CN110188953A - A Prediction Method of O-D Spatio-temporal Distribution Based on Spatial Durbin Model - Google Patents

Abstract

The invention relates to an O-D spatiotemporal distribution prediction method based on a spatial Durbin model, belonging to the technical fields of urban traffic planning and management and intelligent traffic systems. The built environment is added as the explanatory variable of O-D spatiotemporal distribution, and the explanatory power of built environment for O-D spatiotemporal distribution is proved through a case. A method of generating or attracting volume. The effects and benefits of the present invention are to explain the spillover effect of the built environment on the O-D space-time distribution, and decompose the spillover effect into direct effects, indirect effects and total effects, so as to improve the accuracy of urban O-D space-time distribution prediction results .

Description

A Prediction Method of O-D Spatio-temporal Distribution Based on Spatial Durbin Model

technical field

The invention belongs to the technical field of urban traffic planning and management, relates to the temporal and spatial distribution of Origin-Destination (O-D for short) and the field of ITS intelligent transportation system, and is particularly suitable for the interpretation of O-D temporal and spatial distribution based on the urban built environment and the O-D temporal and spatial distribution. Distribution forecast method.

Background technique

The existing research on O-D demand distribution is mainly divided into two steps: O-D data acquisition and O-D matrix construction to obtain the O-D distribution of urban transportation system. Alexander uses mobile phone triangulation data as the daily travel trajectory of individuals and families, and its accuracy and timeliness can replace traditional family travel survey data. Hadavi and Shafahi proposed an O-D estimation based on traffic sensor data, which applied license plate recognition sensors and proposed four location models to obtain O-D flow. O-D matrix construction methods are mainly divided into two categories: statistical methods and mathematical programming methods. Using the principle of maximum entropy, Ge and Fukuda realized job-related travel O-D demand estimation based on mobile phone GPS trajectory data. Lee proposes a robust optimization method for O-D network evaluation, which is able to generate an O-D matrix and overcome the uncertainty of O-D demand distribution.

Whether it is in the process of O-D data acquisition or O-D matrix construction, in the research, more use of historical start and end point data and mathematical methods are used to carry out O-D inverse inference, and there is often no (unable to) analysis of the inherent factors that affect the O-D distribution. Therefore, the present invention adopts the taxi data to propose a O-D spatiotemporal distribution prediction method based on the spatial Durbin model, which fully considers the influence of the urban built environment on the O-D spatiotemporal distribution, thereby proposing a more accurate prediction method.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to first obtain the O-D distribution of each cell by using the GPS data of taxis in each traffic cell, and then build a spatial Durbin model on this basis to estimate the spillover effect of the urban built environment on the O-D distribution. The distribution number of vehicle O-D is a method for estimating the O-D distribution number of neighboring cells.

Technical scheme of the present invention:

A method for predicting O-D spatiotemporal distribution based on the spatial Durbin model, characterized in that the steps are as follows:

(1) Traffic district division

First, the study area is divided into traffic cells, which can be divided into administrative townships and streets or grid division.

(2) Extraction and statistics of urban built environment elements

According to the research needs, extract the indicators of various urban built environment elements in the traffic area, mainly including density, land use diversity, block design, destination accessibility and distance from public transportation facilities, and conduct statistics on the built environment elements in the traffic area. In addition, it is necessary to obtain the traffic generation and attraction of each traffic area during the study period through basic data processing.

(3) The basic form of the space Doberman model

y=ρWy+Xβ+γWX+ε,ε～N(0,σ ² I _n ) (1)

In the formula, n is the number of traffic districts; y is an n×1 vector, which represents the explanatory variable, that is, the traffic generation or attraction in the morning peak of a traffic district; X is an n×k data matrix, representing the explanatory variable, that is, each traffic district is a built-in environment index, k is the number of urban built-up environment elements; W is the spatial weight matrix, ρ is the coefficient of the spatial lag dependent variable Wy, γ is the coefficient of the spatial lag independent variable WX, and β reflects the effect of the explanatory variable on the change of the dependent variable y. , ε is the random error term;

