A shop block site selection method based on space-time big data and machine learning algorithm

Abstract

This invention discloses a method for selecting store district locations based on spatiotemporal big data and machine learning algorithms, belonging to the field of commercial site selection technology. The method includes: calculating commercial site selection characteristic indicators based on a multi-dimensional intelligent grid data layer and a spatiotemporal business element set, and using machine learning weighting to generate a multi-dimensional indicator dataset; based on the multi-dimensional indicator dataset, performing temporal correlation and spatial dependency fusion on the commercial site selection characteristic indicators to generate a site selection potential map; based on the site selection potential map, using machine learning algorithms to predict and dynamically optimize the temporal evolution of the site selection potential at different locations, generating a site selection prediction report; and performing multi-scenario simulation and risk assessment on the site selection prediction report to generate actual deployment site selection and operational configuration information. This invention improves the reliability and feasibility of site selection decisions in actual operating environments by performing multi-scenario simulation and risk assessment on the site selection prediction information.

Description

Technical Field

[0001] This invention relates to the field of commercial site selection technology, and in particular to a method for selecting store blocks based on spatiotemporal big data and machine learning algorithms. Background Technology

[0002] In recent years, with the widespread availability of remote sensing imagery, laser point cloud data, and internet map data, as well as the continuous accumulation of spatiotemporal business elements such as mobile communication data, transportation data, and environmental monitoring data, commercial site selection analysis based on spatiotemporal big data has gradually become a research hotspot. Meanwhile, the application of machine learning methods in pattern recognition and predictive analytics has deepened, enhancing the ability to model complex urban commercial areas and providing a more refined and dynamic technical foundation for shopping district site selection.

[0003] However, existing store location selection technologies based on spatiotemporal data and machine learning still have shortcomings. On the one hand, multi-source geographic data and spatiotemporal business elements are highly heterogeneous in terms of spatial reference system, resolution, and temporal structure, making it difficult to maintain data consistency and temporal continuity under a unified spatial benchmark. On the other hand, most existing location selection analyses remain at the level of static indicator evaluation and single-moment prediction, lacking the ability to predict the changing trends of the location potential of different spatial locations over time, making it difficult to provide reliable dynamic decision support for the actual deployment phase. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for selecting store blocks based on spatiotemporal big data and machine learning algorithms to solve the problems of difficulty in unifying and integrating multi-source spatiotemporal data and lack of temporal evolution prediction of site selection potential.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a method for selecting store district locations based on spatiotemporal big data and machine learning algorithms. The method includes: collecting multi-source geographic data and a spatiotemporal business element set, and generating a spatiotemporal baseline geographic canvas through coordinate unification and geometric registration; performing adaptive grid partitioning and spatiotemporal encoding on the spatiotemporal baseline geographic canvas to obtain a multi-dimensional intelligent grid data layer; calculating commercial location feature indicators based on the multi-dimensional intelligent grid data layer and the spatiotemporal business element set, and using machine learning weighting to generate a multi-dimensional indicator dataset; based on the multi-dimensional indicator dataset, performing temporal correlation and spatial dependency fusion on the commercial location feature indicators to generate a location potential map; based on the location potential map, using machine learning algorithms to predict and dynamically optimize the temporal evolution of location potential at different locations, generating a location prediction report; and performing multi-scenario simulation and risk assessment on the location prediction report to generate actual deployment location and operational configuration information.

[0008] As a preferred embodiment of the store street location selection method based on spatiotemporal big data and machine learning algorithm described in this invention, the multi-source geographic data includes publicly available Internet map tiles, remote sensing images, and laser point clouds.

[0009] The spatiotemporal business element set includes pedestrian behavior data, commercial facility business data, transportation data, and environmental and meteorological monitoring data.

[0010] As a preferred embodiment of the store street location selection method based on spatiotemporal big data and machine learning algorithms described in this invention, the generation of the spatiotemporal reference geographic canvas refers to mapping multi-source geographic data and spatiotemporal business element sets to the same standard spatial reference system, and using a deep learning geometric registration algorithm for geometric registration.

[0011] As a preferred embodiment of the store district site selection method based on spatiotemporal big data and machine learning algorithms described in this invention, the specific steps of performing adaptive grid partitioning and spatiotemporal encoding on the spatiotemporal reference geographic canvas are as follows.

[0012] Based on the spatiotemporal reference geographic canvas, an adaptive grid partitioning algorithm is used to dynamically allocate grid resolution to different spatial regions and perform spatial subdivision to generate a spatial location grid.

[0013] The GeoHash encoding system is used to perform spatiotemporal encoding on the spatial location grid to obtain the spatiotemporal encoding identifier.

[0014] As a preferred embodiment of the store street location selection method based on spatiotemporal big data and machine learning algorithms described in this invention, the acquisition of multi-dimensional intelligent grid data layer refers to the structural aggregation and temporal consistency organization of spatial location grids based on spatiotemporal coding identifiers.

[0015] As a preferred embodiment of the shop block location selection method based on spatiotemporal big data and machine learning algorithms described in this invention, the specific steps for calculating commercial location feature indicators based on a multi-dimensional intelligent grid data layer and a spatiotemporal business element set are as follows:

[0016] Based on the multi-dimensional intelligent grid data layer, the spatiotemporal business element set is mapped to the spatial location grid according to the spatiotemporal coding identifier, generating a business element mapping set.

[0017] Perform spatiotemporal big data analysis on the business element mapping set and calculate business site selection characteristic indicators.

