摘要
针对传统马尔科夫模型在长时序跨度条件下,预测船舶的下一位置精度较低且系统复杂度较高的问题,提出了一种基于长短型记忆循环神经网络(Recurrent Neural Networks-Long Short-Term Memory,RNN-LSTM)的船舶位置预测模型。利用DBSCAN聚类算法提取历史轨迹中的目标位置序列作为预测模型的输入,降低预测系统复杂度。引入Word2Vec模型中的Skip-grams算法将目标位置转换成位置向量,提升目标位置的区分度。实验结果表明,适当调整聚类算法参数可有效提升预测精度,所提模型预测准确度也高于传统预测模型。
Aiming at the problem that the traditional Markov model predicts the lower accuracy of the ship's next position and the complexity of the system is high under the long time span condition,a ship position prediction model is proposed based on Re⁃current Neural Networks-Long Short-Term Memory.The DBSCAN clustering algorithm is used to extract the target position se⁃quence in the historical trajectory as the input of the prediction model,which reduces the complexity of the prediction system.The Skip-grams algorithm in the Word2Vec model is introduced to convert the target position into a position vector,which improves the discrimination of the target position.The experimental results show that the appropriate adjustment of the clustering algorithm param⁃eters can effectively improve the prediction accuracy,and the proposed model prediction accuracy is also higher than the traditional prediction model.
作者
张玉人
龚志猛
ZHANG Yuren;GONG Zhimeng(Fujian Institute of Research on the Structure of Matter Fuzhou,Chinese Academy of Sciences,Fuzhou 350608;Huaxin Consulting Co.,Ltd.,Hangzhou 310052)
出处
《计算机与数字工程》
2021年第2期252-258,共7页
Computer & Digital Engineering
