AI-Powered Cooperative Fleet Management through Explainable Context-Aware Anomaly Detection

Effective cooperative management of maritime fleets is challenged by technologies that cannot distinguish between actual technical faults and normal operational patterns influenced by weather. This ambiguity leads to frequent false alarms, undermining operator trust and limiting collaborative decision-making. To solve this problem, this paper introduces a framework that creates reliable, explainable knowledge from centrally-processed fleet data. The novelty relies in its two-stage process. First, it fuses sensor data with external weather context and assesses the context's reliability using a Weather Confidence Score—a score derived from engine performance, vessel proximity to land, and an assessment of weather data trustworthiness. Second, a dual-autoencoder ensemble performs a counterfactual analysis to generate an explainable Weather Influence Index. The Weather Index quantifies weather's impact, enabling a granular classification of anomalies as technical or weather-driven, moving beyond simple binary flagging. Evaluation on real-world vessel data shows the framework can reliably differentiate these anomaly types, enabling more trustworthy fleet-wide monitoring