基于注意力机制的堆叠LSTM短时船舶交通流预测模型

廉清云, 孙伟, 李润生

doi:10.16411/j.cnki.issn1006-7736.2024.01.007

大连海事大学学报 >

2024 , Vol. 50 >Issue 1: 57 - 65

DOI: https://doi.org/10.16411/j.cnki.issn1006-7736.2024.01.007

基于注意力机制的堆叠LSTM短时船舶交通流预测模型

展开

（1.上海海事大学商船学院,上海 201306;2.上海海事大学信息工程学院,上海 201306）

廉清云（1970 — ），男，硕士，工程师，研究方向：智能交通系统、船舶通信导航。孙伟* （1978 — ），男，博士，硕士生导师，研究方向：智能交通、时空数据挖掘。李润生（1993 — ），男，硕士生，研究方向：智能交通、时空数据挖掘。E-mail：weisun@shmtu.edu.cn。

收稿日期: 2023-07-15

修回日期: 2023-10-09

录用日期: 2023-10-09

网络出版日期: 2023-10-09

基金资助

交通运输部2021年度交通运输行业重点科技项目(2021-ZD6-095)。

收起

Stacked LSTMs Short-Term Ship Traffic Flow Prediction Model based on Attention Mechanism

Expand

（1.College of Merchant Marine, Shanghai Maritime University, Shanghai 201306；2. College of Information Engineering, Shanghai Maritime University, Shanghai 201306）

Received date: 2023-07-15

Revised date: 2023-10-09

Accepted date: 2023-10-09

Online published: 2023-10-09

Fold

摘要

针对短时船舶交通流数据的非线性与非平稳性特征，导致预测精度低的问题，本文提出了基于注意力机制的堆叠LSTM船舶交通流预测模型。该模型主要采用堆叠式LSTM神经网络，用于捕捉短时船舶交通流数据的时序特征，并通过引入注意力机制来更好地学习全局性特征，以提高船舶交通流预测的精度。为了构建船舶交通流数据集，本文提取了长江下游三个航段的船舶AIS数据，并将其用于本文模型的训练和测试。结果表明，相较于HA、ARIMA、GPR、LSTM和Seq2Seq等基线模型，本文模型在交通流量宏观参数的预测中，均方根误差和平均绝对误差两个评价指标均有所降低。与最优基线模型相比，本文模型在船舶交通流预测中表现出更高的精度，其均方根误差降低了4.05%，平均绝对误差降低了4.04%。

关键词： 内河交通；船舶交通流预测；长短时记忆网络；堆叠LSTM；注意力机制

本文引用格式

廉清云, 孙伟, 李润生 . 基于注意力机制的堆叠LSTM短时船舶交通流预测模型[J]. 大连海事大学学报, 2024 , 50(1) : 57 -65 . DOI: 10.16411/j.cnki.issn1006-7736.2024.01.007

Abstract

According to address the issue of poor prediction accuracy caused by the nonlinear and non-stationary characteristics of short-term ship traffic flow data, this paper proposes an attention-based stacked LSTM ship traffic flow prediction model. The model primarily utilizes stacked LSTM neural networks to capture the temporal features of short-term ship traffic flow data, and introduces the attention mechanism to enhance the learning of global features and improve the accuracy of ship traffic flow prediction. The ship Automatic Identification System (AIS) data from three segments in the lower reaches of the Yangtze River are extracted and employed to construct the ship traffic flow datasets for training and testing the model. The results demonstrate that compared to baseline models such as HA, ARIMA, GPR, LSTM, and Seq2Seq, the model proposed in this paper reduces both the root mean square error and mean absolute error for predicting macro traffic flow parameters. This model demonstrates improved accuracy in ship traffic flow prediction compared to the optimal baseline model, achieving a reduction of 4.05% in root mean square error and 4.04% in mean absolute error.

Key words： river transportation; ship traffic flow prediction; LSTM; stacked LSTM; attention mechanism

参考文献

[1]赵龙文, 苌道方, 朱宗良,等. 基于SARIMA-BP模型的港口船舶交通流量预测[J]. 中国航海, 2020, 43(1):50-55.

ZHAO Longwen, CHANG Daofang, ZHU Zongliang, et al. Prediction of ship traffic flow in ports based on SARIMA-BP model[J]. China Navigation, 2020, 43(1):50-55.(in Chinese)

[2]吴康，汪洋，谢磊，等．基于状态空间的连续断面船舶交通流量预测研究[J]．交通信息与安全，2015，33(2):51-56．

WU Kang, WANG Yang, XIE Lei, et al. Research on continuous cross-section vessel traffic flow prediction based on state space [J]. Traffic Information and Safety, 2015, 33(2):51-56.(in Chinese)

[3]刘敬贤，张涛，刘文．船舶交通流组合预测方法研究[J]．中国航海，2009，32(3):80-84．

LIU Jingxian, ZHANG Tao, LIU Wen. Research on Combined Prediction Method of Ship Traffic Flow [J]. China Navigation, 2009, 32(3):80-84.(in Chinese)

[4]张矢宇,杨宇昊,陈尘,等.基于回归-卡尔曼滤波组合模型的航道整治区域船舶交通流时空预测[J].大连海事大学学报,2021,47(1):37-44.

ZHANG Shiyu, YANG Yuhao, CHEN Chen, et al. Spatio-temporal prediction of ship traffic flow in waterway regulation area based on regression-Kalman filter combination model[J]. Journal of Dalian Maritime University, 2021, 47(1):37-44.(in Chinese)

[5]孙立谦. 基于最小二乘支持向量机的深圳市船舶交通流量预测研究[D]. 大连:大连海事大学.2019

SUN Liqian. Research on ship traffic flow prediction in Shenzhen based on least squares support vector machine [D]. Dalian: Dalian Maritime University.2019.(in Chinese)

[6]肖进丽, 刘明俊, 李晓磊. 基于集合经验模态分解和灰色神经网络的船舶交通流预测[J]. 武汉理工大学学报（交通科学与工程版）, 2017, 41(6):938-942.

