Commercial district passenger flow prediction method, system and device based on graph neural network

Abstract

The invention provides a graph neural network-based commercial district passenger flow prediction method, system and device, and relates to the technical field of commercial district passenger flow prediction, and the method mainly comprises the steps: constructing a feature space of commercial district passenger flow prediction based on a traffic cell set and a commercial district set; embedding the traffic zone features into a graph neural network to obtain traffic zone embedded vectors; embedding the business district features into a first multi-layer perceptron to obtain business district embedded vectors; the embedded vectors are connected with the corresponding OD flow features to serve as OD flow comprehensive representation vectors; training and verifying the second multi-layer perceptron; and taking the OD flow comprehensive representation vector which is best in performance on the verification set as input, transmitting the OD flow comprehensive representation vector into a full-connection neural network, and predicting the commercial district passenger flow volume. According to the scheme, the geographical semantic information of the plot is incorporated into the model for feature embedding, a multi-graph structure combining two spatial dependencies is constructed, the OD flow of the business district is predicted through the graph neural network and the graph embedding technology, and the prediction accuracy is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of shopping district passenger flow prediction, and in particular to a shopping district passenger flow prediction method, system and device based on graph neural network. Background Art

[0002] At present, the prediction methods for the passenger flow in shopping districts have developed from traditional models to machine learning models. The most commonly used traditional models, such as gravity model and Poisson regression, have good explanatory power, but due to the simple formula and the parameters derived from empirical values, they cannot capture nonlinear and irregular patterns well, and it is difficult to accurately model complex human flow patterns, nor can they predict passenger flow in different areas of the city.

[0003] Although existing machine learning models can simulate the complex patterns of passenger flow to a certain extent, they often ignore the structural characteristics of the spatial interaction network formed by the flow of people within the city and the spatial proximity effect between locations. They also do not consider the impact of the economic attributes of the source of customers on the passenger flow in the business district. Therefore, they cannot accurately predict the passenger flow from transportation areas to business districts at a fine scale. Summary of the invention

[0004] The purpose of the present invention is to provide a method, system and device for predicting customer flow in a shopping district based on a graph neural network to solve at least one of the above-mentioned technical problems existing in the prior art.

[0005] In the first aspect, in order to solve the above technical problems, the present invention provides a method for predicting customer flow in a shopping district based on a graph neural network, comprising the following steps: Step 1: Based on the set of traffic communities and the set of business districts, a feature space for predicting business district passenger flow is constructed, including OD flow features, business district features, and traffic community features.

[0006] In a feasible implementation manner, the OD flow characteristics include OD flow and OD distance; the OD flow is Traffic Zone To Business district The average number of passengers per hour (i.e., OD pairs); the OD distance is and The Euclidean distance between .

[0007] Preferably, the average number of passengers per hour is obtained by calculating the average passenger flow from each traffic zone to the business district every hour every day based on mobile phone signaling data (of a preset year).

[0008] In a feasible implementation, the business district characteristics include the average daily passenger flow of the business district, the land area of ​​the business district, the business diversity index of the business district, and the number of parking lots; The average daily passenger flow of the business district is based on the user daily residence anonymized information table provided by the big data platform (such as the Smart Footprint Extreme Intelligence DaaS platform), and is obtained through (SQL language) query; the user daily residence anonymized information table includes residence time, residence location, etc.; The commercial area is collected from the building outline data provided by the navigation platform (such as Amap); The business district business diversity index is calculated based on the number of POIs (points of interest) provided by the navigation platform using the HillNumbers index formula. The specific formula is: ; in, It represents the diversity index of business formats in the business district; Indicates the total number of business types in the business district; Indicates the first The proportion of POIs of this business type to the total number of POIs in the business district; It represents the parameter, specifically the inverse of the Simpson index, which can take into account the richness of POIs and the uniformity of different business types; The number of parking lots refers to the number of parking lots within a preset buffer zone of the business district, which can be collected from the navigation platform.

