Particle Swarm Optimisation

PSO

Particle Swarm Optimisation is a new natural algorithm developed by Russell Eberhart and James Kennedy. For more information, please see some of the many PSO web pages:

• Yuhui Shi's Particle Swarm Optimizer page
• Chilukuri K. Mohan's Particle Swarm Optimization Homepage

The Applet

This applies PSO to two of De Jong's test functions (F1 and F5) in two dimensions. You can choose the test function, population size, and three algorithm parameters, as well as set the random seed. A population of 20 suffices for F1, whereas 50 is probably better for F5. A good choice of maximum velocity seems to be 0.5 for F1 and 4.0 for F5. I tend to leave c1 and c2 set to 2.0, altho' feel free to experiment. Constant c1 influences how much each particle is attracted to its previous best position, constant c2 how much it is attracted by the best ever position.

When the applet displays, each particle is shown with a tail indicating its velocity (once they start to move!). If you choose to display their past best positions, then a grey 'ghost' will indicate this for each particle. (Don't expect to see many of these with F1 - it's all downhill!). The best ever position is shown in red, and the target (optimum position) as a blue cross - the particles don't know where this is, of course! Oh, and to let you see what's going on, the algorithm runs in slow motion - it sleeps for 100ms after every move. The source code is available here.