A method for extracting alternative indicators for road traffic accident risk assessment
By dividing the road into equal length section units, analyzing the parameter sequence data related to driving risks, building regression and discrete selection models, and generating alternative indicators, the problems of difficulty in obtaining data and poor applicability in road traffic accident risk assessment are solved, and the problem of data acquisition and poor applicability in road traffic accident risk assessment is realized. Fast and accurate risk assessment and short-term safety assessment.
Patent Information
- Application Number
- CN202210260054.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-16
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing technology has problems in the assessment of road traffic accident risk, which is difficult to obtain data, poor timeliness and difficult to guarantee quality. The existing alternative indicators such as traffic conflict time are poorly applicable in highway scenarios, and the extraction method is There are problems of subjectivity and inefficiency.
By dividing the road into equal length section units, the parameter sequence data related to driving risk is collected, the parameter importance is analyzed, the regression model and discrete selection model are constructed, alternative indicators are generated, and accident risk calculation equations are constructed to evaluate road risk. .
Achieve rapid and accurate extraction of alternative indicators from large-scale original time series data, suitable for any factors related to driving risks, can evaluate road risks instead of accident data, and safety for roads and short-term traffic control programs that lack accident statistics evaluate.
Smart Images

Figure CN114742362B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of highway traffic safety, and in particular to a method for extracting alternative indicators for road traffic accident risk assessment. Background Art
[0002] Road accident prevention is the focus and difficulty of the transportation industry. Current scientific research, especially applied engineering research, mostly uses historical accident statistics as the basis for judging the accident risk of roads. Although it is simple and convenient, the research results show that the accident data has the following insurmountable defects:
[0003] (1) Accident data is difficult to obtain: It is impossible to query the annual accident statistics of a specific road, including the time, location, severity, type, property loss, and road environment and weather when the accident occurred. At present, Chinese scholars tend to use open source data from European and American countries, but due to huge differences in traffic regulations, population, and motor vehicle ownership, the research results cannot be well transplanted into my country's traffic safety practice.
[0004] (2) The acquisition cycle of accident data is long and the timeliness is poor: accidents are low-probability events, and it takes several years of accumulation to obtain a sample of sufficient size to conduct research. This feature makes it difficult to apply it to road safety research on newly built roads and other areas where there is a lack of accident data accumulation, and it is also impossible to conduct short-term safety evaluations on traffic management strategies.
[0005] (3) The quality of accident data is difficult to ensure: accidents are recorded manually, lacking a unified recording standard, and there are many underreporting phenomena.
[0006] In the case of the above defects in accident data, it is necessary to find alternative indicators for road traffic accident risk assessment, but the alternative indicators have the following defects when applied: Taking the most widely used traffic conflict time as an example, it is defined as: two or more road users approach each other in space or time, and there is a risk of collision if their movements remain unchanged. If the movements of each other remain unchanged, this assumption is difficult to meet in practice, and some driving scenes that are not dangerous will be mistakenly identified as traffic conflicts. Traffic conflicts are mainly applied to road nodes with a large number of trajectory intersections, such as intersections, ramp entrances and exits, toll station areas, etc., or urban roads with complex and changeable traffic flow conditions. However, for highways with free flow and few intersections, or tunnels and bridge sections where trajectory data is difficult to obtain, traffic conflicts are less applicable.
[0007] Except for the traffic conflict index, other index extraction methods are mainly based on qualitative judgment. Combined with driving videos, the values of the mutation points of the vehicle speed or acceleration curve or the values of the mutation points of the driver's physiological / psychological index are selected as indicators for determining the level of risk. This extraction method is highly subjective, lacks unified standards, and has low extraction efficiency when the amount of data is large. Summary of the invention
[0008] Purpose of the invention: In view of the above problems, the purpose of the present invention is to provide a method for extracting alternative indicators for road traffic accident risk assessment, which can quickly and accurately extract alternative indicators for road traffic accident risk assessment from large-scale original time series data.