[0018] As a preferred embodiment of the shop street location selection method based on spatiotemporal big data and machine learning algorithms described in this invention, the generation of multidimensional index dataset refers to adaptively adjusting the commercial location feature indicators using a machine learning weighting method.

[0019] As a preferred embodiment of the shop street location selection method based on spatiotemporal big data and machine learning algorithms described in this invention, the step of generating a location potential map by fusing temporal correlation and spatial dependence of commercial location feature indicators based on a multi-dimensional indicator dataset is as follows.

[0020] Based on a multidimensional indicator dataset, the characteristics of commercial site selection are sorted over time, and the changes in indicators between adjacent time points are recorded.

[0021] Extract spatial location information from the multidimensional indicator dataset and summarize the changes in indicators between adjacent time points to generate spatial summary indicator values.

[0022] By mapping the spatial aggregated index values ​​of different spatial locations into a continuous spatial distribution, a site selection potential map is generated.

[0023] As a preferred embodiment of the shop block site selection method based on spatiotemporal big data and machine learning algorithms described in this invention, the specific steps of predicting and dynamically optimizing the temporal evolution of the site selection potential of different locations using machine learning algorithms based on the site selection potential map, and generating a site selection prediction report, are as follows.

[0024] Based on the site selection potential map, the potential distribution corresponding to different spatial locations is sorted over time to construct a site selection potential learning sequence.

[0025] Machine learning algorithms are used to simulate the evolution paths of different business districts and to predict the learning sequence of site selection potential, thereby generating potential evolution prediction information.

[0026] Based on potential evolution prediction information, the machine learning algorithm is dynamically optimized, and the optimized potential evolution prediction information is summarized and organized to generate a site selection prediction report.

[0027] As a preferred embodiment of the store street location selection method based on spatiotemporal big data and machine learning algorithms described in this invention, the specific steps for performing multi-scenario simulations and risk assessments on the location prediction report to generate actual deployment location and operational configuration information are as follows.

[0028] Based on the site selection prediction report, the potential evolution prediction information is mapped into simulable objects, and a virtual reality simulation environment is constructed.

[0029] The performance of the site selection prediction report is simulated in a virtual reality simulation environment using the digital twin method, and the performance data is recorded.

[0030] Based on operational performance data, the probability distribution of return on investment is calculated using the Monte Carlo method, and risk constraints are applied to generate actual deployment site selection and operational configuration information.

[0031] The beneficial effects of this invention are as follows: by predicting and dynamically optimizing the temporal evolution of site selection potential, the trend of site selection conditions changing over time in different spatial locations is characterized; by conducting multi-scenario simulation and risk assessment of site selection prediction information, the reliability and feasibility of machine learning-based site selection decisions in actual operating environments are improved. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart of a method for selecting store district locations based on spatiotemporal big data and machine learning algorithms.

[0034] Figure 2 A flowchart for generating a spatiotemporal reference geographic canvas.

[0035] Figure 3 A flowchart for obtaining multidimensional smart grid data layers.

[0036] Figure 4 This is a flowchart for scenario simulation evaluation.

[0037] Figure 5 This is a schematic diagram illustrating the evolution of hit rate over time.

[0038] Figure 6 This is a schematic diagram of the time evolution of absolute error. Detailed Implementation

[0039] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0040] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0041] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0042] Reference Figures 1-6 This is one embodiment of the present invention, which provides a method for selecting store district locations based on spatiotemporal big data and machine learning algorithms, including the following steps:

[0043] S1: Collect multi-source geographic data and spatiotemporal business element sets, and generate a spatiotemporal benchmark geographic canvas through coordinate unification and geometric registration.

[0044] S1.1: Multi-source geographic data includes publicly available internet map tiles, remote sensing imagery, and laser point clouds.

[0045] Specifically, geographic image data is collected by spatial information sensing sources that continuously observe roads, buildings and landforms in the urban environment. The collected geographic image data is then organized in the form of map tiles. For example, continuous geographic images covering the same urban area are divided into image blocks with a side length of 256×256 pixels, and each image block is assigned a unique spatial location identifier. At the same time, the spatial location relationship and collection time information corresponding to each map tile are recorded to obtain publicly available map tiles on the Internet.

[0046] Earth observation sensors are used to scan the surface area and collect remote sensing image data that reflects the surface morphology and the distribution of land features. During the acquisition process, the spatial coverage, imaging time and image resolution information of the remote sensing images are retained to obtain remote sensing images.

[0047] A laser point cloud is obtained by scanning space using a laser rangefinder sensor to collect point cloud data that reflects the three-dimensional structural features of the area. During the acquisition process, the spatial coordinates, sampling time, and reflection intensity information contained in the point cloud data are recorded.

[0048] S1.2: The spatiotemporal business element set includes pedestrian behavior data, commercial facility business data, transportation data, and environmental and meteorological monitoring data.

[0049] Specifically, spatial behavior sensors are used to sense the activity status of people, collect data on the location and movement of people in urban public spaces, record the dwell time and movement path of people in different spatial locations, and attach corresponding time stamps and spatial location identifiers to the collected records to obtain pedestrian behavior data.

[0050] Spatial recognition sensors are used to collect data on the distribution of shops and changes in business formats within urban commercial areas. The spatial location, business type, and update status of each commercial facility are recorded, and the collected results are organized and compiled to obtain commercial facility business format data.

[0051] Traffic status sensors are used to collect data on road conditions and public transportation operations, recording vehicle traffic status, traffic changes, and time and spatial location information corresponding to travel behavior to obtain traffic travel data.

