Method for judging and visualizing relative risk of severity of urban intersection space accidents

Abstract

The application discloses a kind of urban intersection space accident severity relative risk discrimination and visualization method, comprising: obtaining intersection accident data, intersection design data, meteorological data, traffic flow data;The data obtained are preprocessed;Extract city road network topology, construct intersection adjacency matrix, obtain the spatial correlation of each accident;Establish the spatial prediction model of road intersection accident severity considering intersection space correlation;Extract the spatial residual term of each intersection, and discriminate the relative risk of intersection heavy accident;According to the relative risk value of intersection, draw the accident severity spatial risk hotspot identification diagram, realize risk profile visualization.The application can effectively determine whether a specific intersection in urban road has higher probability of serious injury accident than intersection with similar characteristics, and visualize in the form of heat map, provide theoretical support for accident black spot diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of accident safety analysis, in particular to a method for identifying and visualizing the relative risk of accident severity at urban intersections. BACKGROUND

[0002] Intersections are key nodes connecting urban road networks. Due to the complexity of traffic movement paths and the large amount of traffic conflicts, they have become a hot area for traffic safety management and control. By analyzing and identifying intersection accident black spots, traffic management departments can optimize resource allocation and propose individualized improvement measures for high-accident-prone areas, thereby improving road traffic safety and reducing accident casualties and property losses.

[0003] Existing traffic accident risk hotspot identification is mainly based on accident frequency. However, such methods usually only focus on the number of accidents, ignoring the severity of accidents. This means that some accidents that may cause serious injuries or deaths may not receive enough attention, and the overall level of intersection traffic safety cannot be fully assessed. SUMMARY

[0004] The present application aims to overcome the defects of current intersection accident black spot investigation process, which pays little attention to accident injury levels and intersection spatial topology relationships. A method for identifying and visualizing the relative risk of accident severity at urban intersections is proposed. Based on accident severity, the method focuses on individual accidents, especially fatal accidents, and can better identify low-frequency but high-potential-risk locations that may be overlooked in traditional frequency analysis.

[0005] To achieve the above-mentioned purpose, the technical solution provided by the present application is as follows: a method for identifying and visualizing the relative risk of accident severity at urban intersections, comprising the following steps:

[0006] 1) Obtain data, including intersection accident data, intersection design data, weather data, and traffic flow data;

[0007] 2) Preprocess the obtained data, including filling and deleting missing and abnormal values, and dummy variable processing and variable correlation test and collinearity diagnosis of categorical variables, to form a complete structured accident severity analysis dataset;

[0008] 3) Extract the topology of the urban road network, construct the intersection adjacency matrix, and match the accident records in the accident severity analysis dataset to the nearest intersection according to the spatial position to obtain the spatial correlation of each accident;

[0009] 4) establishing a road intersection accident severity spatial prediction model considering intersection spatial correlation, and measuring the model performance based on two types of indicators of fitting degree and prediction accuracy, wherein the road intersection accident severity spatial prediction model is a binary Logit model framework introducing a spatial residual term to improve the model's adaptability to spatial accident data and the model performance, and for any given observation accident, the probability of a serious accident can be obtained;

[0010] 5) according to the road intersection accident severity spatial prediction model, extracting the spatial residual term of each intersection to determine the relative risk of a serious accident at the intersection, i.e. the RR value (Relative Risk; RR);

[0011] 6) according to the RR value of each intersection, drawing an accident severity spatial risk hotspot identification map to realize risk profile visualization.

[0012] Further, in step 1), the intersection accident data includes accident severity, accident object, accident time, location, driver characteristics and vehicle characteristics, the intersection design data includes intersecting road grade, road speed limit and intersection type, the meteorological data includes weather conditions, wind speed and rainfall, and the traffic flow data includes intersecting road annual average daily traffic AADT (Annual Average Daily Traffic), vehicle proportion and traffic congestion.