XIAO Jinli, LIU Mingjun, LI Xiaolei. Ship traffic flow prediction based on ensemble empirical modal decomposition and grey neural network[J]. Journal of Wuhan University of Technology（Transport Science and Engineering Edition）, 2017, 41(6):938-942.(in Chinese)

[7]高广旭, 刘敬贤, 刘奕,等. 基于矩阵分解的船舶交通流预测方法研究[J]. 武汉理工大学学报（交通科学与工程版）, 2022, 46(1):171-176.

GAO Guangxu, LIU Jingxian, LIU Yi, et al. Research on ship traffic flow prediction method based on matrix decomposition[J]. Journal of Wuhan University of Technology（Transportation Science and Engineering）, 2022, 46(1):171-176.(in Chinese)

[8]刘钊, 崔珑献, 李岩,等. 基于二维矩阵分解的船舶交通流预测[J]. 中国航海, 2021, 44(3):76-83.

LIU Zhao, CUI Longxian, LI Yan, et al. Ship traffic flow prediction based on two-dimensional matrix decomposition[J]. China Navigation, 2021, 44(3):76-83.(in Chinese)

[9]薛晗, 邵哲平, 潘家财,等. 基于文化萤火虫算法-广义回归神经网络的船舶交通流量预测[J]. 上海交通大学学报, 2020, 54(4): 421-429.DOI:10.16183/j.cnki.jsjtu.2020.04.011.

XUE Han, SHAO Zheping, PAN Jiazai, et al. Ship traffic flow prediction based on cultural firefly algorithm-generalised regression neural network[J]. Journal of Shanghai Jiao Tong University, 2020, 54(4): 421-429. DOI:10.16183/j.cnki.jsjtu.2020.04.011.(in Chinese)

[10]LV Y, DUAN Y, KANG W, et al. Traffic flow prediction with big data: a deep learning approach[J]. IEEE Transactions on Intelligent Transportation Systems, 2015, 16(2):865-873.

[11]MA, X, TAO, Z, WANG, Y, et al. Long short-term memory neural network for traffic speed prediction using remote microwave sensor data[J]. Transportation Research Part C: Emerging Technologies, 2015, 54:187-197.

[12]索永峰, 陈文科, 杨神化,等. 基于深度神经网络的船舶交通流预测[J]. 集美大学学报（自然科学版）, 2020, 25(6):430-436.DOI:10.19715/j.jmuzr.2020.06.04.

SUO Yongfeng, CHEN Wenke, YANG Shenhua, et al. Ship traffic flow prediction based on deep neural network[J]. Journal of Jimei University: Natural Science Edition, 2020, 25(6): 430-436. DOI:10.19715/j.jmuzr.2020.06.04.(in Chinese)

[13]王凯, 刘文, 陈进维,等. 基于VMD和LSTM的船舶交通流量预测方法研究[J]. 武汉理工大学学报（交通科学与工程版）, 2022, 46(1):177-182.

WANG Kai, LIU Wen, CHEN Jinwei, et al. Research on ship traffic flow prediction method based on VMD and LSTM[J]. Journal of Wuhan University of Technology（Transportation Science and Engineering Edition）, 2022, 46(1):177-182.(in Chinese)

[14]DONG Z Y. Prediction of ship traffic flow based on wavelet decomposition and LSTM[C]//7th International Conference on Cloud Computing and Big Data(ICCCBDA).Chengdu,China:IEEE,2022:8893.DOI:10.1109/ICCCBDA55098.2022.9778881.

[15]ZHOU H X, ZUO Y, LI T S, et al. Application of PSO-LSTM for forecasting of ship traffic flow[C]//International Conference on Security, Pattern Analysis, and Cybernetics（SPAC).Chengdu,China:IEEE,2021:298-302.

[16]刘敬贤,高广旭,刘奕,等. 基于卷积神经网络及长短时记忆网络的短时船舶交通流量预测[J]. 中国航海, 2022, 45(2):56-61.

LIU Jingxian, GAO Guangxu, LIU Yi, et al. Short-time ship traffic flow prediction based on convolutional neural network and long and short-term memory network[J]. China Navigation, 2022, 45(2):56-61.(in Chinese)

[17]吴莹莹,赵丽宁,袁志鑫,等.基于注意力机制的CNN-GRU船舶交通流预测模型[J].大连海事大学学报,2023,49(1):75-84.

WU Yingying, ZHAO Lining, YUAN Zhixin, et al. CNN-GRU ship traffic flow prediction model based on attention mechanism[J]. Journal of Dalian Maritime University, 2023,49(1):75-84.(in Chinese)

[18]董伟, 张磊磊, 金子恒,等. 基于多特征时空图卷积网络的水运通航密度预测[J]. 物联网学报, 2020, 4(3):78-85.

DONG Wei, ZHANG Leilei, JIN Ziheng, et al. Water transport navigation density prediction based on multi-feature spatio-temporal graph convolutional network[J]. Journal of Internet of Things, 2020, 4(3):78-85.(in Chinese)

[19] LUONG T, PHAM H, MANNING C D. Effective approaches to attention-based neural machine translation[C]//International Conference on Empirical Methods in Natural Language Processing (EMNLP). Lisbon, Portugal: Association for Computational Linguistics,2015: 1412-1421.

Options

文章导航

模态框（Modal）标题

摘要

本文引用格式

Abstract

参考文献