Ship trajectory planning for collision avoidance using hybrid ARIMA-LSTM models. (15th July 2022)

• Record Type: Journal Article
• Authors: Abebe, Misganaw; Noh, Yoojeong; Kang, Young-Jin; Seo, Chanhee; Kim, Donghyun; Seo, Jin
• Is Part Of: Ocean engineering. Volume 256(2022)
• Publication Date: 2022-07-15
• Subjects: Ship trajectory -- Collision avoidance -- Hybrid model -- ARIMA -- LSTM; Ocean engineering -- Periodicals; Ocean engineering; Periodicals; 620.4162
• DOI: 10.1016/j.oceaneng.2022.111527
• Languages: English
• ISSNs: 0029-8018

Abstract

Abstract: In maritime transportation, accurate estimation of ship trajectories has a great impact on collision-free trajectory planning. Previously, many approaches were proposed for ship trajectory estimation, of which multi-step estimation received more attention because it can estimate both position and time in the near future. Nevertheless, those approaches have limitations due to their low accuracy or high complexity. To resolve this problem, this study provides a hybrid Autoregressive Integrated Moving Average (ARIMA) – Long short-term memory (LSTM) model to forecast the near future ship trajectory using automatic identification system (AIS) data for subsequent ship collision avoidance. By using a moving average (MA) filter, the AIS data are decomposed into linear and nonlinear data, and ARIMA and LSTM, respectively, are applied to model the ship's trajectory. The proposed model is tested and validated in terms of accuracy and computational time under different situations and compared with ARIMA, LSTM, and a previously suggested hybrid model. Finally, collision-avoidance simulations are conducted for various collision situations, showing that the proposed model can accurately estimate a near-future trajectory and evaluate collision risks to make proper early decisions to avoid the possibility of a collision. Highlights: A hybrid model (ARIMA-LSTM) is developed to estimate the near future ship trajectory. Moving average filter method is applied to decompose AIS data into linear and nonlinear data. The linear and nonlinear data are used to generate ARIMA and LSTM models, respectively. The proposed model is compared with ARIMA, LSTM, and K-models for various collision situations. The proposed model can make accurate, efficient, and real-time prediction of ship trajectories.