Introduction
The total mileage of highway, use of civilian vehicles and number of motor vehicle drivers in China have gone up by 2.71 times, 15.99 times and 4.97 times over the past 20 years, respectively [1]. The continued improvement in facilities and level of traffic management could not keep up with the dramatically increase in traffic demand, resulting in lots of safety concerns. In 2021, more than 273,000 road traffic accidents and 62,218 deaths were reported in China, up by 38.76 percent and 6.32 percent respectively over that of 2014 [2]. A path-breaking investigation was conducted by Heinrich (1931), and found that the proportion relationship of serious injury accidents slight injury accidents and no injury accidents in a mechanical manufacturing company was counted as 1: 29: 300. In other words, when multiple risk factors interact with each other, a series of errors can occur, which can easily lead to major accidents. Similar findings were found in road traffic accidents, subway construction accidents, chemical accidents and coal mine accidents [3–6]. From 2010 to 2020, there were 151 major road traffic accidents with 10+ fatalities, more than half of which was single-vehicle accidents. Commercial vehicle accidents accounted for more than 80% of the total accidents.
A increasing body of research has investigated the contributing factors of accident frequency and injury severity for single- and multi-vehicle accidents separately [7–9]. The results have shown that the differences in the mechanisms and contributing factors between single- and multi-vehicle accidents are significant. In numerous studies on the contributing factors of single-vehicle accidents, human factor has been recognized as the principal factor influencing the incidence and severity of single-vehicle accidents [10]. Furthermore, a certain number of studies has focused on the impact of vehicle factors, environmental factors and risk factors related to the enterprise management for motor vehicle accidents. Moreover, each major accident show complex risk coupling relationship, resulting from interactions among multiple factors. Many researchers have paid extensive attention to evaluate the joint effect of multiple factors, such as weather and lighting conditions [11], traffic lane and shoulder widths [12], vehicle type and light condition [13], environment, traffic, and population characteristics [14], economic stress and urbanization [15], combination of road infrastructure elements [16], enterprise administration, external environment and driver attitudes to traffic safety [17].
At present, data-driven statistical model and graph theory are the main methods to analyze the coupling relationship among accident risk factors. The former includes random parameter regression model [18, 19], Bayesian model [20], shrinkage regression model [21], finite mixture model [22] and Markov transformation model [23], etc. Data-driven statistical methods are difficult for users to find out the evolution process of risk factors in essence. Even though the interactive impacts of risk factors have been extensively studied by using the above model, these studies had neglected chain reactions between accident risk factors. The latter graphically displays the relationship of the risk factors, such as event tree analysis [24], SDG model [4] and petri nets [25], etc. Comparatively speaking, SDG model not only reveals the risk transmission path, but also comprehensively explains the occurrence regularity of accidents [26], so that it can provide specific analysis on the characteristics of network structure. Compared with general accidents, the joint effect of multiple contributing factors is more complex in the major accident network. The coupling evolution mechanism of multiple factors has not been fully elucidated to date.
Before a major accident occurs, the fluctuation on risk degree of paths will become obvious. If the key risk transmission path can be identified, the accident frequency and loss will be greatly shortened by timely extraction of high risk nodes. In this context, scholars in different fields have proposed path recognition methods based on graph theory to monitor the risk degree of transmission paths. Wu et al. (2021) [27] examined in their study that the combination of multi-threshold method, trend fitting, and SDG model to conduct online monitoring and risk detection for the nuclear power plants was superior in speed and precision to the conventional SDG method. Zhang et al. [28] proposed a deep-first search strategy to identify the evolutionary path of chain faults in power grid, on the basis of the combination of knowledge graph and machine learning method to judge time sequence characteristics. In addition, dijkstra’s algorithm [29], and ant colony algorithm [30] were also applied to the identification of key risk transmission paths of various accidents. So far, the complex mechanism of road traffic accidents has not been revealed.
The goal of this study is to map the interactive effect of accident influencing factors from four aspects, namely human, vehicle, environment and enterprise administration. To that end, the investigation reports of single-vehicle accidents with 10+ fatalities and enterprise safety management database throughout the life cycle of vehicles were statistically analyzed. The main contributions of this study are as follows: (1) identify two categories of relationships inclusive of inter-relationships between various risk factors, intra-relationships between risk factors and accidents within the network-based approach in the SVDRN. (2) propose the algorithm of risk chain identification to identify key risk transmission path, basing on a combination of node importance and its risk state threshold in the SVDRN.