[0052] The system collects data on temperature, humidity, and air quality in the urban environment using environmental monitoring sensors. It records the observation location and time corresponding to each environmental condition, and then organizes and summarizes the collected results to obtain environmental meteorological monitoring data.

[0053] S1.3: Map multi-source geographic data and spatiotemporal business element sets to the same standard spatial reference system, and use a deep learning geometric registration algorithm to perform geometric registration to generate a spatiotemporal benchmark geographic canvas.

[0054] Specifically, the spatial location identifiers contained in multi-source geographic data (including publicly available internet map tiles, remote sensing images, and laser point clouds) and spatiotemporal business element sets (including pedestrian behavior data, commercial facility business data, transportation data, and environmental and meteorological monitoring data) are organized, and a standard spatial reference system is selected as a unified mapping benchmark.

[0055] The values ​​in the spatial location identifier are converted according to a unified mapping reference. For example, when the spatial location value recorded in the spatial location identifier is (120.5000, 30.2500), the corresponding value is converted according to the unified mapping reference to obtain the spatial location value under the unified mapping reference as (120500, 30250), so that the converted spatial location identifier can be consistently represented under the same standard spatial reference system.

[0056] Using the spatial location identifiers contained in publicly available internet map tiles as the basis for spatial alignment, the spatial extent of remote sensing imagery and laser point clouds is correlated under the same standard spatial reference system to form registration data for geometric registration. Based on the registration data, a deep learning geometric registration algorithm is used to calculate the spatial transformation index of remote sensing imagery and laser point clouds relative to publicly available internet map tiles, expressed as: ;

[0057] in, For spatial transformation index, To identify the quantity of spatial locations, For spatial location identification indexing of remote sensing images, For the first in remote sensing image Spatial location coordinates, These are the spatial coordinates of the locations in publicly available internet map tiles that correspond to remotely sensed images. For spatial location identification indexing of laser point clouds, The number of sampling points in the laser point cloud. For the first laser point cloud Spatial coordinates of each sampling point These are the spatial coordinates of the laser point cloud corresponding to the publicly available map tiles on the Internet.

[0058] It should be noted that the expression for calculating the spatial transformation index of remote sensing images and laser point clouds relative to publicly available Internet map tiles uses spatial location offset as the basic metric. After spatial transformation mapping and numerical norm calculation, the overall dimension is length.

[0059] Using spatial transformation indicators as the alignment basis, a deep learning geometric registration algorithm is employed to match and train the spatial distribution features reflecting roads, buildings, and landforms in publicly available internet map tiles, remote sensing images, and laser point clouds. This learns the spatial correspondence between different data sources and continuously adjusts the feature matching results during the training process, so that the positional distribution of roads, buildings, and landforms in remote sensing images gradually converges to the corresponding positions in publicly available internet map tiles.

[0060] Based on the spatial transformation index obtained from training, the positions of each sampling point in the laser point cloud are adjusted as a whole to ensure that the spatial distribution of the laser point cloud reflecting the three-dimensional structural features of the region corresponds to the location of ground features in the publicly available Internet map tiles. Data on pedestrian behavior, business facility formats, transportation, and environmental meteorological monitoring are registered to the same spatial location according to the corresponding spatial location identifiers and time markers, thereby generating a spatiotemporal reference geographic canvas.

[0061] It should be noted that the deep learning geometric registration algorithm obtains spatial transformation indicators between different data by performing multi-layer feature extraction and matching on the spatial structure features in the input data; it can complete geometric alignment calculation based on the data's own features in complex scenarios, and is suitable for geometric registration processing between multi-source spatial data such as remote sensing images and laser point clouds.

[0062] S2: Perform adaptive grid subdivision and spatiotemporal coding on the spatiotemporal reference geographic canvas to obtain a multidimensional intelligent grid data layer.

[0063] S2.1: Based on the spatiotemporal reference geographic canvas, the adaptive grid partitioning algorithm dynamically allocates grid resolution to different spatial regions and performs spatial subdivision to generate a spatial location grid.

[0064] Specifically, the overall spatial range covered by the spatiotemporal reference geographic canvas is expanded according to the spatial coordinate arrangement order, and the spatial range is continuously divided along the spatial coordinate arrangement direction, decomposing the overall spatial range into multiple adjacent spatial segments. An adaptive grid partitioning algorithm is used to assign different grid resolutions to the spatial segments according to their position in the arrangement order. For example, the spatial segment numbered 1 in the arrangement order is divided into a grid with a side length of 10 meters, the spatial segment numbered 2 in the arrangement order is divided into a grid with a side length of 20 meters, and the spatial segment numbered 3 in the arrangement order is divided into a grid with a side length of 10 meters again.

[0065] Based on the grid resolution corresponding to each spatial segment, the spatiotemporal reference geographic canvas is spatially partitioned; the range of spatial location identifier values ​​corresponding to each spatial location grid is used as the coverage area of ​​the spatial location grid to generate the spatial location grid.

[0066] It should be noted that the adaptive grid partitioning algorithm refers to segmenting the spatial range according to the arrangement and distribution of spatial location markers within the spatial range, and using different grid partitioning scales in different spatial segments to discretize the continuous space into a set of ordered spatial location grids, so that the generated spatial location grids can simultaneously meet the requirements of spatial coverage integrity and subsequent spatiotemporal calculation organization.

[0067] S2.2: Use the GeoHash encoding system to perform spatiotemporal encoding on the spatial location grid and obtain the spatiotemporal encoding identifier.

