Fuzzy Self-Tuning PSO (FST-PSO) is a swarm intelligence global optimization method [1] based on Particle Swarm Optimization [2].
FST-PSO is designed for the optimization of real- or discrete-valued multi-dimensional minimization problems.
FST-PSO is settings-free version of PSO which exploits fuzzy logic to dynamically assign the functioning parameters to each particle in the swarm. Specifically, during each generation, FST-PSO determines the optimal choice for the cognitive factor, the social factor, the inertia value, the minimum velocity, and the maximum velocity. FST-PSO also uses an heuristics to choose the swarm size, so that the user must not select any functioning setting.
In order to use FST-PSO, the programmer must implement a custom fitness function and specify the boundaries of the search space for each dimension. The programmer can optionally specify the maximum number of iterations and the swarm size. When the stopping criterion is met, FST-PSO returns the best fitting solution found, along with its fitness value. In the case of discrete problems, FST-PSO also returns the probability distributions of the underlying generative model.
FST-PSO can be used as follows:
from fstpso import FuzzyPSO
def example_fitness( particle ):
return sum(map(lambda x: x**2, particle))
if __name__ == '__main__':
dims = 10
FP = FuzzyPSO()
FP.set_search_space( [[-10, 10]]*dims )
FP.set_fitness(example_fitness)
result = FP.solve_with_fstpso()
print("Best solution:", result[0])
print("Whose fitness is:", result[1])
pip install fst-pso
Fuzzy Time Travel PSO (FFT-PSO) is a variant of FST-PSO that explores different optimization scenarios starting from the same initial population where only the particle that lead to the best solution found is randomly changed [3].
FFT-PSO works under the assumption that premature convergence could be prevented by backtracking to the beginning of an optimization and removing the particle that was ultimately responsible for leading the whole swarm to a stalling condition.
FFT-PSO provides the same interface of FST-PSO, with the only exception in the creation of the object. Indeed, in the object initializer the programmer can specify the additional paramter alpha, that is, either the number or iterations (if it is an int) or a percentage (if it is a float) of iterations after that the swarm is rewinded to the initial state and the particle leading to the stall is randomly re-initialized.
FFT-PSO can be used as follows:
from fstpso import FFTPSO
def example_fitness( particle ):
return sum(map(lambda x: x**2, particle))
if __name__ == '__main__':
dims = 10
FP = FFTPSO()
FP.set_search_space( [[-10, 10]]*dims )
FP.set_fitness(example_fitness)
result = FP.solve_with_fstpso(max_iter=200)
print("Best solution:", result[0])
print("Whose fitness is:", result[1])
FST-PSO has been created by M.S. Nobile, D. Besozzi, G. Pasi, G. Mauri, R. Colombo (University of Milan-Bicocca, Italy), and P. Cazzaniga (University of Bergamo, Italy). The source code is written and maintained by M.S. Nobile.
Please check out the Wiki for additional descriptions.
If you need any information about FST-PSO please write to: marco.nobile@unive.it
FST-PSO requires two packages: miniful and numpy.
[1] Nobile, Cazzaniga, Besozzi, Colombo, Mauri, Pasi, "Fuzzy Self-Tuning PSO: A Settings-Free Algorithm for Global Optimization", Swarm & Evolutionary Computation, 39:70-85, 2018 (doi:10.1016/j.swevo.2017.09.001)
[2] Kennedy, Eberhart, Particle swarm optimization, in: Proceedings IEEE International Conference on Neural Networks, Vol. 4, 1995, pp. 1942–1948
[3] Papetti, Tangherloni, Coelho, Besozzi, Cazzaniga, Nobile, "We Are Sending You Back... to the Optimum! Fuzzy Time Travel Particle Swarm Optimization", in: García-Sánchez, P., Hart, E., Thomson, S.L. (eds) Applications of Evolutionary Computation. EvoApplications 2025. Lecture Notes in Computer Science, vol 15613
http://www.sciencedirect.com/science/article/pii/S2210650216303534
https://link.springer.com/chapter/10.1007/978-3-031-90065-5_10