CN107909247A - A kind of city macroscopic view Road Traffic Safety Factors analysis method based on spatial level Bayesian model - Google Patents
A kind of city macroscopic view Road Traffic Safety Factors analysis method based on spatial level Bayesian model Download PDFInfo
- Publication number
- CN107909247A CN107909247A CN201711017922.3A CN201711017922A CN107909247A CN 107909247 A CN107909247 A CN 107909247A CN 201711017922 A CN201711017922 A CN 201711017922A CN 107909247 A CN107909247 A CN 107909247A
- Authority
- CN
- China
- Prior art keywords
- model
- spatial
- value
- condition
- road traffic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
Abstract
The invention discloses a kind of city macroscopic view Road Traffic Safety Factors analysis method based on spatial level Bayesian model, include the following steps：Road traffic accident data and partition information are collected, based on the existing division result based on road traffic current density, using Q gis softwares by the geolocation mapping that traffic accident occurs to corresponding traffic zone；The spatial informations such as the land use morphology of survey region and category of roads are mapped in corresponding region by Q gis, and statistical analysis is carried out to association attributes；Correlation analysis is carried out to each variable based on test for multi-collinearity；Build spatial level Bayes's condition autoregression model；The structure of spatial level Bayesian model based on two kinds of condition autoregression models and compare；Macroscopical Road Traffic Safety Factors analysis based on two kinds of condition autoregression models.The present invention is compared research to two kinds of condition models, provides the relative risk in each cell, and macro-level traffic safety is influenced to carry out variable analysis.
Description
Technical Field
The invention relates to the technical field of road traffic safety, in particular to an urban macroscopic road traffic safety influence factor analysis method based on a spatial hierarchy Bayesian model.
Background
Macroscopic traffic safety can be defined as: taking active traffic safety as a principle, taking the traffic safety level as a main planning target and an evaluation index, comprehensively considering various safety influence factors such as regional characteristics, road network characteristics, travel characteristics, traffic organization, traffic flow characteristics, road conditions, vehicle conditions, driving behaviors and the like in the whole process of planning, designing, constructing, operating and maintaining of a traffic system, and making and implementing corresponding strategies and schemes around the safety target, thereby reducing the potential of accident occurrence at the source and reducing the risk of accident occurrence in operation. Therefore, compared with the traditional traffic safety research, the traffic safety based on the macroscopic level of traffic planning has three characteristics of initiative, globality and quantization. The content related to traffic safety in each stage of traffic planning is considered qualitatively and quantitatively, so that traffic builders and traffic managers can fully master the road safety level and improvement measures of each layer, traffic participants can know the accident risk of a travel route in advance, and the aim of 'preventing better than treating' is achieved in a real sense.
As the macroscopic traffic accident spatial distribution has the characteristics of nonnegativity, discreteness, abnormal distribution and the like, accidents obey Poisson distribution according to experience, but the assumption that the mean value of the Poisson distribution is equal to the variance is often inconsistent with the reality. The negative binomial distribution regression is based on poisson distribution but the set error follows Gamma distribution, and the negative binomial distribution is widely used in a macroscopic traffic safety analysis model. Simple statistical models require that the objects under study be independently distributed, yet traffic accident data have complex spatial autocorrelation. Spatial dependency or autocorrelation occurs when a particular variable within a region is affected by the same variable of an adjacent region. Therefore, in space statistics, it is important to consider the space effect of variables, and neglecting this characteristic will greatly affect the accuracy and robustness of the security level estimation.
The prior art has the following defects: (1) The spatial analysis needs large-scale multi-element fused spatial data suitable for processing of a Geographic Information System (GIS), which is difficult to obtain by general researchers, and is an important premise of macroscopic traffic safety research, but the current research can really realize the large-scale multi-element fused spatial data; (2) Because traffic accident data is often recorded with reference to geographical positions, the influence of spatial correlation and heterogeneity can occur, and the traditional accident prediction models, such as a negative binomial model and a poisson logarithm model, largely ignore the problem of spatial correlation of the accident data, and violate the gaussian-markov assumption used in regression modeling. However, the existing macroscopic security model has no clear solution at present when explaining the correlation of the space.
Disclosure of Invention
The invention aims to solve the technical problem of providing an urban macroscopic road traffic safety influence factor analysis method based on a spatial hierarchy Bayesian model, comparing and researching two condition models of the hierarchy Bayesian model, giving out relative risk degree in each cell, and carrying out variable analysis on macroscopic traffic safety influence.
In order to solve the technical problem, the invention provides an urban macroscopic road traffic safety influence factor analysis method based on a spatial hierarchy Bayesian model, which comprises the following steps:
(1) Collecting road traffic accident data and partition information, and mapping the geographical position of the traffic accident to a corresponding traffic cell by using Q-gis software based on the existing partition result based on the road traffic flow density;
(2) Mapping the land utilization property of the research region, the road grade and other spatial information into the corresponding region through Q-gis, and performing statistical analysis on the related attributes;
(3) Carrying out correlation analysis on each variable based on multiple collinearity tests;
(4) Constructing a spatial hierarchy Bayes condition autoregressive model;
(5) Constructing and comparing a spatial hierarchy Bayes model based on two condition autoregressive models;
(6) And (3) analyzing macroscopic road traffic safety influence factors based on the autoregressive model under two conditions.
