A Multi-Strategy Adaptive Hybrid Optimization Algorithm for Benchmark Functions and Engineering Design Problems

Engineering optimization problems are increasingly complex, requiring more sophisticated approaches to locate global optima. This paper presents the Adaptive Hybrid Optimization (AHO) algorithm, which addresses the limitations of single-equation update mechanisms and conventional linear decay schedules through three structural contributions. First, the algorithm implements four distinct position-update strategies, selected uniformly at random at each agent update, ensuring continuous structural diversity throughout the optimization process. Second, a dual-leader probabilistic guidance mechanism directs each search agent using asymmetric weighting between the top two solutions. Third, a nonlinear power-changing control parameter D replaces the conventional linear decay schedule used in existing algorithms such as WOA and GWO, providing 43% more total exploration budget while maintaining faster late-phase exploitation. AHO is validated on 23 benchmark functions from the CEC2005 suite and on seven constrained engineering problems, including four space truss structures: the 10-bar Truss Design, the 25-bar Spatial Truss, the 72-bar Spatial Truss, and the 120-bar Dome Truss. Friedman ranking analysis across all 23 benchmark functions ranked AHO first among the eight algorithms tested, with an overall best rank of 1.78 and an overall mean rank of 1.84. Wilcoxon signed-rank test results confirm statistically significant improvement over five of seven competitor algorithms on more than 19 of 23 benchmark functions, with AHO achieving the best overall Friedman rank of 1.78 across all 23 functions. On the four truss optimization problems, AHO achieves the best ranking across all five performance metrics against all seven competitor algorithms under an equal budget of 50,000 structural function evaluations per run. Performance analysis demonstrates competitive convergence behaviour and solution quality compared with the selected benchmark methods.