[0013] Further, the step 2) includes the following steps:

[0014] 2.1) pre-processing the data, including filling and deleting missing values and outliers, and dummy variable processing for classification characteristic variables, to establish an accident severity analysis data set D;

[0015] 2.2) performing variable correlation test and collinearity diagnosis on the processed accident severity analysis data set D, if the Pearson correlation coefficient between two variables X1 and X2 is greater than 0.8 or less than -0.8, it is considered that they have significant correlation, and the calculation is shown in formula (1):

[0016]

[0017] In the formula, is the correlation coefficient between variables X1 and X2, are the standard deviations of variables X1 and X2, respectively;

[0018] In the collinearity diagnosis, if the variance inflation factor is greater than 10, it is considered that there is significant collinearity between the variables, and the calculation is shown in formula (2):

[0019]

[0020] where P is the number of variables, is the square of the complex correlation coefficient of the jth variable to the remaining P-1 variables, and each variable can get its variance inflation factor VIF j The maximum value of the variance inflation factor of all variables in the model is taken as the variance inflation factor VIF of the model. According to the diagnostic results, combined with the representativeness and importance of the variables, the observation variables with significant correlation or significant multicollinearity are removed.

[0021] Further, in step 3), the topological extraction of the road network in the study area is performed, and the correlation between each intersection is constructed in the ArcGIS software to obtain the intersection adjacency matrix W, and the rules are as follows:

[0022] If the intersections m and n are directly connected by a road, and there is no other intersection between the two intersections, it is determined that there is a spatial adjacency relationship between the two intersections, and the weight ω m,n = 1, otherwise ω m,n = 0, forming a square matrix with the number of intersections as the dimension, i.e. the intersection adjacency matrix W, and each accident record in the accident severity analysis data set D is matched to the nearest intersection according to the spatial position.

[0023] Further, in step 4), a road intersection accident severity spatial prediction model is established. For any given observation accident i, the accident severity Y i is divided into two categories: no injury and injury, which are set to 0 and 1 respectively, and in the framework of the binary Logit model, a spatial residual term is introduced, then in any intersection m, the probability of serious injury accident π im of accident i has the following relationship:

[0024]

[0025]

[0026] where φ m is the spatial residual term of the mth intersection, φ n is the spatial residual term of the nth intersection, is the variance parameter of the spatial term, ω m,n indicates the adjacency weight value between intersection m and intersection n, and p is a weight parameter reflecting the correlation strength, x impFor the pth risk variable of intersection m of accident i, β0 is a constant term; for the constructed road intersection accident severity spatial prediction model considering the spatial correlation of intersections, the deviation information criterion (Deviation Information Criterion, DIC), recall, specificity and F1-score indicators are used to measure the performance of the model, and the calculation method is as follows:

[0027]

[0028]

[0029]

[0030]

[0031] In the formula, DIC, Recall, Specificity and Total represent the deviation information criterion, recall, specificity and overall classification accuracy respectively, and p D respectively represent the fitting accuracy and effective variable number of the model, TP represents the number of true heavy injury accidents predicted as heavy injury accidents, FN represents the number of true heavy injury accidents predicted as light injury accidents, TN represents the number of true light injury accidents predicted as heavy injury accidents, and FP represents the number of true light injury accidents predicted as heavy injury accidents.

[0032] Further, in step 5), according to the constructed road intersection accident severity spatial prediction model, the spatial residual term of each intersection is extracted, and the relative risk RR is defined as the expected excess odds ratio to determine whether the mth intersection has a higher probability of heavy injury accident than the intersection with similar characteristics, and the specific formula is as follows:

[0033] RR=exp(φ m ) (9)

[0034] Wherein, the intersection node with RR value greater than 1 is considered to have higher safety improvement potential.

[0035] Further, in step 6), the RR values of each intersection are arranged in ascending order, and the most dangerous and the safest intersections are marked, the accident severity spatial risk hotspot identification diagram is drawn, and the ArcGIS software is used for visual representation.

[0036] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0037] 1. Oriented to urban dense road network, the topological relationship of intersection is considered.

[0038] 2. The severity of the accident was taken into account in the identification of accident black spots.

[0039] 3. Propose relative risk indicators, which can better identify locations with low accident incidence but high casualties while controlling for disturbance variables. These locations are often overlooked in traditional accident black spot screening. Attached Figure Description

[0040] Figure 1 This is a flowchart of the method of the present invention.