Preferably, in the step (3), the correlation analysis of each variable based on the multiple collinearity test is specifically: if some constant c is present 0 、c 1 And c 2 So that the linear equation
c 1 X 1 +c 2 X 2 ＝c 0 (1)
For all samples in the data, two independent variables X are true 1 、X 2 There is precise collinearity;
a common but not entirely suitable X 1 And X 2 The measure of the degree of co-linearity between them is the square r of their sample correlation coefficient 12 2 (ii) a If r is 12 2 =1, then there is exact collinearity; if r is 12 2 =0, then non-collinearity; r is 12 2 The closer to 1, the stronger the approximate collinearity;
for (p)>, 2) independent variables, if constants c are present 0 ，c 1 …c p So that
c 1 X 1 +c 2 X 2 …+c p X p ＝c 0 (2)
If the approximation is true, the p variables have multiple collinearity;
an important metric for measuring the severity of multicollinearity is the matrix X T Condition number of X, i.e.
Wherein λ max (X T X)，λ min (X T X) respectively represent the matrix X T Maximum, minimum eigenvalues of X; number of strips directly portrays X T The magnitude of the difference in the eigenvalues of X is generally k&100, the degree of multiple collinearity is considered to be small; if 100<k&1000, then medium or strong multiple collinearity is considered to exist; if k is> 1000, we consider that severe multicollinearity exists.
Preferably, in the step (4), the constructing of the spatial hierarchy bayesian condition autoregressive model specifically comprises the following steps: (41) constructing a hierarchical Bayesian inference model; in the hierarchical Bayesian inference model, a random effect S i By structural effects U i And non-structural effects V i The structure of the finished model is as follows:
log(SMR i )＝α+X T U+V i +U i +X i (4)
where SMR is the standard mortality, a smooth estimate of the relative risk of a region
y＝(y 1 ,y 2 ,…y n ) Is the observed value of the area i; e = (e) 1 ,e 2 ,…e n ) Is the expected value for zone i; α is the overall risk of the region of interest, also the intercept term; x T To interpret variables or covariates, S i Is a random effect produced by a random factor, X i Is a random error;
the hierarchical Bayesian inference model needs to assign prior distribution to random effect, and in the formula (4), the non-structural effect
V i ～N(0，τ v 2 ) (6)
Using spatial correlation structure to measure U i The aggregation effect of (a);
(42) Constructing a corresponding spatial hierarchy Bayesian model based on an Intrinsic condition autoregressive model and a Cresie condition autoregressive model; intrinsic condition autoregressive model
Conditional expectation u k Is the mean, conditional variance tau of the random effect of the adjacent region u 2 Number n of adjacent regions k In inverse proportion, the expression of the variance proves that strong spatial correlation exists, namely, the more adjacent areas of a certain area, the more random effect information exists; parameter tau u 2 Controlling the variation interval tau of the spatial structure random effect u u 2 Gamma distribution of (0, m) to (0, m);
when each u k Increase by 10, corresponding to τ u 2 The value increases but the representation of the spatial correlation strength (0/1) does not change.
The Cresie condition autoregressive model changes a representation method of space correlation strength, a random effect number set is used in the model, another space correlation parameter is introduced, and the model structure is as follows:
compared with the Intrasic model, the model also has conditional variance, and the conditional expectation value is a weighted average of the mean of adjacent regions of the region and the mean alpha of the whole region; the weight parameter p controls the strength of spatial correlation, when p =0, the spatial correlation is completely independent in correspondence, and the spatial correlation is increased along with the gradual increase of the value of p; the series of full-condition distributions correspond to a more appropriate multivariate Gaussian distribution; if 0 is less than or equal to rho&1, then u is a continuous variable, where the covariance is τ 2 Q C -1 And n is k Same, Q C There are j × k elements; if j and k are adjacent, w jk = - ρ; otherwise, the value is zero;
therefore, rho value is a key parameter of the model, vector C is used in WinBUGS software to represent a weight matrix, namely rho value, the row of the matrix represents a region, and the column represents the correlation of the region with other regions;
(43) WinBUGS software is used for constructing the Bayesian model of the two different levels of space.
Preferably, in the step (5), the construction and comparison of the spatial hierarchy bayesian model based on the two conditional autoregressive models specifically includes the following steps:
(51) Judging whether the model is converged or not by the ratio of the MC error to the standard deviation;
(52) Comparing the complexity and goodness of fit of the models of the two models by using a Pearson Chirson fitness value RSS and a deviation information criterion DIC;
DIC is expressed as follows:
here, theIs a quantity used to measure the deviation, which is equal to negative double the log-likelihood, i.e.
Here, theIs thatPosterior mean value of (1), p m It can be understood as the number of effective parameters in the model m, which is equal to the difference between the dispersion posterior mean and the posterior mean dispersion, i.e. the difference between the dispersion posterior mean and the dispersion posterior mean
WhereinThe mean posterior value of parameters related to the model m is obtained, and a smaller DIC value means that the model fits well; compared with the Intrinsic model, the Cresse model has lower RSS value and DIC value, so the fitting degree of the model is relatively higher and the complexity is equivalent;
(53) Comparing the simulation of the two models to the actual situation using the relative risk RR;
the relative risk map RR, namely the standard mortality SMR, is the ratio of the risk of an exposure group to the risk of a contrast group, indirectly reflects the traffic safety level in the area, and is obtained by calculating macroscopic level traffic influence factors, area of the area and the like in the comprehensive area;
preferably, in the step (6), the analysis of the influence factors of the macroscopic road traffic safety based on the two condition autoregressive models specifically includes the following steps:
(61) Judging whether the variable is significant or not; the posterior mean value is less than the posterior standard deviation value or the 2.95% confidence interval should not cover zero; (62) Calculating the impact of significant variables on accident rates
δ＝(-e ξ -1) (14)
In the formula, delta is the influence degree of the variable on the accident rate, and xi is the posterior standard deviation of the model.