[0009] In a feasible implementation, the traffic community characteristics include the average housing price of the community, the average consumption amount of the community, the distance from the community to the urban center, the residential population of the community and the employed population of the community; The average housing price of the community can be collected from second-hand housing price data of a residential transaction platform (such as Lianjia Company) and processed by an inverse distance weighted interpolation method; The average consumption amount of the community can be collected from consumption statistics data provided by financial institutions (such as UnionPay) (such as consumption statistics data of various categories in a 1km grid in a city, which are spatially aggregated and do not involve personal consumption privacy), and processed by the inverse distance weighted interpolation method; The distance from the community to the urban center can be the Euclidean distance from the center point of each traffic community polygon to the nearest urban center; The residential population of the community and the employed population of the community can be obtained through (SQL language) query based on the user daily residence desensitized information table provided by the big data platform; the residence types of the user daily residence desensitized information table include residence, work and visit.

[0010] Step 2: Embed the traffic zone features into the graph neural network to obtain the traffic zone embedding vector; embed the business district features into the first multi-layer perceptron to obtain the business district embedding vector; connect the embedding vector with the corresponding OD flow features as the OD flow comprehensive representation vector.

[0011] In a feasible implementation, the specific method of embedding the traffic community into the graph neural network includes: The traffic zone features are respectively embedded through a first graph convolutional neural network (GCN) module and a second graph convolutional neural network module set in parallel to obtain a first embedding vector and a second embedding vector respectively; The first graph convolutional neural network includes a neighborhood graph , the second graph convolutional neural network includes the consumption level similarity graph ,in, Represents the nodes of the graph, i.e., the traffic subdistrict; Indicates the number of nodes; Indicates the number of traffic zones, equal ; Represents the edges connecting nodes; An adjacency matrix representing weighted edges in a neighborhood graph; An adjacency matrix representing weighted edges in a consumption level similarity graph; The first embedding vector The specific expression is: ; in, Represents the dimension of the vector; express Matrix of order; The second embedding vector The specific expression is: ; Then and Fusion is performed to obtain the third embedding vector , the specific formula can be: ; in, and are all trainable weight parameters and belong to .

[0012] In one possible implementation, The definition formula of the middle edge is: ; in, represents the trainable distance decay parameter of the neighborhood graph; Indicates Traffic Zone With Traffic Zone In this way, the graph can be constructed by the proximity of two transportation cells (irregular geographical units) and information can be passed through the edges.

[0013] In one possible implementation, The definition formula of the middle edge is: ; in, Trainable distance decay parameters representing the consumption level similarity graph; express consumption level; express consumption level; express and The reason for designing the formula in this way is that two areas with similar consumption levels may have similar shopping options even if they are far apart in space, and the housing prices and consumption amounts in a traffic community can reflect the differences in consumption levels of different groups of people to a certain extent. Therefore, this application uses housing prices and consumption amounts as evaluation indicators of the consumption level of traffic communities, and uses cosine similarity to calculate the similarity of consumption levels of two traffic communities.

[0014] In a feasible implementation, each graph convolutional neural network module is configured with a multi-layer conventional structure, namely, an input layer, a graph convolutional layer, and an output layer connected in sequence, so as to reduce the amount of computation while ensuring the embedding effect.

[0015] In a feasible implementation, the first multi-layer perceptron has a two-layer structure, so as to reduce the amount of computation while ensuring the embedding effect.

[0016] Step 3: Based on the OD flow comprehensive representation vector, a training set and a validation set are constructed; based on the training set, the second multi-layer perceptron is trained to predict the passenger flow between each traffic zone and each business district; based on the validation set, the second multi-layer perceptron is iteratively verified until the performance of the validation set is no longer improved, thereby avoiding overfitting; the OD flow comprehensive representation vector with the best performance on the validation set is used as input and passed into the fully connected neural network (FCN); based on external prediction requirements, a test set is constructed, and the passenger flow of the business district is predicted through the fully connected neural network; In a feasible implementation, the second multi-layer perceptron has a batch-normalized multi-layer structure (eg, four layers) so as to learn embedding through back propagation.