[0068] Specifically, the spatial location range corresponding to each spatial location grid is read one by one and processed according to the encoding order of the GeoHash encoding system. By performing bit-by-bit splitting and character mapping on the spatial location values, the spatial location values ​​are converted into an encoding string composed of letters and numbers, so that different spatial locations correspond to different GeoHash encoding results.

[0069] For example, when the spatial location value corresponding to the spatial location grid is (120500, 30250), according to the processing order of the GeoHash encoding system, the spatial location value is mapped to the character encoding string: wtw3s, which is used to represent the position of the corresponding spatial location in the encoding space.

[0070] Read the time stamp corresponding to the spatial location grid and append the time stamp to the end of the GeoHash encoding result in sequence so that the encoding result contains both spatial location code and time stamp; combine the spatial location code and time stamp to obtain the spatiotemporal code identifier of each spatial location grid.

[0071] It should be noted that the GeoHash encoding system is a method that converts continuous geospatial locations into discrete character encodings through recursive partitioning, enabling spatial locations to be expressed in string form, which facilitates fast indexing, matching, and grouping of spatial locations.

[0072] S2.3: Based on spatiotemporal coding identifiers, the spatial location grid is structurally aggregated and time-series consistent to obtain a multi-dimensional intelligent grid data layer.

[0073] Specifically, the spatiotemporal coding identifier is used as the aggregation basis. The spatial location grid is read line by line and classified according to the spatiotemporal coding identifier. Spatial location grids with the same spatiotemporal coding identifier are grouped into the same aggregated grid.

[0074] The spatial location grids within each aggregated grid are arranged sequentially according to the corresponding time markers, and the arrangement results are checked item by item to keep the spatial location grids under different time markers of the same spatial location in a continuous arrangement. The spatial location grids arranged in chronological order within the same aggregated grid are uniformly organized to form a stable temporal structure expression of the spatial location grids formed under continuous time markers of the same spatial location.

[0075] By processing the spatial location grid in a temporal order, a multi-dimensional intelligent grid data layer (including spatial location dimension and time dimension) is obtained.

[0076] It should be noted that the multidimensional intelligent grid data layer refers to a data set formed by structuring spatial regions based on a spatiotemporal benchmark geographic canvas through spatial location grids and spatiotemporal coding identifiers. Within the same spatial location grid, the multidimensional intelligent grid data layer integrates spatial location identifiers with pedestrian behavior data, commercial facility type data, transportation data, and environmental and meteorological monitoring data according to time stamps. This allows the spatial and temporal dimensions to be expressed collaboratively under a unified coding system, providing a directly computable data structure for subsequent site selection analysis.

[0077] S3: Calculate commercial site selection characteristic indicators based on the multidimensional intelligent grid data layer and the spatiotemporal business element set, and use machine learning to weight them to generate a multidimensional indicator dataset.

[0078] S3.1: Based on the multi-dimensional intelligent grid data layer, the spatiotemporal business element set is mapped to the spatial location grid according to the spatiotemporal coding identifier, generating a business element mapping set.

[0079] Specifically, the spatial location grid information and corresponding spatiotemporal coding identifiers recorded in the multidimensional intelligent grid data layer are read one by one; based on the spatiotemporal coding identifiers, the data on pedestrian behavior, business facility formats, transportation and travel, and environmental meteorological monitoring data recorded in the spatiotemporal business elements are extracted one by one, along with their corresponding spatial location identifiers and time markers.

[0080] The spatial location identifier in the spatiotemporal business element record is compared with the spatial range covered by the spatial location grid, and the time mark in the spatiotemporal business element record is aligned with the time mark corresponding to the spatial location grid. When the spatial location identifier of the spatiotemporal business element record falls within the coverage area of ​​the spatial location grid and the time mark is consistent with the time mark corresponding to the spatial location grid, the corresponding spatiotemporal business element record is registered under the corresponding spatiotemporal code identifier name.

[0081] By comparing and registering each item of the multidimensional intelligent grid data layer and the spatiotemporal business element set, a business element mapping set is generated under each spatiotemporal code identifier, which includes data on pedestrian behavior, business facility formats, transportation data, and environmental meteorological monitoring data.

[0082] S3.2: Perform spatiotemporal big data analysis on the business element mapping set and calculate business site selection characteristic indicators.

[0083] Specifically, after generating the business element mapping set, the set is expanded item by item according to the spatiotemporal coding identifier, and pedestrian behavior data, commercial facility business data, transportation data, and environmental meteorological monitoring data under the same spatiotemporal coding identifier are read separately. Based on the time stamp corresponding to the spatiotemporal coding identifier, the pedestrian behavior data, commercial facility business data, transportation data, and environmental meteorological monitoring data are time-aligned, forming a comparable data set under the same time stamp for data from different sources.

[0084] Based on spatiotemporal coding identifiers, pedestrian traffic data is aggregated in terms of quantity, commercial facility business type data is aggregated in terms of business type distribution, traffic travel data is aggregated in terms of travel status, and environmental meteorological monitoring data is aggregated in terms of environmental status. This generates a comprehensive status value for the spatial location grid under the corresponding time marker, and calculates commercial site selection characteristic indicators, expressed as follows: ;

[0085] in, As a characteristic indicator for commercial site selection, The total number of time stamps, For time stamp index value, For time stamps The overall state value of the spatial location grid below, This represents the magnitude of the combined state value change between adjacent time and spatial location grids.

