Ship collision avoidance anthropomorphic decision-making for structured learning based on AIS with Seq-CGAN. (1st December 2020)

• Record Type: Journal Article
• Authors: Gao, Miao; Shi, Guo-You
• Is Part Of: Ocean engineering. Volume 217(2020)
• Publication Date: 2020-12-01
• Subjects: AIS -- Seq-CGAN -- Anthropomorphic decision-making -- Seq2seq -- Structured learning; Ocean engineering -- Periodicals; Ocean engineering; Periodicals; 620.4162
• DOI: 10.1016/j.oceaneng.2020.107922
• Languages: English
• ISSNs: 0029-8018

Abstract

Abstract: The ship handling decision-making process constitutes the most critical step in intelligent ship collision avoidance. This study utilized the ship encounter azimuth map to extract the data of 12 types of ship encounter modes from the automatic identification system (AIS) big data. The sliding window algorithm was used to identify the ship encounter behaviors, which were used as the training data. The sequence conditional generative adversarial network (Seq-CGAN) was proposed using the sequence-to-sequence (Seq2Seq) model to learn how to generate appropriate anthropomorphic collision avoidance decisions and bypass the process of ship collision risk assessment based on the quantification of a series of functions. The long short-term memory (LSTM) cell was simultaneously combined with Seq-CGAN to improve the memory capacity and the current adaptability of the environment of the overall system. The Seq-CGAN was trained by 2018 Zhoushan Port AIS data to achieve structured learning regarding human experiences. The results indicate that the Seq-CGAN effectively formulates the sequence of ship handling behaviors. This study contributes significantly to the increased efficiency and safety of sea operations. The proposed method could be potentially applied to determine the predictive framework for various intelligent systems, including intelligent collision avoidance, route planning, and operational efficiency estimation. Highlights: Proposed the application of multiple consecutive SHBB sequences as a new representation method for ship handling decision-making. Construct the Seq-CGAN model based on Seq2Seq learning the underlying sequence of handling behaviors implicit in AIS data. Combining LSTM cell with Seq-CGAN to extend Seq-CGANs into deep neural networks, enhancing the stability and memory of the entire model. Proposed the 7-tuple coding and Capsule structure coding transformation to fully express a ship handling behavior unit and a recommended splicing threshold is given.