Methodology
Analysis procedures
The backward inference method along the risk transmission paths was used to extract potential risk factors in all aspects of the vehicle’s daily operation, starting from the result of the accident. The SVDRN was established based on the SDG model, in which the risk factors of single-vehicle accidents and their connectivity were taken as the network nodes and directed edges respectively. The topological structure of SVDRN was analyzed by using five indicators, including degree centrality, eigenvector centrality, closeness centrality, betweenness centrality and clustering coefficient. Considering the associated impact of the importance nodes and variable risk intensity in the SVDRN, this paper proposed a method for identifying key risk transmission path to real-time monitor the risk status of vehicle driving.
Data collection
Compared with minor accidents, the investigation of major single-vehicle accidents with 10+ fatalities is more thorough and comprehensive in China. According to Decree No. 493 of the State Council [31], the accident investigation group established by the people’s government of province where the accident occurred is responsible for writing the accident investigation report. The data of major accident needs to contain the upstream and downstream information of the accident, so that the causal relationship of accident can be systematically sorted out. Major single-vehicle accidents at the number of 83 from 2010 to 2020 were analyzed in this paper, and types of accidents at number of 6 were divided as follows: rollover (marked as R1), plunging into water or a ditch (R2), running over pedestrians (R3), fire (R4), natural disaster (R5), collision with fixed objects or parked vehicle (R6). The explanation of risk factors in the SVDRN is depicted in Table 1.
[Figure omitted. See PDF.]
SDG model for single vehicle driving risk network
The adjacency matrix M was defined, whose element aij was the correlation between node i and node j. The value of aij was equal to 0, when node i was not associated with node j or the value of i was the same as j. Otherwise, the value of aij was equal to 1. Considering the one-way evolution characteristic of accident chain, the SVDRN was a directed graph corresponding to an asymmetric matrix. According to mutual relations of risk factors, the structure diagram of network node based upon the SDG technology is constructed, as shown in Fig 1. The weight of SVDRN represents the occurrence number of two adjacent nodes in the total sample of accidents.
[Figure omitted. See PDF.]
Topological features of the SDG model
In order to estimate whether the network has the features of complexity, scale-free and small world, topology parameters need to be calculated, including degree distribution, clustering coefficient and closeness centrality of the network [43]. The more number of sides connected to node i is, the larger the degree of node i is, explaining that this node is closer to the center of the network, and susceptible to the interaction of neighbor nodes. The degree of a node contains two types in a directed network: in-degree and out-degree, so the total degree value of this node is the sum of both values. Take node M1 in Fig 1 as an example, none of the nodes points to M1, so the in-degree value of M1 is 0. There are three nodes that flow out from M1, including M2, M3, and M4. The out-degree value of M1 is 73, which is equal to the sum of the occurrence number for all the accident samples between M1 and each of three nodes.
The clustering coefficient Ci is the probability that any two neighbors connected to node i are also neighbors to each other in complex network theory. The calculation expression of Ci is as follows:(1)where, ki is the sum total of sides connected to node i, and Ni is the sum total of sides between neighbor nodes connected to node i.
Betweenness centrality BCi points out that a core node should appear on multiple shortest paths of other nodes. The stronger the transmission capacity of node i is, the higher the value of BCi is. The betweenness centrality of node i can be obtained as [44]:(2)where, Pjm is the sum total of shortest paths between node j and node m, Pjm(i) is the sum total of shortest paths through node i, node j and node m simultaneously, and n is the sum total of nodes in the SVDRN.
Closeness centrality describes the proximity effect of any two nodes in the risk network. Sij is used to represent the amount of shortest paths from node i to all other nodes in the SVDRN. The calculation formulas of closeness centrality for node i is defined as [45]:(3)The basic thought of eigenvector centrality is that the node importance is dependent on the amount of its neighbors (i.e., degree distribution), as well as the importance of its neighbor nodes. The more important the neighbors connected to a node are, the higher the numerical value of eigenvector centrality is. Suppose xi represents the importance of node i, the eigenvector centrality ECi can be expressed as in Eq (4). When the steady state is reached after many iterations, it can be written in Eq (5).(4)(5)where, c is the proportionality constant, and equal to the inverse of the principal eigenvalue of matrix M; x is the column vector of (x1, x2, …,xn)T.
Comprehensive evaluation of node importance in the risk network
In view of the limitation of single indicator on evaluation of node importance in practical application, the above-mentioned five indicators were used to construct the multi-attribute comprehensive evaluation model of node importance for SVDRN. The indicator weight of the comprehensive evaluation model was computed by the average weighted standardized matrix, and the value of comprehensive evaluation of node importance was obtained by an ideal scheme.