The invention has the beneficial effects that: (1) two condition models of layer Bayes are compared and researched: in the field of spatial statistical analysis, a Conditional Autoregressive (CAR) model is widely applied to data statistical analysis with spatial dependency, and the previous research only uses an Intrasic conditional autoregressive model, defaults the weight of a spatial adjacent matrix into a discrete 0/1 variable, and is not consistent with the actual situation; thus, an optimized conditional autoregressive model is used herein: the Cresse condition autoregression model compares the advantages and disadvantages of the Cresse condition autoregression model and the Cresse condition autoregression model in terms of model convergence rate, fitting degree and complexity, and the Cresse condition autoregression model is superior to the Intrinsic condition autoregression model; (2) giving the relative risk in each cell: the relative risk of each cell under an Intra-cell traffic safety influence factor is obtained through modeling analysis of the macro traffic safety influence factors in each cell, the intra-cell traffic safety level is indirectly reflected by the relative risk, the relative risk is obtained through calculation of the macro level traffic influence factors, the area of each cell and the like in the comprehensive area, the relative risk approximately reflects the traffic safety level in each cell under the same influence factor, and further the regional traffic safety can be improved; (3) analyzing macroscopic level traffic safety influence variables: the method uses two main types of influence variables of road density and land use properties of each grade, obtains the relation between the influence variables and traffic safety by establishing a hierarchical Bayes condition autoregressive model on the basis of considering a space structural effect and a non-structural effect, and specifically comprises the following steps: the increase of the proportion of the secondary road density and the industrial land can improve the regional traffic safety, and conversely, the primary and secondary land and the low-end residential land can have adverse effect on the traffic safety; the invention fills some defects in the current macroscopic safety research; in terms of practical value, the research result of the invention can better guide the policy and scheme formulation of traffic planning, thereby effectively improving traffic safety.
Drawings
FIG. 1 is a schematic flow diagram of the overall process of the present invention.
Fig. 2 is a schematic diagram of the traffic plot division and the traffic accident location of the research area according to the present invention.
FIG. 3 is a schematic diagram showing the properties of various plots in the research area of the present invention.
FIG. 4 is a schematic diagram of the road network of each level in the research area according to the present invention.
Fig. 5 (a) is a schematic diagram of a historical iteration track based on the inrinsic condition autoregressive model of the present invention.
FIG. 5 (b) is a diagram of the mean trace of the inrinsic-based conditional autoregressive model of the present invention.
Fig. 5 (c) is a schematic diagram of a historical iteration track based on the Cressie conditional autoregressive model of the present invention.
FIG. 5 (d) is a diagram illustrating a mean trajectory of the Cresse-based conditional autoregression model of the present invention.
FIG. 6 (a) is a schematic diagram of the relative risk of the region calculated by the Intrinsic conditional autoregressive model of the present invention.
FIG. 6 (b) is a schematic diagram of the relative risk of the region calculated by the Cresie conditional autoregression model of the present invention.
Detailed Description
As shown in fig. 1, a method for analyzing influence factors of urban macroscopic road traffic safety based on a spatial hierarchy bayesian model includes the following steps:
(1) Collecting road traffic accident data and partition information, and mapping the geographical position of the traffic accident to a corresponding traffic cell by using Q-gis software based on the existing partition result based on the road traffic flow density;
(2) Mapping the land utilization property, road grade and other spatial information of the research region into the corresponding region through Q-gis, and performing statistical analysis on the related attributes;
(3) Carrying out correlation analysis on each variable based on multiple collinearity tests;
(4) Constructing a spatial hierarchy Bayes condition autoregressive model;
(5) Constructing and comparing a spatial hierarchy Bayes model based on the two condition autoregressive models;
(6) And (3) analyzing macroscopic road traffic safety influence factors based on the autoregressive model under two conditions.
S1, collecting road traffic accident data and partition information, and mapping the geographical position of a traffic accident to a corresponding traffic cell by using Q-gis software based on the existing partition result based on the road traffic flow density, as shown in FIG. 2:
and S2, mapping the land utilization property of the research region, the road grade and other spatial information into the corresponding region through Q-gis, and performing statistical analysis on the related attributes. The land use properties selected in the present invention are as follows:
class a land use properties: public management and public service facility land; carrying out A33 independently: land for primary and secondary schools;
the land utilization property of the type B is commercial service industry facility land;
the G-type land utilization property is the land used by greenbelts and squares;
the M-type land utilization property is industrial land;
a high-grade residential land of R1 class, a residential land of R2 class and a residential land of R3 class.
The distribution of the land property of each subarea in the research area is shown in fig. 3, and the proportion statistics of the land property of each type in each traffic cell is shown in table 1.
TABLE 1 summary of the land characteristics of various types in various traffic districts in the research area
The invention divides the road grade in the region according to the road width index classification of the large and medium city roads, which is concretely shown as follows:
trunk line: the road width is 35m-45m;
secondary trunk line: road width is 30m-40m;
branch circuit: road width of 12m-15m
The roads of each level are mapped into the analysis area using Q-gis, as shown in fig. 4, the length of the road of each level in each traffic cell is calculated, and the corresponding road density in the area is calculated, as shown in table 2.
TABLE 2 summary of road networks at various levels within a research area
S3, carrying out correlation analysis on each variable based on multiple collinearity tests, wherein the correlation analysis is specifically described as follows:
if some constant c exists 0 、c 1 And c 2 So that the linear equation
c 1 X 1 +c 2 X 2 ＝c 0 (1)
For all samples in the data, two arguments X are then true 1 、X 2 There is precise collinearity.
