Training method of spacing prediction model and spacing prediction method

Abstract

The invention provides a training method of a spacing prediction model and a spacing prediction method. The training method of the spacing prediction model comprises the steps of obtaining region features of a sample geographic region and spacing labels of target type entity places in the sample geographic region; and training an interval prediction model based on the region features of the sample geographic region and the interval label to obtain a trained interval prediction model, thereby predicting the interval of the entity place based on the trained interval prediction model. The model is adopted to predict the distance, so that the accuracy of determining the distance between the interest points is improved. When the physical place is the retail site, the situation that the space is too small, and consequently the deployment of the newly added retail site influences the sales volume of the existing retail site is effectively avoided; when the type of the entity place is the public service type, incomplete service coverage caused by too large spacing is effectively avoided, and high deployment cost of the entity place caused by too small spacing is avoided.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to a training method for a spacing prediction model and a spacing prediction method. Background Art

[0002] As cities and economies continue to develop, and considering the needs of commercial operations or urban management, relevant institutions or departments will build new physical places such as shops, hospitals, schools, parks, etc. in the real world.

[0003] For physical places related to commercial operations, before building a new physical place, it is often necessary to consider whether the newly added physical place will affect the operation of the existing physical place. Taking a certain brand's store as an example, before building a new store, it is necessary to consider whether the new store will affect the operation of the existing store of the brand. In addition, taking a physical place for public convenience, such as a public toilet, as an example, before building a new public toilet, it is necessary to consider whether the new public toilet can solve the problems that the existing public toilet cannot solve. If the newly built physical place affects the operation of the existing place or cannot solve the problems existing in the existing place, it means that the site selection of the newly built physical place is unreasonable. The inventors of this application found that one of the important bases for site selection is the distance between physical places. If the distance between physical places is unreasonable, the construction of a new physical place is prone to the aforementioned problems. Therefore, it is necessary to provide a technical solution that can objectively evaluate the distance between physical places and provide a basis for the site selection of physical places. Summary of the Invention

[0004] The present application provides a training method for a spacing prediction model and a spacing prediction method, which are used to evaluate the spacing between physical places of the same type and reduce the adverse effects caused by unreasonable spacing between new physical places and existing physical places.

[0005] In a first aspect, the present application provides a method for training a spacing prediction model, comprising:

[0006] Obtaining regional characteristics of a sample geographic area and distance labels of physical places of a target type within the sample geographic area;

[0007] Based on the regional characteristics of the sample geographical area and the distance labels, a distance prediction model is trained to obtain a trained distance prediction model.

[0008] In a second aspect, the present application provides a spacing prediction method, comprising:

[0009] Obtain regional characteristics of the target geographic area;

[0010] Inputting the regional features into a trained distance prediction model, and outputting a predicted distance between physical places of the target type within the target geographic area based on the distance prediction model;

[0011] Among them, the spacing prediction model is a model trained based on the method provided in the first aspect of this application.

[0012] In a third aspect, the present application provides a method for predicting the distance between retail locations, comprising:

[0013] The distance between sales locations to be assessed;

[0014] Obtaining regional characteristics of the geographical area where the newly added retail outlet is located;

[0015] Inputting the regional features into a trained spacing prediction model to obtain a predicted spacing output by the spacing prediction model;

[0016] Based on the comparison result of the to-be-evaluated distance and the predicted distance, generating and feeding back the distance evaluation result of the candidate address;

[0017] Among them, the spacing prediction model is a model trained based on the method provided in the first aspect of this application.

[0018] In a fourth aspect, the present application provides a training device for a spacing prediction model, comprising:

[0019] A sample data collection module is used to obtain regional characteristics of a sample geographical area and distance labels of physical places of target types within the sample geographical area;

[0020] The model training module is used to train a distance prediction model based on the regional characteristics of the sample geographical area and the distance label to obtain a trained distance prediction model.

[0021] In a fifth aspect, the present application provides a spacing prediction device, comprising:

[0022] A regional feature acquisition module is used to obtain regional features of a target geographical area;

[0023] a distance prediction module, configured to use the regional features and the trained distance prediction model to output a predicted distance between physical places of the target type within the target geographic area based on the distance prediction model;

[0024] Among them, the spacing prediction model is a model trained based on the method provided in the first aspect of this application.

[0025] In a sixth aspect, the present application provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the electronic device to perform the method provided in the first, second or third aspect of the present application.

[0026] In the seventh aspect, the present application provides a computer-readable storage medium, which stores computer-executable instructions. When the processor executes the computer-executable instructions, it implements the method provided in the first aspect, second aspect or third aspect of the present application.

[0027] In an eighth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method provided in the first, second or third aspect of the present application.

[0028] The training method of the spacing prediction model and the spacing prediction method provided in the present application implement a scheme for predicting the spacing between physical places of the same type based on the regional characteristics of the area where the physical place is located and a pre-trained spacing prediction model for physical place spacing prediction. In the model training stage, the spacing prediction model is trained through the regional characteristics of the sample geographical area and the spacing labels of the physical places of the target type in the sample geographical area, so that the spacing prediction model has the ability to predict the spacing between physical places of the target type in the corresponding area based on the regional characteristics. By utilizing the powerful analytical ability of the trained spacing prediction model to perform spacing prediction, the accuracy of physical place spacing prediction is improved. At the same time, feature extraction and analysis are performed based on the dimension of the area where the physical place is located, which improves the matching degree between the predicted spacing and the regional characteristics. Through the prediction of the spacing between physical places, a reliable data basis is provided for the layout of physical places in the region. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0030] Figure 1 A schematic diagram of a point-of-interest distance prediction process provided in an embodiment of the present application;

[0031] Figure 2 A flowchart of a method for training a spacing prediction model provided in an embodiment of the present application;

[0032] Figure 3 A flowchart of another method for training a spacing prediction model provided in an embodiment of the present application;

[0033] Figure 4For this application Figure 3 A schematic diagram of the region division result in the illustrated embodiment;

[0034] Figure 5 A schematic diagram of a Voronoi diagram provided in an embodiment of the present application;

[0035] Figure 6 A schematic diagram of the point of interest screening process provided in an embodiment of the present application;

[0036] Figure 7 A flowchart of another method for training a spacing prediction model provided in an embodiment of the present application;

[0037] Figure 8 A flow chart of a spacing prediction method provided in an embodiment of the present application;

[0038] Figure 9 A flowchart of another distance prediction method provided in an embodiment of the present application;

[0039] Figure 10 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0040] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0041] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0042] First, some of the terms involved in the embodiments of this application are explained:

[0043] Point of Interest (POI): It is the expression of a physical place in the real world in a geographic information system or electronic map. It can be a house, a store, a school, a bus stop, etc.