It was assumed that each node has characteristic indicators at number of s, then the jth indicator of the ith node was defined as T = wi(qj), among which the value of i ranges from 1 to n, and the value of j ranges from 1 to s. Due to the different dimensions of indicators, the matrix was normalized to establish a multi-attribute matrix of nodes, and carried out by Eqs (6)–(8). The conventional practice is to assign weight for each indicator of evaluation model according to experience. If it is difficult to find reasonable values, the average method of weighting normalized matrix is commonly used [46, 47]. The normalized matrix is denoted as A = (rij)n×s. Ki in Eq (9) represents the comprehensive importance of node i for SVDRN. An ideal scheme was adopted to evaluate the value of Ki, and the calculation formula of Ki was shown in Eq (9). The values of and are calculated by the euclidean norm in Eqs (10) and (11).(6)(7)(8)(9)(10)(11)where, and represent the highest and the lowest value in the column j of matrix A respectively.
Real-time monitoring of risk status with three-level threshold
“Management Measures for the Evaluation of Safety Production Standardization Construction in Transport Enterprises” is issued by Ministry of Transport of China in 2016, urging transport enterprises to establish a technical management database throughout the life cycle of vehicles. The purpose of above regulation is to facilitate the risk evaluation and supervision of vehicle driving safety. Table 2 shows the hierarchical thresholds of risk status for network nodes. The status of evaluation indicators for all the risk factors was determined by the three-level threshold, namely H (high risk), M (medium risk) and N(normal). The values of three-level threshold were 0.9, 0.5 and 0.1 respectively for subsequent quantitative processing. Each risk factor had at least one and at most three evaluation indicators. For the threshold of a factor with multiple evaluation indicators, the threshold of the factor was equal to 0.1, when the risk status of all the evaluation indicators were normal. The threshold of the factor was equal to 0.9, when one or more indicators were at high risk. The threshold of the factor was equal to 0.75, when more than two evaluation indicators were at medium risk. The node data of SVDRN indicated in Table 2 had both real-time dynamic and static data.
[Figure omitted. See PDF.]
The active safety intelligent prevention and control system of road transport enterprises comes with the increasingly high requirements for vehicle safety, and it is currently applied at a very high rate, especially for “two passengers and one danger” vehicles. Specifically, “two passengers and one danger” vehicles refer to line buses with the third-class or above, tourist chartered buses and road special vehicles carrying hazardous chemical, civil explosive, fireworks and firecrackers [48]. Generally speaking, the intelligent prevention and control system of vehicle active safety is composed of driving records, satellite navigation system, safety assistance driving system, driver behavior monitoring, etc. The real-time monitoring on drivers’ driving behaviors, vehicle driving environment and overloading of vehicle is carried out by using the system, so as to accomplish the warning of driving safety hazards.
Algorithm for identifying key risk chains
The algorithm of risk chain identification for SVDRN was divided into two stages: network structure scanning and the identification of key risk chain. After obtaining the comprehensive importance of each node in the SVDRN, adjacency matrix B was set to store the connection relationship and the direction of the arc among network nodes, and matrix D was used to store the comprehensive importance of nodes, both of which was based on ergodic search algorithm [49]. The first step in the second stage was to read data from technical management database, so as to record the threshold of risk status for network nodes. The reverse search of SVDRN along the directed arc was carried out according to the path number. The node risk degree was defined as the product of the node’s comprehensive importance and its risk state threshold. The risk degree of path i was the product of the risk degree of all nodes on the path, marked as L(i). The path set with top 3 risk degrees was obtained, and all nodes of the key risk chains were output in sequence. Fig 2 illustrates the steps of identification algorithm for key risk chains.
[Figure omitted. See PDF.]
Result discussion
Analysis result of the network topology
As can be shown from Fig 1, there were 30 nodes and 42 edges in the SVDRN, with an mean network density of 0.731 and a standard deviation of 4.155. The steps of all paths in the SVDRN were less than or equal to 4, and the average path length was 2.413. As a matter of fact, most real risk networks have small average path length, which leads to the implications of small world [4]. The average path length is inversely correlated with efficiency of risk transmission. Moreover, the average value of clustering coefficient in the SVDRN was 0.246, indicating that the overall clustering level was more than that of the local. In other words, there were no small groups within the SVDRN. Although the sum total of nodes in the SVDRN was not so large, the coupling relationship from the root node to the leaf node was complex. For example, the number of possible paths from M1 to R6 was 22. Therefore, the structure of SVDRN conformed to the property of complexity and small world. The degree distribution of SVDRN was fitted as a straight line with negative slope, as shown in Fig 3. Y-axis was the natural logarithm of frequency in the Fig 3, and X-axis was natural logarithm of degree value. It had the approximate fit y = -0.6788x + 1.9665 with R2 = 0.5568. The nodes with low degree value accounted for the vast majority in a scale-free network, while the nodes with high degree value were only very few [50]. The result denoted that the cumulative distribution of node degree within the SVDRN conformed to a power-law distribution, hence the SVDRN was scale-free.