[0009] Technical solution: A method for extracting alternative indicators for road traffic accident risk assessment of the present invention comprises the following steps:
[0010] S10, dividing the road to be analyzed into q equal-length road section units, and collecting a sequence data set of parameters associated with driving risks in each road section unit in chronological order;
[0011] S20, analyzing the importance of each parameter character in the parameter sequence data set to the accident frequency, and retaining the characters with an importance exceeding I0;
[0012] S30, constructing a regression model using the retained characters, retaining characters with a confidence level not less than P0 in the regression model and constructing a discrete selection model, retaining characters with a confidence level not less than P1 as alternative indicators;
[0013] S40, using alternative indicators to construct an accident risk calculation equation to evaluate the risk level of the road.
[0014] Furthermore, before step S20, the method further includes:
[0015] The parameter sequence data set C = {c1, c2, ..., c n} is transformed into a standard sequence set C″={c″1,c″2,…,c″ with a mean of 0 and a standard deviation of 1 that conforms to the normal distribution n}, the expression is:
[0016]
[0017] In the formula, μ is the mean of the sequence data set C, and σ is the standard deviation of the sequence data set C;
[0018] Then, the dimension reduction formula is used to reduce the dimension of the standard sequence set C″, converting the standard sequence set of length n into m subsequences of equal length, and then replacing each subsequence with its mean value, and recording it as a string C′={c′1,c′2,…,c′ m}, the smooth curve of the sequence data set C becomes a broken line segment after dimensionality reduction, and the dimensionality reduction formula is:
[0019]
[0020] In the formula Indicates the compression ratio, which is an integer.
[0021] Further, step S20, analyzing the importance of each parameter character in the parameter sequence data set to the accident frequency includes:
[0022] According to the Gaussian distribution table, find the interval splitting point and convert the reduced-dimensional string C′={c′1,c′2,…,c′ m} is divided into k equal probability intervals, each probability interval is set with a character, and the frequency of each character appearing in each road section unit is counted x i,w , i represents the road segment, w represents the character category, and the frequency x i,w To explain the variables, the importance of each character to the accident frequency was analyzed.
[0023] Furthermore, parameters associated with driving risks include vehicle speed, acceleration, tire mechanical response, and driver physiological and psychological indexes.
[0024] Furthermore, the regression model in step S30 includes a regression model that can establish a linear or generalized linear relationship with traffic accidents.
[0025] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:
[0026] 1. The alternative indicators extracted by the present invention can replace accident data to assess road risks. The extraction method is highly portable, and the indicators can be extracted from any time series data of factors related to driving risks;
[0027] 2. The present invention also provides a risk calculation equation based on the alternative indicator, which is convenient for practical application;
[0028] 3. After the alternative indicators are extracted, the indicators and the corresponding risk calculation equations can be directly used to analyze the accident risks on roads lacking accident statistics. It is also suitable for the safety evaluation of short-term traffic control plans and has strong practical significance. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION
[0030] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments.
[0031] The present embodiment describes a method for extracting alternative indicators for road traffic accident risk assessment, the flow chart of which is as follows: Figure 1 As shown, in this embodiment, Jinliwen Expressway is taken as an example to describe the implementation process of the extraction method in detail:
[0032] S10, dividing the road to be analyzed into q road section units of equal length, and collecting a sequence data set of parameters associated with driving risks in each road section unit in chronological order.
