当前位置: X-MOL 学术arXiv.eess.SY › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Estimation of Sea State Parameters from Ship Motion Responses Using Attention-based Neural Networks
arXiv - EE - Systems and Control Pub Date : 2023-01-21 , DOI: arxiv-2301.08949
Denis Selimović, Franko Hržić, Jasna Prpić-Oršić, Jonatan Lerga

On-site estimation of sea state parameters is crucial for ship navigation systems' accuracy, stability, and efficiency. Extensive research has been conducted on model-based estimating methods utilizing only ship motion responses. Model-free approaches based on machine learning (ML) have recently gained popularity, and estimation from time-series of ship motion responses using deep learning (DL) methods has given promising results. Accordingly, in this study, we apply the novel, attention-based neural network (AT-NN) for estimating sea state parameters (wave height, zero-crossing period, and relative wave direction) from raw time-series data of ship pitch, heave, and roll motions. Despite using reduced input data, it has been successfully demonstrated that the proposed approaches by modified state-of-the-art techniques (based on convolutional neural networks (CNN) for regression, multivariate long short-term memory CNN, and sliding puzzle neural network) reduced estimation MSE by 23% and MAE by 16% compared to the original methods. Furthermore, the proposed technique based on AT-NN outperformed all tested methods (original and enhanced), reducing estimation MSE by up to 94% and MAE by up to 70%. Finally, we also proposed a novel approach for interpreting the uncertainty estimation of neural network outputs based on the Monte-Carlo dropout method to enhance the model's trustworthiness.

中文翻译:

使用基于注意力的神经网络从船舶运动响应中估计海况参数

海况参数的现场估计对于船舶导航系统的准确性、稳定性和效率至关重要。对仅利用船舶运动响应的基于模型的估计方法进行了广泛的研究。基于机器学习 (ML) 的无模型方法最近受到欢迎,使用深度学习 (DL) 方法对船舶运动响应的时间序列进行估计已取得可喜的成果。因此,在这项研究中,我们应用新颖的、基于注意力的神经网络 (AT-NN) 从船舶纵摇的原始时间序列数据中估计海况参数(波高、过零周期和相对波向),起伏和滚动运动。尽管使用了减少的输入数据,已成功证明,通过改进的最先进技术(基于用于回归的卷积神经网络 (CNN)、多元长短期记忆 CNN 和滑动拼图神经网络)提出的方法将估计 MSE 降低了 23 % 和 MAE 比原来的方法提高了 16%。此外,所提出的基于 AT-NN 的技术优于所有测试方法(原始的和增强的),将估计 MSE 降低了高达 94%,将 MAE 降低了高达 70%。最后,我们还提出了一种新的方法来解释基于 Monte-Carlo dropout 方法的神经网络输出的不确定性估计,以增强模型的可信度。和滑动拼图神经网络)与原始方法相比,估计 MSE 减少了 23%,MAE 减少了 16%。此外,所提出的基于 AT-NN 的技术优于所有测试方法(原始的和增强的),将估计 MSE 降低了高达 94%,将 MAE 降低了高达 70%。最后,我们还提出了一种新的方法来解释基于 Monte-Carlo dropout 方法的神经网络输出的不确定性估计,以增强模型的可信度。和滑动拼图神经网络)与原始方法相比,估计 MSE 减少了 23%,MAE 减少了 16%。此外,所提出的基于 AT-NN 的技术优于所有测试方法(原始的和增强的),将估计 MSE 降低了高达 94%,将 MAE 降低了高达 70%。最后,我们还提出了一种新的方法来解释基于 Monte-Carlo dropout 方法的神经网络输出的不确定性估计,以增强模型的可信度。
更新日期:2023-01-24
down
wechat
bug