This is a paper published by our research group on the Journal of Transportation Engineering Part A-Systems. The article uses the popular GTWR, GWR, and OLS models for research. The paper is introduced here, and the code of the paper has been open source , can be used directly after modification.
github: https://github.com/RobinYaoWenbin/OpenSource-for-Analysis-of-Spatiotemporal-Factors-Affecting-Traffic-Safety
gitee: https://gitee.com/RobinYaoWenbin/OpenSource-for-Analysis-of-Spatiotemporal -Factors-Affecting-Traffic-Safety
1. Overview of the paper
The article mainly analyzes the spatio-temporal influencing factors of traffic accidents. In general the article can be divided into the following parts:
- Descriptive statistical analysis of traffic accident data. According to the severity, the accident data can be divided into (killed and severe injury) KSI and (property damage only) PDO, which are casualties and no casualties respectively. Then the temporal and spatial characteristics of the two types of accident data are analyzed respectively. The time dimension is to count the frequency of accidents every day, and the spatial dimension is to look at the distribution of accidents. For further analysis, we use the global Moran index and the local Moran index to analyze the spatial autocorrelation.
- The second part is the extraction of spatio-temporal influence factors. Spatio-temporal influencing factors are divided into three categories, the first is traffic status information, the second is road network structure characteristics, and the third is built environment information. A total of 18 variables were extracted from these three categories. In order to carry out subsequent modeling, collinearity analysis and spatial autocorrelation test are required respectively. Traffic status information comes from license plate recognition data, road network structure features come from OSM data, built environment information comes from POI data, and the latter two types of data are open source.
- The third part is the modeling of accident space-time relationship. The model is modeled with GTWR, GWR and OLS models respectively.
- The fourth part is the interpretation of the results. The interpretation of the results of this type of influencing factor analysis often takes up a lot of space, and as a paper, this part is very important, because here is the main conclusion and inspiration for engineering practice. However, as far as methodology is concerned, people often don't think much about this part.
2. Data overview
The paper uses 4 kinds of data, which are traffic accident data , license plate recognition data , road network data (openstreetmap) , and built environment data (POI data) .
The road network data and built environment data in these four types of data are open source, and they are also very common data in traffic-related research. For the acquisition and use of road network data, you can check my blog【OSMnx】 . By crawling openstreetmap and organizing it into a graph structure, you can use complex network methods for analysis. The built environment data is usually POI data. The POI of Gaode can be crawled through the Gaode api, and then the built environment can be roughly analyzed with POI, such as residential areas, commercial areas, etc. For details, please refer to the paper. I have also used the corresponding method in many papers.
As for accident data and license plate recognition data, it needs to cooperate with relevant departments to obtain it, and it seems that no relevant open source data has been seen. Accident data is the structured record of the accident by the traffic police after a traffic accident. License plate recognition data is a commonly used space-time trajectory data. Most of my research is based on this data. For the understanding of this data, you can read my thesis. License plate recognition data can be used to analyze travel behavior and the traffic status of road sections and road networks.
3. Methodology
This section in turn explains the methodology used in the article:
- The spatial analysis of traffic accidents uses global and local Moran indices, which are generally used for exploratory analysis, that is, to see whether the data exhibits spatial autocorrelation.
- Some complex network methods are used to mine road network structure variables. City intersections are used as nodes and roads are used as edges to form a graph, and then in-degree, out-degree, edge density, node importance (pagerank), etc. can be obtained. In addition, some relatively complex complex network indicators such as Orientation entropy are obtained to reflect the orderliness of street orientation.
- When calculating the built environment indicators, the commonly used built environment analysis methods and information entropy are used to measure the land use properties. These methods are detailed in my papers, such as Yao, W., J. Yu, Y. Yang, N. Chen, S. Jin , Y. Hu, and C. Bai. 2022b. “Understanding travel behavior adjustment under COVID-19.” Commun. Transp. Res.2 (Dec): 100068. https://doi.org/10.1016/j.commtr. 2022
.100068., Yao, W., C. Chen, H. Su, N. Chen, S. Jin, and C. Bai. 2022a. "Analysis of key commuting routes based on spatiotemporal trip chain." J. Adv. Transp.2022 (Mar): 31. https://doi.org/10.1155/2022/6044540. - GTWR, GWR, and OLS models are used in the analysis of spatio-temporal influencing factors. These models are also common models. For details, please refer to my GTWR introduction blog .
4 Conclusion
The conclusion is actually the coefficient of GTWR to analyze how each independent variable affects the accident rate. The GTWR model can model spatio-temporal heterogeneity, so it can actually analyze the coefficient changes in the time dimension and the coefficient changes in the space dimension. For example, I have put a picture of the time dimension and a picture of the space dimension below. Combining these two pictures, you can know how the coefficient changes in a day, that is, how the relationship between the independent variable and the dependent variable changes in a day, and you can also know how the relationship between the two changes with space. of. See our paper for details.
5. Open source code
Relevant codes have been open sourced on github and gitee, and those who need them can download them by themselves
references
Xu C, Zhang Z, Yao W, et al. 2023, Analysis of Spatiotemporal Factors Affecting Traffic Safety Based on Multisource Data Fusion.J. Transp. Eng., Part A: Systems, 2023, 149(10): 04023098
论文链接：https://doi.org/10.1061/JTEPBS.TEENG-7990