A trip chain destination selection prediction method considering neighborhood influence

Abstract

The present invention discloses a method for predicting the destination selection of a travel chain that takes into account the influence of neighborhoods. The present invention divides a region into separate traffic zones, and determines the adjacency matrix of different traffic zones through spatial position relationships; the traffic model based on the travel chain framework predicts the total amount of travel for different groups and different purposes in different traffic zones, and calculates the service levels of different traffic zones for different purposes based on different travel purposes. Based on the adjacency matrix, a graph convolutional neural network is constructed between traffic zones to fully learn the characteristics of the neighborhood zones around the zone. Finally, the logit model is used to comprehensively consider the accessibility of the zone, the comprehensive attractiveness level of the zone and the neighborhood, and the K parameter of the zone to which the zone belongs, to calculate the probability of destination selection. The present invention achieves a method for predicting the destination selection of travel chains through a logit model by characterizing the impedance between traffic zones, calculating the locational influence of the zone and its surrounding zones through a graph neural network, and factors such as the K parameter.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a travel chain destination selection prediction method considering neighborhood influence. Background Art

[0002] The research background of destination choice prediction is based on the understanding of human mobility patterns. The choice of individual travel destination is a complex process, which is affected by many factors, including personal habits, traffic conditions, time, weather, etc. With the acceleration of urbanization and the diversification of transportation, accurately predicting individual destination choices is of great significance for alleviating traffic congestion and improving the efficiency of the transportation system.

[0003] In the prior art, the prediction of destination selection is usually calculated using a gravity model, which calculates the road accessibility between different communities and fits the parameters to make predictions. However, this method only considers the accessibility of traffic communities or the absolute amount of attraction generated by traffic communities, and does not describe the mutual influence between communities and the spatial relationship between communities. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a travel chain destination selection prediction method taking neighborhood influence into account.

[0005] A first aspect of the present invention provides a method for predicting travel chain destination selection considering neighborhood influence, the method comprising the following steps:

[0006] Step 1) Based on the urban skeleton road network and land use structure distribution, as well as administrative boundaries, the area is divided into multiple traffic zones and numbered;

[0007] Step 2), based on the geographic spatial distribution of traffic communities, construct a graph network structure;

[0008] Step 3) Cross-classify the population structure and travel purpose to generate basic analysis units for travel entities with different purposes for different groups of people; use big data analysis to fit the travel rates of different groups of people with different purposes in different traffic areas, and calculate the total travel volume in different traffic areas;

[0009] Step 4), based on different travel purposes, associate relevant types of land use and calculate the service level of different transportation areas for different purposes;

[0010] Step 5), using the constructed graph network structure and the service level of the traffic zone, construct a graph convolutional neural network (GCN), obtain the arrival volume of different traffic zones through signaling big data, perform GCN network training, and calculate the attractiveness of the traffic zone;

[0011] Step 6), based on the road network, obtain the comprehensive impedance between different traffic areas and calculate the accessibility between the traffic areas;

[0012] Step 7) Considering the accessibility of the transportation area, the comprehensive attractiveness level of the transportation area and the neighborhood, and the K parameter of the location of the area, the logit model is used to calculate the probability of destination selection.

[0013] According to a second aspect of the present invention, a computer-readable storage medium is provided, on which program instructions are stored, and the program instructions implement the above method when executed.

[0014] According to a third aspect of the present invention, a computer program product is provided, comprising a computer program / instruction, which implements the above method when executed by a processor.

[0015] The beneficial effects of the present invention are as follows: the present invention realizes a method for predicting the destination selection of a travel chain by characterizing the impedance between traffic communities, calculating the location influence of a community and its surrounding communities by using a graph neural network, factors such as K parameters, and using a logit model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flow chart of a method according to an embodiment of the present application;

[0017] Figure 2 This is the graph network structure of an embodiment of the present application. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the patent examples of the present invention to make a clear and complete description of the principles, structures, methods and technical effects of the present invention, so as to fully understand the purpose, characteristics and effects of the present invention. Obviously, the implementation described is only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the patent of the present invention, all other implementations obtained by ordinary technicians in this field without creative work or technical solutions that can be obtained through logical analysis, reasoning or limited experiments on the basis of the prior art of the present invention are within the scope of protection of the patent of the present invention.

