A Meta-Learning-Based Adaptive Hybrid Evolutionary Algorithm for Dynamic Resource Configuration in Regional Industrial Cluster

Abstract

Dynamic resource configuration in regional industrial clusters is pivotal for unlocking their collective efficiency and ensuring resilience in cyber-physical production systems. However, static optimization methods suffer from a fundamental mismatch against volatile industrial rhythms, lacking the situational awareness to adapt strategies, the structural diversity to prevent stagnation, and the precision required for operational execution. To address these limitations, this paper proposes a meta-learning-based adaptive hybrid evolutionary algorithm (MAHEA), establishing a hierarchical observe-configure-execute paradigm. Specifically, MAHEA integrates three core mechanisms: a behavioral-cloning trained meta-controller that perceives high-dimensional system states to dynamically modulate strategic weights; an asymmetric parallel search that balances exploration and exploitation to escape local optima; and a potential-field guided refinement strategy that polishes coarse solutions into executable plans. Validated on a digital twin of the Pearl River Delta supply chain, results demonstrate that MAHEA significantly outperforms classical baselines in adaptive responsiveness and robustly handles demand surges and infrastructure failures. Note to Practitioners—This paper is motivated by the operational friction observed in collaborative regional manufacturing, where static schedules frequently fail due to stochastic disruptions like demand surges or fleet shortages. Existing tools often operate as blind optimizers, lacking the situational awareness to adjust plans in real time. We introduce an autonomous resource configuration approach that mimics the decision logic of an experienced operations manager. It continuously monitors system states and dynamically reconfigures optimization strategies to match operational intensity. Compared with classical schedulers, the proposed approach completes more orders at a lower average per-task cost while operating well within the practical decision interval required by industrial deployment. This hierarchical approach offers practitioners a robust solution to minimize systemic costs and ensure just-in-time compliance even under extreme volatility. While validated on a logistics interface, this blueprint is readily transferable to other supervisory automation tasks, such as dynamic shop-floor scheduling and multi-robot coordination.