In practice, exact collinearity happens by chance, so if the approximation of equation (1) holds for the measured data, there is an approximate collinearity. A common but not entirely suitable X 1 And X 2 The measure of the degree of co-linearity between them is the square r of their sample correlation coefficient 12 2 . If r is 12 2 =1, then there is exact collinearity; if r is 12 2 And =0, then non-collinearity. r is 12 2 The closer to 1, the stronger the approximate collinearity.
For (p)>, 2) independent variables, if constants c are present 0 ，c 1 …c p So that
c 1 X 1 +c 2 X 2 …+c p X p ＝c 0 (2)
If the approximation is true, it means that the p variables have multiple collinearity.
An important indicator for measuring the severity of multicollinearity is the matrix X T Condition number of X, i.e.
Wherein λ is max (X T X)，λ min (X T X) respectively represent a matrix X T Maximum, minimum eigenvalues of X. Number of strips directly portrays X T The magnitude of the variance of the eigenvalues of X, from the experience of practical applications, is generally k<100，The degree of multicollinearity is considered to be small; if 100<k&1000, then the existence of moderate or strong multiple collinearity is considered; if k is>, 1000, severe multicollinearity is considered to exist.
The invention uses scale () function in R language to standardize the data in tables 1 and 2, and then uses Kappa () function to calculate the condition number of the matrix, and judges whether the matrix has multiple linear relations. The macroscopic traffic safety impact variables in table 3 were finally obtained.
TABLE 3 macroscopic traffic safety impact factor descriptive statistics
S4, constructing a spatial hierarchy Bayes condition autoregressive model
S41, constructing a hierarchical Bayesian inference model (BYM)
In BYM model, the random effect S i By structural effects (spatial variation) U i And non-structural effects V i And (4) forming. The completed model structure is as follows:
log(SMR i )＝α+X T U+V i +U i +X i (4)
where SMR is Standard mortality rates (Standard mortality rates) is a smooth estimate of the relative risk of a region
y＝(y 1 ,y 2 ,…y n ) The observed value of the area i; e = (e) 1 ,e 2 ,…e n ) Is the expected value for region i. α is the overall risk of the study area and is also the intercept term. X T To interpret variables or covariates, S i Is a random effect produced by a random factor, X i Is a random error.
The BYM model requires a priori distribution to be assigned to random effects, and in equation (4), the non-structural effects
V i ～N(0，τ v 2 ) (6)
Using spatial correlation structure to measure U i The aggregation effect of (a). In other words, the risk of any one zone depends on the adjacent zones of this zone. A specific method will be described in step S42
S42, constructing corresponding spatial hierarchy Bayesian model based on Intrinsic condition autoregression model and Cresie condition autoregression model
S421 and Intrinsic condition autoregression model
Conditional expectation u k Is the mean, conditional variance tau, of the random effect of the adjacent regions u 2 Number n of adjacent regions k In inverse proportion, the expression of the variance proves that strong spatial correlation exists, namely, the more adjacent areas of a certain area, the more random effect information exists. Parameter tau u 2 Controlling the variation interval tau of the spatial structure random effect u u 2 Gamma distribution of (0, m).
The result of prior distribution in this way is often one-sidedness. Its single parameter does not reflect the strength of spatial correlation between random effects, since when each u is k Increase by 10, corresponding to τ u 2 The value increases but the representation of the spatial correlation strength (0/1) does not change.
S422 and Cressie condition autoregression model
The Cresse conditional autoregressive model can change the representation method of the spatial correlation strength (not only a simple 0,1 discrete binary variable), and a random effect number set is used in the model, and another spatial correlation parameter is introduced. The model structure is as follows:
compared to the Intrinsic model, this model also has a conditional variance, while the conditional expectation is a weighted average of the mean of the neighboring regions of the region and the mean α of the entire region. The weight parameter ρ controls the strength of spatial correlation, which corresponds to complete spatial independence when ρ =0, and the spatial correlation increases as ρ is gradually increased (i.e., IAC model when ρ = 1). The series of all-condition distributions correspond to a more appropriate multivariate Gaussian distribution. If 0 is less than or equal to rho&1, then u is a continuous variable, where the covariance is τ 2 Q C -1 And n is k Same, Q C There are j × k elements. If j and k are adjacent, w jk = - ρ; otherwise it is zero.
The value of p is therefore a key parameter of this model. Vector C is used in WinBUGS software to represent a weight matrix, i.e., ρ values, and the rows of the matrix represent regions and columns represent the correlation of the regions with other regions. A variable cumsung is defined in WinBUGS, which contains information on the number of contiguous regions of all regions. For example, there are 4 regions, cumsum = c (0,3,5,6,8).
Number of adjacent regions of region 1: 3-0=3;
number of adjacent regions of region 2: 5-3=2;
number of adjacent regions of region 3: 6-5=1;
number of adjacent regions of region 4: 8-6=2;
then, the total number of adjacent areas sumnumneigh =3+2+1+2=8, and a matrix pick is established, which has N rows corresponding to areas 1 to N; total number of rows of pick matrix = sumnumneighbor. And counting the adjacent areas of all the areas according to the column order of the pick, wherein the pick function in the above example is as follows:
and finally, solving the inner product of the pick matrix and the expected value E according to the rows:
the k-th row of the pick matrix corresponds to the k-th element in the vector C. and adj is an adjacent matrix of the area and is directly given by a GeoBUGS map module in WinBUGS.
S43, constructing the two different hierarchical spatial Bayesian models by using WinBUGS software
S431 and Intrinsic condition autoregressive distribution model core code under WinBUGS
For spatial structure effect u i Assuming that the prior is an Intrasic conditional autoregressive distribution, the conditional distribution obeys equation 4-1.