[0019] The basic concept of this application is as follows: This application divides the entire study area (generally a city, region, or street) into separate traffic zones, and determines the adjacency matrix of different traffic zones through spatial position relationships; the traffic model based on the travel chain framework predicts the total amount of travel for different groups and different purposes in different traffic zones, and uses multivariate linear regression based on different travel purposes to calculate the service level of different traffic zones for different purposes, and at the same time, calculates the road impedance between different traffic zones. Based on the adjacency matrix, a graph convolutional neural network is constructed between traffic zones to fully learn the characteristics of the neighborhood zones around this zone. Finally, the logit model is used to comprehensively consider the accessibility of the zone (road impedance between zones), the comprehensive attractiveness level of the zone and its neighborhood (calculated by the graph convolutional neural network), and the location of the zone (the location is related to the urban functional zone, such as the core area, suburbs, etc.) K parameter (describes the other unmeasurable variable parts in the process of traffic travel behavior selection, except for parameters such as time, distance, and cost, generally defined as constant terms under traffic zones or groups), to calculate the probability of destination selection. The above technical concept can solve the problem that the traditional gravity model only considers the accessibility of traffic areas, or the absolute amount of attraction generated by traffic areas, without describing the influence of the location between areas. In terms of logical principles, it is more in line with objective reality and has better prediction effects.

[0020] The present application embodiment provides a method for predicting the destination selection of a travel chain considering the influence of neighborhoods. The method is specifically:

[0021] Divide traffic zones and number them: Divide geographical space based on the urban skeleton road network, land use structure distribution, and administrative boundaries. Divide the study area into a separate traffic zone. The main division principles are as follows:

[0022] To facilitate data acquisition and use, subdivision boundaries should refer to administrative boundaries;

[0023] The land use, economic, social and other characteristics within the transportation zone should be as consistent as possible;

[0024] Maintain the integrity of the zoning to avoid the division of land for the same purpose;

[0025] Whenever possible, natural barriers such as railways and rivers should be used as zoning boundaries;

[0026] Zoning takes into account the separation effect of highways, expressways and other roads;

[0027] The core area has a high socioeconomic level and high travel intensity, but the aggregated degree of residents' travel is relatively low, so the division of traffic zoning should be more detailed.

[0028] Graph network structure generation: Based on the geographical distribution of traffic cells, the cell centroids are used as nodes, and edges are generated when the cells are directly connected. The graph network structure G(V,E) and its adjacency matrix A and degree matrix D are generated from the nodes and edges. Figure 2 As shown in the figure: 1 to 6 represent traffic communities, edges represent spatial connections, and with communities as vertices, a graph structure G(V,E) is constructed, where V = {1, 2, 3, 4, 5, 6}, E = {(1, 2), (1, 6), (1, 5), (2, 3), (3, 4), (4, 5), (4, 6)}.

[0029] Generation and calculation of travel volume: Based on cross-classification of population structure (family attributes, personal attributes) and travel purpose, basic analysis units of travel entities with different purposes for different groups of people are generated. Through big data fitting, travel rates of different purposes for different groups of people in different traffic areas are obtained, and the total travel volume in different traffic areas is calculated.

[0030] In one example, the specific calculation formula for the total travel volume is as follows:

[0031]

[0032] Among them: i represents the traffic area; p represents the purpose of travel, and common travel purposes include home-based work, home-based shopping, home-based other, non-home-based, etc.; c represents the travel group, and common group classifications include working people, retired people, primary and secondary school students, college students, farmers, etc.; T ipc represents the total number of trips of type c and type p in traffic zone i throughout the day; ic Indicates the population of group c in community i; Rate ipc It represents the probability of type p travel group in traffic community i based on type c travel purpose.