[Figure omitted. See PDF.]
The degree distribution of 30 nodes in the SVDRN is displayed in Fig 4, which the mean total degree is 13.103. The four nodes of F7 (improper operation), F2(speeding), M3 (inefficient driver management) and F16 (loss of vehicle control) had a higher total degree in the network. The node of F7 had the highest in-degree of 258, indicating that F7 was in a core position in the SVDRN and had a strong correlation with other nodes. The in-degree of F7 was extremely high due to their parent nodes at the number of 17. An accident due to improper operation occurred, involving 1 to 4 parent nodes. The node of M3 was the highest node of the out-degree value, with the value of 82. Measured by the annual training time and assessment results of drivers, it indicated that the driver’s education and awareness levels were the main factors inducing accidents. In addition, the node of F16 was the second highest in-degree with the value at 68, indicating the running state of the vehicle body. Whether ESP works or not was applied to measure the body stability. The node of F2 belonged to the intermediate node, and its total degree value was relatively high. Accordingly, speeding played a important the role of bridge in vehicle driving safety.
[Figure omitted. See PDF.]
Fig 5 demonstrates the distribution of closeness centrality, betweenness centrality, clustering coefficient and eigenvector centrality of 30 nodes in SVDRN. Except for a few nodes, there was little difference in the closeness centrality of nodes, and the mean value of closeness centrality was 50.302. The top three nodes of closeness centrality in the risk network were F7(improper operation), M3(inefficient driver management), and F16 (loss of vehicle control). From the global perspective of the network, these nodes located in the center of the SVDRN, and were close to other nodes, the result of which was not susceptible to be affected by other nodes and played a more obvious role in risk transmission.
[Figure omitted. See PDF.]
The calculation result of the nodes in SVDRN showed that the betweenness centrality among nodes varied tremendously. The mean value of betweenness centrality was 3.654. Almost 30% of nodes had the zero value, so it was hard for them to be hubs in the shortest path between pairs of other nodes. The node of F7 (improper operation) with the largest value of betweenness centrality located in the core position of SVDRN, indicating that it was easier to affect the whole network than other nodes. The following nodes were F16 (loss of vehicle control), M3 (inefficient driver management), F2(speeding) and M4 (negligent vehicle technical management).
The clustering coefficient with the value of non-zero had 13 nodes in SVDRN, and the rest of 17 nodes were zero. According to the concept of the clustering coefficient, the clustering effect for nodes with only one neighbor node did not exist, such as R2 and R5. The values of clustering coefficient of seven nodes, including F1 (driving without a license), F2 (speeding), F5 (drunk driving), F8 (negligent observation), F9 (driving in the unapproved line), R3 (running over pedestrians), and R4 (fire), were relatively remarkable in Fig 4. The nodes of M3, F16 and F7 with low value are worthy of note, which could be explained that these three nodes had many neighbor nodes, but the probability of direct connection between neighbor nodes was low.
The mean values of eigenvector centrality in the SVDRN was 0.223. The node of F7 (improper operation) was the node with the highest eigenvector centrality at the value of 0.766. To sum up, improper operation ranked the first risk factor according to four indicators of topological features except for clustering coefficient. The following nodes with a higher value of eigenvector centrality were F16 (loss of vehicle control), M3 (inefficient driver management) and F2 (speeding). The four nodes were obtained in the light of three aspects: the amount of nodes connected to a node, the frequency acting as a intermediary node, and the importance ranking of neighbor nodes. The numeric value of eigenvector centrality for other nodes in the SVDRN were no more than 0.3.
Evaluation result of comprehensive importance of network nodes
Fig 6 shows the comprehensive importance of nodes in the SVDRN calculated corresponding to Eqs (6)~(11), with the mean value of 0.185 and a standard deviation of 0.119. The larger its value was, the more core it was in the whole network. According to the foregoing evaluation model of the node importance, the order of influencing factors were listed from high to low as follows: (1) the node importance at the value of more than 0.3 for driver factors and vehicle factors: F7 (improper operation), F2 (speeding), F16 (loss of vehicle control), F1(driving without a license), F5 (drunk driving), F9 (driving in the unapproved line), F8 (negligent observation). These 7 nodes were also the top seven nodes importance in the whole risk network. (2) the importance sequence of management factors: M3(inefficient driver management), M4(negligent vehicle technical management), M1(vehicle ownership), and M2(failed vehicle dynamic monitoring); (3) the importance sequence of environmental factors: E3(non-standard road alignment), E4(unfavorable terrain), E2(inadequate road infrastructure), and E1(severe weather).
