Field-Validated Probabilistic Digital Twin of a Shipboard Microgrid for Reliability-Oriented Operation: Applications to Hybrid and Unmanned Vessels

Abstract

The transition toward unmanned and hybrid vessels increases the requirements for diagnosability and continuity of power supply to critical consumers under physical and information-layer faults. The aim of this study is to develop a probabilistic digital twin of a shipboard microgrid that combines nonlinear state estimation, Bayesian mode inference, adaptive measurement-quality accounting, and risk-aware power reconfiguration. Field validation was performed using a long-term operational measurement program. The database comprises 12 classes of operating and fault events, with 30 field records per class, yielding 360 thirty-minute records, 180 h of observations, and 162,360 time samples at a 4 s sampling interval. Three diagnostic and control methods were applied to the same field records, producing 1080 paired algorithmic evaluations. Relative to the deterministic digital twin, the proposed method reduced voltage RMSE by 37.3%, SOC RMSE by 32.7%, and power RMSE by 30.1%, while reducing ENS by 13.3%. Mean fault-detection delay did not differ significantly from that of the deterministic approach; however, the missed-detection rate decreased from 12.7% to 7.3%. Compared with threshold-based protection, detection latency and recovery time were reduced by 25.7% and 30.8%, respectively. The field evidence therefore supports the proposed architecture as a supervisory state-estimation, diagnosis, and energy-reconfiguration layer. Because the field records originate from one operator and predominantly conventional tanker vessels, transfer to hybrid and unmanned platforms requires prospective onboard validation on the target vessel classes.