GIS (Geographic Information System) driving-based commercial plot space correlation visualization analysis method

Abstract

The invention discloses a GIS (Geographic Information System)-driven commercial parcel space correlation visualization analysis method, which belongs to the technical field of commercial parcel analysis, and effectively improves the accuracy and reliability of commercial parcel correlation analysis by integrating multi-source heterogeneous data, realizing time-space accurate alignment, constructing a dynamic network structure and filtering noise interference. By extracting core features to construct a multi-layer matrix and combining with a space-time attention weight fusion algorithm, dynamic association modes among plots under different time scales can be captured, a space interaction rule hidden in complex data is disclosed, an analysis basis closer to an actual operation condition is provided for commercial decision making, and the method is suitable for commercial decision making. By constructing a three-dimensional flow field visualization model and developing a multi-scale interactive exploration interface, abstract associated data is converted into a visual visualization map, and in combination with a continuous application verification and iterative optimization mechanism, the timeliness and practicability of an analysis result are ensured, and scientific and efficient technical support is provided for commercial planning and resource allocation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of commercial land analysis, and particularly relates to a commercial land space correlation visualization analysis method based on GIS driving. BACKGROUND

[0002] Commercial land space correlation analysis is used for studying the spatial distribution characteristics, interaction relationship and influencing factors among commercial lands in a city or region. The analysis integrates land use data, population density, traffic network, economic indicators and other multi-source spatial data, and uses spatial autocorrelation, kernel density estimation, spatial regression and other models to reveal the correlation and dependency of commercial lands in geographical location, functional layout, resource allocation and the like. The core goal is to provide a scientific basis for urban planning, commercial development, policy making and the like, to optimize the commercial space structure, to improve the resource allocation efficiency and to promote the coordinated development of regional economy. With the continuous development of big data and spatial information technology, the commercial land space correlation analysis is increasingly widely applied in city research and commercial decision making, and becomes an important tool for modern city governance and commercial operation.

[0003] However, the existing technology has problems such as insufficient data integration capability, inaccurate dynamic correlation capture, weak visual interaction and single decision support in commercial land space correlation analysis, and is often limited to static data processing and two-dimensional display, which is difficult to effectively integrate multi-source heterogeneous data and reveal the spatio-temporal dynamic correlation law, resulting in deviation of the analysis result from the actual commercial operation demand, and inability to provide accurate and efficient technical support for decision makers. SUMMARY

[0004] The purpose of the present application is to solve the above-mentioned problems, and to provide a commercial land space correlation visualization analysis method based on GIS driving.

[0005] The technical solution adopted by the present application is as follows: the commercial land space correlation visualization analysis method based on GIS driving comprises the following steps:

[0006] S1: integrating commercial land basic geographic data, real-time flow trajectory data, consumption behavior data and surrounding facility POI data by the system, and realizing cross-source data alignment through spatio-temporal coordinate mapping to provide a standardized input basis for subsequent feature engineering;

[0007] S2: based on the multi-dimensional data collected in S1, using an adaptive threshold algorithm to construct an initial land correlation network, and simultaneously filtering non-correlation noise data to provide a high-quality network topology structure for correlation pattern mining;

[0008] S3: fusing the network structure data processed in S2, extracting core features such as land space location, economic vitality and traffic accessibility, and constructing a multi-level feature matrix to provide an input variable set for the innovative correlation mining algorithm;

[0009] S4:innovatively introduce a bidirectional LSTM-attention hybrid model to dynamically associate the feature matrix output by S3, capture the time-varying correlation strength between plots through spatiotemporal attention weights, and break through the limitations of traditional static analysis;

[0010] S5: based on the dynamic correlation weight calculated by S4, construct a three-dimensional flow field visualization model to convert abstract correlation into concrete flow field atlas, providing an intuitive visual carrier for interactive analysis;

[0011] S6: integrate the visualization model generated by S5 to develop an interactive interface supporting micro (single plot) - meso (block) - macro (region) three-level scale switching, allowing users to update the correlation flow field display in real time through parameter adjustment;