[Figure omitted. See PDF.]
As can be shown from Fig 6, driver factors played a leading role in vehicle driving safety, especially the driver’s emergency decisions and behavior habits. When drivers encounter all kinds of emergencies, they need to make disposal decisions in a very short time. The speed and accuracy of driver’s response capability under emergency situation are the main influence factors for the occurrence and severity of traffic accidents [51]. The drivers’ response capability under emergency situation is mainly manifested in four aspects: psychological quality, physiological characteristics, decision-making ability and driving skills. The previous studies was more focused on the evaluation of drivers’ response capability, and found that the driver’s individual differences is significant [52, 53]. At present, the analysis of typical accident cases is mostly adopted to improve the stress response of professional drivers in the process of safety management of road transport enterprises. A few of transport enterprises use the driving simulator to conduct a certain number of danger scene training, but the management effect is insignificant because of not consistent with the real traffic conditions. In fact, the major accident is rarely triggered by a single factor.
The Apriori algorithm is used to analyze the combination correlation of four risk factors from Table 1. The analysis results show that the combination correlation of improper operation (F7) and non-standard road alignment (E3) has the highest, with the support of 51.81% and the confidence of 69.35%. In addition, the combination of speeding (F2) and improper operation (F7) has the second highest support rate at the value of 32.53%, and their confidence is 75.0%.
There were minor differences of 0.017 in the node importance among environmental factors, among which severe weather had the smaller value. However, the accident rate of severe weather (E1) was 31.33% in the single-vehicle accidents with 10+ fatalities. The accident rates of rainy, snowy and foggy days were 30.12%, 8.43% and 4.82%, respectively. A small number of the incidents involved two weather attributes, such as sleet or rain mixed with fog. The combination support of E1-F7 and E1-F2 is 32.53% and 27.71%, respectively.
Results of risk chain identification
The monitoring practice of risk status for three passenger vehicles in a transport enterprise was carried out, so as to prove the scientific validity of the proposed method in this paper. The data came from a technical management database throughout the life cycle of vehicles. If the enterprise has not established an integrated database, the data for this study can also be obtained step-by-step on the premises that the enterprise has the active safety intelligent prevention and control system. The real-time dynamic information of test vehicle is applied to identify risk paths in combination with static daily management information. On the basis of data collection, the proposed algorithm for key risk chain identification was used to calculate the risk degree of all the risk chains. Table 3 demonstrates the analysis results of the risk chain analysis for the test vehicles, and the path set with the highest risk degree is output. The path length of risk chains displayed in Table 3 were short, which explained why single-vehicle accidents happened frequently, to a certain extent. For transport enterprises, the key risk chains of multiple vehicles are comprehensively analyzed to find out the high-frequency risk chains and risk factors. Combined with the comprehensive importance of their nodes, the enterprise can take effective measures before the accident. Any one or more causation nodes in the key risk chains can be removed, the loss of the accident will be reduced.
[Figure omitted. See PDF.]
Conclusions
A comprehensive mode in this paper that integrated SDG model, hierarchical threshold, the multi-attribute comprehensive evaluation model and ergodic search algorithm was proposed to monitor the risk status of SVDRN and identify key risk chains. Among these, SDG model was used to construct the SVDRN, the multi-attribute comprehensive evaluation model were adopted to evaluate the node’s comprehensive importance, jointed with hierarchical threshold of node risk status for measuring the risk intensity of accident causation factors, three of which provided a basis for the algorithm of key risk chain recognition. This study provides novel insights into the identification of high-risk factors and paths for single-vehicle accidents and providing effective and specific countermeasures for mangers, especially when the prevention of major accidents is to be explored. With the gradual improvement of vehicle intelligent level for road transport enterprises, its active safety intelligent prevention and control system can be upgraded to realize the synchronous management of dynamic risk information and static daily management information for each vehicle on the monitoring platform. The model and algorithm proposed in this paper are embedded into the computing software to identify the high-risk risk chain and risk factors. The dynamic data of the active safety intelligent prevention and control system can ensure the real-time update. The change frequency of vehicle daily management data in Table 2 is lower, but they also needs to be updated regularly in the practice of enterprise safety management, so as to ensure the accuracy of risk assessment.
The SDG model of SVDRN and the rationality of evaluation indicators corresponding to risk factors need to be further refined, so that the data of safety management throughout the life cycle of vehicles is fully utilized. Moreover, due to significant differences in the risk mechanism of single- and multi-vehicle accidents, multi-vehicle driving risk network is a subject worth studying. Comparing the similarities and differences between them is of great reference value to the accident prevention in the process of road transportation. In consideration of complexity of risk coupling mechanism, the key risk chain recognition algorithm established in this paper needs further refinement, especially the computational method of path risk degree.