[0033] Calculation of service levels for transportation communities: Based on different purposes, relevant types of land are associated to obtain the service levels of different communities.

[0034] In one example, the service levels of different cells are calculated as follows:

[0035] S jpc =∑E jpc *W jpc

[0036] Where: S jpc E represents the service level for group c and purpose p in cell j; jpc W represents the impact factor coefficient of category p in cell j on category c; jpc It represents the influencing factors of category p of group c within community j. Common factors include total population, total number of jobs, number of degrees, building area, etc. The values ​​are different in different locations.

[0037] Calculation of attractiveness of traffic zones: Based on the graph structure constructed in step 2 and the traffic zone service level obtained in step 4, a GCN graph neural network is constructed. The arrival volume of different traffic zones is obtained based on signaling big data, and the GCN graph neural network is trained. Based on the parameter values ​​obtained from the training, the attractiveness of the traffic zone is calculated.

[0038] In one example, the attractiveness of a traffic zone is calculated as follows:

[0039]

[0040] Where: H (l+1) is the node feature matrix of the l+1th layer, and the initial feature matrix is ​​S pc , which is the initial service level. (l) is a learnable weight matrix and σ is a nonlinear activation function such as ReLU. It is the result of adding self-connection (that is, each node is connected to itself) to the original adjacency matrix. yes The corresponding degree matrix.

[0041] The converged node feature matrix H calculated by the graph neural network n , which is the attractiveness value of different traffic areas, as follows:

[0042]

[0043] Where: n represents the number of neural network layers, A 1pc Indicates the attractiveness to P type of people and c type of purposes.

[0044] In this step, not only the internal variables of a single community, such as population, land use, jobs, and social factors within the community, but also the influencing factors of other traffic communities around the community are fully considered. The characteristics of the community and its neighboring communities are fully learned through the graph neural network. It can better express the relevant impact of the spatial relationship between communities, which is lacking in traditional gravity models and other distribution models.

[0045] Calculation of accessibility between traffic zones: Based on the road network, the comprehensive impedance between different traffic zones is obtained. Specifically:

[0046] Based on different modes of transportation, the time, cost and other parameters between different modes are obtained, and based on the proportion between different modes, the comprehensive impedance is obtained.

[0047] In one example, the comprehensive impedance is calculated as follows:

[0048]

[0049] in: It represents the time impedance of mode m between traffic zones ij; Vot represents the cost impedance of mode m between traffic zones ij; pc is the time value parameter; β pcm is the utility coefficient; U ijpcm represents the utility of the following method m for the c-type group P between traffic communities ij; U ijpc represents the comprehensive utility of category P of c groups between traffic communities ij, that is, the accessibility between communities; P m is the proportion of mode m to all modes.

[0050] Destination selection calculation: The MNL-type logit model is used to calculate the probability of selection between communities based on the accessibility between transportation communities, the total volume of community trips, and the community attraction level.

[0051] In one example, the selection probability calculation formula is as follows:

[0052]

[0053] PA ijpc =T ipc *P ijpc

[0054] Where: P ijpc represents the probability of a group of category c, category p, choosing from cell i to cell j; PA ijpc represents the travel volume from cell i to cell j for group c of category p; T ipc A represents the total number of trips from community i for group c and category p; jpc represents the attraction of category c group p in community j; θ pc represents a scaling factor; K is a constant term based on the zone to which the cell belongs, and the zone reflects the difference in location.

[0055] Based on the same concept as the above method, an embodiment of the present application provides a computer-readable storage medium on which program instructions are stored, and the above method is implemented when the program instructions are executed.