S432 and Cresie condition autoregressive distribution model core code under WinBUGS
For spatial structure effect u i It is assumed that the prior is a Cresie conditional autoregressive distribution, and follows the full-conditional distribution of equation 4-2.
S5, construction and comparison of spatial hierarchy Bayes model based on two condition autoregressive models
S51, judging whether the model converges or not according to the ratio of MC error to standard deviation
In the case, winBUGS software is used to calculate the posterior result of the parameters, including the posterior mean, the posterior standard deviation and the 95% confidence interval. For any variable V, V5 (common management and public service facility) is taken in the invention, and the ratio of MC error to standard deviation under the two models is calculated, as shown in Table 4. The MC error/sd under both models is less than 0.1, so the models converge.
TABLE 4 variable V5 posterior results
Wherein FIGS. 5 (a), and 5 (b) are trajectory diagrams based on the inrinsic conditional autoregressive model; fig. 5 (c), and 5 (d) are trajectory diagrams based on the cresse conditional autoregression model, both models being iterated 10000 times in common.
As can be seen from FIG. 5 (a), the estimated value of the variable V5 tends to stabilize over 3000 iterations; the solid line in FIG. 5 (b) is the mean distribution plot of the variable V5 iterated 10000 times, and the dotted line is the variation interval between 97.5% and 2.5%, and it can be seen that the first 3000 times gradually decrease and stabilize after 3000 times, which is consistent with the information in FIG. 5 (a). As can be seen from FIG. 5 (c), the variable V5 tends to stabilize over 2500 iterations; in FIG. 5 (d), the solid line is the mean distribution graph of the variable V5 iterated 10000 times, and the dotted line is the variation interval between 97.5% and 2.5%, which shows that the fluctuation of the first 2500 times is larger and tends to be stable after 2500 times, which is identical to the information in FIG. 5 (c). Therefore, the convergence speed is faster by using the Cresie condition autoregressive model.
S52, comparing the complexity and goodness of fit of the models of the two models by using Pearson chi-square goodness-of-fit (RSS) and variance Information Criterion (Dvice Information Criterion) DIC
DIC is expressed as follows:
here, theIs a quantity used to measure the deviation, which is equal to negative double of the log-likelihood, i.e.
Here, theIs thatPosterior mean value of (1), p m It can be understood as the number of valid parameters in the model m, which is equal to the difference between the dispersion posterior mean and the posterior mean dispersion. Namely, it is
WhereinIs the posterior mean of the parameters involved in the model m. Smaller DIC values mean better model fit, and DIC is now widely used in comparison of Bayesian hierarchical spatial models and logarithmic spatial models.
Case data analysis As shown in Table 5, the Cresse model has lower RSS and DIC values than the Intrinsic model, and thus the model has relatively high fitting degree and complexity.
TABLE 5 BYM model fitting test
S53, comparing the simulation of the two models to the actual situation by using Relative Risk (Relative Risk) RR
Relative Risk map (Relative Risk) RR, normalized mortality (SMR), refers to the ratio of the Risk of the exposed group to the Risk of the control group. Indirectly reflects the traffic safety level in the region and is obtained by integrating the macroscopic level traffic influence factors, the region area and the like in the region.
Fig. 6 (a) and (b) show the relative risk of the region calculated by the Intrinsic conditional autoregressive model and the cresse conditional autoregressive model, respectively, and the variation range of the relative risk of the two is 6 units, but the grade of the relative risk of the cresse model is more, so that the region variation is more moderate and tends to the true value. No extremes occurred compared to the Intrinsic model. Due to the number of the cells (13) analyzed by the traffic, the phenomenon becomes more obvious when the number of the partitions is larger.
S6, analysis of macroscopic road traffic safety influence factors based on autoregressive model under two conditions
S61, judging whether the variable is significant or not
There are two criteria for determining whether a variable is significant: 1. the posterior mean value is smaller than the posterior standard difference value; the 2.95% confidence interval should not cover zero. According to the judgment of the above criteria, the significant variables in the BYM (Intrinsic) model and the BYM (credit) model in the case are shown in bold numbers in table 6. Wherein the significant variables in the BYM (Cresie) model include five variables (secondary road density, land for primary and secondary schools, industrial site, and low-end residential site), and only the public management and public service facilities are significant in the BYM (Intrinsic) model.
TABLE 6 BYM (Intrinsic) model vs. BYM (Cresie) model variable analysis results
S62, calculating the influence of the significant variable on the accident rate
δ＝(-e ξ -1) (14)
In the formula, delta is the influence degree of the variable on the accident rate, and xi is the posterior standard deviation of the model
Case data show that: the road network density is increased, and the vehicle attraction of main roads in the region is reduced, so that the traffic accident rate can be reduced.
On the groundWith the type aspect, the BYM (credit) model results show: the increase of the proportion of the industrial area in the area can obviously reduce the traffic accident rate, and the traffic accident quantity can be reduced by 2.5 percent (-e) for each unit of industrial area 0.025 -1). And the primary and secondary school land and the low-end residential land have positive correlation with the number of traffic accidents, which is consistent with the actual traffic operation condition. Meanwhile, the land used by the middle and low-end residences in the city also increases the accident rate risk, the land property is mostly distributed in old urban areas, the residential buildings are older in times, the population density of the residences is high, and the roads inside and outside the residential areas are usually narrow and are often parked by occupied roads. The influence on the traffic operation safety is similar to the nature of land used in primary and secondary schools, and is often a safety accident caused by traffic conflict. Therefore, for the land property, from the controllable viewpoint, the traffic risk rate of the land property can be reduced by reasonably regulating the ways of roadside parking in old cities, increasing pedestrian street-crossing signal lights, regional division, time-slot charging and the like for reducing the traffic density.
