Computational engineers and applied mathematicians continually develop nature-inspired metaheuristic algorithms to resolve complex, non-linear, and high-dimensional optimization problems that overwhelm traditional calculus-based or linear programming techniques. In a study published in the Journal of Electrical Systems and Information Technology, a research team comprising Abena Sarpomaa Appiah, Daniel Kwegyir, Francis Boafo Effah, Daniel Opoku, Peter Asigri, Emmanuel Kwaku Anto, Nana Maryam Abdul-Bassit Munagah, and Kelvin Worlanyo Tamakloe introduced the Pelican Optimization Algorithm with Velocity Effect, designated as VEPOA. The classical Pelican Optimization Algorithm, originally formulated in 2022 by Pavel Trojovský and Mohammad Dehghani, models the cooperative hunting procedures of pelican flocks as they locate, approach, and capture aquatic prey. Although standard pelican optimization effectively balances search phases, existing implementations frequently encounter reduced convergence speeds and become vulnerable to local optima entrapment. To mitigate these specific computational limitations, the researchers incorporated aerodynamic principles governing natural pelican flight into the algorithm mathematical framework, demonstrating that simulated wind velocity dynamics can substantially refine trajectory updates across challenging numerical domains.
In biological hunting scenarios, pelicans exploit atmospheric wind currents to sustain efficient gliding trajectories, alter flight orientations dynamically, and conserve metabolic energy while surveying broad surface areas. High wind speeds generate updrafts that expand the birds geographic search reach, whereas shifting headwind and tailwind conditions alter their descent velocities and approach angles during sudden plunging dives. The authors translated these natural behavioral adjustments into a mathematical velocity effect that modifies how candidate solutions navigate multi-dimensional problem spaces. In standard metaheuristics, optimization progresses across two alternating phases comprising global exploration, where agents broadly survey the search territory, and local exploitation, where candidate solutions refine approximations around identified optima. Rather than relying on purely position-driven movements toward identified prey, the newly developed VEPOA assigns a distinct velocity vector to each individual member of the numerical population. This synthetic velocity functions as an adaptive scaling mechanism that dynamically alters update step sizes, expanding trajectory shifts when velocities are low to foster global exploration and contracting step lengths when velocities are high to secure fine-grained local exploitation.
The mathematical formulation of the velocity effect begins by establishing stochastic angular properties for every candidate solution within the optimization population. Each agent is initialized with a forward orientation angle sampled uniformly between zero and a maximum boundary of thirty degrees, alongside an orientation perturbation angle drawn from a uniform distribution bounded between one degree and ten degrees, both of which are converted into radians during implementation. The pelican final flight orientation is subsequently updated using a uniform random variable that dictates leftward or rightward angular shifts relative to the target prey position. Once this orientation is established, the forward velocity is calculated as the product of an initial velocity component and the sine of the updated orientation angle, where initial velocities are randomly distributed between zero point one and one point zero. The resulting velocity term is integrated into the denominator of the position update equation during the exploration phase. Consequently, when velocity values increase, the spatial displacement diminishes to facilitate thorough local search, whereas lower velocities amplify displacement distances, preventing the population from premature stagnation.
Introducing trigonometric velocity terms into iterative numerical denominators creates inherent risks of computational instability, particularly when angular arguments approach zero and threaten division by zero exceptions. To prevent undefined arithmetic states and numerical overflow, the authors embedded rigorous safeguarding mechanisms directly into the VEPOA algorithm architecture. First, a dedicated regularization parameter epsilon, established at ten to the negative eighth power, was permanently added to the denominator of the exploration position update equation, guaranteeing bounded displacement steps regardless of angular magnitude. Second, the researchers established an absolute lower threshold on the orientation angle, enforcing a minimum angle of ten to the negative third radians to eliminate infinitesimally small sine computations that could destabilize sequential updates. Third, because the initial velocity remains strictly confined between zero point one and one point zero, the operational velocity parameter is constrained within a deterministic interval defined by the minimum velocity scaled by the sine of the lower angular bound up to the maximum velocity. These bounded parameters effectively eliminate not-a-number and infinity errors during extensive computing iterations.