[0041] Figure 2 This is a spatial risk hotspot identification map for the severity of intersection accidents in specific implementation cases. Detailed Implementation

[0042] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0043] like Figure 1 As shown in the figure, this embodiment discloses a method for relative risk assessment and visualization of the severity of spatial accidents at urban intersections, including the following steps:

[0044] Step 1: Historical Accident Data Collection and Preprocessing at Intersections. Pedestrian traffic accident data from 489 intersections within a specific area is collected. This data details the severity of the accident, the parties involved, the time and location of the accident, driver and vehicle characteristics, the grade of the intersecting roads, road speed limits and intersection types, weather conditions, wind speed and rainfall, the annual average daily traffic volume (AADT) of the intersecting roads, the proportion of different vehicle types, and traffic congestion information.

[0045] Step 2: Clean and fill the accident data. The categorical variables in Table 1 are processed into Table 2 through dummy variable processing to establish the accident severity analysis dataset D.

[0046] Table 1 Vehicle Type Classification

[0047]

[0048]

[0049] Table 2. Dumb Variable Transformation Operations

[0050] Accident number Car Lorry Coach Bus 1 1 0 0 0 2 1 0 0 0 3 0 1 0 0 4 0 0 1 0 5 1 0 0 0 6 0 0 0 1 7 0 1 0 0 8 0 0 1 0 9 0 0 1 0 10 1 0 0 0

[0051] Choose any column as the reference item. In the example above, when choosing cars as the reference item, you only need to use the data of trucks, buses, and public buses as independent variables.

[0052] The variable correlation test and collinearity diagnosis are performed on the accident severity analysis data set D. If the Pearson correlation coefficient between two variables X1 and X2 is greater than 0.8 or less than -0.8, it is considered that they have significant correlation, and the calculation is shown in formula (1):

[0053]

[0054] In the formula, is the correlation coefficient between variables X1 and X2, is the standard deviation of variable X1 and variable X2, respectively.

[0055] In the collinearity diagnosis, if the variance inflation factor is greater than 10, it is considered that there is significant collinearity between variables, and the calculation is shown in formula (2):

[0056]

[0057] In the formula, P is the number of variables, is the square of the multiple correlation coefficient of the jth variable to the remaining P-1 variables, and each variable can obtain its variance inflation factor VIF j The maximum value of the variance inflation factor of all variables in the model is taken as the variance inflation factor VIF of the model; according to the diagnosis results, the representative and importance of the variables are combined to eliminate the observation variables with significant correlation or significant collinearity.

[0058] In this embodiment, the processed data in step 1 is analyzed for correlation and collinearity using STATA15 software. The results show that the correlation coefficient between the variable “whether it is a rainy day” and the variable “whether the road surface is wet and slippery” is greater than 0.8. The variable “whether the road surface is wet and slippery” is removed, and the accident severity analysis data set D is constructed, as shown in Table 3.

[0059] Table 3 Accident severity analysis data set

[0060]

[0061] Step 3, topology of the road network in the study area is performed, and the correlation between 489 intersections is constructed in ArcGIS software to obtain the intersection adjacency matrix W, as shown in Table 4. The rules are as follows:

[0062] If intersections m and n are directly connected by a road, and there is no other intersection between the two intersections, it is considered that there is a spatial adjacency relationship between the two intersections, and the weight ω m,n = 1, otherwise ω m,n = 0, forming a square matrix with the number of intersections as the dimension, i.e. the intersection adjacency matrix W.

[0063] Each accident record in the accident severity analysis dataset D is matched to the nearest intersection according to the spatial position.