[0012] S7: couple the user interaction parameters of S6 interface with historical correlation data to train a gradient boosting regression model, realize quantitative prediction of the impact of business strategy adjustment on the correlation network, and output multi-scenario simulation results;

[0013] S8: based on the prediction results of S7, construct a three-dimensional decision-making index system including correlation gain, risk warning value, and resource optimization rate to provide quantitative evaluation standards for business layout decisions;

[0014] S9: apply the decision-making indicators generated by S8 to practical scenarios such as business site selection and format adjustment, verify the effectiveness of the method through A / B testing, and use the feedback results to optimize the data collection dimensions of S1 and the attention mechanism parameters of S4.

[0015] In a preferred embodiment, in step S1, first collect multi-source heterogeneous data of commercial plots, including 15-minute interval traffic flow data, daily economic statistics data, 10-meter resolution geographic spatial data, and POI interest point data containing 200+ categories. Through Z-Score standardization processing to remove outliers, use KNN interpolation method to fill in missing values to make the data completeness rate reach more than 95%. Spatial coordinates are converted to UTM projection coordinate system, with the central meridian set to 117 degrees east and the projection zone number set to 50N. On the time dimension, collect data once an hour to ensure that different source data are completely aligned in space-time scale, finally forming a standardized data set with a time span of 12 months and a spatial coverage area of 50 square kilometers.

[0016] In a preferred embodiment, in step S2, a dynamic network structure is constructed based on the aligned data set, with commercial plots as nodes and correlation strength as edge weights. A sliding time window of 24 hours is used to capture the trend of network topology over time, and a correlation strength threshold of 0.15 is set to filter insignificant connections below this value. A Gaussian filter algorithm with a standard deviation of 3 is used to smooth the network weights to eliminate short-term fluctuation noise, and L1 regularization is used to constrain the network complexity to control the node degree within 30.

[0017] In a preferred embodiment, in step S3, a feature extraction method combining principal component analysis and a 3-layer autoencoder is used to select 5 core features from the original data, including economic vitality index, traffic accessibility index, commercial density index, population flow index, and land use type. The input layer node number of the autoencoder is 64, the hidden layer node numbers are 32 and 16 respectively, and the output layer node number is 8. A three-dimensional feature matrix is constructed according to the time dimension, spatial dimension and feature dimension, the time dimension contains 8760 hour-level time slices, the spatial dimension covers 1200 commercial plots, and the feature dimension contains 8 key indicators. The matrix element values are mapped to the interval of 0 to 1 through Min-Max standardization processing, ensuring that different dimension features can be effectively integrated.

[0018] In a preferred embodiment, in step S4, the bidirectional LSTM-attention hybrid model takes the multi-level feature matrix output by S3 as input. First, the bidirectional LSTM network captures the time sequence dynamics of plot correlation.

[0019] The forward LSTM layer of the model extracts the feature sequence from t=1 to t=T, and the reverse LSTM layer captures the historical dependence from t=T to t=1, and the outputs of both are concatenated to form the bidirectional hidden state at each time step. On this basis, a spatio-temporal dual-dimensional attention mechanism is introduced: the time dimension calculates the attention weight of the hidden state sequence through the softmax function, highlighting the contribution of different time points to the current correlation strength; the spatial dimension constructs a distance decay function based on the geographical coordinates of the plots, giving higher weights to the features of adjacent plots. Finally, the spatio-temporal weight Hadamard product generates a dynamic correlation matrix, realizing the quantitative characterization of time-varying correlation strength.

[0020] The spatio-temporal attention weight fusion formula is:

[0021]

[0022] In the formula:

[0023] α i,j,t represents the spatio-temporal attention weight of plot i to plot j at time t;

[0024] h i,t denotes the hidden state vector of land parcel i output by bidirectional LSTM at time t;

[0025] d k denotes the dimension of the hidden state vector;

[0026] x i ,x j denotes the geographical coordinates of land parcels i and j;

[0027] σ 2 denotes the spatial distance decay coefficient;

[0028] N denotes the total number of land parcels in the study area;

[0029] T denotes the length of the time series.