[0044] Distance between points of interest: refers to the distance between points of interest of the same type, which can be straight-line distance, route distance or other distances.

[0045] Voronoi Diagram: Also known as Thiessen polygons or Dirichlet diagrams, it is a spatial segmentation algorithm that evolved from a tree diagram and consists of continuous polygons formed by the perpendicular bisectors of the lines connecting two adjacent points. Voronoi diagrams have the following characteristics:

[0046] 1. Each V polygon has a generator (such as a point of interest in the embodiment of the present application);

[0047] 2. The distance from each point in the V polygon to the generator inside the polygon is shorter than the distance to other generators;

[0048] 3. Points on the boundary of the polygon V are equidistant from the generators inside the polygon.

[0049] It should be noted that the user information (including but not limited to user device information, user attribute information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0050] When new physical locations, such as parks, public restrooms, retail outlets, supermarkets, etc., are needed in an area, or when considering whether to add new physical locations in an area, the distance between the new physical locations and existing physical locations needs to be considered.

[0051] Taking retail locations as an example, as a region develops and consumer spending increases, new retail locations are needed. When adding new physical locations, such as retail locations, it is often necessary to consider the distance between the new location and existing locations of the same type to avoid impacting the sales of existing locations.

[0052] Taking public service physical venues as an example, in order to provide more comprehensive and convenient services to the region, new physical venues need to be added within the region. Taking into account the cost and comprehensiveness of the services, it is necessary to consider the distance between the new physical venues and existing physical venues of the same type to avoid waste of resources due to too close distances, and low regional coverage of public services due to too large distances.

[0053] In summary, when adding a new physical location or considering whether to add a new physical location within an area, the distance between the physical locations needs to be used as a reference.

[0054] Based on this, the present application provides a training method for a spacing prediction model and a spacing prediction method. The training method for the spacing prediction model provides a method for training the spacing prediction model through the regional characteristics of the sample geographical area and the spacing labels of the target type of physical places therein. The spacing prediction model trained based on the training method and the regional characteristics of the target geographical area is used to predict the spacing of the target type of physical places in the target geographical area. The model's powerful analytical capabilities and rich, multi-dimensional regional characteristics are used to perform spacing prediction, thereby improving the accuracy of physical place spacing prediction.

[0055] As mentioned above, in an electronic map or a geographic information system, the above-mentioned physical places may also be referred to as points of interest. Figure 1 A schematic diagram of a point of interest distance prediction process provided in an embodiment of the present application is shown as follows: Figure 1 As shown, when it is necessary to add a specified type of point of interest in the area, or to view the distance between the specified type of points of interest in the area, the user can select the area through the electronic map application of the user terminal and move the selected area to the map. Figure 1 The target area and the type of interest points, such as the target type, are sent to the server. The server extracts regional features of the target area, such as population characteristics, traffic volume, and distribution characteristics of interest points, from the regional features collected by the feature collection terminal. Based on the pre-trained spacing prediction model deployed on the server, the server analyzes the regional features of the target area to predict the spacing of interest points of the target type in the target area and returns the predicted spacing to the user terminal.

[0056] Figure 2 A flow chart of a training method for a spacing prediction model provided in an embodiment of the present application is provided. The training method can be executed by any electronic device with corresponding data processing capabilities, such as the above-mentioned server, which can be an object server, a cloud server, a high-performance computer, etc. Figure 2 As shown, the training method of the spacing prediction model includes the following steps:

[0057] Step S201 : obtaining regional features of a sample geographical area and distance labels of physical places of a target type within the sample geographical area.

[0058] The target type is the type of physical location corresponding to the distance prediction model, and can be any type of physical location. Sample geographic areas are preferably selected where physical locations of the target type are more mature, such as areas with a large number of physical locations of the target type.

[0059] The sample geographic area may be an area containing a large number of physical venues of the target type and having a high score of the physical venues of the target type. The score of the physical venue may be determined based on the customer flow, sales, rating, etc. of the physical venue.

[0060] For example, taking the retail outlets selling product A as an example, the sample area should meet the following requirements: the number of retail outlets selling product A included is at least a preset number, such as 10, 20, 50, 100 or other values, and the average sales of the retail outlets selling product A is at least a preset amount. The average sales can be daily average sales, monthly average sales or average sales for other periods, and the preset amount can be 10,000, 100,000, 500,000 or other amounts.

[0061] The boundaries of the sample geographical area may be natural boundaries. For example, the entire area may be divided based on one or more of a road network, a water system, a railway, etc., to obtain multiple sample geographical areas with natural boundaries.

[0062] Regional characteristics include one or more characteristics of population, physical location distribution, vehicle flow, passenger flow, etc. within the corresponding area, such as the sample geographical area.

[0063] Optionally, the regional characteristics include at least one of the following:

[0064] Characteristics of residents within the area, characteristics of mobile users within the area, passenger flow within the area, vehicle flow within the area, and the distribution characteristics of various types of physical locations within the area. Characteristics of residents and mobile users may include age, gender, occupation, income level, and consumption level. Residents within the area are defined as those whose residences are located within the area, while mobile users within the area may include those whose work addresses or rental housing are located within the area.

[0065] Passenger flow describes the number of people entering physical locations within an area during each period, such as the number of people entering physical locations of the target type. Vehicle flow describes the number of vehicles entering an area during each period, specifically the number of vehicles passing through the physical location or the number of vehicles entering and exiting the road where the physical location is located.

[0066] Passenger and vehicle flow can be collected at the hourly level, such as once per hour, or at other intervals. Statistics can also be collected based on different seasons, dates, and time periods. For example, hourly passenger / vehicle flow on weekdays, hourly passenger / vehicle flow on holidays, every 15 minutes of morning rush hour, and every 15 minutes of evening rush hour.