To verify that the introduced mathematical complexity does not impose an unmanageable computational burden, the researchers conducted a formal asymptotic complexity analysis of the VEPOA pipeline. The computational footprint was systematically dissected across four core operational phases: population initialization, objective function evaluation, prey generation, and iterative candidate solution updates. Initializing a population of size N across an m-dimensional search landscape generates an initial computational cost of order N multiplied by m. During each optimization iteration over a maximum iteration limit T, every individual candidate solution evaluates the objective function throughout both the exploration and exploitation stages, contributing an evaluation complexity of order two multiplied by T and N. Generating and assessing the prey position occurs once per iteration and entails m-dimensional calculations, yielding a cumulative complexity order of T plus T multiplied by m plus T. Finally, updating candidate positions across all dimensions during both distinct hunting phases requires an operational complexity order of two multiplied by T, N, and m, producing an overall algorithmic complexity expressed mathematically as big-O of N times m plus T times one plus m times one plus two N.
Experimental evaluation of the VEPOA architecture began with extensive benchmarking across twenty-two classical functions adopted from the Congress on Evolutionary Computation literature, categorized into unimodal, high-dimensional multimodal, and fixed-dimensional multimodal topologies. Unimodal functions F1 through F6, which possess single global optima, were employed to evaluate exploitation capabilities, while multimodal functions F7 through F22 tested exploratory behavior and resistance to deceptive local peaks. Simulations were executed in MATLAB on an Intel Core i7-7500U processor operating between 2.70 and 2.90 gigahertz with 12 gigabytes of random-access memory, running for one thousand iterations across twenty independent trials per benchmark. In baseline comparative evaluations, VEPOA was benchmarked against the standard Pelican Optimization Algorithm alongside two previously published enhanced variants, namely the Logistic Tent Chaotic Map variant known as LTCMPOA and the Chaotic Inference Factor variant known as CIFPOA. Across the complete twenty-two function suite, VEPOA achieved optimal values in nineteen functions, secured the lowest optimal values in fifteen functions, and recorded an overall mean absolute error of 50.44, representing a forty-two percent error reduction over competing pelican variants.
To evaluate broader competitiveness, VEPOA was subsequently benchmarked against seven widely recognized metaheuristic algorithms: the African Vulture Optimization Algorithm, Gravitational Search Algorithm, Grey Wolf Optimizer, Particle Swarm Optimization, Teaching-Learning-Based Optimization, Tunicate Swarm Algorithm, and Whale Optimization Algorithm. On unimodal exploitation benchmarks, VEPOA secured five optimal values out of six functions and achieved zero standard deviation across five functions, tying for top overall mean absolute error ranking alongside AVOA, PSO, and TLBO with an error score of zero point zero zero. Across high-dimensional multimodal exploration functions F7 through F12, VEPOA achieved zero optimal values on two functions and delivered superior cumulative results on four functions, ultimately ranking third in mean absolute error behind AVOA and WOA with a recorded error value of 1.72E+02. On fixed-dimensional multimodal functions F13 through F22, VEPOA tied for the first-place mean absolute error ranking with AVOA, consistently reaching optimal target values. Furthermore, Wilcoxon signed-rank hypothesis testing confirmed that VEPOA performance gains were statistically significant compared to AVOA, standard POA, and TSA at a significance threshold below zero point zero five.
Beyond absolute error metrics, convergence speed evaluations demonstrated distinct operational advantages and trade-offs inherent to the velocity-augmented pelican framework. In unimodal optimization tasks, VEPOA demonstrated the fastest convergence rate among all evaluated methods, requiring an average of only 169 iterations to reach convergence, whereas AVOA averaged 520 iterations and alternative algorithms such as TSA and WOA required between 719 and 964.5 iterations, indicating that standard methods approached the one-thousand iteration ceiling and risked local entrapment. In high-dimensional multimodal tests, VEPOA converged at a moderate pace averaging 548.33 iterations; while Particle Swarm Optimization exhibited faster average convergence at 337.50 iterations, PSO failed to secure global optimal values due to premature convergence. Similarly, on fixed-dimensional multimodal landscapes, VEPOA and AVOA exhibited deliberate convergence trajectories but successfully secured superior final objective values. This empirical divergence underlines an essential optimization trade-off, demonstrating that rapid iteration decline does not guarantee accurate landscape navigation if exploration mechanisms fail to sufficiently probe surrounding search bounds.
