针对基于马尔可夫模型在真实时间上进行位置预测时,需要通过对时间进行等值划分来确定位置转移时间点,从而导致预测结果粗糙的问题,提出一种基于高斯分析的马尔可夫位置预测方法.该方法首先利用高斯混合模型拟合连续时间下地点之间的转移概率,从而发现可能的位置转移时间点,并将这些时间点作为马尔可夫模型的状态转移点,建立马尔可夫模型;然后通过用户在这些时间点的转移概率流向,计算用户位于某一位置的概率值,从而得到最终的位置预测结果.在数据集GeoLife上的实验结果表明,该方法相对于传统马尔可夫模型和高斯混合模型的预测准确率分别提升了约10％和12％.%To solve the problem that the prediction results based on Markov model are rough due to the equivalent partition of time for deter-mining of transition time point,we propose a new location prediction method of Markov based on Gaussian analysis. First,it finds out the possible transition time points by using Gaussian mixed model fitting the transition probability of locations with continuous time,and establi-shes the Markov model by making these points to be the state transition points of the traditional Markov model. Finally it predicts the user' s location by calculating the probability of transition between states. The experiment on GeoLife dataset shows that the precision can be im-proved respectively by about 10％ and 12％ compared with Markov model and Gaussian mixture model.
展开▼