[0067] The various types of physical places can be of specified types, such as food, public services, travel, beauty, clothing, etc., or they can be associated types of the target type.

[0068] A correlation type table may be pre-established, in which various types of correlation types are recorded, so that the correlation type of the target type can be determined based on the correlation type table, thereby collecting the distribution characteristics of the entity locations of the correlation type of the target type. A type of correlation type includes the type itself.

[0069] Taking tobacco sales as an example, the associated types may include food, public service, and travel.

[0070] When there are multiple sample geographic areas, for each sample geographic area, the regional characteristics of the sample geographic area are collected, and based on the addresses or locations of the physical places of the target type in the sample geographic area, the actual distance between the physical places of the target type in the sample geographic area is calculated; based on the actual distance, the distance label of the physical places of the target type in the sample geographic area is obtained.

[0071] The actual distance between two physical places may be the straight-line distance between the two physical places, or the route distance between the two physical places, that is, the length of the route from one physical place to the other physical place.

[0072] The distance label may be the average, minimum, median, or other value of the actual distances between physical locations of the target type within the sample geographic area.

[0073] In some embodiments, the spacing labels may be manually marked.

[0074] In some embodiments, after obtaining the regional features corresponding to the target area, the regional features can be preprocessed, such as normalization or standardization, missing value supplementation, data cleaning, etc., so as to train the spacing prediction model based on the preprocessed regional features.

[0075] Step S202 : training a distance prediction model based on the regional characteristics of the sample geographical area and the distance label to obtain a trained distance prediction model.

[0076] For each sample geographic area, a distance prediction model is trained based on the regional characteristics and distance labels of the sample geographic area. Based on the deviation between the distances predicted by the distance prediction model and the distance labels, such as the loss value, the parameters of the distance prediction model are continuously adjusted until the training end conditions are met. The trained distance prediction model is then outputted to facilitate prediction of the distances between physical locations of the target type within the target geographic area based on the trained distance prediction model.

[0077] Exemplarily, the training end condition may be: the training duration or the number of training rounds reaches a corresponding limit, or the deviation between the predicted intervals and the interval labels for multiple consecutive times is less than a preset threshold.

[0078] The distance prediction model can be a regression model, a neural network model or other models.

[0079] When the spacing prediction model is a regression model, after the training end conditions are met, the trained spacing prediction model, that is, the posterior model, needs to be subjected to a Bayesian test. If the test passes, the trained spacing prediction model is output to perform spacing prediction using the trained spacing prediction model.

[0080] Bayesian testing, also known as Bayesian inference, is a process of analyzing a model based on Bayesian theorem and combining it with prior knowledge. If the posterior results output by the model are consistent with the prior knowledge, the model passes the Bayesian test.

[0081] In some embodiments, the spacing predicted by the spacing prediction model may be one or more spacings, or a spacing interval, such as [first spacing, second spacing].

[0082] In some embodiments, before step S202 , it is necessary to initialize the distance prediction model, specifically to initialize the parameters of the distance prediction model, so as to train the initialized distance prediction model based on the regional characteristics of the sample geographic area and the distance labels.

[0083] If the spacing prediction model is a regression model, the regional characteristics can be iteratively tested based on the stepwise regression algorithm. By removing or adding variables (regional characteristics), a combination of regional characteristics that significantly affects the spacing can be obtained. The regional characteristics in the combination are used as variables to construct a regression model, and the parameters of the regression model are initialized. After obtaining the regional characteristics of the sample geographic area under the combination and the spacing labels of the target type of physical places in the sample geographic area, the initialized regression model is trained based on the regional characteristics and spacing labels under the combination.

[0084] The training method for the spacing prediction model provided in this embodiment trains the spacing prediction model during the model training phase using the regional characteristics of a sample geographic area and the spacing labels of target physical locations within the sample geographic area. This enables the spacing prediction model to predict the spacing between target physical locations within the corresponding area based on the regional characteristics. By utilizing the powerful analytical capabilities of the trained spacing prediction model to perform spacing prediction, the accuracy of physical location spacing prediction is improved. At the same time, feature extraction and analysis based on the region where the physical location is located is performed, which improves the degree of matching between the predicted spacing and regional characteristics. The prediction of physical location spacing provides a reliable data basis for the layout of physical locations within the region.

[0085] Optional, Figure 3A flow chart of another method for training a distance prediction model provided in an embodiment of the present application. In this embodiment, the distance prediction model is a regression model and the sample geographical areas are multiple as an example. The model training method provided in this embodiment is Figure 2 On the basis of the illustrated embodiment, step S201 is further limited, and steps related to regional feature screening and spacing label t-test are added after step S201.

[0086] like Figure 3 As shown, the model training method may specifically include the following steps:

[0087] Step S301 : Divide a preset geographical area along a road network, a water system and / or a railway to obtain a plurality of sample geographical areas.

[0088] The preset geographical area may be any geographical area, such as a specified area.

[0089] The data of electronic maps can be used to divide the preset geographical area into multiple sample geographical areas with natural boundaries, along the road network, water system, railway and other features.

[0090] The preset geographical area may be a provincial area, a municipal area, or other larger areas, such as the North China area.

[0091] For example, Figure 4 For this application Figure 3 The schematic diagram of the region division result in the embodiment shown is as follows: Figure 4 As shown, for a preset geographical area, the preset geographical area is divided based on architectural features and natural features such as road networks, water washes, and railways contained in the preset geographical area in the electronic map, and the preset geographical area is divided into multiple sub-areas, such as sub-areas 401 to 411. The boundaries between the sub-areas are natural boundaries. Subsequently, one or more sub-areas are regarded as sample geographical areas, and regional features are collected to obtain a training set for training the spacing prediction model.

[0092] Step S302 : For each sample geographical area, a Voronoi diagram of the sample geographical area is constructed using the physical places of the target type in the sample geographical area as generators.

[0093] The Voronoi diagram of the sample geographical area is constructed using the physical places of each target type in the sample geographical area as generators. The algorithm for constructing the Voronoi diagram can be arbitrary and is not limited in this application.