[0030] In a preferred embodiment, in step S5, the dynamic correlation weight matrix is converted into a three-dimensional flow field visualization atlas. The fourth-order Runge-Kutta streamline generation algorithm is used to calculate the flow direction and flow rate parameters according to the weight values. The streamline starting point density is set to 20 per square kilometer, the color coding uses the HSV color system, red represents a correlation intensity of 0.8-1.0, yellow represents 0.4-0.8, and blue represents 0-0.4. By adjusting the transparency parameter to 0.6, the correlation relationships at different levels are distinguished, and the three-dimensional spatial display of the flow field is realized using volume rendering technology. The view rotation step is set to 15 degrees, and the zoom factor is set to 1.2, ensuring that the visualization results can intuitively reflect the spatiotemporal dynamic changes in the correlation intensity between land parcels, and support real-time interactive operations.

[0031] In a preferred embodiment, in step S6, an interactive exploration interface for the correlation flow field supporting multi-scale switching is developed based on WebGL technology, realizing stepless scaling from 1:50,000 to 1:500, with a response time controlled within 300 milliseconds. The correlation path tracking tool is designed to allow users to click on any land parcel to view other land parcels with significant correlation and the correlation intensity change curve in the past 30 days. The time axis control component is integrated to support users to drag the time slider to observe the dynamic changes of the flow field within 24 hours, while providing data export functions in PNG and CSV formats. The image resolution is set to 1920x1080 pixels, and the data sampling interval can be selected as 5 minutes, 15 minutes, or 1 hour, meeting different analysis needs.

[0032] In a preferred embodiment, in step S7, the business policy implementation data and the corresponding associated network change data of the past 5 years are collected to construct a training set containing 10,000+ samples. An XGBoost gradient boosting tree algorithm is used to construct a strategy impact prediction model, the input layer contains 4 types of characteristic variables including policy type, implementation intensity, implementation range and economic environment, a total of 28 characteristics, and the output layer is the change rate of the key indicators of the associated network. Through 5-fold cross-validation, the model hyperparameters are optimized, the learning rate is set to 0.01, the maximum tree depth is 8, the number of leaf nodes is 64, and the L2 regularization coefficient is 0.001.

[0033] In a preferred embodiment, a three-dimensional decision-making index system is constructed, including economic and social benefits, environmental sustainability and implementation feasibility. The economic dimension includes investment return rate, tax growth expectation and business vitality index, the social dimension includes employment growth rate, resident satisfaction and public service coverage, the environmental dimension evaluates traffic pressure index, carbon emissions and green coverage, and the implementation feasibility dimension analyzes policy landing difficulty, execution cost and policy synergy. Each dimension index is weighted by the analytic hierarchy process, wherein the economic dimension weight is 0.4, the social dimension weight is 0.3, the environmental dimension weight is 0.2, and the implementation feasibility dimension weight is 0.1. Finally, a comprehensive decision-making score of 0-100 is generated and visualized in the form of a three-dimensional radar chart.

[0034] In a preferred embodiment, in step S9, three different types of commercial areas, including city center business district, emerging development zone and community business circle, are selected for field application verification, and 3 experimental groups and 1 control group are set up in each area. The actual operation data and policy implementation effect feedback of the past 6 months are collected, compared with the model prediction results, and the deviation rate of the two is calculated. When the deviation exceeds 8%, the potential problems in the data collection, feature extraction and model training stages are traced back, and the model is iteratively optimized by adjusting the abnormal value judgment threshold in the data preprocessing stage, optimizing the penalty coefficient in the feature selection algorithm and updating the model training samples. The model performance is evaluated every quarter to ensure that the long-term prediction accuracy is stable above 90% and form an optimization report to guide subsequent business decisions.

