본문 바로가기

Operational-Mode-Aware Framework for Ship-Load Forecasting

문서 유형
연구출판물 · 학술지 논문
등재
SCIEScopus
저널 순위
JCR Q2Scopus Q1
저널
IEEE ACCESS14 · pp.98253 · 2026
피인용
Web of Science 0Scopus 0
FWCI (Scopus)
0
JCR IF
JCR 2024 3.6
ISSN
2169-3536

저자

  1. 문창재제1저자Pusan Natl Univ, Dept Elect & Elect Engn, Busan 46241, South Korea
  2. 정상훈Changwon Natl Univ, Dept Stat, Chang Won 51140, South Korea
  3. Park, Ki-DoKorean Register, Alternat Fuel Technol Res Team, Busan, South Korea
  4. 황평익교신저자Pusan Natl Univ, Dept Elect & Elect Engn, Busan 46241, South Korea

초록

Accurate short-term load forecasting is essential for the energy management of hybrid-electric ships, which coordinate diverse power sources under greenhouse-gas regulations. Conventional predictors trained on non-segmented navigation data fail to accurately predict abrupt load swings, such as those during transitions between seagoing and harboring operations. This study proposes an operational-mode-aware load-forecasting framework that couples offline mode segmentation with online mode estimation. Historical power profiles are partitioned offline by ship function (harboring, seagoing) and further segmented by constant power mode (CPM) and variable power mode (VPM), yielding operational-modes such as Harboring-CPM and Seagoing-VPM, with a dedicated predictor trained for each function and mode combination. Online, a mode estimator identifies the current operational-mode and activates the corresponding model. The modular framework integrates two online estimators sharing a common offline segmentation, online change-point (OCP) detection and automated machine-learning (AutoML) classification, together with a fully data-driven K-means baseline. It was verified on long short-term memory (LSTM) and transformer backbones over four voyages of a 9,196-ton training vessel, with the uniform Masan voyage retained as an unseen test. The framework reduces the root mean square error (RMSE) by up to 23.2% on VPM segments with LSTM and 11.9% with the transformer. Mode-aware gains, confirmed as statistically significant by the Diebold-Mariano test, were achieved on most segments with negligible real-time overhead under both architectures. These accurate forecasts can be consumed directly by the onboard EMS, enabling improved system stability, efficiency, and operational performance, supporting lower greenhouse-gas emissions in line with IMO MEPC targets.

키워드

AutoMLchange-point detectionK-meansLoad forecastingLSTMmode-aware modeloperational-mode estimationtransformer

식별자

Scopus EID
2-s2.0-105042699298
WoS UID
001850541000002