[0094] Since the distance from a point in the Thiessen polygon in the Voronoi diagram to the generator (the physical location of the target type) within the Thiessen polygon is shorter than the distance between the point and other generators, the Thiessen polygon where the generator is located, that is, the polygon associated with the generator, can be used to represent the influence area of ​​the generator.

[0095] For example, Figure 5 This is a schematic diagram of a Voronoi diagram provided in an embodiment of the present application. The distribution of physical locations of target types within a sample geographic area is shown in FIG. Figure 5 As shown, the target type of physical places in the sample geographical area are used as generators, and the sample geographical area is divided based on the perpendicular bisectors of the lines connecting the adjacent generators. The Voronoi diagram is as follows: Figure 5 As shown, Figure 5 The Thiessen polygons in the Voronoi diagram shown include quadrilaterals, pentagons, and hexagons.

[0096] Step S303: Obtain the regional characteristics of the area covered by each Thiessen polygon included in the Voronoi graph in the sample geographical area.

[0097] Step S304: constructing the regional features of the sample geographic area based on the regional features of each Thiessen polygon included in the Voronoi graph.

[0098] For each Thiessen polygon (also known as V-polygon) in the Voronoi diagram or the polygon associated with each generator in the Voronoi diagram, the regional characteristics of the area covered by the polygon are collected. The regional characteristics of the sample geographical area include the regional characteristics of each Thiessen polygon in the Voronoi diagram, and may also include characteristics calculated based on the regional characteristics of multiple Thiessen polygons, such as the average value, cumulative value, maximum value, etc. of the same regional characteristics of multiple Thiessen polygons.

[0099] Since the areas of different Thiessen polygons in the Voronoi diagram are not the same, in order to improve the consistency of indicators in regional features and thus improve the accuracy of spacing prediction, it is necessary to normalize the regional features based on the area, such as converting the regional features into features under unit area, that is, features in the density dimension.

[0100] Optionally, constructing the regional features of the sample geographic area based on the regional features of each Thiessen polygon included in the Voronoi diagram includes:

[0101] Based on the area of ​​each Thiessen polygon included in the Voronoi diagram, the regional characteristics of each Thiessen polygon included in the Voronoi diagram are normalized; based on the regional characteristics of each Thiessen polygon included in the Voronoi diagram after normalization, the regional characteristics of the sample geographical area are obtained.

[0102] The normalized regional features of the Thiessen polygons included in the Voronoi diagram can be used as the regional features of the sample geographic area or a part of the regional features of the sample geographic area.

[0103] The regional characteristics of the area covered by each Thiessen polygon in the Voronoi diagram are counted. For each Thiessen polygon included in the Voronoi diagram, based on the area of ​​the area covered by the Thiessen polygon, the regional characteristics of the area covered by the Thiessen polygon are characterized as characteristics under unit area, that is, characteristics of density dimension.

[0104] Normalizing regional features based on the area covered by the Thiessen polygons can be understood as converting regional features into features per unit area, or density. For example, pedestrian flow within the Thiessen polygon coverage area can be converted to pedestrian flow per unit area by dividing by the area of ​​the Thiessen polygons. Similarly, vehicle flow within a Thiessen polygon coverage area for a given hour can be converted to vehicle flow per unit area by dividing by the area of ​​the Thiessen polygons.

[0105] Each Thiessen polygon in a Voronoi graph consists of multiple Voronoi edges, and each Thiessen polygon contains a generator. Traverse each Thiessen polygon in the Voronoi graph and, for each traversed Thiessen polygon, obtain the area of ​​the Thiessen polygon and the regional features of the area covered by the Thiessen polygon. For any dimension in the regional features, based on the area of ​​the Thiessen polygon, obtain the feature of that dimension per unit area, and obtain the normalized regional features of the area covered by the Thiessen polygon.

[0106] In order to improve the efficiency and quality of model training, at least one of the physical location, distance label and regional feature can also be filtered.

[0107] Before filtering the distance labels, you can also filter the physical places within the sample geographic area to improve the accuracy of the distance label setting.

[0108] By screening regional features, we can obtain regional features that have a more significant impact on the spacing of physical places, reduce the dimensions of regional features, thereby reducing the complexity of the regression model used and shortening the model training time.

[0109] For the screening of regional features, it can be determined whether the spacing predicted by the regression model of the spacing prediction model changes significantly after removing or adding a certain regional feature. If so, the sample feature is a significant sample feature and is retained; if not, the sample feature is deleted.

[0110] Physical venues of a target type within a sample geographic area may be filtered based on at least one attribute of the physical venue, such as area, registration or establishment time, customer flow, sales volume, etc.

[0111] Different types of physical venues use different attributes when screening. A correspondence between each type of physical venue and the attributes used in screening can be established in advance. Based on this correspondence, the attributes used when screening physical venues of the target type are determined. Then, based on the attributes used when screening physical venues of the target type, physical venues of the target type within the sample geographic area are screened.

[0112] For example, for physical venues of sales type, physical venues can be screened based on one or more parameters such as order volume, sales volume, area, etc.; for physical venues of public service type, physical venues can be screened based on one or more parameters such as passenger flow, area, satisfaction, etc.

[0113] Optionally, when the target type is a physical location for sales, the method further includes:

[0114] The physical places of the target type in the sample geographic area are screened based on at least one of the order volume, sales amount and area of ​​the physical places of the target type to obtain the actual distance between the screened physical places of the target type in the sample geographic area.

[0115] The strategies adopted for screening include, but are not limited to, deleting the physical place if one of the attributes is not within the preset range; calculating the score of the physical place by combining the values ​​of multiple attributes, and screening according to the score, such as deleting physical places with scores lower than the preset score, or retaining a certain number of physical places with higher scores.

[0116] For example, Figure 6 A schematic diagram of the point of interest screening process provided in an embodiment of the present application is shown in FIG. Figure 6 As shown, the target type is a sales type, such as tobacco retail outlets. For the full sample (sample refers to a physical location), that is, each physical location of the target type within the sample geographical area, based on parameters such as order volume, sales volume, and area, a part of the samples are filtered from the full sample (that is, other samples except high-quality samples) to obtain high-quality samples, that is, the physical locations of the target type that are retained.

[0117] Step S305 : For the physical places of the target type within the sample geographic area, determine the actual distance between the physical places based on the location of the physical place and the locations of other physical places of the target type within the sample geographic area.