[0035] In summary, due to the adoption of the above technical solutions, the present application has the following advantages:

[0036] 1、In the present application, by integrating multi-source heterogeneous data and realizing spatio-temporal precise alignment, constructing dynamic network structure and filtering noise interference, the accuracy and reliability of business plot correlation analysis are effectively improved. By extracting core features to construct multi-layer matrix, combining with spatio-temporal attention weight fusion algorithm, the dynamic correlation mode between plots under different time scales can be captured, and the spatial interaction law hidden in complex data can be revealed, providing more actual operation analysis basis for business decision.

[0037] 2、In the present application, by constructing three-dimensional flow field visualization model and developing multi-scale interactive exploration interface, abstract correlation data is converted into intuitive visualization atlas, supporting multi-dimensional analysis from macro to micro, helping decision makers quickly identify key correlation path and potential business opportunities. At the same time, by training prediction model and generating three-dimensional decision index system, quantitative evaluation and comprehensive decision support of business policy influence are realized, combined with continuous application verification and iterative optimization mechanism, ensuring the timeliness and practicality of analysis results, providing scientific and efficient technical support for business planning and resource allocation. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 The flow principle schematic diagram of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application, and are not used to limit the present application.

[0040] Referring to Figure 1 , the method comprises the following steps:

[0041] S1: The system integrates business plot basic geographic data, real-time flow trajectory data, consumer behavior data and surrounding facility POI data, realizes cross-source data alignment through spatio-temporal coordinate mapping, and provides standardized input basis for subsequent feature engineering;

[0042] S2: Based on the multi-dimensional data collected in S1, an adaptive threshold algorithm is used to construct an initial plot correlation network, and non-correlation noise data is filtered at the same time, providing a high-quality network topology structure for correlation pattern mining;

[0043] S3: Fusion of network structure data processed in S2, extraction of plot spatial position, economic vitality, traffic accessibility and other core features, construction of multi-level feature matrix, and provision of input variable set for innovative correlation mining algorithm;

[0044] S4: innovatively introduce a bidirectional LSTM-attention hybrid model to dynamically correlate the feature matrix output by S3, capture the time-varying correlation strength between plots through spatiotemporal attention weights, and break through the limitations of traditional static analysis;

[0045] S5: based on the dynamic correlation weights calculated by S4, construct a three-dimensional flow field visualization model to convert abstract correlation relationships into concrete flow field maps, providing an intuitive visual carrier for interactive analysis;

[0046] S6: integrate the visualization model generated by S5 to develop an interactive interface that supports micro (single plot) - meso (block) - macro (region) three-level scale switching, allowing users to update the correlation flow field display in real time through parameter adjustment;

[0047] S7: couple the user interaction parameters of S6 interface with historical correlation data to train a gradient boosting regression model, realize quantitative prediction of the impact of business strategy adjustment on the correlation network, and output multi-scenario simulation results;

[0048] S8: based on the prediction results of S7, construct a three-dimensional decision-making index system including correlation gain, risk warning value, and resource optimization rate to provide quantitative evaluation standards for business layout decisions;

[0049] S9: apply the decision-making indexes generated by S8 to practical scenarios such as business site selection and format adjustment, verify the effectiveness of the method through A / B testing, and use the feedback results to optimize the data collection dimensions of S1 and the attention mechanism parameters of S4.

[0050] In step S1, first collect multi-source heterogeneous data of commercial plots, including 15-minute interval traffic flow data, daily economic statistics data, 10-meter resolution geographic spatial data, and POI interest point data containing 200+ categories. Through Z-Score standardization processing to remove outliers, use KNN interpolation method to fill in missing values to make the data completeness rate reach more than 95%. Spatial coordinates are converted to UTM projection coordinate system, with the central meridian set to 117 degrees east and the projection zone number set to 50N. On the time dimension, strictly collect data once an hour to ensure that different source data are completely aligned in space-time scale, finally forming a standardized data set with a time span of 12 months and a spatial coverage area of 50 square kilometers.

