【作者】 赵鹏；

【导师】 张万松； 赖初荣；

【作者基本信息】 哈尔滨工程大学 ， 工程硕士（专业学位）， 2017， 硕士

【摘要】 随着移动计算的普及,不同领域中存在大量的轨迹数据,对轨迹的研究越来越重要,如社交网络、智慧城市中对位置、移动趋势、流动方向等的预测。现有的预测模型一般是基于一阶、高阶马尔可夫模型、扩展马尔可夫模型、动态贝叶斯等方式进行预测。在轨迹预测过程中,针对GPS采集数据存在误差、无法直接进行线路匹配问题,本文提出改进的基于权重的路网匹配算法,对轨迹中的轨迹点进行路网匹配,通过与现有方法进行对比实验,证明了本文所提出的方法的有效性。此外,针对路网匹配后的无用路段过多的问题,提出了热门行驶路线发现算法,对轨迹进行进一步建模,根据该方法将轨迹从路段的序列转化为热门路段序列。转化后的轨迹相对于原轨迹,有效地提高了轨迹语义信息。与传统轨迹处理方式不同,本文结合深度学习技术,提出基于卷积神经网络加多层感知器的复合模型。通过轨迹特征提取、轨迹图像化转换、轨迹目的地标签化等方法,形成卷积神经网络的输入,配合每条轨迹附带的相对应的元数据,经过大规模轨迹的训练,实现对目的地的预测。预测结果与传统预测模型相比,在各种条件下都具有一定的优势,也侧面反应了本文提出模型的可靠性。更多还原

【Abstract】 With the popularity of mobile computing,there are a lot of trajectory data in different fields,and the research on trajectory is becoming more and more important,such as social network,intelligent city,position,movement trend,flow direction and so on.The existing prediction model is generally based on first-order,high-order Markov model,extended Markov model,dynamic Bayesian and other methods to predict.In the process of trajectory prediction,there is a problem that data collected by GPS devices cannot be used directly to match the map.In this paper,an improved weight-based road network matching algorithm is proposed to match the track points in the trajectory and compare with the existing methods.Experiments show the effectiveness of the proposed method.In addition,aiming at the problem of excessive use of road network matching,a popular driving route discovery algorithm is proposed,and the trajectory is further modeled.According to this method,the trajectory is transformed from the sequence of the road segment to the hot road sequence.The trajectory of the trajectory is improved compared with the original trajectory,which effectively improves the trajectory semantic information.Different from the traditional trajectory processing method,this paper predict the destination based on the convolution neural network(CNN)and the multi-layer perceptron(MLP),which is based on the deep learning technology.In the paper,the feature of trajectory will be extracted by using deep learning technology and general pattern of the trajectory also will be found.After a large-scale trajectory training,and the output of MLP is the prediction result.Compared with the traditional prediction model,the model in this paper get the better result,and also shows the reliability of the model.更多还原