An efficient hybrid framework with knowledge transfer for solving capacitated vehicle routing problems

Recent progress in evolutionary multi-tasking algorithms has demonstrated potential for solving capacitated vehicle routing problems (CVRPs), though challenges persist regarding computational efficiency and knowledge transfer reliability. This study proposes a hybrid framework (HF) that addresses these limitations through four key components: (1) An improved Clarke–Wright savings algorithm for rapid initialization of high-quality solutions, (2) an incremental auxiliary task formulation that optimizes knowledge transfer while minimizing computational overhead, (3) a similarity prediction strategy enabling efficient offspring selection without additional evaluations, and (4) the framework integrates enhanced search strategy and diversity maintenance strategy to prevent premature convergence. Compatible with conventional meta-heuristics, the HF demonstrates competitive optimization capability across seven CVRP benchmark suites and six real-world JD.com logistics scenarios, outperforming state-of-the-art alternatives in solution quality.