Research On Visual Analytics Method For Local Patterns In Urban Traffic Accident Data

Traffic accident safety is a major issue affecting human life and social development.With the popularity of information technology,intelligent transportation and other technologies,urban transportation departments have collected and accumulated a large amount of traffic accident data,which contains the patterns of traffic accidents.The analysis and mining of traffic accident data has attracted the attention of city administrations,industry and academia,as the extraction of patterns hidden in the data can provide powerful support to decision makers.Compared to the general global pattern,the local pattern can provide more fine-grained information that is not easy for users to perceive in the data set,but is also more difficult to analyze and find.Domain experience shows that there are many valuable local patterns in traffic accident data,so this dissertation studies methods and key techniques to efficiently explore and discover local patterns in traffic accident data sets.The Visual analytics combines human intelligence and machine intelligence through visual interactive interface,which has outstanding advantages in exploratory pattern analysis,therefore this dissertation studies the problem of local pattern analysis in urban traffic accident data based on the visual analytics method,the main contents include:(1)Traffic accident data enhancement based on multi-source data fusion: The original traffic accident data set contains less attribute information,which does not allow a more comprehensive and in-depth analysis of traffic accidents,and enhancements to the data set can build a more complete and rich analysis data set to meet the diverse analysis requirements.In this dissertation,we first enhance the meteorological data and road network data to obtain a multi-source data set,and then correlate and fuse the time domain,road domain and the original traffic accident data set to build a complete and unified analytical data set with stronger semantics.(2)Local correlation analysis based on accident-prone segments: Accident-prone segments are the direct targets of the relevant departments for accident management,with a significant correlation with weather and other factors,and analysis of the local correlation of accident-prone segments can reveal local patterns in traffic accident data.Therefore,in order to analyze the influence of weather and other factors on traffic accidents in different regions,a local correlation visual analytics method based on traffic accident-prone segments is proposed.The histogram of the factors to be analyzed is used to describe the accident-prone segments,a cluster-supported visual analytics system with multi-correlation views is designed to explore and analyse the local patterns existing in different accident-prone segments.(3)Local spatio-temporal patterns analysis based on tensor decomposition:Large-scale multidimensional spatio-temporal data sets have a large number of local spatio-temporal patterns that are difficult to explore manually.Tensor decomposition can automatically extract the patterns in the data set.This dissertation proposes a visual analytics method of local spatio-temporal patterns based on tensor decomposition,which integrates intelligent algorithms such as tensor decomposition and clustering in a visual analytics system.Users interactively explore and analyze local spatio-temporal patterns that exist in data sets in a semi-automated manner to solve the difficulty of local pattern exploration and analysis in large-scale traffic accident data sets.