[0086] It should be noted that the expression for calculating the commercial site selection characteristic index is constrained by introducing the cumulative change over adjacent time periods on the basis of the time average state value, so that the commercial site selection characteristic index reflects both the overall level of commercial activities and the stability of time changes, and the parameter dimension is dimensionless.

[0087] S3.3: Adaptively adjust the business site selection feature indicators using machine learning weighting methods to generate a multi-dimensional indicator dataset.

[0088] Specifically, the commercial site selection characteristic indicators are organized chronologically to form a sequence of commercial site selection characteristic indicators arranged continuously over time. The organized sequence of commercial site selection characteristic indicators is then used as the processing object of the machine learning weighting method, and the values ​​of each commercial site selection characteristic indicator are proportionally converted. For example, the pedestrian flow intensity indicator, which ranges from 0 to 500, is proportionally converted to a value range of 0 to 1, and the business density indicator, which ranges from 0 to 50, is converted to a value range of 0 to 1.

[0089] Within a continuous time scale, the changes in the values ​​of commercial site selection characteristic indicators between adjacent time points are statistically analyzed. When the changes in commercial site selection characteristic indicators remain unchanged under continuous time scales, the numerical contribution of the corresponding commercial site selection characteristic indicators in the machine learning weighting method remains unchanged. When the changes in commercial site selection characteristic indicators decrease under continuous time scales, the numerical contribution of the corresponding commercial site selection characteristic indicators in the machine learning weighting method is reduced accordingly.

[0090] The adaptively adjusted commercial site selection feature indicators are integrated with the corresponding spatiotemporal coding identifiers and time stamps to generate a multidimensional indicator dataset.

[0091] It should be noted that the machine learning weighting method is a method that uses machine learning algorithms to automatically adjust the role of multiple analytical indicators in the calculation process. By analyzing the changes and stability of commercial site selection characteristic indicators in the calculation results, it dynamically determines the influence ratio of commercial site selection characteristic indicators in the comprehensive calculation, reduces the bias caused by manually setting weights, and enables the comprehensive analysis results to more accurately reflect the changing characteristics of the data itself.

[0092] S4: Based on a multidimensional indicator dataset, the temporal correlation and spatial dependence of commercial site selection characteristic indicators are fused to generate a site selection potential map.

[0093] S4.1: Based on a multidimensional indicator dataset, sort the commercial site selection characteristic indicators by time and record the changes in indicators between adjacent time points.

[0094] Specifically, based on the multidimensional indicator dataset, the spatiotemporal coding identifiers, time stamps, and corresponding business location feature indicators recorded in the multidimensional indicator dataset are extracted, and the business location feature indicators corresponding to the same spatiotemporal coding identifier are arranged and organized according to the order of the time stamps.

[0095] The difference between the commercial site selection characteristic indicators corresponding to adjacent time markers is statistically analyzed to obtain the change in indicators between adjacent time points in the same spatiotemporal code identifier (such as the numerical difference formed when the commercial site selection characteristic indicator corresponding to the same spatial location changes from 0.62 to 0.68 under two adjacent time markers). The change in indicators is recorded together with the corresponding time markers and commercial site selection characteristic indicators to form a time-related record reflecting the change of commercial site selection characteristic indicators over time.

[0096] S4.2: Extract spatial location information from the multidimensional indicator dataset and summarize the changes in indicators between adjacent time points to generate spatial summary indicator values.

[0097] Specifically, the spatiotemporal coding identifiers recorded in the multidimensional indicator dataset are extracted, and the commercial site selection feature indicators are spatially grouped and organized according to the spatial location identifiers contained in the spatiotemporal coding identifiers, so that the changes in indicators under the same spatial location identifier form corresponding spatial sets.

[0098] In the spatial set, the changes in indicators at adjacent time points corresponding to the same spatial location identifier are summarized in chronological order to obtain numerical results reflecting the overall changes of commercial site selection characteristic indicators at the corresponding spatial location over time. These results are then associated with the corresponding spatial location identifiers to generate spatial summary indicator values.

[0099] S4.3: Map the spatial aggregate index values ​​of different spatial locations into a continuous spatial distribution form to generate a site selection potential map.

[0100] Specifically, according to the arrangement order of spatial location identifiers in the spatiotemporal reference geographic canvas, the spatial aggregated index values ​​are spatially arranged and organized, and the spatial aggregated index values ​​corresponding to adjacent spatial locations are formed into a continuous correspondence in the spatial dimension.

[0101] Based on the order of adjacent spatial location markers in the spatial arrangement, the arithmetic mean of the spatial summary index values ​​corresponding to adjacent spatial location markers is taken as the intermediate value of the spatial summary index values ​​of adjacent spatial location markers. All intermediate values ​​of spatial summary index values ​​are summarized and arranged in order to obtain continuously distributed spatial summary index values.

[0102] The continuously distributed spatial aggregated index values ​​are mapped to the corresponding location range in the spatiotemporal reference geographic canvas according to the spatial location identifier, generating a site selection potential map.

[0103] S5: Based on the site selection potential map, machine learning algorithms are used to predict and dynamically optimize the temporal evolution of the site selection potential of different locations, and generate a site selection prediction report.

[0104] S5.1: Based on the site selection potential map, sort the potential distribution corresponding to different spatial locations by time and construct a site selection potential learning sequence.

[0105] Specifically, the site selection potential values ​​corresponding to different spatial location identifiers in the site selection potential map are extracted under multiple time markers; using the spatial location identifiers as the sorting benchmark, the site selection potential values ​​in the site selection potential map are merged so that the site selection potential values ​​corresponding to the same spatial location identifier form an independent data set.

