Semantic fuzzy bat optimization for regional manufacturing supply chain resilience

Regional manufacturing supply chains are experiencing more frequent supplier closures, logistics bottlenecks, severe weather and cascading failures of secondary suppliers. The existing disruption-response systems generally separate textual risk perception from optimisation-based decision-making and are thus unable to convert unstructured warning information into timely recovery plans. This paper introduces a semantic-driven resilience decision engine that combines BERT-based event extraction with a fuzzy-controlled bat algorithm in a perception-decision-execution framework. The Perception Layer converts news reports, enterprise announcements and other operational text into structured disruption vectors that contain information such as event type, intensity, affected entities and estimated duration. The decision layer uses the above semantic signals, as well as population diversity and stagnation status, to adaptively adjust the loudness and pulse rate of the bat algorithm to optimise recovery time, extra cost and service-level maintenance. The execution layer connects the selected strategy with ERP, MES, WMS and transportation interfaces to realise a closed-loop operation. Based on experiments with five manufacturing enterprises, 10,500 labelled disruption records and three representative disruption scenarios were used, and the proposed engine achieved an F1 score of 91.0% for entity extraction and 84.6% for relation extraction. It reduces the number of convergence generations by 37.1% compared to the standard Bat algorithm, is about 37.2% faster than manual decision-making, and has an end-to-end latency of less than four seconds. Based on the above results, a deployable and robust solution for high-frequency supply chain disruption management can be achieved by combining semantic perception with adaptive swarm optimisation.