[0051] In step S2, a dynamic network structure is constructed based on the aligned data set, with business blocks as nodes and correlation strength as edge weights. A sliding time window of 24 hours is used to capture the trend of network topology over time, and a correlation strength threshold of 0.15 is set to filter insignificant connections below this value. A Gaussian filter algorithm with a standard deviation of 3 is used to smooth the network weights and eliminate short-term fluctuation noise, and L1 regularization is used to constrain the network complexity to control the node degree within 30.

[0052] In step S3, a feature extraction method combining principal component analysis and a 3-layer autoencoder is used to select 5 core features from the original data, including economic vitality index, traffic accessibility index, business density index, population flow index, and land use type. The input layer node number of the autoencoder is 64, the hidden layer node numbers are 32 and 16 respectively, and the output layer node number is 8. A three-dimensional feature matrix is constructed according to the time dimension, spatial dimension and feature dimension, with 8760 hour-level time slices in the time dimension, 1200 business blocks in the spatial dimension, and 8 key indicators in the feature dimension. The matrix element values are processed by Min-Max standardization to map to the interval of 0 to 1, ensuring that different dimension features can be effectively integrated.

[0053] In step S4, the bidirectional LSTM-attention hybrid model takes the multi-level feature matrix output by S3 as input. First, the bidirectional LSTM network captures the temporal dynamics of block correlation.

[0054] The forward LSTM layer extracts the feature sequence from t = 1 to t = T, and the reverse LSTM layer captures the historical dependence from t = T to t = 1. The outputs of both are concatenated to form the bidirectional hidden state at each time step. On this basis, a spatio-temporal attention mechanism is introduced: the time dimension calculates the attention weight of the hidden state sequence through the softmax function, highlighting the contribution of different time points to the current correlation strength; the spatial dimension constructs a distance decay function based on the geographical coordinates of the blocks, giving higher weights to the features of neighboring blocks. Finally, the spatio-temporal weight Hadamard product generates a dynamic correlation matrix, realizing the quantitative characterization of the time-varying correlation strength.

[0055] The spatio-temporal attention weight fusion formula is:

[0056]

[0057] In the formula:

[0058] α i,j,t represents the spatio-temporal attention weight of block i to block j at time t;

[0059] h i,tdenotes the hidden state vector of parcel i output by the bidirectional LSTM at time t;

[0060] d k denotes the dimension of the hidden state vector;

[0061] x i ,x j denotes the geographical coordinates of parcels i and j;

[0062] σ 2 denotes the spatial distance decay coefficient;

[0063] N denotes the total number of parcels within the study area;

[0064] T denotes the length of the time series.

[0065] In step S5, the dynamic correlation weight matrix is converted into a three-dimensional flow field visualization map. The fourth-order Runge-Kutta streamline generation algorithm is used to calculate the flow direction and flow rate parameters based on the weight values. The streamline starting point density is set to 20 per square kilometer, and the color coding uses the HSV color system, with red representing a correlation intensity of 0.8-1.0, yellow representing 0.4-0.8, and blue representing 0-0.4. By adjusting the transparency parameter to 0.6, the correlation relationships at different levels are distinguished, and the three-dimensional spatial display of the flow field is realized using volume rendering technology. The view rotation step is set to 15 degrees, and the scaling factor is set to 1.2, ensuring that the visualization results can intuitively reflect the spatiotemporal dynamic changes in the correlation intensity between parcels, supporting real-time interactive operations.

[0066] In step S6, an interactive exploration interface for the correlation flow field supporting multi-scale switching is developed based on WebGL technology, realizing stepless scaling from 1:50,000 to 1:500, with a response time controlled within 300 milliseconds. The correlation path tracking tool is designed to allow users to click on any parcel and view other parcels with significant correlation and the correlation intensity change curve over the past 30 days. The time axis control component is integrated to support users in dragging the time slider to observe the dynamic changes in the flow field within 24 hours, while providing data export functions in PNG and CSV formats, with an image resolution of 1920x1080 pixels and data sampling intervals of 5 minutes, 15 minutes, or 1 hour, meeting different analysis needs.