[0017] In a feasible implementation manner, the loss function of the second multi-layer perceptron is a mean square error function, and the specific formula is: ; in, Indicates the number of matches between transportation communities and business districts; Indicates Traffic Zone To Business district The actual calculated value of the average number of passengers per hour; Indicates Traffic Zone To Business district The predicted value of the average number of passengers per hour.

[0018] In the second aspect, based on the same inventive concept, the present application also provides a shopping district passenger flow prediction system based on graph neural network, including a data receiving module, a data processing module and a result generating module; The data receiving module is used to receive a set of traffic areas, a set of business districts and external forecasting requirements; The data processing module includes a feature space unit, a feature embedding unit and a traffic prediction unit; The feature space unit constructs a feature space for predicting passenger flow in a business district based on a set of traffic districts and a set of business districts, including OD flow features, business district features, and traffic district features; The feature embedding unit embeds the traffic zone feature into the graph neural network to obtain the traffic zone embedding vector; embeds the business district feature into the first multi-layer perceptron to obtain the business district embedding vector; connects the embedding vector with the corresponding OD flow feature to obtain the OD flow comprehensive representation vector; The traffic prediction unit constructs a training set and a validation set based on the OD flow comprehensive representation vector; trains the second multi-layer perceptron based on the training set to predict the passenger flow between each traffic zone and each business district; it iteratively verifies the second multi-layer perceptron based on the validation set until the performance of the validation set is no longer improved; uses the best OD flow comprehensive representation vector on the validation set as input and passes it into a fully connected neural network (FCN); constructs a test set based on external prediction requirements, and predicts the passenger flow of the business district through the fully connected neural network; The result generation module is used to send the customer flow of the business district outward.

[0019] On the third aspect, based on the same inventive concept, the present application also provides a shopping district passenger flow prediction device based on graph neural network, including a processor, a memory and a bus, the memory stores instructions and data read by the processor, the processor is used to call the instructions and data in the memory to execute the shopping district passenger flow prediction method based on graph neural network as described above, and the bus connects the functional components for transmitting information.

[0020] By adopting the above technical solution, the present invention has the following beneficial effects: The present invention provides a method, system and device for predicting passenger flow in a business district based on a graph neural network, which incorporates geographic semantic information such as the socioeconomic attributes of a plot of land into the model for feature embedding, constructs a multi-graph structure combining two spatial dependencies, and predicts the OD flow of a business district through graph neural networks and graph embedding technology, which can achieve a higher prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 A flow chart of a method for predicting customer flow in a shopping district based on a graph neural network provided by an embodiment of the present invention; Figure 2 A diagram illustrating the principle of the CDFPM model provided by an embodiment of the present invention; Figure 3 A diagram of a shopping district passenger flow prediction system based on a graph neural network provided by an embodiment of the present invention; Figure 4 A comparison chart of the MAE values ​​of the CDFPM model and the baseline model provided in the embodiment of the present invention; Figure 5 A comparison chart of the RMSE values ​​of the CDFPM model provided in the embodiment of the present invention and the baseline model; Figure 6 A comparison chart of CPC values ​​between the CDFPM model provided in an embodiment of the present invention and the baseline model. DETAILED DESCRIPTION

[0023] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0025] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0026] To facilitate understanding of the following embodiments, the specific inventive concept of the present application is briefly described as follows: Based on graph neural networks and graph embedding algorithms, the present application constructs a prediction model for passenger flow in a business district (OD flow), referred to as CDFPM; first, the feature selection of the prediction model is performed from three dimensions: business district OD flow, business district, and traffic district; then, the graph neural network is used to represent the spatial proximity relationship and consumption similarity of the traffic district, and two GCN modules are used for feature embedding respectively, and then the final traffic district embedding vector is obtained through weight fusion; and a 2-layer multi-layer perceptron is used to embed the business district; finally, a 4-layer multi-layer perceptron network with batch normalization is used to predict the passenger flow in the business district, and the model is trained by the OD flow data of the business district obtained from the mobile phone signaling data. In this way, the present application takes into account the structural characteristics of the spatial interaction network formed by the flow of people in the city, the spatial proximity effect between places, and the influence of the economic attributes of the source of customers on the passenger flow in the business district, which is conducive to the accurate prediction of the passenger flow from the traffic district to the business district at a fine scale.