[0033] Parameters associated with driving risks include vehicle speed, acceleration, tire mechanical response, and driver physiological and psychological indexes. In this embodiment, the six-component force of the tire, a factor closely related to driving risk, is selected. The left line (Lishui direction) and the right line (Wenzhou direction) of the Jinliwen Expressway K117-K189 section are divided into 144 equal-length section units, where q=144, denoted as i=1,2,…144. In each section unit, the six-component force sequence data set of the Jinliwen Expressway K117-K189 section is collected in chronological order, denoted as C={c1,c2,…,c n In this example, the tire six-component force sequence data set is collected based on real vehicle tests. The experimental vehicle is equipped with a six-component force measurement system, which collects the forces and moments of the tire in three directions (X, Y, and Z) during driving, which are recorded as longitudinal force Fx, lateral force Fy, vertical force Fz, overturning moment Mx, rolling moment My, and return moment Mz.
[0034] The tire six-component force sequence data set C = {c1, c2, …, c n} is transformed into a standard sequence set C″={c″1,c″2,…,c″ with a mean of 0 and a standard deviation of 1 that conforms to the normal distribution n}, the expression is:
[0035]
[0036] Where μ is the mean of the sequence data set C, and σ is the standard deviation of the sequence data set C.
[0037] The standard sequence set C″ is reduced in dimension using the dimensionality reduction formula. The standard sequence set of length n is converted into m subsequences of equal length. Then, the mean of each subsequence is used to replace the subsequence, and recorded as a string C′={c′1,c′2,…,c′ m}, the smooth curve of the sequence data set C becomes a broken line segment after dimensionality reduction, and the dimensionality reduction formula is:
[0038]
[0039] In the formula is the compression ratio, which is an integer. In this example, the value is 50.
[0040] S20, analyzing the importance of each parameter character in the parameter sequence data set to the accident frequency, and retaining characters with an importance exceeding I0.
[0041] According to the Gaussian distribution table in Table 1, find the interval splitting point and convert the reduced-dimensional string C′={c′1,c′2,…,c′ m} is divided into k equal probability intervals, each probability interval is set with a character, in this example k = 8, the character calibration results are shown in Table 2, a total of 48 types of characters are used in this embodiment, the longitudinal force Fx includes characters a1-h1, the lateral force Fy includes characters a2-h2, the vertical force Fz includes characters a3-h3, the turning moment Mx includes characters a4-h4, the rolling moment My includes characters a5-h5, and the self-centering moment Mz includes characters a6-h6.
[0042] Table 1 Gaussian distribution table
[0043] k 3 4 5 6 7 8 9 10 <![CDATA[β1]]> -0.43 -0.67 -0.84 -0.97 -1.07 -1.15 -1.22 -1.28 <![CDATA[β2]]> 0.43 0 -0.25 -0.43 -0.57 -0.67 -0.76 -0.84 <![CDATA[β3]]> / 0.67 0.25 0 -0.18 -0.32 -0.43 -0.52 <![CDATA[β4]]> / / 0.84 0.43 0.18 0 -0.14 -0.25 <![CDATA[β5]]> / / / 0.97 0.57 0.32 0.14 0 <![CDATA[β6]]> / / / / 1.07 0.67 0.43 0.25 <![CDATA[β7]]> / / / / / 1.15 0.76 0.52 <![CDATA[β8]]> / / / / / / 1.22 0.84 <![CDATA[β9]]> / / / / / / / 1.28
[0044] Count the frequency of each character in each road section unit on 144 road section units x i,w , i represents the road segment, w represents the character category, and the frequency x i,w To explain the variables, the importance of each character to the accident frequency is analyzed. In this embodiment, the importance of each character to the accident frequency is calculated using the Shapley value method, taking I=0.6, and there are 4 characters with importance exceeding I, as shown in Table 3.