[0118] The actual distance can be represented by the distance between the physical venue and the nearest physical venue of the same type within the same area (such as the sample geographic area or the subsequent target geographic area). It can also be represented by the average distance between the physical venue and multiple physical venues of the same type within the same area, such as the average distance between the physical venue and the three nearest physical venues of the same type.

[0119] The locations of physical places can be represented by two-dimensional or three-dimensional coordinates, and the distance between physical places can be represented by Euclidean distance, or by the route distance between points of interest corresponding to two physical places on an electronic map.

[0120] Step S306 : determining the distance labels of the physical places of the target type within the sample geographical area based on the actual distances between the physical places of the target type within the sample geographical area.

[0121] The interval label can be the average value, minimum value, median value or other statistical indicators of each actual interval in the sample geographic area.

[0122] After obtaining the spacing labels, in order to improve the accuracy of the spacing labels and thus improve the quality of model training, the spacing labels need to be tested. Statistical test methods such as t-test (abbreviated as t-test) and Anova (Analysis of Variance) test can be used to delete abnormal spacing labels.

[0123] Step S307 : performing a t-test on the distance labels of the physical places of the target type within multiple sample geographic areas, and screening the regional characteristics of the multiple sample geographic areas based on a stepwise regression algorithm.

[0124] An initial regression model can be constructed, including the previously filtered regional features. After filtering the regional features, the parts of the initial regression model corresponding to the filtered or deleted regional features are deleted, so that the regression model only uses the filtered regional features as variables, thus constructing a spacing prediction model (a type of regression model). The parameters of the constructed spacing prediction model are initialized.

[0125] Step S308 : Based on the distance labels that pass the t-test and the regional features of the sample geographical areas that are screened, the distance prediction model is trained to obtain a trained distance prediction model.

[0126] In this embodiment, the accuracy of regional division is improved by dividing the region by natural boundaries, and the physical place is prevented from being divided into two parts by the boundary of the region, thereby causing one physical place to appear in two regions; the regional features are constructed in combination with the Voronoi diagram, and the Thiessen polygons in the Voronoi diagram are used to approximate its internal generators, that is, the influence area of ​​the physical place, thereby improving the accuracy of regional feature construction; by screening the spacing labels, regional features, etc. corresponding to the points of interest and sample areas, the accuracy of spacing label setting is improved, the samples and labels on which the training model depends are optimized, and the efficiency and quality of model training are improved. At the same time, by screening the regional features, the input dimension of the model is reduced, the spacing prediction model is simplified, and the efficiency of model training is further improved and the cost of model training is reduced.

[0127] Figure 7 A flow chart of another method for training a spacing prediction model provided in an embodiment of the present application is provided. In this embodiment, the spacing prediction model is a GPR (Gaussian Process Regression) model as an example. Figure 7 As shown, the model training method provided in this embodiment mainly includes the following steps:

[0128] Data processing stage: The interval labels and regional characteristics of the collected sample geographic areas are subjected to t-test and stepwise test respectively to obtain sample data, i.e., training set, including the interval labels and regional characteristics of the screened sample geographic areas.

[0129] Model construction and training phase: Select the kernel function initialization function to initialize the GPR and obtain the GPR prior model; based on the sample data, perform model training and hyperparameter tuning on the GPR prior model to obtain the GPR posterior model; input the prediction set, that is, input the regional characteristics of the target geographic area of ​​the interval to be predicted into the GPR posterior model to obtain the output of the GPR posterior model, such as the predicted interval or interval.

[0130] Figure 8 This is a flow chart of a spacing prediction method provided in an embodiment of the present application. The method can be executed by an electronic device with corresponding data processing capabilities, such as the above-mentioned server, which can be a computer, a physical server, a cloud server, etc.

[0131] like Figure 8 As shown, the spacing prediction method includes the following steps:

[0132] Step S801: Acquire regional characteristics of a target geographical area.

[0133] The target geographical area may be any area, such as any administrative area, or an area selected or circled on an electronic map.

[0134] Regional features are used to characterize one or more characteristics of residents, mobile objects, distribution of various types of physical places, vehicle flow, passenger flow, etc. within a corresponding area (such as a target geographical area).

[0135] The regional characteristics of the target geographical area can be obtained by referring to the method of obtaining or collecting the regional characteristics of the sample geographical area provided in the above embodiment. The object is only replaced by "sample geographical area" with "target geographical area", which will not be repeated here.

[0136] In some embodiments, the feature collection terminal pre-collects regional features of the target geographical area under multiple physical place types. Then, based on the type of physical place of the distance to be predicted, that is, the target type, the features of the target geographical area under the target type can be obtained from the features describing the target geographical area, that is, the regional features of the target geographical area can be obtained.

[0137] In some embodiments, the type of regional characteristics of the target geographic area that needs to be obtained can be determined based on the target type; then, features corresponding to the target type are obtained from pre-collected features describing the target geographic area to obtain the regional characteristics of the target area.

[0138] During the training of the distance prediction model, the regional characteristics of the sample geographic area can be pre-tested and / or screened. Based on the results of the test and / or screening, regional characteristics used for model training are obtained. During the model utilization phase, regional characteristics are collected according to the regional characteristics used for model training during the training phase to obtain regional characteristics of the target area.

[0139] For example, before testing and / or screening, the regional features include F1 to F9. After testing and / or screening, F6 fails the test and F7 is deleted. Therefore, the regional features ultimately used for model training include F1 to F5, as well as F8 and F9. After model training is complete, for the target geographic area for which distances between physical venues of the same type need to be predicted, the values ​​of the features in the seven dimensions of F1 to F5, F8, and F9 are collected to obtain the regional features of the target geographic area.

[0140] Step S802: input the regional features into a trained distance prediction model, and output the predicted distances of physical places of the target type within the target geographic area based on the distance prediction model.

[0141] The spacing prediction model is a model trained based on the training method provided in the above embodiment.

[0142] After obtaining the regional characteristics of the target geographic area, these regional characteristics can be preprocessed, such as normalization or standardization, missing value filling, and data cleaning. The preprocessed regional characteristics are input into the trained distance prediction model, which analyzes the input data and outputs the predicted distance between physical places of the target type within the target geographic area.

