Discussing the article: "Population optimization algorithms: Bird Swarm Algorithm (BSA)"

New comment
 

Check out the new article: Population optimization algorithms: Bird Swarm Algorithm (BSA).

The article explores the bird swarm-based algorithm (BSA) inspired by the collective flocking interactions of birds in nature. The different search strategies of individuals in BSA, including switching between flight, vigilance and foraging behavior, make this algorithm multifaceted. It uses the principles of bird flocking, communication, adaptability, leading and following to efficiently find optimal solutions.

Bird Swarm Algorithm (BSA) is an exciting bioinspired evolutionary algorithm using swarm intelligence based on social interactions and behavior of bird flocks. Developed by Meng and colleagues in 2015, BSA is a unique optimization approach that combines three key aspects of bird behavior: flight, foraging and vigilance. Among the electronic flocks, where each "bird" has individual tactics and strategies, a unique system of collective interaction is born, filled with algorithmic intelligence and creativity. What is important here is not only personal effort, but also the ability to cooperate, exchange and support each other in pursuit of the common goal of optimization.

Different individuals in the BSA may have different search strategies. Birds can randomly switch between flight, vigilance and foraging behavior. The bionic design algorithm includes foraging based on global and individual fitness. Birds also try to move to the center of the population (which can lead to competition with other birds) or to move away from the flock. Bird behavior includes regular flight and migration, as well as switching between the roles of producer and beggar. In the BSA world, each individual at a given iteration has its own search strategy, making the algorithm multifaceted and capable of exerting its power.

Author: Andrey Dik

 
Which AO converges the fastest (number of FF calculations)? It doesn't matter where it converges to. As long as there is a minimum of steps.
 
fxsaber #:
Which AO converges the fastest (number of FF calculations)? It doesn't matter where it converges to. As long as there are a minimum of steps.
Any of the top 5, they converge very quickly.
 
Andrey Dik #:
Any of the top 5, very quick to converge.

I wish there was a numerical value for fast.

 
fxsaber #:

Too bad there's no numerical value for quickness.

You could do it, make several runs of tests, save the FF values at each epoch, calculate the average improvement at each corresponding epoch. Of course, there will be different values for each number of variables. This is if you get very fussy with numerical indicators of "convergence speed".

In each first test for all three test functions (10 parameters), the Top 5 of the list will be very close to the theoretical maximum already around the 100th epoch (with a population of 50).

 
Andrey Dik #:

Of course, you can do it, do several runs of tests, save the FF values at each epoch, calculate the average improvement at each corresponding epoch. Of course, for each number of variables there will be different indicators. This is if you are very fussy with numerical indicators of "convergence speed".

In each first test for all three test functions (10 parameters), the Top 5 of the list will be very close to the theoretical maximum already around the 100th epoch (with a population of 50).

~5000 FF?

 
fxsaber #:

~5,000 FF?

Yes. Even at 50th epoch will be already around 70-80% of the theoretical max.

Well, this is of course with parameter step 0 (as it is done by me when testing). If the step is different from 0, the convergence is even higher.

New comment