Particle Swarm Optimization (PSO) in C

An implementation of the Particle Swarm Optimization (PSO) algorithm [1,2] in C that can be "plugged into" your code as a small library. PSO is used for problems involving global stochastic optimization of a continuous function (called the objective function). PSO can also be used for discrete optimization problems, but this behaviour is not implemented in the current version of this library.

There's also an implementation in Go which can be found here

USAGE

Just include pso.h and pso.c in your code - no other dependencies are necessary apart from the standard C library.

In order to use pso_solve(), you need :

1. an objective function to be minimized (see defined type pso_obj_fun_t in pso.h),

2. a pso_results_t object with a properly initialized (malloc'd) gbest buffer. This is where the best position discovered will be stored, along with the minimum error (stored in member error).

3. a pso_settings_t object with properly initialized values -- use pso_settings_new() to allocate and initialize the object. Do not forget to free the object using pso_settings_free(), especially if you're doing multiple optimization attempts over a loop (otherwise memory will leak).

FEATURES

NEIGHBORHOOD TOPOLOGIES

pso provides three different strategies for determining each particle's neighborhood attractor:

1. global topology (PSO_NHOOD_GLOBAL) where each particle is informed by every other particle in the swarm

2. ring topology (PSO_NHOOD_RING) where there exists a fixed ring topology and each particle is informed by its adjacent particles

3. random topology (PSO_NHOOD_RANDOM) where a random topology is used that is updated when the error is not improved in two successive iterations [3,4]. Use nhood_size in pso_settings_t to adjust the average number of informers for each particle.

INERTIA WEIGHT STRATEGIES

The value of the inertia weight (w) determines the balance between global and local search. Two different strategies are implemented:

1. Constant value (PSO_W_CONST) using w=0.7298 [5]

2. Linearly decreasing inertia weight (PSO_W_LIN_DEC) [6]. Use w_max and w_min in pso_settings_t to control the starting and ending point respectively.

OTHER SETTINGS

The following can be set in the pso_settings_t struct:

dim : problem dimensionality which should be equal to the size of the gbest buffer in pso_result_t as well as to the size of the first argument (pointer to a position buffer) of the objective function (pso_obj_fun_t)

size : the number of particles in the swarm (the function pso_calc_swarm_size() is also provided for the automatic calculation of th swarm size based on the problem dimensionality).

range_lo, range_hi : array boundaries for particle positions; these are arrays (double *) whose length is determined by dim. They are allocated and filled with their supplied values automatically by pso_settings_new.

clamp_pos : if TRUE then the position of a particle that has exceeded a boundary is set to the value of that boundary and its velocity becomes 0 (along the dimension where the boundary was exceeded). If it's FALSE, then periodic boundary conditions are enforced.

print_every : if greater than zero then this value specifies how many steps should pass before information about the state of the search is printed on screen

c1, c2 : cognitive and social coefficients respectively. The default values are c1 = c2 = 1.496 [5].

steps : the maximum number of steps to run the algorithm for.

goal : if the objective function returns a value lower than this goal the search will stop

EXAMPLES

A file demo.c with its Makefile are provided for your convenience. demo.c provides instructions on how to setup pso in your application. Type make for building the demo.

DISCLAIMER

Feel free to use the code as you see fit; I accept no responsibility for any damage/catastrophy that might occur as a result :)

REFERENCES

[1] Kennedy J and Eberhart R, "Particle Swarm Optimization." Proc. IEEE Int’l Conf. Neural Networks, vol. 4, pp. 1942-1948, 1995.

[2] Poli R, Kennedy J and Blackwell T. "Particle swarm optimization." Swarm intelligence 1.1 (2007): 33-57.

[3] Xu, R., Wunsch II, D., & Frank, R. (2007). Inference of genetic regulatory networks with recurrent neural network models using particle swarm optimization. IEEE/ACM Transactions on Computational Biology and Bioinformatics (TCBB), 4(4), 681-692.

[4] http://clerc.maurice.free.fr/pso/random_topology.pdf

[5] Clerc, M., & Kennedy, J. (2002). The particle swarm-explosion, stability, and convergence in a multidimensional complex space. Evolutionary Computation, IEEE Transactions on, 6(1), 58-73.

[6] Shi, Y., & Eberhart, R. (1998). Parameter selection in particle swarm optimization. In Evolutionary Programming VII (pp. 591-600). Springer Berlin/Heidelberg.