=========================================================================
YAPSO: Yet Another Particle Swarm Optimization
Author: Maurizio Clemente Center for Automotive Research (CAR) Department of Mechanical and Aerospace Engineering The Ohio State University
Citation: Maurizio Clemente, Marcello Canova, "Quadratic Surrogate Attractor for Particle Swarm Optimization", 2026, arXiv:2603.17163. https://doi.org/10.48550/arXiv.2603.17163
Description:
This class implements an enhanced Particle Swarm Optimizer (PSO) in which
the classical global-best attractor is replaced by the minimizer of a quadratic
surrogate model.
The objective is to mitigate premature convergence and sensitivity to noise
observed in standard PSO, while incurring only minimal computational overhead.
Rather than relying on a single global best attractor, the method leverages
multiple elite particles to construct a surrogate that better reflects first
the global and later the local landscape to identify the optimum.
The improvement is particularly pronounced for quasi-convex functions, where the
surrogate can exploit the underlying convex-like structure of the landscape.
Provided that a sufficient number of particles is available to construct the surrogate
(Np ≥ NQ), the approach remains effective even for higher-order functions, as the local
behavior near the optimum is well approximated by a quadratic model.
Inputs:
obj: N-dimensional objective function to be optimized (callable)
bounds: List of [lower, upper] values defining the search space per dimension
Ns: (Optional) number of independent swarms (multi-swarm configuration)
Np: (Optional) number of particles per swarm
w: (Optional) inertia weight
c1: (Optional) cognitive coefficient (attraction to personal best)
c2: (Optional) social coefficient (attraction to global/surrogate best)
Imax: (Optional) maximum number of PSO iterations
X0: (Optional) Initial particle position
stats: (Optional) True/False enable stats
animate: (Optional) True/False enable (AND SAVE!) runtime animation
min: (Optional) True for minimization, False for maximization
verbose: (Optional) enable console logging and diagnostics
epsilon: (Optional) false-alarm threshold for convergence P(S* >= kappa | H0) <= epsilon
tau: (Optional) exploration gain factor
vmax: (Optional) velocity clipping threshold
ST: (Optional) self-termination True/False (early stopping logic)
QS: (Optional) quadratic surrogate attraction mechanism True/False
Outputs:
opt_x: (ndarray) Location of the optimal solution
opt_f: (float) Value of the optimal solution
if stats enabled: (Optional)
results.swarms: Final swarm states for all Ns swarms
results.obj_arr: Objective function history over iterations
results.y_mean: Mean objective value across swarms per iteration
results.y_min: Minimum objective value registered per iteration
results.y_max: Maximum objective value registered per iteration
results.t: Iteration timeline
results.q25: 25th percentile of objective distribution
results.q75: 75th percentile of objective distribution