[0143] In some embodiments, there may be multiple distance prediction models, one for each type of physical location. After obtaining the type of physical location for which distance prediction is to be made, such as the target type, and the regional characteristics of the target geographic area, the regional characteristics or pre-processed regional characteristics are input into a trained distance prediction model for the target type. The distance prediction model then outputs the predicted distance for physical locations of the target type within the target geographic area.

[0144] In some embodiments, the interval prediction model is a regression model, and the output predicted intervals are two, such as a first predicted interval and a second predicted interval, and the two predicted intervals constitute a interval interval.

[0145] Optionally, the spacing prediction model is a regression model, and the spacing prediction model is used to output a predicted spacing interval of physical places of the target type in the target area based on the input regional characteristics of the target geographical area.

[0146] The prediction interval of the regression model output can be expressed as: [prediction interval × (1-standard error), prediction interval × (1+standard error)].

[0147] Exemplarily, the regression model may be a Gaussian Process Regression model (GPR), a Ridge Regression model, a Lasso Regression model, or the like.

[0148] Since the regression model has a fast modeling speed and low training complexity, it can greatly reduce the cost of model-based physical venue spacing prediction and shorten the training cycle.

[0149] Furthermore, a distance assessment report for target type physical places can be generated based on the predicted distance of target type physical places within at least one target geographical area output by the distance prediction model and the actual distance of existing target type physical places within the at least one target geographical area.

[0150] Furthermore, for the scenario where a new point of interest is added in the target geographical area, the distance between the new point of interest and the existing points of interest of the same type can be obtained based on the address where the new point of interest is planned to be deployed, which is recorded as the deployment distance; based on the predicted distance output by the distance prediction model and the deployment distance, it is determined whether the address where the new point of interest is planned to be deployed meets the distance requirements, such as whether the deployment distance is greater than or equal to the predicted distance, or whether the deployment distance is within the interval formed by the predicted distances. If so, the distance requirements are met and the new point of interest can be deployed at the address where the new point of interest is planned to be deployed.

[0151] The spacing prediction method provided in the present application realizes the predicted spacing of target type physical places in the target geographic area based on the collected regional characteristics of the target geographic area and the pre-trained spacing prediction model, thereby assisting the site selection of new physical places based on the obtained predicted spacing, avoiding the new physical places from being too close or too far away from existing physical places of the same type, improving the accuracy of determining the spacing of new physical places, and reducing the adverse impact on existing points of interest while meeting user needs; through the spacing predicted based on the model, relevant personnel can also be assisted in judging whether the actual spacing between existing physical places in the target geographic area is appropriate, providing a reliable data basis for the layout of physical places in the area.

[0152] Figure 9 This is a flow chart of another distance prediction method provided in an embodiment of the present application. In this embodiment, the physical place is represented by a point of interest. This embodiment is aimed at the scene of adding a new point of interest. Figure 8 Based on the illustrated embodiment, step S801 is further refined, and after step S802, steps related to feeding back the distance assessment results of the newly added physical places are added.

[0153] like Figure 9 As shown, the distance prediction method provided in this embodiment may specifically include the following steps:

[0154] Step S901 : determining, based on candidate addresses of a newly added point of interest of a target type within a target geographical area, a distance to be evaluated between the newly added point of interest and existing points of interest of the target type within the target geographical area.

[0155] When a new POI of the target type exists within the target geographic area, a candidate address (i.e., the planned deployment address) for the new POI is obtained. Based on the candidate address of the new POI and the addresses of existing POIs of the same type within the target area, a distance to be assessed for the new POI is determined.

[0156] There may be one or more candidate addresses for a newly added POI. For each candidate address, the distance to be evaluated between the newly added POI at the candidate address and similar existing POIs within the target geographical area is calculated.

[0157] In some embodiments, the distance to be evaluated for a newly added point of interest may be the distance between the newly added point of interest and the existing point of interest that is closest to the newly added point of interest among the existing points of interest of the target type within the target geographical area.

[0158] Specifically, for points of interest of the same type as the newly added point of interest in the target geographical area, the distance between the point of interest and the newly added point of interest is calculated based on the address of the point of interest and the alternative address of the newly added point of interest; multiple smaller distances are selected from the distances between the points of interest of the same type and the newly added point of interest and are averaged to obtain the distance to be evaluated for the new point of interest.

[0159] Illustratively, the distance to be evaluated of the newly added interest point is the average of three smaller distances between the same type of interest points in the target area and the newly added interest point.

[0160] The distance between the points of interest can be represented by a straight-line distance, such as Euclidean distance, or can be represented by the length of a route from one point of interest to another point of interest in an electronic map.

[0161] Step S902 : constructing a Voronoi diagram of the target geographical area by using the points of interest of the target type in the target geographical area as generators.

[0162] The Voronoi diagram of the target geographical area is constructed using the points of interest of each target type in the target geographical area, specifically the existing points of interest of the target type, as generators. The algorithm for constructing the Voronoi diagram can be arbitrary and is not limited in this application.

[0163] Step S903 , obtaining the regional characteristics of the area covered by each Thiessen polygon included in the Voronoi graph in the target geographical area, and constructing the regional characteristics of the target geographical area based on the regional characteristics of each Thiessen polygon included in the Voronoi graph.

[0164] This step can be performed with reference to step S303 in the aforementioned embodiment, except that the sample geographical area is replaced with the target geographical area, and details will not be repeated here.

[0165] Step S904: input the regional features into a trained distance prediction model, and output a predicted distance between points of interest of the target type within the target geographic area based on the distance prediction model.

[0166] The regional features of the target geographic area, or the features obtained after preprocessing the regional features of the target geographic area, are input into the trained spacing prediction model, and the input features are analyzed by the spacing prediction model to obtain the predicted spacing of the target type of interest points in the target geographic area.

[0167] Step S905 : generating a distance evaluation result of the candidate address based on a comparison result between the to-be-evaluated distance and the predicted distance.

[0168] Compare the distance to be assessed of the newly added POI in the target geographic area with the predicted distance of the POI of the target type in the target geographic area output by the model. Based on the comparison results, such as whether the distance to be assessed is greater than or equal to the predicted distance, or whether the distance to be assessed is within the predicted distance range, generate and return the distance assessment result of the alternative address of the newly added POI to assist the user in deciding whether to deploy the new POI at the alternative address.