[0045] Table 2 Character calibration results
[0046] character Fx(kN) character Fy(kN) character Fz(kN) <![CDATA[a1]]> (-0.65,-0.38] <![CDATA[a2]]> (-0.45,-0.17] <![CDATA[a3]]> (4.45,4.64] <![CDATA[b1]]> (-0.38,-0.32] <![CDATA[b2]]> (-0.17,-0.12] <![CDATA[b3]]> (4.64,4.7] <![CDATA[c1]]> (-0.32,-0.28] <![CDATA[c2]]> (-0.12,-0.08] <![CDATA[c3]]> (4.7,4.73] <![CDATA[d1]]> (-0.28,-0.24] <![CDATA[d2]]> (-0.08,-0.04] <![CDATA[d3]]> (4.73,4.77] <![CDATA[e1]]> (-0.24,-0.21] <![CDATA[e2]]> (-0.04,0] <![CDATA[e3]]> (4.77,4.8] <![CDATA[f1]]> (-0.21,-0.16] <![CDATA[f2]]> (0,0.04] <![CDATA[f3]]> (4.8,4.84] <![CDATA[g1]]> (-0.16,-0.11] <![CDATA[g2]]> (0.04,0.1] <![CDATA[g3]]> (4.84,4.89] <![CDATA[h1]]> (-0.11,0.15] <![CDATA[h2]]> (0.1,0.3] <![CDATA[h3]]> (4.89,5.12] character Mx(N·m) character My(N·m) character Mz(N·m) <![CDATA[a4]]> (-0.01,0.1] <![CDATA[a5]]> (-0.17,-0.05] <![CDATA[a6]]> (-0.02,0] <![CDATA[b4]]> (0.1,0.12] <![CDATA[b5]]> (-0.05,-0.03] <![CDATA[b6]]> (0,0.003] <![CDATA[c4]]> (0.12,0.13] <![CDATA[c5]]> (-0.03,-0.02] <![CDATA[c6]]> (0.003,0.006] <![CDATA[d4]]> (0.13,0.14] <![CDATA[d5]]> (-0.02,-0.01] <![CDATA[d6]]> (0.006,0.009] <![CDATA[e4]]> (0.14,0.15] <![CDATA[e5]]> (-0.01,0] <![CDATA[e6]]> (0.009,0.011] <![CDATA[f4]]> (0.15,0.17] <![CDATA[f5]]> (0,0.01] <![CDATA[f6]]> (0.011,0.014] <![CDATA[g4]]> (0.17,0.18] <![CDATA[g5]]> (0.01,0.03] <![CDATA[g6]]> (0.014,0.018] <![CDATA[h4]]> (0.18,0.24] <![CDATA[h5]]> (0.03,0.18] <![CDATA[h6]]> (0.018,0.04]
[0047] Table 3 Characters retained after importance analysis
[0048] Importance Ranking character 1 <![CDATA[e5]]> 2 <![CDATA[d4]]> 3 <![CDATA[g2]]> 4 <![CDATA[b4]]>
[0049] S30, constructing a regression model using the retained characters, retaining characters with a confidence level not less than P0 in the regression model and constructing a discrete selection model, retaining characters with a confidence level not less than P1 as alternative indicators;
[0050] The characters that remain are The four characters are used as explanatory variables to construct a regression model. In this embodiment, a negative binomial regression model is used, and the expression of the model is:
[0051] γ i =EXP(β0+∑β w x i,w )
[0052] Where γ i is the accident frequency prediction value of the i-th road section, βw is the coefficient to be estimated, and β0 is a fixed term.
[0053] The negative binomial regression model was fitted using STATA software, and the characters with a confidence level of P ≥ 95% were retained. The fitting results of the model are shown in Table 4. The characters retained are and
[0054] Table 4 Estimation results of negative binomial regression model
[0055] character <![CDATA[β w ]]> Keep <![CDATA[e5]]> -0.003 no <![CDATA[d4]]> 0.049 yes <![CDATA[g2]]> 0.036 yes <![CDATA[b4]]> 0.007 no
[0056] The frequency x of the retained 2-type characters d4 and g2 i,w is the explanatory variable: and Use these two explanatory variables to build a discrete selection model, retain the characters in the model with a confidence level of not less than 95%, and the characters d4 and g2 that are retained are indicators that can replace historical accidents for road section risk assessment. In this example, a binomial logistic regression model is used, and the structure of the model is as follows:
[0057] U i =β w x i,w +ε i,w
[0058] U i is the utility function representing the accident risk of the i-th road section, β w is the coefficient to be estimated, ε i,w is a random disturbance term. Let the road sections with accident frequency exceeding 75% quantile be high-risk sections, marked as 1, and the rest of the sections be marked as 0. Then when When , the output value of the model is the probability that section i is a high-risk section, which can be expressed by a mathematical equation as follows:
[0059]
[0060] The binomial logistic regression model was fitted using STATA software. Only characters with a confidence level of not less than 95% were retained in the model. The results are shown in Table 5.