[0027] The present invention is further explained below in conjunction with specific implementation modes.

[0028] It should also be noted that the following specific embodiments or specific implementations are a series of optimized settings listed in the present invention to further explain the specific content of the invention, and these settings can be used in combination or in association with each other.

[0029] Embodiment 1: like Figure 1-2As shown, this embodiment provides a method for predicting customer flow in a shopping district based on a graph neural network, comprising the following steps: Step 1: Based on the set of traffic communities (1,374 traffic communities in a city) and the set of business districts (198 business districts in the city), a feature space for predicting business district passenger flow is constructed, including OD flow features, business district features, and traffic community features.

[0030] Furthermore, the OD flow feature includes two features, namely: Traffic Zone To Business district The average number of passengers per hour and and The Euclidean distance between .

[0031] Preferably, the average number of passengers per hour is based on the mobile phone signaling data (of China Unicom) in 2019, and is obtained by calculating the average flow of people from each traffic zone to the business district every hour of every day.

[0032] Furthermore, the business district characteristics include four characteristics, namely: average daily passenger flow of the business district, land area of ​​the business district, business diversity index of the business district and number of parking lots; The average daily passenger flow of the business district is based on the user daily residence anonymized information table provided by the big data platform (Smart Footprint Extreme Intelligence DaaS Platform), which is obtained through the common command query of the SQL language; the user daily residence anonymized information table includes residence time, residence location, etc.; The commercial area is collected from the 2019 building outline data provided by the navigation platform (Amap); The business district business diversity index is calculated based on the number of POIs (points of interest) provided by the navigation platform using the HillNumbers index formula. The specific formula is: ; in, It represents the diversity index of business formats in the business district; Indicates the total number of business types in the business district (e.g. catering services, shopping services, education and training, life services, cultural and sports leisure services and accommodation services, a total of 6 types); Indicates the first The proportion of POIs of this business type to the total number of POIs in the business district; The parameter is the inverse of the Simpson index, which can take into account the richness of POIs and the uniformity of different business types. For example, The value is 2; The number of parking lots refers to the number of parking lots within a preset buffer zone of the business district, which can be collected from a navigation platform. Exemplarily, the preset buffer zone is a 500-meter circular zone centered on the business district.

[0033] Furthermore, the traffic community characteristics include five characteristics, namely: average housing price in the community, average consumption amount in the community, distance from the community to the urban center, residential population in the community and employed population in the community; The average housing price of the community can be collected from the second-hand housing price data of the residential transaction platform (Lianjia), and processed by the inverse distance weighted interpolation method; The inverse distance weighted interpolation method specifically includes: Step a1: Calculate the Euclidean distance from the current traffic zone to other traffic zones. The specific formula is: ; in, Indicates the current cell to The Euclidean distance between traffic zones; and Respectively represent The two-dimensional coordinates of the traffic zone; and They respectively represent the two-dimensional coordinates of the cells to be found; Step a2: Calculate the inverse distance weight. The specific formula is: ; in, Indicates The inverse distance weight of each traffic zone; Indicates the number of traffic zones; Step a3: Calculate the average housing price of the community at the interpolation point. The specific formula is: ; in, Indicates the average house price of the community to be sought; Indicates The average housing price of each transportation community; in this way, the average housing price of the community can be assigned to the transportation community, and it can be reflected that the closer the distance between communities is, the closer the housing prices are; The average consumption amount in the community can be collected from the 2019 consumption statistics provided by the financial institution (UnionPay Company) (the consumption statistics of each category in the 1km grid of the city, which are spatially aggregated and do not involve personal consumption privacy), and processed by the inverse distance weighted interpolation method; the specific calculation method can refer to the formula in the average housing price of the community; The distance from the community to the urban center can be the Euclidean distance from the center point of each traffic community polygon to the nearest urban center; The residential population of the community and the employed population of the community can be obtained through common SQL language command queries based on the user daily residence desensitized information table provided by the big data platform; the residence types of the user daily residence desensitized information table include residence, work and visit.

