当无需揭示船舶操纵运动机理过程，而只需对输入输出建立映射时，黑箱建模成为一种有效途径。本文基于长短期记忆-循环神经网络（Long Short-term Memory-recurrent Neural Network，LSTM-RNN）构建船舶航向-舵角黑箱模型，LSTM 网络为10-10-1结构，误差指标为RMSE，参数学习采用Adam算法。开展实船Z型操纵实验获取了航向-舵角数据。前70%用于模型训练，后30%用于模型测试。训练后的模型使得RMSE达到设计目标。对测试集数据，训练后模型拟合优度在0.98以上，表明其具有良好的有效性和泛化性。文中航向-舵角LSTM-RNN黑箱模型结构简明清晰，参数明确，易于实际操作使用，为航向-舵角关系建模提供了一种可行方法。
Black box modeling becomes an effective approach when it is not necessary to reveal the mechanism of ship maneuvering motion, but only to establish a mapping of input and output. The paper focuses on building a black model of heading angle - rudder angle based on LSTM-RNN. The LSTM layer has a structure of 10-10-1 nodes. Error indicator for measuring the degree of fit is RMSE. The Adam algorithm is adopted for parameter learning. Zigzag manoeuvering experiments by using a full-scale ship maneuvering experiments are conducted to obtain Input and output data, namely the heading angle - rudder angle. The first 70% of the data is used for model training, while the last 30% is used for model testing. The trained model enables RMSE to achieve the hoped-for goals. For the testing data, the goodness of fit of the trained model is above 0.98, indicating that the final LSTM-RNN model has good effectiveness and generalization. The LSTM-RNN black box model for heading angle - rudder angle has a concise and clear structure, clear parameters, and is easy to use in practical operations. It provides a feasible method for modeling the relationship between heading angle and rudder angle for full scale ship.
2024,46(11): 80-84 收稿日期：2023-07-19
DOI：10.3404/j.issn.1672-7649.2024.11.015
分类号：U661.33
基金项目：浙江省教育厅资助项目（Y202147772）
作者简介：田延飞(1983-),男,博士,讲师,研究方向为智能航海及其仿真理论与技术
参考文献：
[1] 徐锋, 邹早建, 徐小卡, 等. 基于支持向量机的船舶操纵运动黑箱建模[J]. 北京航空航天大学学报, 2013(11): 1553-1557.
[2] NIE Z H, SHEN F, XU D J, et al. An EMD-SVR model for short-term prediction of ship motion using mirror symmetry and SVR algorithms to eliminate EMD boundary effect[J]. Ocean Engineering, 2020, 217: 107927.
[3] 刘娇, 史国友, 杨学钱, 等. 基于DE-SVM的船舶航迹预测模型[J]. 上海海事大学学报, 2020, 41(1): 34-39.
[4] ZHANG C K, BIN J C, WANG W, et al. AIS data driven general vessel destination prediction: a random forest based approach[J]. Transportation Research Part C, 2020, 118: 102729.
[5] MURRAY B, PERERA L P. A dual linear autoencoder approach for vessel trajectory prediction using historical AIS data[J]. Ocean Engineering, 2020, 209: 107478.
[6] GAO M, SHI G Y. Ship-handling behavior pattern recognition using AIS sub-trajectory clustering analysis based on the T-SNE and spectral clustering algorithms[J]. Ocean Engineering, 2020, 205: 106919.
[7] 胡玉可, 夏维, 胡笑旋, 等. 基于循环神经网络的船舶航迹预测[J]. 系统工程与电子技术, 2020, 42(4): 871-877.
[8] LIU Y C, DUAN W Y, HUANG L M, et al. The input vector space optimization for LSTM deep learning model in real-time prediction of ship motions[J]. Ocean Engineering, 2020, 213: 107681.
[9] 任宇翔, 赵建森, 刘卫, 等. 基于 AIS 数据和 LSTM 网络的船舶航行动态预测[J]. 上海海事大学学报, 2019, 40(3): 32-37.
[10] SUO Y F, CHEN W K, CLAR AMUNT C, et al. A ship trajectory prediction framework based on a recurrent neural network[J]. Sensors, 2020(1): 29-36.
[11] ZHOU H, CHEN Y J, ZHANG S M. Ship trajectory prediction based on BP neural network[J]. Journal on Artificial Intelligence, 2019(1): 29-36.
[12] 徐婷婷, 柳晓鸣, 杨鑫. 基于BP神经网络的船舶航迹实时预测[J]. 大连海事大学学报, 2012, 38(1): 9-11.
[13] 游兰, 韩雪薇, 何正伟, 等. 基于改进 Seq2Seq 的短时AIS轨迹序列预测模型[J]. 计算机科学, 2020, 47(9): 169-174.