[0169] For example, if the distance prediction model outputs a predicted distance, if the distance to be evaluated is less than the predicted distance, the distance evaluation result of the newly added POI is unqualified, or the distance between the newly added POI and the existing POI is too close. If the distance to be evaluated is greater than or equal to the predicted distance, the distance evaluation result of the newly added POI is qualified, or the distance between the newly added POI and the existing POI meets the requirements.

[0170] For example, if the distance prediction model outputs a predicted distance interval (an interval formed by two predicted distances), if the distance to be evaluated exceeds the predicted distance interval, the distance evaluation result of the newly added POI is unqualified, or the distance between the newly added POI and the existing POI is too close or too far. If the distance to be evaluated is within the predicted distance, the distance evaluation result of the newly added POI is qualified, or the distance between the newly added POI and the existing POI meets the requirements.

[0171] When there are multiple alternative addresses, the distance evaluation results of each alternative address of the new point of interest can be generated based on the comparison results of the distance to be evaluated under each alternative address and the recommended distance, so that the user can select an address from the multiple alternative addresses as the deployment address of the new point of interest based on the distance evaluation results.

[0172] Furthermore, based on the distance evaluation results of each candidate address of the newly added point of interest, the candidate addresses may be screened to filter out candidate addresses with unqualified distance evaluation results, and the candidate addresses retained after screening may be returned to the user.

[0173] In some embodiments, a recommended deployment address for a newly added point of interest may be determined from the candidate addresses retained after screening, and the recommended deployment address may be returned to the user.

[0174] Illustratively, the recommended deployment address may be an alternative address with the smallest absolute value of the difference between the to-be-evaluated distance and the predicted distance among the alternative addresses with qualified distance evaluation results, or an alternative address with the smallest to-be-evaluated distance.

[0175] In this embodiment, the Voronoi diagram is introduced when constructing regional features, and the area affected by the point of interest is represented by the Voronoi diagram; the distance between points of interest is estimated based on the features corresponding to each polygon in the Voronoi diagram through a pre-trained model, so that the model can better learn the area affected by the point of interest in the Voronoi diagram, thereby improving the accuracy of determining the distance between points of interest; for newly added points of interest, the accuracy of the site selection of new points of interest is improved by comparing the actual distance corresponding to the alternative addresses with the recommended distance output by the model, and the rationality of the layout of points of interest in the region is improved.

[0176] The present application also provides a method for predicting the distance between retail locations, including:

[0177] Based on the alternative addresses of the newly added retail points, determine the distance to be evaluated between the newly added retail points and existing retail points of the same type in the same geographical area; obtain the regional characteristics of the geographical area where the newly added retail points are located; input the regional characteristics into a trained distance prediction model to obtain a predicted distance output by the distance prediction model; based on the comparison result of the distance to be evaluated and the predicted distance, generate and feedback the distance evaluation result of the alternative address.

[0178] The spacing prediction model is obtained by training based on the training method provided in any of the aforementioned embodiments.

[0179] It should be noted that, in order to improve efficiency, some steps mentioned in the aforementioned method embodiments of the present application can be executed in parallel, provided that they are logical; or, in accordance with logic, the execution order of the steps can be adjusted, such as switching parallel steps to serial execution, or adjusting the order in which the steps are executed. This application does not impose any restrictions on this.

[0180] The present application also provides a training device for a spacing prediction model, including:

[0181] A sample data acquisition module is used to obtain the regional characteristics of the sample geographical area and the distance labels of the target type of physical places in the sample geographical area; a model training module is used to train the distance prediction model based on the regional characteristics of the sample geographical area and the distance labels to obtain a trained distance prediction model.

[0182] Optional, sample data collection module, including:

[0183] A Voronoi diagram construction unit is used to construct a Voronoi diagram of a sample geographical area with the physical places of the target type in the sample geographical area as generators; a polygon feature acquisition unit is used to obtain the regional features of the area covered by each Thiessen polygon included in the Voronoi diagram in the sample geographical area; a regional feature construction unit is used to construct the regional features of the sample geographical area based on the regional features of each Thiessen polygon included in the Voronoi diagram; a spacing label acquisition unit is used to obtain the spacing labels of the physical places of the target type in the sample geographical area.

[0184] Optionally, a regional feature construction unit is used to:

[0185] Based on the area of ​​each Thiessen polygon included in the Voronoi diagram, the regional characteristics of each Thiessen polygon included in the Voronoi diagram are normalized; based on the regional characteristics of each Thiessen polygon included in the Voronoi diagram after normalization, the regional characteristics of the sample geographical area are obtained.

[0186] Optional, spacing label acquisition unit, specifically used for:

[0187] For physical places of the target type within the sample geographic area, determining an actual distance between the physical places based on a location of the physical place and locations of other physical places of the target type within the sample geographic area;

[0188] Based on the actual distances between physical places of the target type within the sample geographic area, distance labels of the physical places of the target type within the sample geographic area are determined.

[0189] Optionally, the target type is a physical location for sales, and the device further includes a sample screening module for:

[0190] The physical places of the target type in the sample geographic area are screened based on at least one of the order volume, sales amount and area of ​​the physical places of the target type to obtain the actual distance between the screened physical places of the target type in the sample geographic area.

[0191] Optionally, the spacing prediction model is a regression model, and the device further includes:

[0192] A t-test module is used to perform a t-test on the distance labels of the target type of physical places in the multiple sample geographic areas after obtaining the regional characteristics of the multiple sample geographic areas and the distance labels of the target type of physical places in the sample geographic areas, so as to train the distance prediction model based on the distance labels after passing the t-test; and / or a feature screening module is used to screen the regional characteristics of the multiple sample geographic areas based on a stepwise regression algorithm, so as to train the distance prediction model based on the screened regional characteristics.

[0193] The training device for the spacing prediction model provided in the embodiment of the present application can be used to execute the technical solution of the training method for the spacing prediction model provided in any of the above embodiments of the present application. Its implementation principle and technical effects are similar, and will not be repeated here in this embodiment.