While the invention has been shown and described with respect to the preferred embodiments, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the scope of the invention as defined in the following claims.
Claims (5)
1. A city macroscopic road traffic safety influence factor analysis method based on a spatial hierarchy Bayesian model is characterized by comprising the following steps:
(1) Collecting road traffic accident data and partition information, and mapping the geographical position of the traffic accident to a corresponding traffic cell by using Q-gis software based on the existing partition result based on the road traffic flow density;
(2) Mapping the land utilization property of the research region, the road grade and other spatial information into the corresponding region through Q-gis, and performing statistical analysis on the related attributes;
(3) Carrying out correlation analysis on each variable based on multiple collinearity tests;
(4) Constructing a spatial hierarchy Bayes condition autoregressive model;
(5) Constructing and comparing a spatial hierarchy Bayes model based on two condition autoregressive models;
(6) And (3) analyzing macroscopic road traffic safety influence factors based on the autoregressive model under two conditions.
2. The urban macroscopic road traffic safety influence factor analysis method based on the spatial hierarchy Bayesian model as recited in claim 1, wherein in the step (3), the correlation analysis of each variable based on the multiple collinearity test is specifically: if some constant c exists 0 、c 1 And c 2 So that a linear equation
c 1 X 1 +c 2 X 2 ＝c 0 (1)
For all samples in the data, two independent variables X are true 1 、X 2 There is precise collinearity;
a common but not entirely suitable X 1 And X 2 The measure of the degree of co-linearity between them is the square r of their sample correlation coefficient 12 2 (ii) a If r is 12 2 =1, then there is exact collinearity; if r is 12 2 =0, then non-collinearity; r is 12 2 The closer to 1, the stronger the approximate collinearity;
for (p)>, 2) independent variables, if constants c are present 0 ，c 1 …c p So that
c 1 X 1 +c 2 X 2 …+c p X p ＝c 0 (2)
If the approximation is true, the p variables have multiple collinearity;
an important indicator for measuring the severity of multicollinearity is the matrix X T Condition number of X, i.e.
Wherein λ is max (X T X)，λ min (X T X) respectively represent the matrix X T Maximum, minimum eigenvalues of X; number of strips directly portrays X T The magnitude of the difference in the eigenvalues of X is generally k&100, the degree of multiple collinearity is considered to be small; if 100<k&1000, then the existence of moderate or strong multiple collinearity is considered; if k is> 1000, we consider that severe multicollinearity exists.
3. The urban macroscopic road traffic safety influence factor analysis method based on the spatial hierarchy Bayesian model as claimed in claim 1, wherein in the step (4), the construction of the spatial hierarchy Bayesian conditional autoregressive model specifically comprises the following steps:
(41) Constructing a hierarchical Bayesian inference model; in the hierarchical Bayesian inference model, a random effect S i By structural effects U i And non-structural effects V i The structure of the finished model is as follows:
log(SMR i )＝α+X T U+V i +U i +X i (4)
where SMR is the standard mortality, a smooth estimate of the relative risk of a region
y＝(y 1 ,y 2 ,…y n ) Is the observed value of the area i; e = (e) 1 ,e 2 ,…e n ) Is the expected value of zone i; α is the overall risk of the study area, and is also the intercept term; x T To interpret variables or covariates, S i Is a random effect produced by a random factor, X i Is a random error;
the hierarchical Bayesian inference model needs to assign prior distribution to random effect, and in the formula (4), the non-structural effect
V i ～N(0，τ v 2 ) (6)
Using spatial correlation structure to measure U i The aggregation effect of (a);
(42) Constructing a corresponding spatial hierarchy Bayesian model based on an Intrinsic condition autoregressive model and a Cresie condition autoregressive model; intrinsic condition autoregressive model
Conditional expectation u k Is the mean, conditional variance tau, of the random effect of the adjacent regions u 2 Number n of adjacent regions k In inverse proportion, the expression of the variance proves that strong spatial correlation exists, namely, the more adjacent areas of a certain area, the more random effect information exists; parameter tau u 2 Controlling the variation interval tau of the spatial structure random effect u u 2 Gamma distribution of (0, m) to (0, m);
when each u k Increased by 10, corresponding to τ u 2 The value will increase, but the representation of the spatial correlation strength (0/1) is not changed;
the Cresie condition autoregressive model changes a representation method of space correlation strength, a random effect number set is used in the model, another space correlation parameter is introduced, and the model structure is as follows:
compared with the Intrasic model, the model also has conditional variance, and the conditional expectation value is a weighted average of the mean of adjacent regions of the region and the mean alpha of the whole region; the weight parameter p controls the strength of spatial correlation, when p =0, the spatial correlation is completely independent in correspondence, and the spatial correlation is increased along with the gradual increase of the value of p; the series of full-condition distributions correspond to a more appropriate multivariate Gaussian distribution; if 0 is less than or equal to rho&1, then u is a continuous variable, where the covariance is τ 2 Q C -1 And n is k In the same way, the first and second,Q C there are j × k elements; if j and k are adjacent, w jk = - ρ; otherwise, the value is zero;
therefore, rho value is a key parameter of the model, vector C is used in WinBUGS software to represent a weight matrix, namely rho value, the row of the matrix represents a region, and the column represents the correlation of the region with other regions;
(43) WinBUGS software is used for constructing the Bayesian model of the two different levels of space.