[0056] Based on the same concept as the above method, an embodiment of the present application also provides a computer program product, including a computer program / instruction, which implements the above method when executed by a processor.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A trip chain destination selection prediction method considering neighborhood influence, characterized by The method comprises the following steps: Step 1) Based on the urban skeleton road network and land use structure distribution, as well as administrative boundaries, the area is divided into multiple traffic zones and numbered; Step 2), based on the geographic spatial distribution of traffic communities, construct a graph network structure; Step 3) Cross-classify the population structure and travel purpose to generate basic analysis units for travel entities with different purposes for different groups of people; use big data analysis to fit the travel rates of different groups of people with different purposes in different traffic areas, and calculate the total travel volume in different traffic areas; Step 4), based on different travel purposes, associate relevant types of land use and calculate the service level of different transportation areas for different purposes; Step 5), using the constructed graph network structure and the service level of the traffic zone, construct a graph convolutional neural network, obtain the arrival volume of different traffic zones through signaling big data, perform graph convolutional neural network training, and calculate the attractiveness of the traffic zone; Step 6), based on the road network, obtain the comprehensive impedance between different traffic areas and calculate the accessibility between the traffic areas; Step 7) Considering the accessibility of the transportation community, the comprehensive attractiveness of the transportation community and its neighborhood, and the K parameter of the location of the community, the logit model is used to calculate the probability of destination selection, where the K parameter represents an unmeasurable variable in the process of transportation travel behavior selection and is a constant term; Wherein step 2) specifically comprises: 2.1) Determine the center point of each traffic zone as a node of the graph network structure; 2.2) Determine the edges based on the spatial direct connection relationship between traffic zones; 2.3) Generate a graph network structure G(V,E), where V represents the node set and E represents the edge set; 2.4) Construct the adjacency matrix A and degree matrix D to describe the connection relationship between nodes and the degree of nodes; The service level of different traffic zones for different purposes in step 4) is expressed as: S jpc =∑E jpc *W jpc S jpc E represents the service level for group c and purpose p in cell j; jpc W represents the impact factor coefficient of category p in cell j on category c; jpc represents the impact factor of category p on category c in cell j; Step 5) is specifically: 5.1) Construct a graph convolutional neural network model based on graph network structure; 5.2) Using signaling big data, a graph convolutional neural network model is trained to learn the attractiveness of traffic zones; 5.3) The node feature matrix calculated by the graph convolutional neural network model represents the attractiveness of different traffic zones; In the constructed graph convolutional neural network model, the node feature matrix of the l+1th layer is represented as: Where: σ is a nonlinear activation function, W (l) is a learnable weight matrix, It represents the result of adding self-connection to the adjacency matrix A. yes The corresponding degree matrix; the initial feature matrix of the node feature matrix is ​​S pc , S pc It represents the service level for category c group and category p purpose; The comprehensive impedance in step 6) is calculated as follows: in: It represents the time impedance of mode m between traffic zones ij; Vot represents the cost impedance of mode m between traffic zones ij; pc is the time value parameter; β pcm is the utility coefficient; U ijpcm represents the utility of the following method m for the c-type group P between traffic communities ij; U ijpc represents the comprehensive utility of category P of category c between transportation communities ij, that is, the accessibility between transportation communities; P m is the proportion of mode m to all modes; The calculation formula for the destination selection probability in step 7) is as follows: PA ijpc =T ipc *P ijpc Where: P ijpc represents the probability of a group of category c, category p, choosing from cell i to cell j; PA ijpc represents the travel volume from cell i to cell j for group c of category p; T ipc A represents the total number of trips from community i for group c and category p; jpc represents the attraction of category c group p in community j; θ pc represents the scaling factor; K is a constant term based on the zone to which the cell belongs.

2. A method for predicting travel chain destination selection considering neighborhood influence according to claim 1, characterized in that: The population structure in step 3) includes family attributes and personal attributes, and the travel purposes include home-based work, home-based shopping, home-based other and non-home-based.

3. The method for predicting travel chain destination selection considering neighborhood influence according to claim 1, characterized in that: Step 4) The service level is based on land use type, total population, total number of jobs and number of degrees.

4. A computer-readable storage medium, characterized in that: Program instructions are stored thereon, and when the program instructions are executed, the method according to any one of claims 1 to 3 is implemented.

5. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the method according to any one of claims 1 to 3 is implemented.

Images