[0064] Table 4 Intersection adjacency matrix

[0065]

[0066]

[0067] Step 4, a road intersection accident severity spatial prediction model considering intersection spatial correlation is constructed in WinBUGS software. For any given observed accident i, the accident severity Y i is divided into two categories: no injury and injury, which are set to 0 and 1 respectively. In the framework of the binary Logit model, a spatial residual term is introduced to improve the adaptability of the model to spatial accident data and the performance of the model. Then, for any intersection m, the probability of a serious accident π im has the following relationship:

[0068]

[0069]

[0070] In the formula, φ m is the spatial residual term of the mth intersection, φ n is the spatial residual term of the nth intersection, is the variance parameter of the spatial term, ω m,n represents the adjacency weight value between intersection m and intersection n, ρ is the weight parameter reflecting the correlation strength, x imp is the pth risk variable of intersection m of accident i, and β0 is the constant term. For the constructed road intersection accident severity spatial prediction model considering intersection spatial correlation, the model performance is measured by the deviation information criterion (Deviation Information Criterion; DIC), recall, specificity and overall classification accuracy index, which are calculated as follows:

[0071]

[0072]

[0073]

[0074]

[0075] In the formula, DIC, Recall, Specificity and Total represent the deviation information criterion, recall, specificity and overall classification accuracy respectively, and pD respectively represent the fitting accuracy and effective variable number of the model, TP represents the number of true heavy injury accident prediction, FN represents the number of true heavy injury accident prediction as light injury accident, TN represents the number of true light injury accident prediction, and FP represents the number of true light injury accident prediction as heavy injury accident.

[0076] Step 5, the spatial residual term of each intersection in the study area is calculated in sequence, and the relative risk RR (Relative Risk) of the intersection heavy accident is defined as the expected excess odds ratio to determine whether the mth intersection has a higher probability of heavy injury accident than the intersection with similar characteristics, and the specific formula is as follows:

[0077] RR = exp (φ m ) (9)

[0078] Wherein, the intersection node with the RR value greater than 1 is considered to have higher safety improvement potential.

[0079] Step 6, the RR values of the intersections are arranged in ascending order, and a number of most dangerous and safest intersections are listed respectively, and the accident severity spatial risk hotspot identification diagram is drawn, and the visualization is represented by using ArcGIS, as shown in Figure 2

[0080] The above embodiment is the preferred embodiment of the present application, but the embodiment of the present application is not limited by the above embodiment, and any change, modification, substitution, combination, simplification made without departing from the spirit and principle of the present application should be equivalent replacement mode, and all are included in the protection scope of the present application.​

Claims

1. A method for assessing and visualizing the relative risk of spatial accidents at urban intersections, characterized in that... Includes the following steps: 1) Acquire data, including: intersection accident data, intersection design data, meteorological data, and traffic flow data; 2) Preprocess the acquired data, including: filling and deleting missing and outlier values, as well as handling dummy variables for categorical variables and performing correlation and collinearity tests to form a complete structured dataset for accident severity analysis. 3) Extract the urban road network topology, construct the intersection adjacency matrix, and match the accident records in the accident severity analysis dataset to the nearest intersection according to their spatial location to obtain the spatial correlation of each accident; 4) Establish a spatial prediction model for the severity of road intersection accidents that considers the spatial correlation of intersections, and measure the model performance based on two indicators: good fit and prediction accuracy. The spatial prediction model for the severity of road intersection accidents introduces a spatial residual term under the framework of a binary Logit model to improve the model's adaptability to spatial accident data and model performance. For any given observed accident, the probability of serious injury can be obtained. Establish a spatial prediction model for the severity of accidents at road intersections. For any given observed accident i, its severity Y is... i The accident is divided into two categories: no injury and injury, denoted as 0 and 1 respectively. Under the framework of a binary Logit model, a spatial residual term is introduced. Then, within any intersection m, the probability of a serious injury accident i is π. im The following relationship exists: In the formula, φ m Let φ be the spatial residual term for the m-th intersection. n For the spatial residual term of the nth intersection, Let ω be the variance parameter of the spatial residual term. m,n This represents the adjacency weight between intersection m and intersection n, where ρ is a weight parameter reflecting the strength of the correlation, and x... imp Let be the p-th risk variable at intersection m where accident i is located, P be the number of variables, and β0 be a constant term. For the constructed spatial prediction model of road intersection accident severity considering spatial correlation at intersections, the performance of the model is measured by the bias information criterion, recall rate, specificity, and overall classification accuracy. The calculation methods are as follows: In the formula, DIC, Recall, Specificity, and Total represent the bias information criterion, recall, specificity, and overall classification accuracy, respectively. and p D TP represents the model's fitting accuracy and the number of effective variables, respectively. TP represents the number of serious injuries that are actually predicted to be serious injuries. FN represents the number of serious injuries that are actually predicted to be minor injuries. TN represents the number of minor injuries that are actually predicted to be serious injuries. FP represents the number of minor injuries that are actually predicted to be serious injuries. 5) Based on the spatial prediction model of the severity of accidents at road intersections, extract the spatial residual terms of each intersection and determine the relative risk of major and serious accidents at the intersections, i.e., the RR value. 6) Based on the RR values ​​of each intersection, draw a spatial risk hotspot identification map of accident severity to visualize the risk overview.