[0067] In step S7, the implementation data of business policies and the corresponding correlation network change data over the past 5 years are collected to construct a training set containing more than 10,000 samples. The XGBoost gradient boosting tree algorithm is used to build a policy impact prediction model, with the input layer containing 4 types of characteristic variables, including policy type, implementation intensity, implementation range, and economic environment, totaling 28 features, and the output layer being the change rate of key indicators of the correlation network. The model hyperparameters are optimized through 5-fold cross-validation, with a learning rate of 0.01, a maximum tree depth of 8, a leaf node number of 64, and an L2 regularization coefficient of 0.001.

[0068] In step S8, a three-dimensional decision-making index system including economic and social benefits, environmental sustainability and implementation feasibility is constructed. The economic dimension includes investment return rate, tax growth expectation and business vitality index, the social dimension covers employment growth rate, resident satisfaction and public service coverage, the environmental dimension evaluates traffic pressure index, carbon emissions and green coverage, and the implementation feasibility dimension analyzes policy landing difficulty, execution cost and policy synergy. Each dimension index is given a weight through the analytic hierarchy process, wherein the economic dimension weight is 0.4, the social dimension weight is 0.3, the environmental dimension weight is 0.2, and the implementation feasibility dimension weight is 0.1. Finally, a comprehensive decision-making score of 0-100 is generated and visualized in the form of a three-dimensional radar chart.

[0069] In step S9, three different types of commercial areas, namely urban central business district, emerging development zone and community business circle, are selected for field application verification, and three experimental groups and one control group are set up in each area. Actual operation data and policy implementation effect feedback are collected for 6 months, and the deviation rate of the model prediction results is calculated through comparison analysis. When the deviation exceeds 8%, the potential problems in data collection, feature extraction and model training are traced back, and the model is iteratively optimized by adjusting the abnormal value judgment threshold in the data preprocessing stage, optimizing the penalty coefficient in the feature selection algorithm and updating the model training samples. The model performance is evaluated every quarter to ensure that the long-term prediction accuracy is stable above 90% and form an optimization report to guide subsequent business decisions.

[0070] From the above, it can be seen that:

[0071] In the present application, by integrating multi-source heterogeneous data and achieving spatio-temporal precise alignment, constructing dynamic network structure and filtering noise interference, the accuracy and reliability of commercial plot correlation analysis are effectively improved. By extracting core features to construct multi-layer matrices and combining with the spatio-temporal attention weight fusion algorithm, the dynamic correlation patterns between plots at different time scales can be captured, and the spatial interaction rules hidden in complex data can be revealed, providing more actual operation analysis basis for business decisions.

[0072] In the present application, by constructing a three-dimensional flow field visualization model and developing a multi-scale interactive exploration interface, abstract correlation data is converted into intuitive visual atlas, supporting multi-dimensional analysis from macro to micro, helping decision-makers quickly identify key correlation paths and potential business opportunities. At the same time, by training the prediction model and generating a three-dimensional decision-making index system, the quantitative evaluation and comprehensive decision support of the influence of business policy are realized, combined with the continuous application verification and iterative optimization mechanism, ensuring the timeliness and practicality of the analysis results, providing scientific and efficient technical support for business planning and resource allocation.

[0073] It has to be noted that, in the present document, the terms "first", "second", etc. merely serve to identify a subject or action, without necessarily requiring or implying any actual relationship or order between the subjects or actions, unless explicitly defined otherwise. Furthermore, the term "comprising" or any other variant thereof is intended to cover the non-exclusive inclusion of elements, such that a process, method, article or apparatus that comprises elements not expressly listed is not excluded. In other words, it is intended that the process, method, article, or apparatus that comprises a set of elements includes not only those elements expressly listed, but also other elements not expressly listed or even inherent to such process, method, article, or apparatus.

