Olive Algorithm: ORS Optimization

• Olive Algorithm is a biologically inspired meta-heuristic that models Olive Ridley sea turtle hatchling survival to drive robust optimization strategies.
• It employs a dual-phase approach integrating environmental and trajectory impacts to update solution velocities based on dynamic parameters.
• Benchmark tests demonstrate ORS’s competitive performance against state-of-the-art methods with low variance across classical and engineering problems.

The term "Olive Algorithm" has been used to denote distinct research advances across different domains. In meta-heuristic optimization, the Olive Ridley Survival (ORS) algorithm—also called the "Olive Algorithm"—models biological principles from the survival process of Olive Ridley sea turtle hatchlings to drive robust and competitive optimization strategies. Separately, algorithmic and hardware co-design under the name OliVe presents a high-performance quantization method for deep learning accelerators. This entry focuses on the Olive Algorithm as introduced in ORS, outlining its biological inspiration, mathematical modeling, computational flow, benchmark performance, and current research context.

1. Biological and Algorithmic Foundations

The Olive Algorithm (ORS) derives its design from empirical studies of Olive Ridley sea turtle hatchlings, where survival rates are dominated by severe environmental hazards—only 1 in 1,000 typically reaches open water. The model abstracts each candidate solution as a "hatchling," with its quality (fitness) reflecting not just its static parameter values, but its dynamic momentum in traversing the solution space. The analogy is operationalized through two coupled algorithmic phases at each generation:

• The environmental-impact phase, modulating solution velocity as a function of simulated sand temperature, emergence order, and time-of-day;
• The trajectory-impact phase, introducing path curvature and obstacle avoidance as curvilinear velocity updates (Panigrahi et al., 2024).

This dual-phase framework enables a balance between stochastic exploration (diversification) and directed search (intensification) using interpretable operators analogous to real ecological scenarios.

2. Mathematical Modeling and Algorithmic Flow

Each population element is encoded as a tuple $h_i = (m_i, \vec v_i)$, where $m_i$ is a scalar "mass" and $\vec v_i \in \mathbb{R}^d$ is a velocity vector in $d$-dimensional space. The solution fitness is defined as $f_i = m_i \|\vec v_i\|$. The algorithm begins with random uniform initialization of masses and velocities within problem-specific bounds.

Phase I (Environmental Impact):

Velocity updates aggregate three effects:

• Sand temperature: $\vec v^{\,t+1}=\omega_1\ \vec v^t$ for $S_{temp} \le T_{tol}$, $\vec v^{\,t+1}=\omega_2\ \vec v^t$ for $T_{tol} < S_{temp} < T_{max}$, $\vec v^{\,t+1}=-\infty$ for $S_{temp} \ge T_{max}$.
• Emergence order: Early: $k\,\vec v^t + k_1$, Middle: $k\,\vec v^t$, Late: $k\,\vec v^t - k_2$.
• Time of day: Piecewise scaling by $\omega_3, \omega_4, \omega_5$ within three day intervals.

Their sum is stochastically weighted by $p_1 \sim U(0,1)$: $$\Delta\vec v_{env} = \Delta\vec v_{temp} + \Delta\vec v_{em} + \Delta\vec v_{time}$$, $r_1 = p_1\,\Delta\vec v_{env}$.

Phase II (Trajectory Impact):

The curvilinear path operator computes velocity and angular updates between consecutive path points, also stochastically weighted: $\Delta v_{traj} = v_B^{t+1} - v_A^t$, $r_2 = p_2 \Delta v_{traj}$.

The net update $\Delta\vec v_{res} = r_1 + r_2$ governs fitness adjustment.

Fitness Update Policy:

Survival is based on normalized fitness factor $S_f^t(i)$. If $S_f^t(i) < \tau$, the velocity is increased by $\Delta\vec v_{res}$ and the best-so-far $\vec v_{best}$, otherwise decreased. Here, $\tau = 0.3$ by default.