[0106] The location potential values ​​in each dataset are arranged according to the chronological order of their timestamps, so that the location potential values ​​show a continuous and orderly change relationship in the time dimension. The location location identifier is used as the primary index and the time stamp as the secondary index, and the corresponding location potential values ​​are registered as ordered records in sequence. This makes the location potential values ​​under each spatial location form a directly readable structured data entry in chronological order, forming a location potential learning sequence that reflects the change process of the location potential of the same spatial location over time.

[0107] S5.2: Employ machine learning algorithms to simulate different business district evolution paths and predict the learning sequence of site selection potential to generate potential evolution prediction information.

[0108] Specifically, machine learning algorithms are used to process the location potential learning sequence, and the location potential values ​​corresponding to adjacent time markers in the location potential learning sequence are used as a set of input-output sample pairs (including input location potential values ​​and output location potential values).

[0109] The numerical difference between the input and output site selection potential values ​​is statistically analyzed, and the calculation parameters in the machine learning algorithm (the numerical transformation coefficients that convert the site selection potential value corresponding to the previous time mark to the site selection potential value of the next time mark) are gradually adjusted based on the numerical difference. For example, when the site selection potential value corresponding to a certain time mark in the potential evolution prediction information is 0.80, while the actual site selection potential value corresponding to the same time mark in the site selection potential map is 0.65, the calculation parameters are updated once in the direction of decreasing, so that the prediction result of the next round is closer to 0.65.

[0110] Using the known or predicted site selection potential value at the previous time mark as input, and combining it with the adjusted calculation parameters, the site selection potential value corresponding to the next time mark is recursively calculated; the site selection potential values ​​within the future time range are arranged in chronological order to obtain the changes in the site selection potential values.

[0111] The numerical changes in the site selection potential of each spatial location are summarized and organized to generate potential evolution prediction information.

[0112] It should be noted that machine learning algorithms are methods that analyze existing data, extract patterns of change between data points, and extrapolate results for subsequent data. By continuously adjusting calculation parameters using known results during the calculation process, it is possible to predict and optimize data with temporal continuity or changing trends without relying on manually set rules.

[0113] S5.3: Based on the potential evolution prediction information, the machine learning algorithm is dynamically optimized, and the optimized potential evolution prediction information is summarized and organized to generate a site selection prediction report.

[0114] Specifically, the potential evolution prediction information and the corresponding time-marked site selection potential values ​​in the site selection potential map are used as a reference. The numerical differences between the potential evolution prediction information and the site selection potential values ​​are statistically analyzed based on the reference, and the numerical differences are used as the basis for adjustment to update the calculation parameters used in the machine learning algorithm to generate the potential evolution prediction information.

[0115] The latest time-marked location potential value is read from the location potential learning sequence based on the updated calculation parameters. The product of the updated calculation parameters and the location potential value is used as the location potential prediction value for the next time-marked time. For example, when the location potential value for the latest time-marked time is 0.65 and the updated value transformation coefficient is 0.92, the location potential prediction value for the next time-marked time is 0.598 obtained by inputting it into the machine learning algorithm. The machine learning algorithm is then inputted again to obtain the corrected potential evolution prediction information.

[0116] The revised potential evolution prediction information is uniformly organized and summarized according to spatial location identifiers and time markers to generate a site selection prediction report.

[0117] like Figure 5The figure illustrates the overall performance changes of different site selection potential prediction methods. The horizontal axis represents time step, and the vertical axis represents site selection prediction evaluation indicators. The comparison shows that the first control group (no time evolution prediction) uses a site selection potential assessment method based on static spatial features and historical statistics, performing only one-time feature calculations on multi-source geographic data and business elements, without modeling the evolution of site selection conditions over time; the second control group (fixed parameter prediction) introduces time series modeling on top of static features to predict changes in site selection potential over time, but the model parameters remain fixed during the prediction process and are not updated or adjusted according to changes in spatiotemporal business elements; the method of this invention constructs a site selection potential prediction model based on a multi-dimensional intelligent grid data layer, continuously predicting the temporal evolution of site selection potential, and combining machine learning algorithms to perform dynamic optimization during the prediction process, enabling the model parameters to adaptively adjust with changes in spatiotemporal features.

[0118] Depend on Figure 5 The results show that the prediction results of the first control group fluctuated greatly over the overall time period, and the second control group had insufficient stability in the long-term evolution process. However, the prediction curve of the method of this invention maintains a more stable trend over the overall time period, reducing the overall deviation level. Thus, it can more accurately depict the long-term change characteristics of the site selection conditions of the shopping district on a global scale, providing a more reliable overall reference for subsequent site selection decisions.

[0119] S6: Perform multi-scenario simulations and risk assessments on the site selection prediction report to generate actual deployment site selection and operational configuration information.

[0120] S6.1: Based on the site selection prediction report, map the potential evolution prediction information into simulable objects and construct a virtual reality simulation environment.

[0121] Specifically, the potential evolution prediction information that has been compiled in the site selection prediction report is read, and the potential evolution prediction information is expanded item by item according to the spatial location identifier and time marker. The potential evolution prediction information corresponding to each spatial location identifier is broken down into the numerical change process.