4. The urban macroscopic road traffic safety influence factor analysis method based on the spatial hierarchy Bayesian model as recited in claim 1, wherein in the step (5), the construction and comparison of the spatial hierarchy Bayesian model based on the two conditional autoregressive models specifically comprises the following steps:
(51) Judging whether the model converges or not according to the ratio of the MC error to the standard deviation;
(52) Comparing the complexity and goodness of fit of the models of the two models by using a Pearson Chirson fitness value RSS and a deviation information criterion DIC;
DIC expression is as follows:
here, theIs a quantity used to measure the deviation, which is equal to negative double the log-likelihood, i.e.
Here, theIs thatPosterior mean value of (1), p m It can be understood as the number of effective parameters in the model m, which is equal to the difference between the dispersion posterior mean and the posterior mean dispersion, i.e. the difference between the dispersion posterior mean and the dispersion posterior mean
WhereinThe mean posterior value of parameters related to the model m is obtained, and a smaller DIC value means that the model fits well; compared with the Intrasic model, the Cressie model has lower RSS value and DIC value, so that the fitting degree of the model is relatively higher and the complexity is equivalent;
(53) Comparing the simulation of the two models to the actual situation by using the relative risk degree RR;
the relative risk map RR, namely the standard mortality SMR, is the ratio of the risk of an exposure group to the risk of a contrast group, indirectly reflects the traffic safety level in the area, and is obtained by calculating macroscopic level traffic influence factors, area of the area and the like in the comprehensive area;
5. the urban macroscopic road traffic safety influence factor analysis method based on the spatial hierarchy Bayesian model as recited in claim 1, wherein in the step (6), the macroscopic road traffic safety influence factor analysis based on the two conditional autoregressive models specifically comprises the following steps:
(61) Judging whether the variable is significant or not; the posterior mean value is less than the posterior standard deviation value or the 2.95% confidence interval should not cover zero;
(62) Calculating the impact of significant variables on accident rates
δ＝(-e ξ -1) (14)
In the formula, delta is the influence degree of the variable on the accident rate, and xi is the posterior standard deviation of the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711017922.3A CN107909247A (en) | 2017-10-26 | 2017-10-26 | A kind of city macroscopic view Road Traffic Safety Factors analysis method based on spatial level Bayesian model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711017922.3A CN107909247A (en) | 2017-10-26 | 2017-10-26 | A kind of city macroscopic view Road Traffic Safety Factors analysis method based on spatial level Bayesian model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107909247A true CN107909247A (en) | 2018-04-13 |
Family
ID=61841842
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711017922.3A Pending CN107909247A (en) | 2017-10-26 | 2017-10-26 | A kind of city macroscopic view Road Traffic Safety Factors analysis method based on spatial level Bayesian model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107909247A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110400471A (en) * | 2019-07-24 | 2019-11-01 | 杭州电子科技大学 | A kind of control method of Traffic Systems safe operation |
| CN111260264A (en) * | 2020-05-06 | 2020-06-09 | 成都四方伟业软件股份有限公司 | Resource type city ecological safety early warning method and device |
| CN111489008A (en) * | 2019-08-24 | 2020-08-04 | 天津职业技术师范大学(中国职业培训指导教师进修中心) | Traffic accident influence factor spatial effect analysis method and application thereof |
| CN112150816A (en) * | 2020-10-26 | 2020-12-29 | 哈尔滨工业大学 | Multi-vehicle interaction risk assessment system |
| CN112508392A (en) * | 2020-12-02 | 2021-03-16 | 云南省交通规划设计研究院有限公司 | Dynamic evaluation method for traffic conflict risk of hidden danger road section of mountain area double-lane highway |
| CN113269960A (en) * | 2021-05-17 | 2021-08-17 | 郑州航空工业管理学院 | Method for analyzing accident risk of highway tunnel group traffic |
| CN113887569A (en) * | 2021-09-09 | 2022-01-04 | 同济大学 | Region scale-based accident-prone region discrimination method |
| CN113920725A (en) * | 2021-09-30 | 2022-01-11 | 同济大学 | Urban main road accident prediction method based on mesoscopic level and storage medium |
| CN114241753A (en) * | 2021-12-03 | 2022-03-25 | 东南大学 | Road safety evaluation method and system based on multi-dimensional influence factors |
| CN114495501A (en) * | 2022-01-27 | 2022-05-13 | 东南大学 | Road safety analysis method based on combined geographic autoregressive matching |
| CN115830873A (en) * | 2023-01-10 | 2023-03-21 | 西南交通大学 | Urban road traffic event classification method, device, equipment and readable storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106530706A (en) * | 2016-12-02 | 2017-03-22 | 清华大学 | Regional traffic safety evaluation method based on connectivity-based spatial weight matrix |
| CN107145964A (en) * | 2017-04-10 | 2017-09-08 | 同济大学 | Multiple regression forecasting model optimization method based on genetic programming |
-
2017
- 2017-10-26 CN CN201711017922.3A patent/CN107909247A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106530706A (en) * | 2016-12-02 | 2017-03-22 | 清华大学 | Regional traffic safety evaluation method based on connectivity-based spatial weight matrix |
| CN107145964A (en) * | 2017-04-10 | 2017-09-08 | 同济大学 | Multiple regression forecasting model optimization method based on genetic programming |