The spatial weight matrix W reveals the interactions between spatial units in the form:

Each element in the space weight matrix is a space weight, and the calculation method of the space weight adopts the inverse distance weight matrix, and the form is as follows:

(2) Effect decomposition of the spatial Doberman model

In order to explain the spillover effects in the spatial Durbin model, the direct, indirect and total effects of independent variables on the dependent variable are introduced, that is, the change of a single built environment element associated with the amount of traffic generation or attraction in any traffic district, It not only affects the traffic generation or attraction of the community, but also indirectly affects the traffic generation and attraction of other adjacent areas;

In order to measure the direct effect and indirect effect, formula (1) is rewritten as the following formula:

(I _n -ρW)y=ι _n α+Xβ+WXθ+ε (4)

y=(I _n -ρW) ^-1 ι _n α+(I _n -ρW) ^-1 X(I _n β+Wθ)+(I _n -ρW) ^-1 ε (5)

Let (I _n -ρW) ^-1 =V(W), S(W)=V(W)(In β+ _Wθ ), then formula (6) is obtained, the matrix form is formula (7), the parameter effect The matrix is expressed as formula (8):

y=V(W)ι _n α+S(W)X+V(W)ε (6)

The sum of the main diagonal elements in formula (8) divided by n is the average direct effect; the sum of all elements of S _r (W) in the matrix divided by n is the average total effect; the difference between the average total effect and the average direct effect is recorded as Average indirect effects; as follows:

In the above formula, r=1, 2, 3,...,k, is the number of urban built-up environment elements, ι _n represents an n×1-order matrix, and M(k) _Direct represents the urban built-up environment elements to the traffic in this traffic area. The direct effect of generation or attraction, M(k) _Indirect represents the indirect effect of urban built-up environment elements on the traffic generation or attraction in adjacent traffic areas, M(k) _total represents the urban built-environment factors on traffic generation in the traffic area The total effect of the amount or the amount of attraction.

Beneficial effects of the present invention: the O-D spatiotemporal distribution prediction method based on the spatial Durbin model of the present invention adds the built environment as an explanatory variable for the traffic generation and attraction of the traffic area, which proves the explanatory power of the built environment for the O-D spatiotemporal distribution; On this basis, the spillover effect of the impact of the urban built environment on the O-D distribution is proved, and it is quantified. This spillover effect is decomposed into direct effect, indirect effect and total effect, and the detailed information of spatial correlation is improved. The planning department provides reference for urban land planning.

Detailed ways

The specific embodiments of the present invention are described in detail below in conjunction with the technical solutions, and the implementation effects of the present invention are simulated.

(1) Research objects

Selecting the city of Shenzhen as the research scope, Shenzhen is not only the economic center of the country, but also has well-established infrastructure, rich and wide distribution of urban built environment elements; it is also the center of population agglomeration in the Pearl River Delta region, and its population flows both internally and externally. Quite large and easy to carry out research.

(2) Basic data

The ArcGIS software is used to complete the division of traffic cells. Considering the richness of the built environment inside the traffic cells and the operability of grid data, this study finally selects a 1.5km*1.5km grid as the unit traffic cell scale, and obtains 1031 traffic cells. . Using Oracle programming to select the starting and ending points of the floating car trajectory from 6:00-8:30 every day from June 9th to 14th, 2014, spread the matched GPS points of the floating car into the traffic area, and extract each GPS point in ArcGIS. The traffic generation and attraction of the community are the dependent variables. Select the hotel density, restaurant density, supermarket density, drugstore density, building density, school density, hospital density, bank density, government unit density, bus station density, intersection density, subway station density, land use diversity in the study area A total of 14 built environment elements are used as built environment independent variables. Statistical values for each built environment attribute are given in Table 1.

Table 2.1 Statistical values of main independent variables in each traffic district

(3) Global regression to determine significant independent variables

For most spatial empirical analysis, spatial econometric modeling generally starts with a non-spatial linear regression model, and then further discusses whether the model needs to be extended to consider spatial interaction effects and then establish a spatial econometric model. Therefore, this study first established a global regression model. The model serves as a benchmark reference for spatial econometric analysis.