[0122] The numerical change process is arranged in chronological order, so that the potential evolution prediction information forms a simulation object in the continuous time dimension. Within the same spatiotemporal reference geographic canvas, each simulation object is matched one-to-one with the corresponding spatial location identifier, and the simulation object is placed at the center of the spatial location area corresponding to the spatial location identifier in the spatiotemporal reference geographic canvas. The simulation objects are synchronized according to a unified time mark, thereby constructing a virtual reality simulation environment under a unified spatial range and a unified time axis.

[0123] S6.2: Simulate the operational performance of the site selection prediction report in a virtual reality simulation environment using the digital twin method, and record the operational performance data.

[0124] Specifically, in the virtual reality simulation environment, according to the spatial location identifiers and time markers already determined in the site selection prediction report, the simulation object is marked time by time and the simulation process is iteratively advanced according to the time markers; under each time marker, the potential evolution prediction information associated with the corresponding spatial location identifier is used as the input value to drive the simulation object to generate corresponding changes in its operating status in the virtual reality simulation environment.

[0125] During the continuous time-marking process, the changes in the operational status of each spatial location identifier under different time markers are continuously read, and the data is organized and recorded according to the spatial location identifier and time marker to obtain operational performance data.

[0126] It should be noted that the digital twin method is a way of constructing a virtual representation that corresponds to a real-world object in terms of structure and operational characteristics. By mapping real-world state changes, behavioral processes, and operational results to a virtual environment, it is possible to repeatedly simulate and analyze changes under different operating conditions without directly interfering with the real-world scenario, providing verifiable reference for site selection decisions.

[0127] S6.3: Based on operational performance data, the probability distribution of return on investment is calculated using the Monte Carlo method, and risk constraint screening is performed to generate actual deployment site selection and operational configuration information.

[0128] Specifically, the spatial location identifiers, time markers, and corresponding changes in operating status recorded in the operational performance data are used as the basis for random sampling; based on each random sampling, a set of operating status changes corresponding to spatial location identifiers and time markers are selected from the operational performance data, and the return on investment is calculated; multiple sets of return on investment corresponding to the same spatial location identifier are collected and organized to form a probability distribution reflecting the return on investment.

[0129] The expression for calculating the rate of return on investment is: ;

[0130] in, For return on investment, The number of time stamps corresponding to the same spatial location identifier in the performance data. For the first in the performance data A time marker, Time stamps in performance data The following records the changes in the business status.

[0131] It should be noted that the expression for calculating the rate of return on investment is calculated by combining the numerical values ​​of changes in operating status with the same dimension. Both the numerator and denominator are composed of the same changes in operating status, and the calculation process is dimensionless. The rate of return on investment is also dimensionless.

[0132] Based on the probability distribution of return on investment (ROI), sampling results where ROI deviates from the risk constraints are eliminated, and only ROI that meets the risk constraints is retained. For example, if the ROI results obtained from multiple random samplings are 3%, 5%, 6%, 7%, and 25%, the 25% ROI result, which only appears in a very small number of samples, is eliminated, and only the ROI concentrated in the range of 5% to 7% is retained. The corresponding spatial location identifiers and operational status configuration results are summarized and organized to generate actual deployment site selection and operational configuration information.

[0133] It should be noted that the risk assessment generates operational performance data through multi-scenario simulation, uses the Monte Carlo method to perform probability analysis on the uncertainty of the rate of return on investment, and then combines risk constraint screening to achieve the assessment and control of the risk level of the deployment plan.

[0134] The Monte Carlo method is a numerical analysis method based on random sampling and repeated computation. By randomly sampling uncertain factors multiple times, it transforms complex problems into a large number of computable random trial processes. It can approximate the probability distribution, fluctuation range and risk characteristics of the results without exhaustively listing all possible situations.

[0135] Risk constraints refer to screening criteria used to limit the acceptable range of return on investment results. By setting the return on investment to fall within a range of frequently occurring values ​​and eliminating extreme values ​​that only appear in random calculations, incorrect site selection and business decisions can be avoided.

[0136] like Figure 6As shown, a local magnified comparative analysis was conducted on the time intervals where the differences in the overall prediction results were concentrated. The upper subplot marks the time intervals where the fluctuations of the site selection prediction evaluation indicators differ from those of the method of this invention, while the lower subplot provides the magnified results, analyzing the response characteristics of different site selection potential prediction methods at key stages. Specifically, the first control group (prediction without time evolution) still uses a site selection potential assessment method based on static spatial characteristics and historical statistics. Because it does not model the continuous evolution of site selection conditions over time, the prediction results lag in response to environmental changes within local intervals. The second control group (prediction with fixed parameters) uses a time series prediction model to characterize changes in site selection potential, but the model parameters remain unchanged throughout the prediction process. When spatiotemporal business elements undergo structural changes in a short period, the prediction results are difficult to adjust in a timely manner, resulting in prediction bias within local intervals. In contrast, the method of this invention, based on a multi-dimensional intelligent grid data layer, continuously predicts the temporal evolution of site selection potential and dynamically optimizes the model parameters using machine learning algorithms during the prediction process, enabling the prediction model to adaptively adjust according to changes in local spatiotemporal characteristics. Figure 6 It can be seen that, within the critical time interval, the prediction difference between the method of this invention and the two control group methods reaches its maximum at the local extreme value position, and the method of this invention maintains a more stable and continuous trend of change, indicating that it has a stronger adaptability in characterizing short-term fluctuations and key change stages of site selection potential, which helps to improve the reliability and feasibility of site selection decisions in actual business scenarios.

[0137] In summary, this invention achieves the characterization of the changing trends of site selection conditions over time by predicting and dynamically optimizing the temporal evolution of site selection potential; and improves the reliability and feasibility of machine learning-based site selection decisions in actual operating environments by conducting multi-scenario simulations and risk assessments of site selection prediction information.