The practical applicability of VEPOA was subsequently verified on three constrained real-world mechanical engineering design benchmarks: pressure vessel design, welded beam design, and tension-compression spring design. In the pressure vessel problem, which minimizes fabrication expenditure across four decision variables comprising shell thickness, head thickness, inner radius, and cylindrical section length, VEPOA achieved an optimal design cost of 5881.576, outperforming AVOA, Genetic Algorithms, Grey Wolf Optimizer, and Particle Swarm Optimization while recording the lowest mean and median costs. In the welded beam optimization problem, which regulates weld thickness, weld length, beam thickness, and beam width under shear, bending, buckling, and deflection constraints, VEPOA recorded the minimum manufacturing cost of 1.473162 and a minimal standard deviation of 0.000548, representing an approximate fourteen point six percent cost reduction compared to standard POA and improving upon GSA and WOA. In the spring design task, which minimizes wire weight subject to shear stress, surge frequency, and minimum deflection limits, VEPOA achieved the lowest design cost of 0.012665 with a standard deviation of 1.0488E-05, outperforming standard POA and competing metaheuristics.
Finally, the researchers evaluated VEPOA against advanced evolutionary algorithms using four benchmark functions from the CEC 2022 competition suite: Shifted and Rotated Zakharov (F1), Shifted and Rotated Rosenbrock (F2), Hybrid Function 1 (F3), and Composition Function 1 (F4), spanning search boundaries of negative one hundred to positive one hundred. Benchmarked against Differential Evolution, Success-History Adaptation Differential Evolution, Linear Population Reduction SHADE, and Covariance Matrix Adaptation Evolution Strategy, VEPOA reached theoretical optima on F1 with a value of 300 and an error of 1.8E-14, matched optima on F2 at 400, and achieved competitive values of 1817.48 on F3 and 2529.28 on F4. Furthermore, MATLAB runtime assessments via tic-toc timers demonstrated that VEPOA delivered the fastest execution time across all four CEC 2022 functions, recording a simulation runtime of 0.5841 seconds on F1 compared to 1.1702 seconds for Differential Evolution. Despite these demonstrated advantages, the authors noted an explicit operational limitation, acknowledging that integrating orientation shifts, trigonometric calculations, and regularized velocity scaling introduces potential computational overhead. Future research aims to deploy VEPOA across expanded engineering domains and complex system optimizations.
Subject of Research: Engineering optimization and nature-inspired metaheuristic algorithm design
Article Title: An improved Pelican Optimization Algorithm with velocity effect
Article References: Appiah, A. S., Kwegyir, D., Effah, F. B., Opoku, D., Asigri, P., Anto, E. K., Munagah, N. M. A.-B., & Tamakloe, K. W. (2026). An improved Pelican Optimization Algorithm with velocity effect. Journal of Electrical Systems and Information Technology, 13(1), Article 70. https://doi.org/10.1186/s43067-026-00356-8
Image Credits: AI Generated
DOI: 10.1186/s43067-026-00356-8
Keywords: Pelican Optimization Algorithm, VEPOA, metaheuristic algorithms, wind velocity effect, engineering optimization, benchmark functions, pressure vessel design, welded beam design, spring design problem, CEC 2022 functions, swarm intelligence, computational complexity, numerical stability, exploration exploitation balance, MATLAB simulation, Wilcoxon signed-rank test
Cite Scienmag News
Denise Maddox. (November 21, 2026). Modified Pelican Optimization Algorithm Incorporates Wind Velocity Dynamics to Boost Computational Precision. Scienmag. https://scienmag.com/modified-pelican-optimization-algorithm-incorporates-wind-velocity-dynamics-to-boost-computational-precision/
Denise Maddox. "Modified Pelican Optimization Algorithm Incorporates Wind Velocity Dynamics to Boost Computational Precision." Scienmag, 21 November 2026, https://scienmag.com/modified-pelican-optimization-algorithm-incorporates-wind-velocity-dynamics-to-boost-computational-precision/. Accessed 21 November 2026.
Denise Maddox. "Modified Pelican Optimization Algorithm Incorporates Wind Velocity Dynamics to Boost Computational Precision." Scienmag. November 21, 2026. https://scienmag.com/modified-pelican-optimization-algorithm-incorporates-wind-velocity-dynamics-to-boost-computational-precision/