Supporting information
S1 File. Matrix M-data.
https://doi.org/10.1371/journal.pone.0302216.s001
(XLSX)
S2 File. The data of Apriori algorithm.
https://doi.org/10.1371/journal.pone.0302216.s002
(TXT)
S3 File. Apriori algorithm code.
https://doi.org/10.1371/journal.pone.0302216.s003
(TXT)
Citation: Li F, Wang X, Feng Z, Wang J, Li M, JIANG K, et al. (2024) Risk chain identification of single-vehicle accidents considering multi-risk factors coupling effect. PLoS ONE 19(5): e0302216. https://doi.org/10.1371/journal.pone.0302216
About the Authors:
Fangyuan Li
Contributed equally to this work with: Fangyuan Li, Xia Wang
Roles: Methodology, Software, Writing – original draft
E-mail: [email protected]
Affiliation: School of Automobile Engineering, Shandong Jiaotong University, Jinan, China
ORICD: https://orcid.org/0000-0002-3911-6938
Xia Wang
Contributed equally to this work with: Fangyuan Li, Xia Wang
Roles: Software
Affiliation: School of Transportation, Shandong University of Engineering and Vocational Technology, Jinan, China
Zenglei Feng
Roles: Investigation, Writing – review & editing
Affiliation: School of Transportation, Shandong University of Engineering and Vocational Technology, Jinan, China
Jian Wang
Roles: Methodology, Project administration
Affiliation: School of Automobile Engineering, Shandong Jiaotong University, Jinan, China
Mengdi Li
Roles: Investigation, Writing – review & editing
Affiliation: School of Automobile Engineering, Shandong Jiaotong University, Jinan, China
Kun JIANG
Roles: Investigation, Project administration
Affiliation: School of Automobile Engineering, Shandong Jiaotong University, Jinan, China
Changli ZHAO
Roles: Investigation, Writing – review & editing
Affiliation: School of Automobile Engineering, Shandong Jiaotong University, Jinan, China
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
[/RAW_REF_TEXT]
1. National Bureau of Statistics. China statistical yearbook-2022. Beijing: China statistics press; 2022.
2. The Traffic Management Bureau of Ministry of Public Security. Annual report of road traffic accidents in China-2021; 2022.
3. Zhang Y.Y., Dong C.T., Guo W.Q., Dai J.B., Zhao Z.M. Systems theoretic accident model and process (STAMP): a literature review, Safety Science. 2021: e105596.
4. Zhou Z.P., Irizarry J., Guo W.Y. A network-based approach to modeling safety accidents and causations within the context of subway construction project management. Safety Science. 2021; 139(10): e105261.
5. Yang Y.X., Wang Z., Xu Y.F., Hu Y.Y., Jia X.P., Wang F. Research on important nodes and key propagation paths of accident chains in chemical industry parks. Journal of Chemical Engineering of Chinese Universities. 2020; 34(2): 503–511.
6. Zhang J.J., Xu K.L., You G., Wang B.B., Zhao L. Causation Analysis of Risk Coupling of Gas Explosion Accident in Chinese Underground Coal Mines. Risk Analysis. 2019; 39 (7): 1634–1646.
7. Kazumoto M., Michiaki S. Analysis of Single-Vehicle Accidents in Japan Involving Elderly Drivers. SAE International Journal of Transportation Safety. 2018; 6(1): 19–28.
8. Islam S., Jones S.L. Dye D. Comprehensive Analysis of Single- and Multi-Vehicle Large Truck At-Fault Crashes on Rural and Urban Roadways in Alabama. Accident Analysis and Prevention. 2014; 67: 148–158.
9. Chen F., Chen S.R. Injury severities of truck drivers in single-and multi-vehicle accidents on rural highways. Accident Analysis and Prevention. 2011; 43(5):1677–1688.
10. Alnawmasi N., Mannering F. A statistical assessment of temporal instability in the factors determining motorcyclist injury severities. Analytic Methods in Accident Research. 2019; 22: e100090.
11. Fountas G., Fonzone A., Gharavi N., Rye T. The joint effect of weather and lighting conditions on injury severities of single-vehicle accidents. Analytic Methods in Accident Research. 2020; 27(5): e100124.
12. Pokorny P., Jensen J.K., Gross F., Pitera K. Safety effects of traffic lane and shoulder widths on two-lane undivided rural roads: A matched case-control study from Norway. Accident Analysis and Prevention. 2020; 144: 1–11.
