CLA-GWO: A Bio-Inspired Closed-Loop Adaptive Grey Wolf Optimizer for the Traveling Salesman Problem

The standard Grey Wolf Optimizer (GWO) suffers from rapid population diversity loss when solving the Traveling Salesman Problem (TSP), leading to premature convergence. This paper proposes the Closed-Loop Adaptive Grey Wolf Optimizer (CLA-GWO), which constructs a five-layer closed-loop adaptive framework comprising: (i) discrete GWO evolution, (ii) dual-threshold diversity detection, (iii) three-level adaptive catastrophe, (iv) two-phase recovery, and (v) self-learning feedback control. The algorithm autonomously perceives population homogenization and restores diversity through scale-adaptive intervention. A two-stage parameter optimization strategy, combining the tree-structured Parzen estimator (TPE) with one-factor-at-a-time (OAT) sensitivity analysis, identified the optimal configuration. Component ablation on seven variants revealed that performance is principally driven by 2-opt local search and nearest-neighbor (NN) initialization, while the catastrophe lockout period (CLP) and multi-level catastrophe mechanisms safeguard diversity. Benchmark comparison on 20 TSPLIB instances demonstrated that CLA-GWO achieves competitive or superior performance under the adopted benchmark protocol against eight state-of-the-art algorithms across primary and secondary comparison groups. A Friedman test confirmed that the overall performance difference among the three independent baselines and CLA-GWO is statistically significant. The framework’s perceive–decide–intervene–recover–learn principle suggests a candidate transferable design pattern for other population-based metaheuristics; its current validation is limited to symmetric TSP, and asymmetric or non-Euclidean variants remain to be examined.