A Novel Golf Optimization Algorithm with Paddleboard Childcare and Industry Development–Inspired Strategy for Global Optimization and Practical Engineering Challenges

To address the limitations of the original Golf Optimization Algorithm (GOA) in high-dimensional complex optimization problems, including insufficient population diversity, limited adaptability of search parameters, and late-stage stagnation, this study proposes a Paddleboard-Inspired Adaptive Memory and Elite Refinement Golf Optimization Algorithm (PAMER-GOA). The proposed algorithm incorporates three complementary mechanisms. First, the Paddleboard Care and Industrial Development-Inspired Strategy (PCIDIS) improves population distribution and enhances global exploration through stratified mirrored initialization, differential search, and elite guidance. Second, the Dual-Memory Adaptive Industrial Investment Strategy (DMAIIS) dynamically adjusts the scaling factor and crossover probability using successful historical information, thereby improving the correspondence between search parameters and different evolutionary stages. Third, the Elite Subspace Principal-Axis Refinement and Stagnation Reallocation Strategy (ESPRS) integrates elite-region refinement, differential perturbation, and sparse Cauchy mutation to strengthen local exploitation and assist stagnant individuals in escaping local optima. Ablation experiments and parameter sensitivity analyses verify the complementary contributions of these strategies and the rationality of the key parameter settings. On the CEC2017 benchmark suite, PAMER-GOA achieves the lowest mean fitness values on 19 and 14 of the 29 functions in the 30- and 100-dimensional settings, respectively. It also ranks first in the Friedman tests, with mean ranks of 1.79 and 2.24, respectively. On the CEC2020 benchmark suite, PAMER-GOA again ranks first, attaining Friedman mean ranks of 1.20 and 1.40 in the 10- and 20-dimensional settings, respectively. The Wilcoxon rank-sum test results further indicate that PAMER-GOA significantly outperforms the comparison algorithms on most test functions. The runtime experiment shows that PAMER-GOA requires an average of 0.5333 s on the 30-dimensional CEC2017 benchmark suite, representing an increase of only 0.0648 s over the 0.4685 s required by the original GOA and indicating that the performance improvements incur only limited additional computational overhead. PAMER-GOA is further applied to the wireless sensor network node deployment problem, achieving the best mean objective values of 0.1553 and 0.1402 in the 30- and 40-node scenarios, respectively, thereby demonstrating its ability to address practical engineering optimization problems. Overall, the results show that PAMER-GOA substantially improves the solution accuracy, convergence behavior, and overall statistical ranking of the original GOA while maintaining strong overall competitiveness across different benchmark suites, dimensional settings, and node deployment scenarios.