13. Wiratama BS, Hsu LM, Yeh YS, Chen CC, Saleh W, Liu YH, Pai CW. Joint Effect of Heavy Vehicles and Diminished Light Conditions on Paediatric Pedestrian Injuries in Backover Crashes: A UK Population-Based Study, International Journal of Environmental Research and Public Health. 2022;
14. Su J.B., Sze N.N., Bai L. A joint probability model for pedestrian crashes at macroscopic level: Roles of environment, traffic, and population characteristics, Accident Analysis and Prevention. 2021; 150(1): e105898.
15. Kitara D.L., Karlsson O. The effects of economic stress and urbanization on driving behaviours of Boda-boda drivers and accidents in Gulu, Northern Uganda: a qualitative view of drivers, Pan African Medical Journal. 2020; 36: e21382.
16. Papadimitriou E., Filtness A., Theofilatos A., Ziakopoulos A., Quigley C., Yannis G. Review and ranking of crash risk factors related to the road infrastructure, Accident Analysis and Prevention. 2019; 125: 85–97.
17. Wang L.Z., Wang Y.P., Shi L.Y., Xu H.Z. Analysis of risky driving behaviors among bus drivers in China: The role of enterprise management, external environment and attitudes towards traffic safety. Accident Analysis and Prevention. 2022;
18. Olowosegun A., Babajide N., Akintola A., Fountas G., Fonzone A. Analysis of pedestrian accident injury-severities at road junctions and crossings using an advanced random parameter modelling framework: The case of Scotland, Accident Analysis and Prevention. 2022;
19. Ulak M.B., Ozguven E.E., Vanli O.A. Multivariate random parameter Tobit modeling of crashes involving aging drivers, passengers, bicyclists, and pedestrians: Spatiotemporal variations. Accident Analysis and Prevention. 2018; 121: 1–13.
20. Andrés F.R., Carlos V. Spatio-temporal correlation study of traffic accidents with fatalities and injuries in Bogota (Colombia). Accident Analysis and Prevention. 2021; 149: e105848.
21. Vicente G., Ana S., Isabel M. et al. Analysis of occupational accidents in Spain using shrinkage regression methods. Safety Science. 2021; 133: e105000. 133.
22. Park B.J., Lord D., Wu L.T. Finite mixture modeling approach for developing crash modification factors in highway safety analysis. Accident Analysis and Prevention. 2016; 9: 274–287.
23. Xiong Y.G., Tobias J.L., Mannering F.L. The analysis of vehicle crash injury-severity data: A Markov switching approach with road-segment heterogeneity. Transportation Research Part B: Methodological. 2014; 67: 109–128.
24. Hosseini N., Givehchi S., Maknoon R. Cost-based fire risk assessment in natural gas industry by means of fuzzy FTA and ETA. Journal of Loss Prevention in the Process Industries. 2020:e104025.
25. Chang Y.J., Wu X.F., Chen G.M. et al. Comprehensive risk assessment of deepwater drilling riser using fuzzy Petri net model. Process Safety and Environmental Protection. 2018; 117: 483–497.
26. Chen J., Li H., and Sheng D. A hybrid data-driven modeling method on sensor condition monitoring and fault diagnosis powers plants. Int. J. Electrical Power and Energy Systems. 2015; 71: 274–284.
27. Wu G.H., Yuan D.P., Yin J.Y., Xiao Y.Q., Ji D.X. A framework for monitoring and fault diagnosis in nuclear power plants based on signed directed graph methods. Frontiers in Energy Research. 2021:e641545.
28. Zhang X.H., Xu W., Wu F., Ren X.C., Feng C.Y. Intelligent method for characteristic event tracing and prediction of cascading failures in AC/DC hybrid power grid. Automation of Electric Power Systems. 2021; 45(10): 17–24.
29. Meng X.K., Zhu J.Y., Fu J.Y., Li T.S., Chen G.M. An accident causation network for quantitative risk assessment of deepwater drilling. Process Safety and Environmental Protection. 2021; 148: 1179–1190.
30. Bai C.F., Gao W.S., Ding D.W. Power transformer risk assessment method using small world network. High Voltage Engineering. 2010; 25(4): 869–872.
31. The State Council of the P.R.C. Regulations on the Reporting, Investigation and Handling of production Safety Accidents (No. 493). Beijing, China. 2007.
32. The Ministry of Public Security of the P.R.C. Measures for the Administration of Points Assigned for Road Traffic Violations (No.163). Beijing, China. 2022.
33. The President of the P.R.C. Law of the People’s Republic of China on Road Traffic Safety (No. 81). Beijing, China. 2021.
34. The State Council of the P.R.C. Regulations for the Implementation of the Road Traffic Safety Law(No. 405). Beijing, China. 2017.
