计算机科学与探索 ›› 2019, Vol. 13 ›› Issue (1): 23-34.DOI: 10.3778/j.issn.1673-9418.1801029

• 数据库技术 • 上一篇    下一篇

基于LSTM的移动对象位置预测算法

高  雅+,江国华,秦小麟,王钟毓   

  1. 南京航空航天大学 计算机科学与技术学院,南京 210016
  • 出版日期:2019-01-01 发布日期:2019-01-09

Location Prediction Algorithm of Moving Object Based on LSTM

GAO Ya+, JIANG Guohua, QIN Xiaolin, WANG Zhongyu   

  1. College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China
  • Online:2019-01-01 Published:2019-01-09

摘要: 移动对象位置预测是基于位置服务的重要组成部分。现有的移动对象位置预测算法有基于马尔可夫链的算法、基于隐马尔可夫模型的算法、基于神经网络的算法等,然而这些算法都无法解决移动对象轨迹数据中位置过多带来的维数灾难问题。为了解决这一问题,提出了位置分布式表示模型(location distributed representation model,LDRM)。该模型将难以处理的表示位置的高维one-hot向量降维成包含移动对象运动模式的低维位置嵌入向量。随后,将该模型与基于长短期记忆网络(long short-term memory,LSTM)的位置预测算法结合为LDRM-LSTM移动对象位置预测算法。真实数据集上的实验表明,与现有算法相比LDRM-LSTM算法在预测准确性上有较大的提升。

关键词: 位置预测, 降维, 移动对象, 长短期记忆网络(LSTM)

Abstract: Location prediction of moving object is an important part in location based service. Existing location prediction algorithms of moving object include Markov chain, hidden Markov model, neural network, etc. However, existing algorithms cannot solve the problem of dimensionality disaster caused by too many positions of the moving object??s trajectory data. In order to overcome this problem, this paper proposes a location distributed representation model (LDRM). The model reduces the dimension of one-hot vector which represents each location to a low dimension location embedding vector which concludes the moving object??s moving pattern. After that, LDRM is combined with location prediction algorithm based on long short-term memory (LSTM) neural network to get an overall algorithm called LDRM-LSTM. Experiment results on real dataset show that, there has been a major improvement of the LDRM-LSTM algorithm compared with the existing ones, in terms of prediction accuracy.

Key words: location prediction, dimension reduction, moving object, long short-term memory (LSTM)