In the global regression model, the morning peak traffic generation and attraction (O-D) of 1031 traffic districts were used as dependent variables, and the urban built environment attributes were used as independent variables. The model calibration was completed in SPSS software. The estimation results are shown in Table 2. When the absolute value of the t value is greater than 1.96, it means that the independent variable has a significant impact on the traffic attraction or occurrence.

Table 2 Global regression model results

Note: ***, **, * indicate significance levels of 99%, 95% and 90%, respectively.

When the explained variable in the model results is traffic generation, R _a ² is 0.610, indicating that the independent variables in the model can explain 61.0% of the changes in traffic generation; when the explained variable is traffic attraction, R _a ² is 0.640, It shows that the independent variables in the model can explain 64.0% of the changes in traffic attraction.

It can be seen from Table 2 that the number of hotels, buildings and bus stops have a significant positive correlation with traffic generation; while the number of supermarkets, diversity and distance from transportation hubs have a significant negative correlation with traffic generation. There is a significant positive correlation between the number of buildings, subway stations, bus stations and government and traffic attraction, and a significant negative correlation between diversity and traffic attraction.

4 Substitute significant independent variables into the spatial Doberman model

Based on Matlab programming, the results of the spatial Durbin model are shown in Tables 3 and 4 below.

Table 3 Parameter estimation results of spatial Durbin model (traffic generation)

Note: ***, **, * indicate significance levels of 99%, 95% and 90%, respectively.

Table 4. Parameter estimation results of spatial Durbin model (traffic attraction)

Note: ***, **, * indicate significance levels of 99%, 95% and 90%, respectively.

From Table 3 and Table 4, it can be seen that hotel density, building density and bus station density have a significant positive correlation with traffic generation, and supermarket density has a significant negative correlation with traffic generation; building density, subway station density, bus station density There is a significant positive correlation between density and government density and traffic attraction, which is consistent with the results of the global regression model. However, the effect of diversity on traffic attraction and traffic generation is not significant. This is because the existence of the spatial weight matrix makes the effect of diversity on the explained variable decomposed, and the effect of diversity on its own traffic generation or attraction is not significant. .

The variables with W in Table 4 and Table 4 represent the spatial lag term, which means that the traffic generation or traffic attraction of the traffic area is affected by the built environment variables of the surrounding traffic areas. For traffic generation, except for the spatial lag variables of supermarket density and bus stop which are not significant, all the other spatial lag variables are significant. Specifically, the density of hotels and buildings in the surrounding community has a significant positive impact on the traffic generation in this community, while the diversity of surrounding communities has a significant negative impact on the traffic generation in this community. For traffic attraction, the spatial lag variables of other variables are all significant except the spatial lag variables of bus stops which are not significant. The density of buildings and subway stations in the surrounding community has a significant positive impact on the traffic attraction of the community, while the density and diversity of government in the surrounding community have a significant negative impact on the traffic attraction of the community.

The coefficient ρ of the spatial lag variable is positive and statistically significant, indicating that the spatial spillover effect cannot be ignored when studying the impact of the built environment on the O-D distribution. And if the trip or attraction of the surrounding traffic area increases by 1%, the traffic generation or attraction of this area will increase by about 0.24% and 0.25%.

The model parameters in Tables 3 and 4 are decomposed into direct effects and indirect effects, and then the total effect is obtained. The decomposition calculation results are shown in Table 5 and Table 6.

Table 5. Direct, indirect and total effects of the spatial Durbin model (traffic generation)

Note: ***, **, * indicate significance levels of 99%, 95% and 90%, respectively.

Table 6 Direct, indirect and total effects (traffic attraction) of the spatial Durbin model

Note: ***, **, * indicate significance levels of 99%, 95% and 90%, respectively.