35. Standardization Administration of China. China national standard: Blood-breath Alcohol Concentration and Examination for Driving (GB19522-2010). Beijing, China. 2010.
36. Ministry of transport of the P.R.C. Commercial Vehicle Running Dangerous Warning System Technical Requirements and Test Procedures (JT/T883-2014). Beijing, China. 2014.
37. Standardization Administration of China. Performance Requirements and Test Methods of Tire Pressure Monitoring System for Passenger Cars(GB 26149–2017). Beijing, China. 2017.
38. Ministry of transport of the P.R.C. Technical Requirement and Test method of Tire Pressure Monitoring System for Commercial Vehicle (JT/T1429-2022). Beijing, China. 2022.
39. Standardization Administration of China. Grade of precipitation (GB/T28592-2012). Beijing, China. 2012.
40. Ministry of transport of the P.R.C. Design Specifications for Highway Safety Facilities (JTG D81-2017). Beijing, China. 2017.
41. Ministry of transport of the P.R.C. Design Specification for Highway Alignment (JTG D20-2017). Beijing, China. 2017.
42. Ministry of transport of the P.R.C. Continuing Education Scheme for Road Transport Drivers (No.18). Beijing, China. 2019.
43. De Nooy W., Mrvar A., Batagelj V. Exploratory Social Network Analysis with Pajek (Structural Analysis in the Social Science). Cambridge: Cambridge University Press; 2005.
44. Freeman L C. A Set of Measurs of Centrality Based on Betweenness. Sociometry. 1977; 40(1): 35–41.
45. Bavelas A. Communication Patterns in Task-Oriented Groups. The Journal of Acoustical Society of America. 1950; 22: 725–730.
46. Hu L.W., Yang H.F., He Y.R., Zhao X.T., Yin Y., Tian H.L., et al. Driving risk identification of commercial trucks based on complex network theory, Journal of Transportation Engineering and Information. 2022; 20(01): 128–134.
47. Yu H., Liu Z., Li Y.J. Key nodes in complex networks identified by multi-attribute decision-making method. Acta Physica Sinica. 2013; 62(2): 9.
48. Qi W.W., Zhu S.F., Hu J.S. Correlation analysis of real-time warning factors for construction heavy trucks based on electrified supervision system. Sustainability. 2022; 14 (7): 10944.
49. Pronzato L., Wynn H.P., Zhigljavsky A.A. Finite sample behaviour of an ergodically fast line-search algorithm. Computational Optimization and Applications. 1999; 14 (1): 75–86.
50. Zhou J., Xu W.X., Guo X., Ding J. A method for modeling and analysis of directed weighted accident causation network (DWACN). Physica A-statistical Mechanics And Its Applications. 2015; 437:263–277.
51. Robinson D., Campbell R. Contributory factors to road accidents. UK Department for Transport. 2006.
52. Davoodia S.R., Hamid H., Pazhouhanfar M. et al. Motorcyclist perception response time in stopping sight distance situations. Safety Science. 2012; 50(3): 371–377.
53. Liu Y.T., Hua J., Wei L. Emergency Response Ability of Drivers Under Risk Guidance Situations. Journal of Transport Information and Safety. 2019; 37(3): 35–41.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
© 2024 Li et al. This is an open access article distributed under the terms of the Creative Commons Attribution License: http://creativecommons.org/licenses/by/4.0/ (the “License”), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Abstract
The real-time monitoring on the risk status of the vehicle and its driver can provide the assistance for the early detection and blocking control of single-vehicle accidents. However, complex risk coupling relationship is one of the main features of single-vehicle accidents with high mortality rate. On the basis of investigating the coupling effect among multi-risk factors and establishing a safety management database throughout the life cycle of vehicles, single-vehicle driving risk network (SVDRN) with a three-level threshold was developed, and its topology features were analyzed to assessment the importance of nodes. To avoid the one-sidedness of single indicator, the multi-attribute comprehensive evaluation model was applied to measure the comprehensive effect of characteristic indicators for nodes importance. A algorithm for real-time monitoring of vehicle driving risk status was proposed to identify key risk chains. The result revealed that improper operation, speeding, loss of vehicle control and inefficient driver management were the sequence of top four risk factors in the comprehensive evaluation result of nodes importance (mean value = 0.185, SD = 0.119). There were minor differences of 0.017 in the node importance among environmental factors, among which non-standard road alignment had the larger value. The improper operation and non-standard road alignment were the highest combination correlation of factors affecting road safety, with the support of 51.81% and the confidence of 69.35%. This identification algorithm of key risk chains that combines node importance and its risk state threshold can effectively determine the high-frequency risk transmission paths and risk factors through multi-vehicle test, providing a basis for centralization management of transport enterprises.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer