optimization-algorithms-wiki

Grey Wolf Optimizer (GWO)

GWO is a swarm metaheuristic that mimics the social hierarchy and hunting of grey wolves (Mirjalili et al., 2014). The pack ranks its solutions: alpha (α) = best, beta (β) = second, delta (δ) = third, and omega (ω) = the rest, which update toward the geometric center of the three leaders. Source: andrey-dik‘s MQL5 implementation (extended to allow a configurable number of leader wolves beyond the canonical three).

How it works

Three hunting stages — search (random exploration), encirclement (refine around leaders), attack (converge) — are governed by a coefficient a decreasing non-linearly from 2 toward 0:

• Ai = 2·a·r1 − a, Ci = 2·r2 (r1,r2 random in [0,1]).
• When |a| > 1 wolves explore; as a → 0 they exploit, tightening on the best-found “prey.” See exploration-vs-exploitation.

Benchmark (context-relative)

In this article’s early scoring, GWO scored 0.4158, ranking 4th of 6 tested. Reasonable on smooth high-dim Skin, but weakest on the sharp-peaked Forest function and poor on discrete Megacity at scale (perfect 1.0 at 2 vars → 0.022 at 1000). Dik notes “low scalability on discrete and non-differentiable functions.” Earlier, small-field methodology — not directly comparable to the mature %-of-MAX rating (see population-optimization-benchmark).

Related

metaheuristic-optimization · exploration-vs-exploitation · population-optimization-benchmark · no-free-lunch-theorem · andrey-dik · particle-swarm-optimization · artificial-bee-colony