[0034] Step 2: Embed the traffic zone features into the graph neural network to obtain the traffic zone embedding vector; embed the business district features into the first multi-layer perceptron to obtain the business district embedding vector; connect the embedding vector with the corresponding OD flow features as the OD flow comprehensive representation vector.

[0035] Furthermore, the specific method of embedding the traffic community into the graph neural network includes: The traffic zone features are respectively embedded through a first graph convolutional neural network (GCN) module and a second graph convolutional neural network module set in parallel to obtain a first embedding vector and a second embedding vector respectively; The first graph convolutional neural network includes a neighborhood graph , the second graph convolutional neural network includes the consumption level similarity graph ,in, Represents the nodes of the graph, i.e., the traffic subdistrict; Indicates the number of nodes; Indicates the number of traffic zones, equal ; Represents the edges connecting nodes; An adjacency matrix representing weighted edges in a neighborhood graph; An adjacency matrix representing weighted edges in a consumption level similarity graph; The first embedding vector The specific expression is: ; in, Represents the dimension of the vector; express Matrix of order; The second embedding vector The specific expression is: ; Then and Fusion is performed to obtain the third embedding vector , the specific formula can be: ; in, and are all trainable weight parameters and belong to .

[0036] Further, The definition formula of the middle edge is: ; in, represents the trainable distance decay parameter of the neighborhood graph; Indicates Traffic Zone With Traffic Zone In this way, the graph can be constructed by the proximity of two transportation cells (irregular geographical units) and information can be passed through the edges.

[0037] Further, The definition formula of the middle edge is: ; in, Trainable distance decay parameters representing the consumption level similarity graph; express consumption level; express consumption level; express and The reason for designing the formula in this way is that two areas with similar consumption levels may have similar shopping options even if they are far apart in space, and the housing prices and consumption amounts in a traffic community can reflect the differences in consumption levels of different groups of people to a certain extent. Therefore, this application uses housing prices and consumption amounts as evaluation indicators of the consumption level of traffic communities, and uses cosine similarity to calculate the similarity of consumption levels of two traffic communities; The cosine similarity The specific calculation formula includes: ; in, It represents the average housing price of the community; Indicates the average consumption amount in the community.

[0038] Furthermore, each graph convolutional neural network module is set up with a multi-layer conventional structure, that is, an input layer, a graph convolution layer (multi-layer stacked) and an output layer connected in sequence, so as to reduce the amount of calculation while ensuring the embedding effect.

[0039] Furthermore, the first multi-layer perceptron has a two-layer structure, namely an input layer, a hidden layer and an output layer connected in sequence, so as to reduce the amount of calculation while ensuring the embedding effect.

