Semantic Mining And Visualization Of Large-scale Urban Traffic Data

In recent years,with the continuous development of data acquisition technology and the popularization of wireless communication equipment,massive amounts of urban data have been collected,and the construction of “smart cities” and “intelligent transportation” driven by big data has become a hot research direction.Although the city data contains a wealth of potentially useful information,the data has the characteristics of large scale,multi-source heterogeneity,high complexity,and low value density.How to mine the latent semantics,understand the knowledge and characteristics of the data,and make reasonable use of the useful information mined is a meaningful and challenging research.In this context,big data visual analysis has become a key research area.Visual analysis and visualization technology can combine the powerful computing and processing capabilities of computers and the powerful visual processing and pattern recognition capabilities of the human eye.It can quickly extract potential semantic information from large-scale data through algorithms such as data mining and machine learning.The data is converted into intuitive and easy-to-understand graphical symbols for display,and the data is explored iteratively by designing a visual analysis interface for human-computer interaction.It plays an irreplaceable role in various fields.Of course,it is also widely used to process large-scale traffic data,and it plays a role as a “navigator” in the process of understanding and analyzing traffic data.This dissertation aims to explore the application of visual analysis and visualization technology in the field of intelligent transportation.We conduct semantic mining on large-scale traffic data based on topic models,and propose an interactive topic model that solves the problem of selecting the optimal number of topics,and then meets the task needs of different users through interaction.We noticed that the topic model can only consider the association between the two-dimensional attributes of the data,while the spatio-temporal traffic data has multi-dimensional attributes.Inspired by this,we propose to model the spatial-temporal traffic data as a three-dimensional tensor,which is capable of handling the association between multiple attribute dimensions of the data.And propose a novel augmented nonnegative tensor factorization method and an interactive visual analysis system to better visually explore the urban functional areas that dynamically change with time.In addition,we analyzed and explained how the visual analysis technology enhances the parallel intelligent transportation system,and proposed a visual analysis enhanced parallel intelligent transportation system framework to achieve the enhancement effect of “1+1>2”.The main contributions of this paper are listed as follows:·Visual analysis of large-scale taxi trajectory data: This research designs and implements an interactive visual analysis prototype system to extract the latent semantics hidden in large-scale taxi trajectory data.Our system first creates a clear traffic flow map by gathering taxi trajectory data in time and space,showing the dynamic changes of traffic flow over time.Based on the nonnegative matrix factorization model to mine the potential traffic semantics of the trajectory data,and then based on the semi-supervised nonnegative matrix factorization algorithm to propose an interactive topic model,which has the ability to overcome the problem of selecting the optimal number of topics.Support multiple types of user interaction at the same time.Finally,we designed multiple visual analysis views in order to better display and in-depth analysis of data knowledge.·Visual exploration of urban functional zones based on spatiotemporal data: This study implements a novel interactive visual analysis system based on augmented nonnegative tensor factorization to explore urban functional areas while considering human movement semantics and inherent location information.We divide the selected urban area into local area units based on the adaptive blue noise sampling method,extract the inherent location information from the POI data as prior knowledge,model the multi-dimensional spatio-temporal taxi trajectory data as a three-dimensional tensor,and propose augmented nonnegative tensor factorization method combines mobile semantics and location information to effectively determine urban functional zones.At the same time,we designed a set of visual views to support users' in-depth understanding and interpretation of the city's functional areas.·Visual analysis enhances the parallel intelligent transportation system framework: This research proposes a parallel intelligent transportation system framework enhanced by visual analysis.By discussing the importance of visual analysis in the new generation of artificial intelligence,and the process of transforming data or information into a knowledge system by visual analysis.It is proved that the seamless combination of visual analysis and parallel intelligent transportation system can better analyze large-scale traffic data and solve traffic problems more effectively,thereby achieving enhanced effects.