2. The method for relative risk assessment and visualization of the severity of spatial accidents at urban intersections according to claim 1, characterized in that, In step 1), the intersection accident data includes: accident severity, accident subject, accident time, location, driver characteristics, and vehicle characteristics; the intersection design data includes: intersecting road grade, road speed limit, and intersection type; the meteorological data includes: weather conditions, wind speed, and rainfall; and the traffic flow data includes: annual average daily traffic volume (AADT) of intersecting roads, proportion of various vehicle types, and traffic congestion.

3. The method for relative risk assessment and visualization of the severity of spatial accidents at urban intersections according to claim 2, characterized in that, Step 2) includes the following steps: 2.1) Preprocess the data, including filling and deleting missing and outlier values, and dummy variable processing for categorical feature variables to establish an accident severity analysis dataset D; 2.2) Perform variable correlation tests and collinearity diagnosis on the processed accident severity analysis dataset D. If the Pearson correlation coefficient between two variables X1 and X2 is greater than 0.8 or less than -0.8, they are considered to have a significant correlation. The calculation is shown in formula (1): In the formula, The correlation coefficient between variables X1 and X2. These are the standard deviations of variables X1 and X2, respectively. In the diagnosis of collinearity, if the variance inflation factor is greater than 10, then significant collinearity is considered to exist between the variables, and the calculation is shown in formula (2): In the formula, The variance inflation factor (VIF) of each variable is the square of the multiple correlation coefficient between the j-th variable and the remaining P-1 variables. j The maximum variance inflation factor (VIF) of all variables in the model is taken as the variance inflation factor (VIF) of the model. Based on the diagnostic results, and considering the representativeness and importance of the variables, observed variables that are significantly correlated or significantly collinear are removed.

4. The method for relative risk assessment and visualization of the severity of spatial accidents at urban intersections according to claim 3, characterized in that, In step 3), the road network topology within the study area is extracted, and the relationships between various intersections are constructed in ArcGIS software to obtain the intersection adjacency matrix W. The rules are as follows: If intersections m and n are directly connected by a road, and there are no other intersections between the two intersections, then the two intersections are considered to have a spatial adjacency relationship, and are assigned a weight ω. m,n =1, otherwise ω m,n =0, forming a square matrix with the number of intersections as the dimension, namely the intersection adjacency matrix W, which matches each accident record in the accident severity analysis dataset D to the nearest intersection according to its spatial location.

5. The method for relative risk assessment and visualization of the severity of spatial accidents at urban intersections according to claim 4, characterized in that, In step 5), based on the constructed spatial prediction model for the severity of accidents at road intersections, the spatial residual terms of each intersection are extracted, and their relative risk (RR) is defined as the expected excess advantage ratio, to determine whether the m-th intersection has a higher probability of serious injury accidents than intersections with similar characteristics. The specific formula is as follows: RR=exp(φ m ) (9) Among them, intersection nodes with an RR value greater than 1 are considered to have a higher potential for safety improvement.

6. The method for relative risk assessment and visualization of the severity of spatial accidents at urban intersections according to claim 5, characterized in that, In step 6), the RR values ​​of each intersection are sorted in ascending order, and the most dangerous and safest intersections are marked. A spatial risk hotspot identification map of accident severity is drawn and visualized using ArcGIS software.

Images