[0040] Step 3, based on the OD flow comprehensive representation vector, construct a training set and a validation set; based on the training set, train the second multi-layer perceptron to predict the passenger flow between each traffic zone and each business district; based on the validation set, iteratively verify the second multi-layer perceptron until the performance of the validation set is no longer improved, thereby avoiding overfitting; the OD flow comprehensive representation vector with the best performance on the validation set is used as input and passed into the conventional fully connected neural network; based on external prediction requirements, construct a test set, and predict the passenger flow of the business district through the fully connected neural network; the second multi-layer perceptron is a four-layer structure with batch normalization, that is, the input layer, the first hidden layer, the second hidden layer and the output layer are connected in sequence, so as to learn embedding through back propagation; the loss function of the second multi-layer perceptron is the mean square error function, and the specific formula is: ; in, Indicates the number of matches between transportation communities and business districts; Indicates Traffic Zone To Business district The actual calculated value of the average number of passengers per hour; Indicates Traffic Zone To Business district The predicted value of the average number of passengers per hour. Embodiment 2: like Figure 3 As shown, this embodiment provides a shopping district passenger flow prediction system based on graph neural network, including a data receiving module, a data processing module and a result generating module; The data receiving module is used to receive a set of traffic areas, a set of business districts and external forecasting requirements; The data processing module includes a feature space unit, a feature embedding unit and a traffic prediction unit; The feature space unit constructs a feature space for predicting passenger flow in a business district based on a set of traffic districts and a set of business districts, including OD flow features, business district features, and traffic district features; The feature embedding unit embeds the traffic zone feature into the graph neural network to obtain the traffic zone embedding vector; embeds the business district feature into the first multi-layer perceptron to obtain the business district embedding vector; connects the embedding vector with the corresponding OD flow feature to obtain the OD flow comprehensive representation vector; The traffic prediction unit constructs a training set and a validation set based on the OD flow comprehensive representation vector; trains the second multi-layer perceptron based on the training set to predict the passenger flow between each traffic zone and each business district; it iteratively verifies the second multi-layer perceptron based on the validation set until the performance of the validation set is no longer improved; takes the OD flow comprehensive representation vector with the best performance on the validation set as input and passes it into the fully connected neural network; constructs a test set based on external prediction requirements, and predicts the passenger flow of the business district through the fully connected neural network; the second multi-layer perceptron has a batch standardized four-layer structure; the loss function of the second multi-layer perceptron is a mean square error function, and the specific formula is: ; in, Indicates the number of matches between transportation communities and business districts; Indicates Traffic Zone To Business district The actual calculated value of the average number of passengers per hour; Indicates Traffic Zone To Business district The predicted value of the average number of passengers per hour; The result generation module is used to send the customer flow of the business district outward.

[0041] Embodiment three: The present embodiment provides a device for predicting passenger flow in a shopping district based on a graph neural network, including a processor, a memory and a bus. The memory stores instructions and data read by the processor, and the processor is used to call the instructions and data in the memory to execute the method for predicting passenger flow in a shopping district based on a graph neural network as described above. The bus connects the functional components for transmitting information.

[0042] In another embodiment, the present solution can be implemented by an integrated device, which may include corresponding modules for performing each or several steps in the above-mentioned embodiments. The module may be one or more hardware modules specially configured to perform the corresponding steps, or implemented by a processor configured to perform the corresponding steps, or stored in a computer-readable medium for implementation by a processor, or implemented by some combination.

[0043] The processor performs the various methods and processes described above. For example, the method implementation in the present solution can be implemented as a software program, which is tangibly contained in a machine-readable medium, such as a memory. In some embodiments, part or all of the software program can be loaded and / or installed via a memory and / or a communication interface. When the software program is loaded into the memory and executed by the processor, one or more steps in the method described above can be executed. Alternatively, in other embodiments, the processor can be configured to perform one of the above methods by any other appropriate means (for example, by means of firmware).

[0044] The device can be implemented using a bus architecture. The bus architecture can include any number of interconnecting buses and bridges, depending on the specific application and overall design constraints of the hardware. The bus connects various circuits including one or more processors, memories, and / or hardware modules together. The bus can also connect various other circuits such as peripherals, voltage regulators, power management circuits, external antennas, etc.

[0045] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0046] Embodiment 4: A comparative experiment was conducted between the model in Example 1 and the baseline model.

[0047] Experimental environment: Operating system: Windows 10; Development language: python, using PyTorch library; Development environment: PyCharm; CPU: i9; Memory: 32G; Graphics card: NVIDIA RTX2060.

[0048] Test set: contains 4 fields, namely traffic community ID, business district ID, OD flow (true value) and OD distance.

