A Fine-Grained Behavior Classification Method for Oil Tankers Based on Bidirectional Long Short-Term Memory Network

The Gulf of Mexico, a globally important oil tanker passage, presents a highly complex maritime traffic environment that imposes stringent requirements on supervision and early risk warning. Traditional monitoring methods have limitations in identifying complex behavioral patterns of ships. To address this gap, this study leverages Automatic Identification System (AIS) data to construct a fine-grained behavior recognition model for oil tankers based on a Bidirectional Long Short-Term Memory (BiLSTM) network. First, systematic AIS data preprocessing methods are proposed to correct anomalies and interpolate missing values, thereby improving data integrity and reliability. Second, using navigation knowledge constraints and trajectory features, nine types of tanker behaviors are manually annotated, including straight-line acceleration, straight-line deceleration, high-speed cruising (speed > 8 knots), low-speed cruising (speed ≤ 8 knots), accelerating turn, decelerating turn, constant-speed turn, berthing, and anchoring, thus establishing a high-quality, fine-grained behavioral dataset. Finally, the BiLSTM network is introduced to build a recognition model capable of classifying these behaviors with high accuracy. An empirical study in the Gulf of Mexico shows that the proposed model achieves an overall recognition accuracy of 98.27%. F1-scores for key behaviors such as anchoring, berthing, and high-speed cruising all exceed 0.98, confirming the model’s effectiveness for practical maritime supervision. Compared with the multi-layer perceptron (MLP), LSTM, and temporal fusion transformer (TFT) models, the BiLSTM model achieves the highest scores in both Macro-F1 (0.936) and Weighted-F1 (0.981), thereby demonstrating the optimal performance in oil tanker behavior classification. The proposed framework provides reliable support for abnormal ship behavior recognition and intelligent maritime supervision, with potential for extension to other vessel types and high-risk sea areas. • Propose a fine-grained classification method for oil tanker behaviors using a BiLSTM network. • Develop a systematic framework for AIS data cleaning and trajectory interpolation. • Accurately identify nine distinct categories of oil tanker behaviors. • Demonstrate and validate the classification framework in the Gulf of Mexico scenario.