[0061] Table 5 Estimation results of binomial logistic regression model
[0062] Explanatory variables The estimated values of the coefficients Confidence Fixed Items -1.946 99% <![CDATA[d4]]> 0.175 99% <![CDATA[g2]]> 0.129 95%
[0063] The extracted indicators that can replace historical accidents for road section risk assessment are shown in Table 6.
[0064] Table 6 Indicators that can replace historical accidents for road section risk assessment
[0065] index meaning <![CDATA[g2]]> 0.04kN <Fy≤0.1kN <![CDATA[d4]]> 0.13N·m<Mx≤0.14N·m
[0066] S40, using alternative indicators to construct an accident risk calculation equation to evaluate the risk level of the road.
[0067] Substitute the extracted indicators and estimated values of the indicator coefficients into The accident risk probability calculation formula for section i is:
[0068]
[0069] At this point, the indicators obtained by the present invention can be used to replace accident data to evaluate the risk level of the road.
Claims
1. A method for extracting alternative indicators for road traffic accident risk assessment, It is characterized in that The steps include: S10, dividing the road to be analyzed into q equal-length road section units, and collecting a sequence data set of parameters associated with driving risks in each road section unit in chronological order; S20, analyze the importance of each parameter character in the parameter sequence data set to the accident frequency, and retain the characters with importance exceeding I. 0 Characters; S30, using the retained characters to build a regression model, retaining the confidence level in the regression model not less than P 0 The characters are used to construct a discrete choice model, retaining the confidence level not less than P 1 Characters as alternative indicators; S40, constructing an accident risk calculation equation using alternative indicators to assess the risk level of the road; Before step S20, the method further includes: The parameter sequence data set C = {c 1 ,c 2 ,…,c n } is transformed into a standard sequence set C″={c 1 ″,c 2 ″,…,c′ n ′}, the expression is: In the formula, μ is the mean of the sequence data set C, and σ is the standard deviation of the sequence data set C; Then, the dimension reduction formula is used to reduce the dimension of the standard sequence set C″, converting the standard sequence set of length n into m subsequences of equal length, and then replacing each subsequence with its mean value, and recording it as a string C′={c 1 ′,c 2 ′,…,c′ m }, the smooth curve of the sequence data set C becomes a broken line segment after dimensionality reduction, and the dimensionality reduction formula is: In the formula is the compression ratio, which takes an integer value; Step S20, analyzing the importance of each parameter character in the parameter sequence data set to the accident frequency includes: According to the Gaussian distribution table, find the interval split point and convert the reduced-dimensional string C′={c 1 ′,c 2 ′,…,c′ m } is divided into k equal probability intervals, each probability interval is set with a character, and the frequency of each character appearing in each road section unit is counted x i,w , i represents the road segment, w represents the character category, and the frequency x i,w To explain the variables, the importance of each character to the accident frequency was analyzed.
2. The method for extracting alternative indicators according to claim 1, It is characterized in that Parameters associated with driving risks include vehicle speed, acceleration, tire mechanical response, and driver physiological and psychological indexes.
3. The method for extracting alternative indicators according to claim 1, It is characterized in that The regression model in step S30 includes a regression model that can establish a linear or generalized linear relationship with traffic accidents.