[0049] Baseline models: Gravity model, Poisson regression, and Geographic Context Multi-Task Embedding Learning (GMEL). The gravity model is the Huff model, which is a gravity model based on spatial interaction and is often used to predict the probability of customers visiting a certain business district. The attractiveness of the business district is represented by the average of the daily passenger flow, area, business diversity and number of parking lots. The resistance is represented by the average of the Euclidean distance of the OD pair and the distance to the nearest urban center. and Set to 1 and 2 respectively; The Poisson regression is a generalized linear model widely used in count data prediction. It is often used for the prediction of count data and can flexibly handle various types of feature variables. OD flow features, traffic area features and business district features will be directly used as input features for each OD pair. The validation set is merged into the training set to participate in the experiment together. The GMEL model uses a graph attention network (GAT) to generate embeddings for the supply and demand of the region respectively, and mostly adopts a multi-task learning framework to improve the prediction performance; the embedding dimension of the two GATs is set to 48, and the final OD flow regression is implemented using a fully connected layer; the GMEL-noMul variant is specifically used in this experiment, which does not use multi-task learning, mainly because in the context of business district OD flow prediction, the inflow and outflow flows of the region are unknown, thereby simplifying the complexity of the model.

[0050] Evaluation criteria: Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Common Part of Commuters (CPC).

[0051] Specific comparison results, such as Figure 4-6 As shown; among them, the RMSE value of the model of this scheme (CDFPM) is 1.2204, which is the lowest among all the compared models, and is 42.84% lower than that of GMEL-noMul, indicating that the model has high accuracy in prediction; the MAE value of the model of this scheme is 0.6913, which is also the lowest among all the compared models; the CPC value of the model of this scheme is 0.9490, which is the highest among all the compared models; in summary, the model of this scheme performs best among all the compared models.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting customer flow in a shopping district based on graph neural network, characterized in that: include: Step 1: Based on the traffic community set and the business district set, construct the feature space for business district passenger flow prediction, including OD flow features, business district features and traffic community features; Step 2: embed the traffic zone features into the graph neural network to obtain the traffic zone embedding vector; embed the business district features into the first multi-layer perceptron to obtain the business district embedding vector; connect the embedding vector with the corresponding OD flow features as the OD flow comprehensive representation vector; Step 3. Based on the OD flow comprehensive representation vector, construct a training set and a verification set; based on the training set, train the second multi-layer perceptron to predict the passenger flow between each traffic zone and each business district; based on the verification set, iteratively verify the second multi-layer perceptron until the performance of the verification set no longer improves; use the OD flow comprehensive representation vector with the best performance on the verification set as input and pass it into the fully connected neural network; based on external prediction needs, construct a test set, and predict the passenger flow of the business district through the fully connected neural network.

2. The method according to claim 1, characterized in that The OD flow characteristics include OD flow and OD distance; the OD flow is Traffic Zone To Business district The average number of passengers per hour; the OD distance is and The Euclidean distance between .

3. The method according to claim 1, characterized in that The business district characteristics include the average daily passenger flow of the business district, the land area of ​​the business district, the business diversity index of the business district and the number of parking lots; The average daily passenger flow of the business district is obtained by querying the user daily residence desensitized information table provided by the big data platform; The commercial area is collected from the building outline data provided by the navigation platform; The business district business diversity index is calculated based on the number of points of interest provided by the navigation platform using the Hill Numbers index formula. The specific formula is: ； in, It represents the diversity index of business district formats; Indicates the total number of business types in the business district; Indicates the first The proportion of the number of points of interest of a certain business type to the total number of points of interest in the business district; Represents a parameter, specifically the inverse of the Simpson index; The number of parking lots refers to the number of parking lots within a preset buffer zone of a business district.

