A Covariance-Guided Pursuit Birds of Prey-Based Optimization for global optimization and UAV path planning

Birds of Prey-Based Optimization (BPBO) may suffer from insufficient directional guidance, undifferentiated population learning, and premature stagnation when solving rotated, multimodal, and high-dimensional optimization problems. To address these limitations, this study proposes a Covariance-Guided Pursuit Birds of Prey-Based Optimization (CGP-BPBO) algorithm integrating three complementary mechanisms. Elite Memory Covariance Rotation Guidance (EMCRG) uses the covariance structure of elite individuals to introduce anisotropic correction into the best-guided branch. Rank-Based Dynamic Pursuit Learning (RBDPL) assigns different learning behaviors according to individual fitness ranks. Stagnation-Aware Gaussian-Cauchy Mutation (SAGCM) applies state-dependent perturbations to restore diversity when search improvement ceases. CGP-BPBO was evaluated using the IEEE Congress on Evolutionary Computation 2017 (CEC2017) and 2022 (CEC2022) benchmark suites under 30 independent runs. After excluding the officially deprecated CEC2017 F2 function, CGP-BPBO achieved the best mean result on 17 of the remaining 29 CEC2017 functions and outperformed the original BPBO on all 29 functions in terms of the reported mean values. On the 10- and 20-dimensional CEC2022 suites, it outperformed BPBO on 12 of 12 and 11 of 12 functions, respectively. Based on the mean results reported in the benchmark tables, CGP-BPBO obtained the lowest Friedman mean ranks of 1.79, 1.79, and 2.96 for CEC2017-30D, CEC2022-10D, and CEC2022-20D, respectively. In the three-dimensional unmanned aerial vehicle (UAV) path-planning experiment, CGP-BPBO achieved a mean path cost of 239.8048 and a standard deviation of 26.2686, representing reductions of 21.74% and 59.68%, respectively, compared with BPBO. These results demonstrate the competitiveness of CGP-BPBO in numerical optimization and constrained path planning, while its covariance decomposition introduces additional computational cost in high-dimensional problems.