improves candidate solutions until no further improvement is detected.
- Firefly algorithm: Meta-heuristic performance depends on balancing exploration of diverse solutions with exploitation near the best solutions.Control parameters affect this balance.
- Firefly algorithm: FA is population-based and can combine building blocks from different candidate solutions through crossover.When both parents share a variable value, the offspring also has that value.
- Review scope: The review surveys FA applications, including engineering and real-world problems, to broaden its potential user base.It also discusses algorithm characteristics, behavior, and future development directions.
2. Firefly algorithm
The firefly algorithm abstracts biological flashing and attraction into a population-based optimization process. Its formulation links fitness, distance-dependent attractiveness, and randomized movement, with parameter settings spanning different search behaviors.
- Biological foundations: FA is inspired by firefly flashing, whose intensity decreases as distance increases and whose signals serve mating and predator-warning functions.The review uses these physical and biological characteristics as foundations for the algorithm.
- Firefly algorithm: The algorithm idealizes fireflies as unisex agents whose attractiveness is proportional to light intensity determined by the fitness landscape.These assumptions simplify the biological model for optimization.
- Algorithm structure: FA initializes a population, evaluates and orders solutions, identifies the best solution, and moves fireflies until the evaluation limit is reached.The search is controlled by the maximum number of fitness function evaluations, MAX FES.
- Algorithm formulation: The standard formulation makes light intensity proportional to fitness and models its distance dependence with an absorption coefficient γ.Attractiveness is modeled analogously and is proportional to perceived light intensity.
- Algorithm formulation: Each firefly moves from its current position toward a more attractive firefly while adding a randomized walk.The movement combines current position, attraction, and randomization; γ strongly affects convergence speed.
- Algorithm behavior: FA spans constant-attractiveness behavior as γ → 0 and random-walk behavior as γ → ∞.The review associates these limiting behaviors with PSO and a parallel version of simulated annealing, respectively.
3. Studies on Firefly Algorithms: Classifications and Analysis
The review classifies firefly algorithms by their components, modification strategies, control scope, and parameter settings. It shows that modified and hybrid variants extend classical FA across problem types, with reported gains that depend on the task and design.
- Classification criteria: Firefly algorithms are classified by what is modified, how modifications are made, and the scope of those modifications.The reviewed components include representation, population scheme, fitness evaluation, best-solution determination, and firefly movement.
- Parameter settings: Parameter control is categorized as deterministic, adaptive, or self-adaptive, while parameter tuning balances exploration against exploitation.The classification also distinguishes how algorithm parameters are selected and controlled during the search.
- Modified and hybrid variants: Classical firefly algorithms are divided into modified and hybrid forms because classical FA mainly addresses continuous optimization problems.Reported modification directions include elitist, binary, Gaussian, Lévy-flight, chaos-based, and parallel algorithms.
- Modified firefly algorithms: Modified FA variants reportedly outperform classical FA on seven benchmark functions, while GPU parallelization is more accurate and faster only for multi-modal functions.The classical algorithm remains well suited to unimodal functions, where few fireflies reduce calculation time.
- Hybrid firefly algorithms: Hybrid variants combine FA with methods including Lévy-flight search, differential evolution, local search, neural networks, learning automata, and genetic algorithms.Examples report improved accuracy, speed, recognition, convergence, exploration, or competitiveness against classical and other optimization methods.
- Comparative findings: The review reports FA superiority over exponential particle swarm optimization in both efficiency and success in extensive experiments.Other reported comparisons include gains over genetic algorithms, particle swarm optimization, evolutionary programming, and classical FA in specific applications.
4. Applications of Firefly Algorithms
The review documents firefly algorithm applications across continuous, combinatorial, constrained, multi-objective, dynamic, noisy, and classification problems, including engineering tasks. Results are generally promising but vary by problem and comparator, motivating modified, hybrid, and parameter-adapted variants.
- Constraint optimization: Constraint-optimization results were mixed: firefly algorithms outperformed several meta-heuristics in structural designs, but were outperformed by particle swarm optimization on 14 benchmark problems.Penalty functions were used to incorporate constraints into the fitness function.
- Combinatorial optimization: 36,000 task-graph scheduling benchmarks showed lower computing time than competing meta-heuristics, but less promising results on other performance measures.
- Other optimization settings: Firefly variants were reported as effective for multi-objective, dynamic, and noisy optimization, including comparisons favoring FA in dynamic environments and at higher noise levels.Variants used adaptive parameters, chaotic maps, self-adaptive mutation, or multi-swarm populations.
- Classification: In classification, hybrid FA-Rough Set feature selection outperformed the compared methods on time and optimality, while FA training of radial-basis-function networks was satisfactory but not superior to PSO and ABC.
5. Discussion and further work
The discussion emphasizes FA’s broad practical applicability, rapid convergence, and dependence on exploration–exploitation balance, while identifying unresolved issues in discrete search, theoretical analysis, and parameter tuning. Hybridization and adaptive or population-based variants are presented as important directions for extending the method.
- Main characteristics: FA is suitable for multi-modal and combinatorial optimization and often converges quickly, but losing population diversity can limit large-scale search.
- Hybridization: The review identifies hybridization as a common strategy, with FA used as a local search or combined with neural networks, learning automata, genetic algorithms, and other heuristics.
- Open algorithmic issues: Combinatorial search can find solutions faster in some cases, whereas in others the solution cannot be found, making exploration–exploitation balance a central development problem.
- Problem-specific modifications: Discrete FA requires mapping permutation-based variables into a continuous search space, while constrained FA commonly represents constraint violations through penalty functions.
- Adaptive variants: Multi-swarm populations and adaptive control parameters are commonly used for dynamic environments, while parameter adaptation and classification-heuristic hybrids support other problem classes.
- Further work: Mathematical analysis of FA remains limited, including a lack of theoretical characterization of convergence speed and stability across major variants.
- Further work: The review calls for better parameter-tuning studies and potentially automatic adaptive schemes to broaden applicability with fewer manual adjustments.
6. Conclusions
The review concludes that the firefly algorithm is a simple, flexible, and versatile optimizer applicable across diverse problem domains. It highlights multimodal capability, fast convergence, broad solver roles, and extensive real-world applicability while identifying open challenges for further research.
- The firefly algorithm has expanded into practically every application domain since its establishment in 2008.
- The review identifies multimodal characteristics and efficient handling of multimodal problems as core strengths.
- The algorithm has a fast convergence rate and can function as both a general global solver and a local search heuristic.
- Firefly algorithms address continuous, combinatorial, constraint, multi-objective, dynamic, noisy, and classification problems, including engineering applications.
- The review presents FA as simple, flexible, versatile, and efficient for diverse real-world problems, while highlighting open questions for future research.