[0194] The present application also provides a spacing prediction device, including:

[0195] A regional feature acquisition module is used to obtain regional features of a target geographic area; a spacing prediction module is used to combine the regional features and a trained spacing prediction model to output a predicted spacing of physical places of the target type within the target geographic area based on the spacing prediction model; wherein the spacing prediction model is a model trained based on the method provided in any embodiment of the present application.

[0196] Optionally, the spacing prediction device further includes a spacing evaluation result generating module, configured to:

[0197] Based on the alternative addresses of the newly added physical places of the target type within the target geographical area, determine the distance to be evaluated between the newly added physical places and the existing physical places of the target type within the target geographical area; and based on the comparison result of the distance to be evaluated and the predicted distance, generate the distance evaluation result of the alternative addresses.

[0198] The spacing prediction device provided in the embodiment of the present application can be used to execute the technical solution of the spacing prediction method provided in any of the above embodiments of the present application. Its implementation principle and technical effects are similar and will not be repeated here in this embodiment.

[0199] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 10 As shown, the electronic device of this embodiment may include: at least one processor 1001; and a memory 1002 communicatively connected to the at least one processor; wherein the memory 1002 stores instructions that can be executed by the at least one processor 1001, and the instructions are executed by the at least one processor 1001 to enable the electronic device to execute the method described in any of the above embodiments.

[0200] Optionally, the memory 1002 may be independent or integrated with the processor 1001 .

[0201] The implementation principle and technical effects of the electronic device provided in this embodiment can be found in the aforementioned embodiments and will not be described in detail here.

[0202] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the method described in any of the above embodiments is implemented.

[0203] An embodiment of the present application further provides a computer program product, including a computer program, which implements the method described in any of the aforementioned embodiments when executed by a processor.

[0204] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules is merely a logical function division. In actual implementation, other division methods may be used. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not implemented.

[0205] The above-mentioned integrated module implemented in the form of a software functional module can be stored in a computer-readable storage medium. The above-mentioned software functional module is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or a processor to perform some steps of the method described in each embodiment of the present application.

[0206] It should be understood that the above-mentioned processor can be a processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The memory may include RAM (Random Access Memory), and may also include NVM (Non-Volatile Memory), such as at least one disk storage, and can also be a USB flash drive, a mobile hard disk, a read-only memory, a disk or an optical disk, etc.

[0207] The storage medium may be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium may be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0208] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and the storage medium can be located in an application-specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0209] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0210] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0211] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0212] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A training method for a spacing prediction model, characterized in that: include: Obtaining regional characteristics of a sample geographic area and distance labels of physical places of a target type within the sample geographic area; Based on the regional characteristics of the sample geographical area and the distance labels, a distance prediction model is trained to obtain a trained distance prediction model.

2. The method according to claim 1, characterized in that Get regional characteristics of the sample geographic area, including: For a sample geographical area, constructing a Voronoi diagram of the sample geographical area using physical places of the target type within the sample geographical area as generators; Obtaining regional characteristics of the area covered by each Thiessen polygon included in the Voronoi diagram in the sample geographic area; Based on the regional feature of each Thiessen polygon included in the Voronoi graph, the regional feature of the sample geographic area is constructed.

3. The method according to claim 2, characterized in that Constructing the regional features of the sample geographic area based on the regional features of each Thiessen polygon included in the Voronoi diagram includes: Based on the area of ​​each Thiessen polygon included in the Voronoi diagram, normalizing the regional features of each Thiessen polygon included in the Voronoi diagram; Based on the normalized regional features of each Thiessen polygon included in the Voronoi diagram, the regional features of the sample geographic area are obtained.

4. The method according to any one of claims 1 to 3, characterized in that Obtaining distance labels of physical places of the target type within the sample geographic area, including: For physical places of the target type within the sample geographic area, determining an actual distance between the physical places based on a location of the physical place and locations of other physical places of the target type within the sample geographic area; Based on the actual distances between physical places of the target type within the sample geographic area, distance labels of the physical places of the target type within the sample geographic area are determined.

5. The method according to claim 4, characterized in that The target type is a physical location for sales, and the method further includes: The physical places of the target type in the sample geographic area are screened based on at least one of the order volume, sales amount and area of ​​the physical places of the target type to obtain the actual distance between the screened physical places of the target type in the sample geographic area.

6. The method according to any one of claims 1 to 3, characterized in that The distance prediction model is a regression model. After obtaining regional features of a plurality of sample geographic areas and distance labels of target-type physical places within the sample geographic areas, the method further includes: Performing a t-test on the distance labels of the target type of physical places within the multiple sample geographic areas, so as to train the distance prediction model based on the distance labels that pass the t-test; and / or, Based on a stepwise regression algorithm, the regional features of the plurality of sample geographic areas are screened, so as to train the spacing prediction model based on the screened regional features.

7. The method according to any one of claims 1 to 3, characterized in that The regional characteristics include at least one of the following: characteristics of residents in the corresponding area, characteristics of mobile objects in the corresponding area, passenger flow in the corresponding area, vehicle flow in the corresponding area, and distribution characteristics of various types of physical places in the corresponding area.

8. A spacing prediction method, characterized in that: include: Obtain regional characteristics of the target geographic area; Inputting the regional features into a trained distance prediction model, and outputting a predicted distance between physical places of the target type within the target geographic area based on the distance prediction model; The spacing prediction model is a model trained based on the method provided in any one of claims 1 to 7.

9. The method according to claim 8, characterized in that The method further comprises: Determining, based on candidate addresses of a newly added physical venue of the target type within the target geographic area, a distance to be assessed between the newly added physical venue and existing physical venues of the target type within the target geographic area; Based on the comparison result of the to-be-evaluated distance and the predicted distance, a distance evaluation result of the candidate address is generated.

10. A method for predicting the distance between retail locations, characterized in that: include: Determining, based on candidate addresses for the newly added retail location, the distance to be assessed between the newly added retail location and existing retail locations of the same type within the same geographic area; Obtaining regional characteristics of the geographical area where the newly added retail outlet is located; Inputting the regional features into a trained spacing prediction model to obtain a predicted spacing output by the spacing prediction model; Based on the comparison result of the to-be-evaluated distance and the predicted distance, generating and feeding back the distance evaluation result of the candidate address; The spacing prediction model is a model trained based on the method provided in any one of claims 1 to 7.

Images