[0074] The above description enables one of ordinary skill in the art to make and use the application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A GIS-driven business plot spatial correlation visualization analysis method, characterized in that: The method comprises the following steps: S1: The system integrates the commercial plot basic geographic data, real-time people flow trajectory data, consumption behavior data and surrounding facility POI data, realizes cross-source data alignment through space-time coordinate mapping, provides standardized input basis for subsequent feature engineering; S2: Based on the multi-dimensional data collected in S1, an adaptive threshold algorithm is used to construct an initial network associated with the plot, and non-associated noise data is filtered synchronously to provide a high-quality network topology structure for associated pattern mining; S3: Fusion of network structure data processed in S2, extraction of core features such as plot spatial position, economic vitality and traffic accessibility, construction of multi-level feature matrix, and provision of input variable set for innovative association mining algorithm; S4: Innovative introduction of a bidirectional LSTM-attention hybrid model for dynamic association calculation of the feature matrix output by S3, and capture of time-varying association strength between plots through space-time attention weight; S5: Based on the dynamic association weight calculated in S4, a three-dimensional flow field visualization model is constructed to convert abstract association relationship into a concrete flow field atlas, providing an intuitive visual carrier for interactive analysis; S6: Integration of the visualization model generated in S5, development of an interactive interface supporting micro (single plot) - meso (block) - macro (region) three-level scale switching, and allowing users to update the associated flow field display in real time through parameter adjustment; S7: Coupling of user interaction parameters in S6 interface with historical association data, training of gradient boosting regression model, realization of quantitative prediction of the influence of business strategy adjustment on the associated network, and output of multi-scenario simulation results; S8: Based on the prediction results of S7, a three-dimensional decision-making index system including association gain, risk warning value and resource optimization rate is constructed to provide quantitative evaluation standards for business layout decision-making; S9: Application of the decision-making indexes generated in S8 to actual scenarios such as commercial site selection and format adjustment, verification of the effectiveness of the method through A / B testing, and feedback of the results to optimize the data collection dimensions of S1 and the attention mechanism parameters of S4.

2. The GIS-driven business plot spatial correlation visualization analysis method of claim 1, wherein: In the step S1, first, multi-source heterogeneous data of commercial plots are collected, including 15-minute interval traffic flow data, daily three times economic statistics data, 10-meter resolution geographic spatial data and POI interest point data containing more than 200 categories; through Z-Score standardization processing, abnormal values are removed, KNN interpolation method is used to fill in missing values to make the data integrity rate reach more than 95%; spatial coordinates are uniformly converted into UTM projection coordinate system, the central meridian is set to 117 degrees east, and the projection zone number is 50N; on the time dimension, data is collected once an hour to ensure that different source data are completely aligned on the space-time scale, and finally a standardized data set with a time span of 12 months and a spatial coverage area of 50 square kilometers is formed.

3. The GIS-driven business plot spatial correlation visualization analysis method of claim 1, wherein: In the step S2, a dynamic network structure is constructed based on the aligned data set, taking commercial plots as nodes and correlation strength as edge weight; a sliding time window with a size of 24 hours is used to capture the change trend of the network topology over time, and a correlation strength threshold of 0.15 is set to filter insignificant connections below the threshold; a Gaussian filtering algorithm with a standard deviation of 3 is used to smooth the network weight and eliminate short-term fluctuation noise, and L1 regularization is used to constrain the network complexity to control the node degree within 30.

4. The GIS-driven based commercial plot spatial correlation visualized analysis method according to claim 1, wherein: In the step S3, a feature extraction method combining principal component analysis and a three-layer self-encoder is used to select five core features, including economic vitality index, traffic accessibility index, commercial density index, population flow index and land use type, from the original data. The number of input layer nodes of the self-encoder is 64, the number of hidden layer nodes is 32 and 16 respectively, and the number of output layer nodes is 8.