Non-Patent Citations (2)
| Title |
|---|
| DUNCAN LEE: "A comparison of conditional autoregressive models used in Bayesian disease mapping", 《SPATIAL AND SPATIO-TEMPORAL EPIDEMIOLOGY 2》 * |
| 吴杏薇 等: "基于美国佛罗里达州Orange县的宏观交通安全建模分析", 《合肥工业大学学报（自然科学版）》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110400471B (en) * | 2019-07-24 | 2020-09-22 | 杭州电子科技大学 | Control method for safe operation of urban traffic system |
| CN110400471A (en) * | 2019-07-24 | 2019-11-01 | 杭州电子科技大学 | A kind of control method of Traffic Systems safe operation |
| CN111489008B (en) * | 2019-08-24 | 2023-09-19 | 天津职业技术师范大学(中国职业培训指导教师进修中心) | Traffic accident influencing factor space effect analysis method and application thereof |
| CN111489008A (en) * | 2019-08-24 | 2020-08-04 | 天津职业技术师范大学(中国职业培训指导教师进修中心) | Traffic accident influence factor spatial effect analysis method and application thereof |
| CN111260264A (en) * | 2020-05-06 | 2020-06-09 | 成都四方伟业软件股份有限公司 | Resource type city ecological safety early warning method and device |
| CN112150816A (en) * | 2020-10-26 | 2020-12-29 | 哈尔滨工业大学 | Multi-vehicle interaction risk assessment system |
| CN112508392A (en) * | 2020-12-02 | 2021-03-16 | 云南省交通规划设计研究院有限公司 | Dynamic evaluation method for traffic conflict risk of hidden danger road section of mountain area double-lane highway |
| CN112508392B (en) * | 2020-12-02 | 2021-11-02 | 云南省交通规划设计研究院有限公司 | Dynamic evaluation method for traffic conflict risk of hidden danger road section of mountain area double-lane highway |
| CN113269960A (en) * | 2021-05-17 | 2021-08-17 | 郑州航空工业管理学院 | Method for analyzing accident risk of highway tunnel group traffic |
| CN113887569A (en) * | 2021-09-09 | 2022-01-04 | 同济大学 | Region scale-based accident-prone region discrimination method |
| CN113887569B (en) * | 2021-09-09 | 2022-10-25 | 同济大学 | Accident frequent region distinguishing method based on region scale |
| CN113920725A (en) * | 2021-09-30 | 2022-01-11 | 同济大学 | Urban main road accident prediction method based on mesoscopic level and storage medium |
| CN114241753A (en) * | 2021-12-03 | 2022-03-25 | 东南大学 | Road safety evaluation method and system based on multi-dimensional influence factors |
| CN114241753B (en) * | 2021-12-03 | 2022-11-01 | 东南大学 | Road safety evaluation method and system based on multi-dimensional influence factors |
| WO2023098076A1 (en) * | 2021-12-03 | 2023-06-08 | 东南大学 | Multidimensional influence factor-based road safety evaluation method and system |
| US11887472B2 (en) | 2021-12-03 | 2024-01-30 | Southeast University | Method and system for evaluating road safety based on multi-dimensional influencing factors |
| CN114495501A (en) * | 2022-01-27 | 2022-05-13 | 东南大学 | Road safety analysis method based on combined geographic autoregressive matching |
| CN115830873A (en) * | 2023-01-10 | 2023-03-21 | 西南交通大学 | Urban road traffic event classification method, device, equipment and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107909247A (en) | A kind of city macroscopic view Road Traffic Safety Factors analysis method based on spatial level Bayesian model | |
| Abbasabadi et al. | An integrated data-driven framework for urban energy use modeling (UEUM) | |
| CN111081016B (en) | Urban traffic abnormity identification method based on complex network theory | |
| CN108492555A (en) | A kind of city road net traffic state evaluation method and device | |
| Firat | Comparison of artificial intelligence techniques for river flow forecasting | |
| Schultz et al. | Analysis of distribution and calibration of car-following sensitivity parameters in microscopic traffic simulation models | |
| CN112506990B (en) | Hydrological data anomaly detection method based on spatiotemporal information | |
| CN110288825B (en) | Traffic control subregion clustering division method based on multi-source data fusion and SNMF | |
| CN106373390B (en) | Traffic state evaluation method based on Adaptive Neuro-fuzzy Inference | |
| Andrade et al. | Development of transport mode choice model by using adaptive neuro-fuzzy inference system | |
| CN112966871A (en) | Traffic jam prediction method and system based on convolution long-short term memory neural network | |
| AlKheder et al. | Roadway traffic noise modelling in the hot hyper-arid Arabian Gulf region using adaptive neuro-fuzzy interference system | |
| Tu et al. | Research on intelligent calculation method of intelligent traffic flow index based on big data mining | |
| Aidan et al. | Predicting earned value indexes in residential complexes’ construction projects using artificial neural network model | |
| Li et al. | New model of travel-time prediction considering weather conditions: case study of urban expressway | |
| KR101703972B1 (en) | System and method for predicting groundwater potential area using spatial information | |
| CN113344470A (en) | Intelligent power supply system management method based on block chain | |
| Qiao et al. | An improved multi-sensor D–S rule for conflict reassignment of failure rate of set | |
| CN113642162B (en) | Urban road traffic emergency plan simulation comprehensive analysis method | |
| CN114943464A (en) | Expressway reconstruction and extension traffic safety risk assessment method based on improved D-S | |
| Dubey et al. | Route choice modelling using fuzzy logic and adaptive neuro-fuzzy | |
| Rassafi et al. | Predicting urban trip generation using a fuzzy expert system | |
| Wang et al. | An improved method for multisensor high conflict data fusion | |
| Feng et al. | Traffic Flow Prediction of Urban Intersection Based on Environmental Impact Factors and Markov Chains | |
| CN116433051B (en) | Urban area police strategy dynamic adjustment method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180413 |
|
| RJ01 | Rejection of invention patent application after publication |