[0138] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for selecting store district locations based on spatiotemporal big data and machine learning algorithms, characterized in that: include, Collect multi-source geographic data and spatiotemporal business element sets, and generate a spatiotemporal benchmark geographic canvas through coordinate unification and geometric registration; Adaptive grid subdivision and spatiotemporal coding are performed on the spatiotemporal reference geographic canvas to obtain a multi-dimensional intelligent grid data layer. Commercial site selection characteristic indicators are calculated based on the multidimensional intelligent grid data layer and the spatiotemporal business element set, and machine learning is used to weight them to generate a multidimensional indicator dataset. Based on a multidimensional indicator dataset, the temporal correlation and spatial dependence of commercial site selection characteristic indicators are fused to generate a site selection potential map. Based on the site selection potential map, machine learning algorithms are used to predict and dynamically optimize the temporal evolution of the site selection potential of different locations, and generate a site selection prediction report. Perform multi-scenario simulations and risk assessments on the site selection forecast report to generate actual deployment site selection and operational configuration information.

2. The method for selecting store districts based on spatiotemporal big data and machine learning algorithms as described in claim 1, characterized in that: The multi-source geographic data includes publicly available internet map tiles, remote sensing imagery, and laser point clouds; The spatiotemporal business element set includes pedestrian behavior data, commercial facility business data, transportation data, and environmental and meteorological monitoring data.

3. The method for selecting store district locations based on spatiotemporal big data and machine learning algorithms as described in claim 2, characterized in that: The aforementioned spatiotemporal reference geographic canvas refers to mapping multi-source geographic data and spatiotemporal business element sets to the same standard spatial reference system, and using a deep learning geometric registration algorithm for geometric registration.

4. The method for selecting store districts based on spatiotemporal big data and machine learning algorithms as described in claim 3, characterized in that: The specific steps for performing adaptive grid partitioning and spatiotemporal coding on the spatiotemporal reference geographic canvas are as follows. Based on the spatiotemporal reference geographic canvas, an adaptive grid partitioning algorithm is used to dynamically allocate grid resolution to different spatial regions and perform spatial subdivision to generate a spatial location grid. The GeoHash encoding system is used to perform spatiotemporal encoding on the spatial location grid to obtain the spatiotemporal encoding identifier.

5. The method for selecting store districts based on spatiotemporal big data and machine learning algorithms as described in claim 4, characterized in that: The multidimensional intelligent grid data layer is obtained by structural aggregation and temporal consistency processing of spatial location grids based on spatiotemporal coding identifiers.

6. The method for selecting store district locations based on spatiotemporal big data and machine learning algorithms as described in claim 5, characterized in that: The specific steps for calculating commercial site selection characteristic indicators based on the multi-dimensional intelligent grid data layer and the spatiotemporal business element set are as follows: Based on the multi-dimensional intelligent grid data layer, the spatiotemporal business element set is mapped to the spatial location grid according to the spatiotemporal coding identifier, generating a business element mapping set. Perform spatiotemporal big data analysis on the business element mapping set and calculate business site selection characteristic indicators.

7. The method for selecting store districts based on spatiotemporal big data and machine learning algorithms as described in claim 6, characterized in that: The multidimensional indicator dataset was obtained by adaptively adjusting the commercial site selection characteristic indicators using a machine learning weighting method.

8. The method for selecting store districts based on spatiotemporal big data and machine learning algorithms as described in claim 7, characterized in that: The method involves fusing temporal correlation and spatial dependency of commercial site selection characteristic indicators based on a multidimensional indicator dataset to generate a site selection potential map. The specific steps are as follows: Based on a multidimensional indicator dataset, the characteristics of commercial site selection are sorted over time, and the changes in indicators between adjacent time points are recorded. Extract spatial location information from the multidimensional indicator dataset and summarize the changes in indicators between adjacent time points to generate spatial summary indicator values. By mapping the spatial aggregated index values ​​of different spatial locations into a continuous spatial distribution, a site selection potential map is generated.

9. The method for selecting store districts based on spatiotemporal big data and machine learning algorithms as described in claim 8, characterized in that: The process involves using machine learning algorithms to predict and dynamically optimize the temporal evolution of site selection potential at different locations based on the site selection potential map, generating a site selection prediction report. The specific steps are as follows: Based on the site selection potential map, the potential distribution corresponding to different spatial locations is sorted over time to construct a site selection potential learning sequence. Machine learning algorithms are used to simulate the evolution paths of different business districts and to predict the learning sequence of site selection potential, thereby generating potential evolution prediction information. Based on potential evolution prediction information, the machine learning algorithm is dynamically optimized, and the optimized potential evolution prediction information is summarized and organized to generate a site selection prediction report.

10. The method for selecting store district locations based on spatiotemporal big data and machine learning algorithms as described in claim 9, characterized in that: The process of performing multi-scenario simulations and risk assessments on the site selection prediction report to generate actual deployment site selection and operational configuration information involves the following steps: Based on the site selection prediction report, the potential evolution prediction information is mapped into simulable objects, and a virtual reality simulation environment is constructed. The performance of the site selection prediction report is simulated in a virtual reality simulation environment using the digital twin method, and the performance data is recorded. Based on operational performance data, the probability distribution of return on investment is calculated using the Monte Carlo method, and risk constraints are applied to generate actual deployment site selection and operational configuration information.

Images