基于混合多步Markov模型的位置预测方法研究

发布时间：2018-06-28 11:54

本文选题：基于位置服务 + 移动定位技术　；参考：《东北大学》2014年硕士论文

【摘要】：近年来,随着移动定位技术的成熟和流行,基于位置服务越来越受到人们的关注。位置预测技术是其中重要部分,并有着广泛的应用。目前,位置预测方法中较为常用的是基于Markov模型的预测方法,然而这类方法存在很多问题,例如：没有有效的划分区域、没有考虑用户特性以及只基于当前位置进行预测。因此,基于Markov模型的新的位置预测方法的研究迫在眉睫。本文以传统的基于Markov模型的位置预测理论为基础,并针对其中存在的问题加以改进。建立了包括离线数据处理、线下模型训练和线上位置预测的完整位置预测方案,使得预测准确率更高,系统适用范围更广。本文的主要贡献如下：首先,针对GPS数据远比其他数据更容易获取的特点,提出只基于用户历史GPS数据进行位置预测的方案,使得位置预测方法更符合实际、应用范围更广其次,针对传统位置预测中将地图网格化的方法所存在的问题,提出一种新的划分方案,从GPS数据中提取兴趣点,并依据兴趣点将地图进行更有意义的划分。再次,针对传统Markov模型预测方法没有考虑用户特性的问题,提出一种聚类算法,将用户聚类为用户组并为每个用户组建立预测模型,实验证明可以有效的提高预测的准确率。最后,针对传统基于Markov模型的位置预测方法存在的问题,提出了建立混合多步Markov模型的方法,考虑了轨迹上的多个位置,并给出了每个位置对预测的影响系数。另外,因为用户每次的移动行为并不一定完全符合用户的习惯,所以提出一种贝叶斯方法,可以仅根据当前轨迹选择最符合该轨迹的模型进行预测。同时,解决了难以为新加入系统的用户和数据稀疏的用户进行预测的问题。通过理论分析和实验评估,证明了本文提出的基于混合多步Markov模型位置预测方法符合理论上的可行性和操作上的正确性。
[Abstract]:In recent years, with the maturity and popularity of mobile location technology, location-based services have attracted more and more attention. Position prediction technology is an important part of it, and has a wide range of applications. At present, the prediction method based on Markov model is commonly used in the location prediction method. However, there are many problems in this method, such as no effective division of regions, no consideration of user characteristics and only prediction based on the current location. Therefore, it is urgent to study a new position prediction method based on Markov model. This paper is based on the traditional position prediction theory based on Markov model and improves the existing problems. A complete position prediction scheme including off-line data processing, offline model training and on-line position prediction is established, which makes the prediction accuracy higher and the system more applicable. The main contributions of this paper are as follows: firstly, in view of the fact that GPS data are far easier to obtain than other data, a position prediction scheme based on user history GPS data is proposed, which makes the location prediction method more practical. Secondly, aiming at the problems existing in the traditional method of gridding map in position prediction, a new partition scheme is proposed to extract the interest points from GPS data, and to divide the map more meaningfully according to the points of interest. Thirdly, aiming at the problem that the traditional Markov model prediction method does not consider the characteristics of users, a clustering algorithm is proposed, which can cluster users into user groups and establish prediction models for each user group. Experiments show that the prediction accuracy can be improved effectively. Finally, aiming at the problems of the traditional Markov model-based location prediction method, a hybrid multistep Markov model is proposed, which considers multiple locations on the trajectory and gives the influence coefficients of each location on the prediction. In addition, because the user's mobile behavior does not always conform to the user's habits, a Bayesian method is proposed, which can only be predicted by selecting the most suitable model according to the current trajectory. At the same time, it solves the problem that it is difficult to predict the new users and the users with sparse data. Through theoretical analysis and experimental evaluation, it is proved that the proposed position prediction method based on mixed multistep Markov model is feasible in theory and correct in operation.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN929.5

【相似文献】