The Narwhal Optimizer (NWOA) is a novel nature-inspired metaheuristic that has shown competitive results on continuous optimisation problems. However, its weak local exploitation capability limits convergence speed and solution quality. This paper proposes four hybrid variants of the NWOA within a unified memetic computing framework, each integrating a complementary local search strategy: Nelder–Mead simplex method (NWOA-NM), Stochastic Hill Climbing (NWOA-HC), Hooke–Jeeves Pattern Search (NWOA-PS), and Simulated Annealing (NWOA-SA). Comprehensive experimental evaluation on the CEC2017 benchmark suite (30 functions, \(D=30\) ) demonstrates average fitness improvements of 43–49% over the baseline, with NWOA-HC and NWOA-PS each achieving 14 wins out of 30 functions and mean Friedman ranks of 1.73 and 1.90 respectively. The evaluation is further extended to the CEC2022 benchmark suite (12 functions, \(D=10\) and \(D=20\) ), where all hybrid variants achieve large Cliff’s delta effect sizes ( \(|\delta |>0.97\) ) against the baseline at \(D=20\) , confirming that improvements are practically substantial across benchmark suites and dimensionality settings. An ablation study isolates the contribution of each local search component, and population diversity analysis reveals that the baseline NWOA’s primary failure mode – premature diversity collapse within 20–30 iterations – is directly prevented by local search hybridisation. Statistical validation using Wilcoxon signed-rank tests with Bonferroni correction and Friedman test with Nemenyi post-hoc analysis confirms significant superiority of all hybrid variants ( \(p < 0.05\) ). A capacitated vehicle routing problem case study demonstrates practical effectiveness, with NWOA-PS achieving a 14% travel distance reduction over the baseline. The hybrid framework incurs a modest 20–27% computational overhead while delivering substantial solution quality improvements, confirming its suitability for deployment in engineering design optimisation and logistics applications.
