基于自优化神经网络的船舶运动模型辨识

PDF下载HTML

Title:

Identification of ship motion model based on self-optimizing neural network

作者:

张浩晢¹, 杨智博¹, 焦绪国¹, 吕成兴¹, 朱齐丹²

1. 青岛理工大学 信息与控制工程学院, 山东 青岛 266520;
2. 哈尔滨工程大学 智能科学与工程学院, 黑龙江 哈尔滨 150001

Author(s):

ZHANG Haozhe¹, YANG Zhibo¹, JIAO Xuguo¹, LYU Chengxing¹, ZHU Qidan²

1. School of Information and Control Engineering, Qingdao University of Technology, Qingdao 266520, China;
2. College of Intelligent Systems Science and Engineering, Harbin Engineering University, Harbin 150001, China

关键词:

船舶运动建模; 改进雪融优化器; 双向时间卷积网络; 注意力机制; 优化; 超参数; 预测; 辨识

Keywords:

ship motion modeling; improved snow ablation optimizer; bidirectional temporal convolutional network; attention mechanism; optimize; hyperparameter; predict; identification

分类号:

TP183; U661.3

DOI:

10.11992/tis.202408004

摘要:

精确的船舶运动模型是船舶自主系统的核心。为提高船舶运动建模精度，引入了改进的雪融优化器(improved snow ablation optimizer, ISAO)；提出一种结合双向时间卷积网络(bidirectional temporal convolutional network, Bi-TCN)与注意力机制(attention mechanism, AM)的网络模型，即BITCA。进一步地，将ISAO与BITCA相结合，建立ISAO-BITCA船舶运动辨识混合模型。该模型利用Bi-TCN深度挖掘船舶运动序列在双向时间及空间维度下的隐藏特征，并引入AM以减少信息损失；基于ISAO，自主搜索并优化了BITCA模型的超参数组合。仿真实验结果表明，经过ISAO优化的BITCA模型在船舶航向角、偏航角速度、横摇角和总速度预测上的均方根误差(root mean square error, RMSE)分别降低了54.1%、28.21%、5.88%和40%，为船舶运动模型的准确辨识提供了一种有效手段。

Abstract:

An accurate ship motion model stands as the cornerstone of autonomous ship systems. To enhance the precision of ship motion modeling, an improved snow ablation optimizer (ISAO) is first introduced. Subsequently, a network model, BITCA, which integrates a bidirectional temporal convolutional network (Bi-TCN) with the attention mechanism (AM), is proposed. Furthermore, by combining the ISAO with BITCA, a hybrid identification model for ship motion, termed ISAO-BITCA, is established. This model initially leverages the Bi-TCN to deeply explore the hidden features of ship motion sequences across both temporal and spatial dimensions, while introducing the AM to mitigate information loss. Utilizing the ISAO, the hyperparameter combination for the BITCA model is autonomously searched and optimized. Simulation results demonstrate that the BITCA model optimized by the ISAO achieves reductions in the root mean square error for ship heading angle, yaw rate, roll angle, and total speed predictions by 54.1%, 28.21%, 5.88%, and 40%, respectively, providing an effective means for the accurate identification of ship motion models.