4. The method according to claim 1, characterized in that The characteristics of the traffic community include the average housing price of the community, the average consumption amount of the community, the distance from the community to the urban center, the residential population of the community and the employed population of the community; The average housing price of the community is collected from the second-hand housing price data of the residential transaction platform and processed by the inverse distance weighted interpolation method; The average consumption amount of the community is collected from consumption statistics data provided by financial institutions and processed by inverse distance weighted interpolation method; The distance from the community to the urban center is the Euclidean distance from the center point of each traffic community polygon to the nearest urban center; The residential population of the community and the employed population of the community are obtained through query based on the user daily residence desensitized information table provided by the big data platform.

5. The method according to claim 1, characterized in that The specific methods of embedding traffic communities into graph neural networks include: The traffic zone features are respectively embedded through a first graph convolutional neural network module and a second graph convolutional neural network module set in parallel to obtain a first embedding vector and a second embedding vector respectively; The first graph convolutional neural network includes a neighborhood graph , the second graph convolutional neural network includes the consumption level similarity graph ,in, Represents the nodes of the graph, i.e., the traffic subdistrict; Indicates the number of nodes; Indicates the number of traffic zones, equal ; Represents the edges connecting nodes; An adjacency matrix representing weighted edges in a neighborhood graph; Represents the adjacency matrix of weighted edges in the consumption level similarity graph; The first embedding vector The specific expression is: ； in, Represents the dimension of the vector; express Matrix of order; The second embedding vector The specific expression is: ； Then and Fusion is performed to obtain the third embedding vector , the specific formula is: ； in, and are all trainable weight parameters and belong to .

6. The method according to claim 5, characterized in that The definition formula of the middle edge is: ; in, represents the trainable distance decay parameter of the neighborhood graph; Indicates Traffic Zone With Traffic Zone The Euclidean distance between The definition formula of the middle edge is: ； in, Trainable distance decay parameters representing the consumption level similarity graph; express consumption level; express consumption level; express and The similarity of consumption levels between them.

7. The method according to claim 1, characterized in that The second multi-layer perceptron is a four-layer structure with batch normalization, including an input layer, a first hidden layer, a second hidden layer and an output layer connected in sequence.

8. The method according to claim 1, characterized in that: The loss function of the second multi-layer perceptron is a mean square error function, and the specific formula is: ； in, Indicates the number of matches between transportation communities and business districts; Indicates Traffic Zone To Business district The actual calculated value of the average number of passengers per hour; Indicates Traffic Zone To Business district The predicted value of the average number of passengers per hour.

9. A shopping district passenger flow prediction system based on graph neural network, characterized in that: It includes a data receiving module, a data processing module and a result generating module; The data receiving module is used to receive a set of traffic areas, a set of business districts and external forecasting requirements; The data processing module includes a feature space unit, a feature embedding unit and a traffic prediction unit; The feature space unit constructs a feature space for predicting passenger flow in a business district based on a set of traffic districts and a set of business districts, including OD flow features, business district features, and traffic district features; The feature embedding unit embeds the traffic zone feature into the graph neural network to obtain the traffic zone embedding vector; embeds the business district feature into the first multi-layer perceptron to obtain the business district embedding vector; connects the embedding vector with the corresponding OD flow feature to obtain the OD flow comprehensive representation vector; The traffic prediction unit constructs a training set and a validation set based on the OD flow comprehensive representation vector; trains the second multi-layer perceptron based on the training set to predict the passenger flow between each traffic zone and each business district; it iteratively verifies the second multi-layer perceptron based on the validation set until the performance of the validation set is no longer improved; takes the best OD flow comprehensive representation vector on the validation set as input and passes it into the fully connected neural network; constructs a test set based on external prediction requirements, and predicts the passenger flow of the business district through the fully connected neural network; The result generation module is used to send the customer flow of the business district outward.

10. A device for predicting customer flow in a shopping district based on graph neural network, characterized in that: It includes a processor, a memory and a bus, wherein the memory stores instructions and data read by the processor, the processor is used to call the instructions and data in the memory to execute any method as claimed in claim 1-8, and the bus connects the functional components for transmitting information.

Images