3. Algorithm Structure and Parameterization

Pseudocode for one generation is as follows (simplified for exposition):

for t in range(1, T): for i in range(n): Δv_env = env_impact(...) Δv_traj = traj_impact(...) Δv_res = r1 + r2 if S_f[i] < tau: v_i = v_i + Δv_res + v_best else: v_i = v_i - Δv_res + v_best f_i = m_i * ||v_i|| update S_f[i] update h_opt if f_i improves

Main parameters:

• $n$: population size (30–100 typical)
• $T$: maximum iterations (500–2000)
• $m_i$, $\vec v_i$: initialized by uniform random sampling within prescribed bounds
• Environment scalars: $\omega_{1\dots 5}$, $k$, $k_1$, $k_2$
• $p_1$, $p_2$: uniformly sampled in $(0,1)$ each iteration
• $\tau$: threshold for the fitness update policy

Parameter selection is problem-centric; continuous parameters are sampled, discrete ones set by empirical tuning (Panigrahi et al., 2024).

4. Benchmark Performance and Comparative Evaluation

The Olive Algorithm was rigorously evaluated on 14 classical 30-dimensional functions from CEC 2005/2008/2010 and 10 complex CEC 2019 benchmarks.

Key results from (Panigrahi et al., 2024) are summarized below:

Benchmark	ORS (Mean ± StdDev)	Best Rival (Method: Mean ± StdDev)
Sphere ($F_1$)	$1.22\times 10^{-1} \pm 1.39\times 10^{-1}$	GWO: $231.0 \pm 19.3$
Rastrigin ($F_8$)	$1.55\times 10^{-1} \pm 1.04\times 10^{-1}$	WOA: $10.17 \pm 4.18$
Griewank ($F_{10}$)	$6.18\times 10^{-3} \pm 3.13\times 10^{-3}$	WOA: $2.04 \pm 0.25$
Rosenbrock ($F_5$)	(not explicitly shown)

In 12 out of 14 tests, ORS achieved the lowest mean and standard deviation. For CEC 2019, it outperformed other algorithms on CEC01 & CEC02, tied on CEC03 & CEC10, and was suboptimal on CEC04–09. Wilcoxon signed-rank testing confirmed the significance (p < $10^{-6}$, except CEC03 where the difference was not significant).

Engineering benchmarks—including pressure vessel, welded beam, and spring design—were solved to match or surpass published solutions.

5. Analysis of Exploration, Exploitation, and Limitations

The Olive Algorithm's curvilinear trajectory operator facilitates exploitation by fine-tuning motion around the current best. Meanwhile, the environment-induced random perturbations (sand temperature, emergence timing, diurnal cycles) supply robust exploration, allowing efficient escape from local minima and enhanced global search capability.

Strengths explicitly documented (Panigrahi et al., 2024):

• Outperforms or ties with seven state-of-the-art meta-heuristics on major test suites.
• Clear interpretability of operators and parameter effects.
• Applicability across both unconstrained and constrained optimization tasks.
• Low empirical variance on standard benchmarks.

Limitations include:

• Suboptimal (premature) convergence on certain highly multimodal or rotated functions from CEC 2019, highlighting insufficient diversity maintenance in these settings.
• Need for problem-specific tuning of environmental parameters to leverage optimal performance.

A plausible implication is that hybridization with additional diversity control operators or evolutionary neighborhood sampling could further enhance ORS performance on complex multimodal landscapes.

6. Context in Meta-Heuristic Research and Prospective Directions

The Olive Algorithm advances meta-heuristic optimization paradigms via biologically faithful abstraction, integrating both intrinsic (trajectory) and extrinsic (environmental) noise. It complements leading population-based algorithms such as Grey Wolf Optimizer (GWO), Whale Optimization Algorithm (WOA), and Differential Evolution (DE), but offers a distinctive two-phase momentum update.

The clear documentation of parameter impact, rigorous statistical benchmarking, and application to classical engineering challenges position ORS as a credible candidate for future research in interpretable, adaptively tunable stochastic optimizers.

Potential future directions include:

• Enhanced parameter adaptation for automated tuning across problem classes.
• Integration with constraint-handling strategies for expanded engineering design applicability.
• Expansion of diversity preservation operators for improved multimodal search.

The Olive Algorithm's combination of stochastic and trajectory-based control establishes it as a competitive, biologically inspired optimizer with well-understood mechanisms and performance characteristics (Panigrahi et al., 2024).