融合滑动窗口和Informer网络的渔船轨迹预测方法-《计算机技术与发展》

文章信息/Info

Title:: Fishing Vessel Trajectory Prediction Method Based on Integration of Sliding Window and Informer Network

作者:: 郭洋1; 李想2; 李响凝3; 1. 成都东软学院,四川成都 611844;
2. 大连东软信息学院,辽宁大连 116023;
3. 大连海洋大学信息工程学院,辽宁大连 116023

Author(s):: GUO Yang1; LI Xiang2; LI Xiang-ning3; 1. Chengdu Neusoft University,Chengdu 611844,China;
2. Dalian Neusoft University of Information,Dalian 116023,China;
3. School of Information Engineering,Dalian Ocean University,Dalian 116023,China

Keywords:: undersea cable protection; fishing vessel trajectory prediction; deep learning; Informer model; sliding window; AIS data

摘要:: 为了实现渔船作业行驶轨迹的精准预测,进而完成对驶入禁锚区的渔船进行警告的需求,该文提出基于融合滑动窗口和 Informer 网络的渔船轨迹预测方法(SIFP)。首先,对 AIS 数据进行预处理,通过在窗口内计算均值或中值等统计量,从而降低噪声的影响,获得更稳定和准确的轨迹信息;其次,采用滑动窗口扩充预测模型的数据量,满足预测模型对数据量的需求;最后,基于 Informer 模型完成渔船轨迹的精准预测,为禁锚预警提供数据支持。实验结果表明,SIFP 模型的MAE、MAPE 较 Transformer 网络模型分别提高了 0. 02% 和 0. 08% ,较 LSTM 网络模型分别提高了 0. 04% 和 0. 18% ,较 BP网络模型分别提高了 1. 47% 和 2. 14% ,证明了 SIFP 模型在轨迹预测问题上的有效性。

Abstract:: To achieve precise prediction of fishing vessel operation trajectories and further fulfill the need for warning fishing vessels entering prohibited anchorage areas,we propose a fishing vessel trajectory prediction method (SIFP) based on the integration of sliding window and Informer network. Firstly,the AIS data is preprocessed by calculating statistical values such as mean or median within the window to reduce the influence of noise and obtain more stable and accurate trajectory information. Secondly,the sliding window is employed to augment the data volume for the prediction model,meeting the data requirements of the prediction model. Finally, the Informer model is utilized to accomplish precise prediction of fishing vessel trajectories,providing data support for prohibited anchorage warnings. Experimental results show that the MAE and MAPE of the SIFP model are improved by 0. 02% and 0. 08% compared to the Transformer network model,0. 04% and 0. 18% compared to the LSTM network model,and 1. 47% and 2. 14% compared to the BP network model,demonstrating the effectiveness of the SIFP model in trajectory prediction problems.