From a quantitative point of view, the impact of bus stop density on traffic generation is significantly positively correlated, and the direct effect accounts for 51% of the total effect, and the indirect effect accounts for 49% of the total effect, indicating that the increase of bus stop density on traffic generation comes from explanation. The double role of variables. This reflects that the daily travel of residents is centered on the bus station and spreads to the surrounding area, and the diffusion range is considerable, suggesting that traffic managers should appropriately add taxi pick-up points near the bus station to meet this part of travel needs. In addition, regardless of whether diversity has a direct effect on traffic generation or attraction, the indirect effect and total effect are both statistically significant and negative, and the indirect effect is an absolute contribution to the total effect, indicating that diversity is not a direct determinant of traffic generation or attraction. . It is generally believed that a certain area has complete supporting facilities, various land use types, and a scientific and reasonable block design. The community is considered to be a mature community. Consistent with previous research, the higher the degree of diversity in a mature community, the increase in the diversity of surrounding communities. This results in less traffic generation and attraction.

Claims (1)

1. a kind of O-D spatial and temporal distributions prediction technique based on space Du's guest's model, which is characterized in that steps are as follows:

(1) traffic zone divides

Traffic zone division is carried out to survey region first, using the division of administrative small towns street or rasterizing division mode；

(2) city built environment elements recognition and statistics

According to needs, the index of various city built environment elements in traffic zone, including density, land use multiplicity are extracted Property, design of street, destination accessibility and apart from public transport facility distance, and carry out built environment element in traffic zone and count； In addition, also to obtain the traffic generation and traffic attraction of each traffic zone in the research period by basic data processing；

(3) citation form of space Du guest's model

Y=ρ Wy+X β+γ WX+ ε, ε~N (0, σ²I_n) (1)

In formula, n is traffic zone quantity；Y is the vector of n × 1, indicates that explained variable i.e. certain traffic zone morning peak traffic is raw At amount or traffic attraction；X is n × k data matrix, represents the various built environment indexs in explanatory variable i.e. certain traffic zone, k is city Built environment element number；W is Spatial weight matrix, and ρ is the coefficient of Spatial lag dependent variable Wy, and γ is Spatial lag independent variable The coefficient of WX, β reflect the influence that explanatory variable generates dependent variable y variation, and ε is stochastic error；

Spatial weight matrix W discloses the interaction between space cell, and form is as follows:

Each element in Spatial weight matrix is space weight, and the calculation method of space weight uses anti-distance weighting matrix, Form is as follows:

(2) effect of space Du guest's model is decomposed

In order to version space Du guest's model in spills-over effects, introduce independent variable to the direct effect of dependent variable, indirect effect and Gross effect, i.e., the variation of single built environment element associated with traffic generation in any traffic zone or traffic attraction, no Only this cordon traffic production quantity or traffic attraction are had an impact, at the same also influence indirectly other adjoining areas traffic generation and Traffic attraction；

For the ease of estimating direct effect and indirect effect, formula (1) is rewritten as following formula:

(I_n- ρ W) y=ι_nα+Xβ+WXθ+ε (4)

Y=(I_n-ρW)^-1ι_nα+(I_n-ρW)^-1X(I_nβ+Wθ)+(I_n-ρW)^-1ε (5)

Enable (I_n-ρW)^-1=V (W), S (W)=V (W) (I_nβ+W θ), then formula (6) are obtained, matrix form is formula (7), the square of parameter effects Battle array is expressed as formula (8):

Y=V (W) ι_nα+S(W)X+V(W)ε (6)

The sum of the elements in a main diagonal is average direct effect divided by n in formula (8)；S in matrix_r(W) the sum of all elements are divided by n Average overall effect；The difference of average overall effect and average direct effect is denoted as average indirect effect；It is as follows:

In above formula, r=1,2,3 ..., k are city built environment element number, ι_nIndicate the rank matrix of n × 1, M (k)_DirectIt indicates City built environment element is to the direct effect of traffic generation or traffic attraction in this traffic zone, M (k)_IndirectIndicate city Built environment element is to the indirect effect of traffic generation or traffic attraction in neighbouring traffic zone, M (k)_totalIndicate that city is built up Gross effect of the environmental element to traffic generation in traffic zone or traffic attraction.