A Review of the Family of Artificial Fish Swarm Algorithms: Recent Advances and Applications

Farhad Pourpanah, Ran Wang, Chee Peng Lim, Danial Yazdani

Abstract

Since its introduction in 2002, the Artificial Fish Swarm Algorithm (AFSA) has established itself as a robust member of the Swarm Intelligence (SI) family. Unlike more rigid optimization methods, AFSA leverages biological behaviors—such as foraging and swarming—to navigate complex search spaces effectively. This review synthesizes the major developments in AFSA research from 2013 to 2022, highlighting how adaptive parameter control, behavioral innovations, and hybridization with other SI methods have addressed the exploration-exploitation trade-off. We examine its performance against standard benchmark functions and its diverse applications in engineering, medical imaging, and network optimization.

Introduction

Optimization serves as the cornerstone of modern computational science, underpinning everything from neural network training to the design of resilient infrastructure. However, many contemporary problems are NP-hard, necessitating heuristic approaches that can find near-optimal solutions within reasonable timeframes. Swarm Intelligence (SI) algorithms, inspired by the collective behavior of social animals, have proven particularly effective in this regard.

The Artificial Fish Swarm Algorithm (AFSA), first proposed by Li et al. in 2002, has seen a significant surge in academic interest over the last two decades. As shown in Figure 1, the volume of research documents related to AFSA has grown steadily, peaking in recent years as researchers seek more diverse and robust alternatives to standard methods like Particle Swarm Optimization (PSO).

Growth of AFSA research publications from 2004 to 2021.

The Core AFSA Framework

The fundamental strength of AFSA lies in its decentralized, bottom-up approach. Each "artificial fish" (AF) in the swarm operates based on local information within a defined "Visual" range. The algorithm typically employs three primary behaviors:

  1. Preying (Foraging): An AF explores its immediate vicinity. If it detects a higher concentration of food (a better objective function value), it moves in that direction.
  2. Swarming: To ensure safety and information sharing, fish tend to move toward the center of their local group, provided the area is not overly congested.
  3. Following: When a neighboring fish discovers a superior food source, others in the vicinity move toward that individual.
The movement mechanism of an artificial fish, showing the relationship between visual range, step size, and the transition from state x_i(t) to x_i(t+1).

As illustrated in Figure 2, the movement of a fish is governed by its current state, its visual range, and its step size. This local interaction allows the swarm to explore the search space dynamically while maintaining a degree of independence that prevents premature convergence.

Recent Advances and Modifications

To overcome limitations such as slow convergence in the later stages of optimization, researchers have introduced several "upgrades" to the base AFSA model:

  • Adaptive Parameter Tuning: Modern AFSA variants often use dynamic functions (linear, exponential, or fuzzy logic) to adjust the "Visual" and "Step" parameters. By starting with a large visual range for global exploration and gradually shrinking it for local exploitation, the algorithm achieves a better balance in the search process.
  • Behavioral Innovations: New behaviors like "Leaping" allow fish to randomly re-initialize their positions if the swarm becomes stagnant, effectively jumping out of local optima. Conversely, the "Swallow" behavior removes underperforming agents to reduce computational overhead and memory usage.
  • Hybridization Strategies: One of the most fruitful areas of research is the merging of AFSA with other SI techniques. For instance, hybrid PSO-AFSA models utilize PSO’s rapid velocity updates for movement while retaining AFSA’s "Crowding Factor" to ensure the swarm doesn't collapse into a single point too early.

Performance and Applications

The efficacy of these modified AFSA models is typically validated using benchmark functions like the Sphere, Ackley, and Griewank functions. Results indicate that advanced variants, such as Normative AFSA (NAFSA), can achieve near-perfect global minima on unimodal functions and maintain high success rates on complex multimodal landscapes where traditional algorithms often fail. For example, on the Sphere function, NAFSA has been recorded achieving a mean error as low as $1.14 \times 10^{-52}$.

Beyond mathematics, AFSA has found practical success in diverse fields:

  • Medical Imaging: Identifying metastasis and segmenting complex biological structures.
  • Fault Diagnosis: Detecting anomalies in industrial machinery and power grids.
  • Wireless Sensor Networks (WSN): Optimizing node layouts for maximum coverage and energy efficiency.

Conclusion and Future Directions

AFSA remains a powerful tool due to its inherent ability to maintain population diversity through the crowding factor and its global "bulletin board" memory. However, challenges remain. The $O(n^2)$ computational complexity of distance calculations makes it less suitable for extremely large swarms, and its performance on high-dimensional problems (thousands of variables) requires further refinement. Future research should focus on establishing more rigorous mathematical proofs for convergence and exploring the potential of AFSA in Large-Scale Optimization Problems (LSOPs).

Summary updated

BibTeX
@article{pourpanah2023review,
  author = {Farhad Pourpanah and Ran Wang and Chee Peng Lim and Danial Yazdani},
  title = {{A Review of the Family of Artificial Fish Swarm Algorithms: Recent Advances and Applications}},
  journal = {Artif. Intell. Rev.},
  year = {2023},
  doi = {10.1007/s10462-022-10214-4},
}
Figure