Sine–Cosine Enhanced Producer Strategy

• The paper introduces a modified SSA where producers use sine–cosine oscillations and a time-adaptive inertia weight to bias movement toward the global best.
• It employs a dual-mode update—sine for larger exploratory leaps and cosine for localized adjustments—to dynamically balance global and local search.
• Experimental results demonstrate accelerated convergence, reduced variance, and improved performance in applications like UAV path planning and complex engineering tasks.

The Sine–Cosine Enhanced Producer Position Update Strategy is a modification to the Sparrow Search Algorithm (SSA) that replaces the original undirected or exponentially shrinking moves of the producer subgroup with an oscillatory, directionally biased advance toward the global best individual. Introduced within the Geometric Sparrow Search Algorithm (GeoSSA), this update mechanism utilizes sine and cosine transformations, random scaling, and a time-adaptive inertia weight to balance global exploration and local exploitation throughout the optimization process. This results in improved convergence rates, lower solution variance, and robust performance across standard benchmarks and practical engineering applications (Wei et al., 27 Jan 2026).

1. Mathematical Formulation

The update step for producer individuals in GeoSSA is governed by the following equations: $X_{t+1}^{\,i,j} = \begin{cases} \omega\,X_{t}^{\,i,j} + r_1\,\sin(r_2)\;|r_3\,X_{\mathrm{best}} - X_{t}^{\,i,j}|, & R_2 < ST,\[1em] \omega\,X_{t}^{\,i,j} + r_1\,\cos(r_2)\;|r_3\,X_{\mathrm{best}} - X_{t}^{\,i,j}|, & R_2 \ge ST. \end{cases}$ where

$\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$

and

• $X_{t}^{\,i,j}$ is the $j$-th dimension of the $i$-th sparrow at iteration $t$,
• $X_{\mathrm{best}}$ is the current global best position,
• $R_2$ is a uniformly distributed alarm variable,
• $ST\in[0.5,1]$ is the safety threshold,
• $r_1, r_2$ are independent uniform random variables with $\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$0 serving as the oscillation angle,
• $\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$1 is a scaling coefficient amplifying the distance to the best,
• $\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$2 is a time-adaptive inertia weight, following a sigmoid curve from near 0 (early) to nearly 1 (late).

This construction instantiates a step that is always biased in the direction of $\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$3 but modulates its character (exploratory vs. exploitative) according to both a stochastic alarm regime and the epoch-adaptive inertia term.

2. Rationale and Design Intuition

The original SSA producer movement exhibits two main behaviors: shrinking toward the origin via exponential decay or making Gaussian-like random jumps, corresponding to perceived “safe” or “danger” periods as modeled by $\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$4 and $\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$5. Both regimes are susceptible to premature convergence or excessive randomness due to lack of directed search toward high-quality solutions.

The Sine–Cosine Enhanced Producer Update injects directional oscillations toward the current global best, with:

• Sine-based updates ($\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$6) encouraging larger leaps (enhanced exploration),
• Cosine-based updates ($\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$7) supporting smaller, more local adjustments (intensified exploitation).

The inertia weight $\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$8 starts low, ensuring that initial updates are dominated by the stochastic, oscillatory term, and gradually increases to privilege incremental exploitation as the run proceeds. $\omega = \frac{1}{1 + \exp[-25(t/T-0.5)]}$9 retains the alarm-driven behavioral switch, preserving the predator-avoidance metaphor intrinsic to SSA.

3. Algorithmic Implementation

A high-level procedural description for the producer update in GeoSSA is:

$i$2 All variables and parameter regimes strictly mirror those in the formal definition above.

4. Parameter Settings and Practical Tuning

A summary of key parameters and recommended settings as evaluated in (Wei et al., 27 Jan 2026) is presented below.

Parameter	Typical Range/Setting	Role and Effect
$X_{t}^{\,i,j}$0	$X_{t}^{\,i,j}$1 ($X_{t}^{\,i,j}$2)	Safety threshold for alarm-based switch
$X_{t}^{\,i,j}$3	$X_{t}^{\,i,j}$4	Random amplitude scaling
$X_{t}^{\,i,j}$5	$X_{t}^{\,i,j}$6 or $X_{t}^{\,i,j}$7	Oscillation angle for sine/cosine
$X_{t}^{\,i,j}$8	$X_{t}^{\,i,j}$9	Distance amplification/contraction
$j$0	Sigmoid (see above)	Epoch-adaptive inertia; slope/midpoint tunable

The inertia-weight sigmoid ($j$1) can be tuned via its steepness (constant $j$2) and midpoint ($j$3) for more rapid or delayed transitions from exploration to exploitation.

The paper recommends $j$4, a producer-to-population ratio (PD ratio) of $j$5, and the use of Good-Nodes-Set initialization alongside the sine–cosine producer update for consistently optimal outcomes.

5. Comparison to the Original SSA Producer Update

The original SSA producer step is given by: $j$6 where movement is either an undirected exponential shrinkage (without bias toward the best) or a random Gaussian jump.

The Sine–Cosine Enhanced version offers:

• Explicit, oscillatory motion toward (or around) the current global best,
• Dynamic balance of global and local search via sine–cosine switching and inertia adaptation,
• Empirically superior convergence, confirmed by lower average fitness values and variances across multiple benchmarks.

Experimentally, enhancements yielded:

• Accelerated convergence on unimodal test problems,
• Improved capability to avoid local minima for multimodal functions,
• Better mean and variance across a suite of 23 benchmarks.

6. Empirical Performance and Ablation Results

Ablation studies in (Wei et al., 27 Jan 2026) demonstrate the impact of removing the sine–cosine producer module:

• Eliminating the enhancement (GeoSSA2 variant) raises the average Friedman rank from $j$7 (full GeoSSA) to $j$8,
• Slower convergence is pronounced on challenging problems such as Rosenbrock and Quartic functions,
• Full Sine–Cosine producer update enables attainment of target fitness $j$9 faster than the original SSA or GeoSSA2,
• Measurably lower standard deviation on difficult benchmarks, indicating enhanced stability.

These results validate the efficacy of the oscillatory, directed update regime in managing the exploration–exploitation trade-off and ensuring robust optimization dynamics.

7. Applications and Practical Significance

In UAV path planning scenarios (Sec. 7, (Wei et al., 27 Jan 2026)), the Sine–Cosine update is directly credited for producing shorter, smoother three-dimensional paths with markedly reduced run-to-run variance (standard deviation ≈ $i$0 vs $i$1 for original SSA).

In diverse engineering design contexts (including Corrugated Bulkhead, Piston Lever, Reactor Network, Refrigeration System), the enhanced producer step is tied to GeoSSA’s ability to meet and often surpass best-in-literature benchmarks with minimal variance across repeated trials.

The authors recommend the Sine–Cosine approach for tasks where a deliberate, adaptive drive toward the best-known solution is desired, alongside sustained oscillatory motion to preclude premature convergence.

A plausible implication is that this mechanism is particularly advantageous for real-world, high-dimensional problems where dynamic adaptability and robustness are requisite alongside efficient solution quality.

---

In summary, the Sine–Cosine Enhanced Producer Position Update Strategy in GeoSSA systematically replaces undirected or exponential producer dynamics with a bounded, oscillatory, best-biased motion pattern, resulting in improved global optimization performance, faster and more stable convergence on classic benchmarks, and documented superiority in complex engineering and path planning applications (Wei et al., 27 Jan 2026).