5. The GIS-driven based commercial plot spatial correlation visualized analysis method according to claim 1, wherein: In the step S4, a bidirectional LSTM-attention hybrid model takes the multi-level feature matrix output by S3 as input, first captures the time sequence dynamics of plot correlation through a bidirectional LSTM network; The forward LSTM layer extracts the feature sequence from t=1 to t=T, and the reverse LSTM layer captures the historical dependence from t=T to t=1, and the outputs of the two are concatenated to form a bidirectional hidden state at each time step; On this basis, a spatio-temporal double-dimensional attention mechanism is introduced: the time dimension calculates the attention weight of the hidden state sequence through the softmax function, highlighting the contribution of different time points to the current correlation strength; The spatial dimension constructs a distance decay function based on the geographical coordinates of the plots, giving higher weight to the features of adjacent plots; Finally, the Hadamard product of the spatio-temporal weight generates a dynamic correlation matrix, realizing the quantitative description of the time-varying correlation strength; The spatio-temporal attention weight fusion formula is: In the formula: a i,j,t denotes the spatio-temporal attention weight of patch i to patch j at time t; h i,t denotes the hidden state vector of the cell i at time t output by the bidirectional LSTM; d k denotes the hidden state vector dimension; x i ,x j represent the geographical coordinates of the sites i and j; σ 2 denotes the spatial distance attenuation coefficient; N represents the total number of plots in the study area; T represents the length of the time series.

6. The GIS-driven based commercial plot spatial correlation visualized analysis method according to claim 1, wherein: In the step S5, the dynamic correlation weight matrix is converted into a three-dimensional flow field visualization atlas, and the fourth-order Runge-Kutta streamline generation algorithm is used to calculate the flow direction and flow rate parameters according to the weight value; the streamline starting point density is set to 20 per square kilometer, and the color coding uses the HSV color system, with red representing a correlation strength of 0.8-1.0, yellow representing 0.4-0.8, and blue representing 0-0.4; By adjusting the transparency parameter to 0.6, the correlation relationship at different levels is distinguished, and the three-dimensional space display of the flow field is realized by using volume rendering technology, with a view rotation step of 15 degrees and a scaling factor of 1.2, ensuring that the visualization result can intuitively reflect the spatio-temporal dynamic changes of the correlation strength between plots, supporting real-time interactive operation.

7. The GIS-driven based commercial plot spatial correlation visualized analysis method according to claim 1, wherein: In step S6, a correlation flow field interaction exploration interface supporting multi-scale switching is developed based on WebGL technology, realizing stepless scaling from 1:50,000 to 1:500, and the response time is controlled within 300 milliseconds; a correlation path tracking tool is designed to allow users to click on any plot to view other plots with significant correlation and the correlation intensity change curve in the past 30 days; a time axis control component is integrated to support users to drag the time slider to observe the flow field dynamic changes within 24 hours, and data export functions in PNG and CSV formats are provided, with a picture resolution of 1920x1080 pixels and data sampling intervals of 5 minutes, 15 minutes or 1 hour to meet different analysis needs.

8. The GIS-driven based commercial plot spatial correlation visualization analysis method of claim 1, wherein: In step S7, the past 5 years of business policy implementation data and the corresponding correlation network change data are collected to build a training set containing 10,000+ samples; an XGBoost gradient boosting tree algorithm is used to build a strategy impact prediction model, the input layer includes 4 types of characteristic variables, including policy type, implementation strength, implementation range and economic environment, a total of 28 characteristics, and the output layer is the change rate of the correlation network key indicators; the model hyperparameters are optimized through 5-fold cross-validation, with a learning rate of 0.01, a maximum tree depth of 8, a leaf node number of 64, and an L2 regularization coefficient of 0.

001.

9. The GIS-driven based commercial plot spatial correlation visualized analysis method according to claim 1, wherein: In step S8, a three-dimensional decision-making index system is constructed, including economic and social benefits, environmental sustainability and implementation feasibility; the economic dimension includes investment return rate, tax growth expectation and business vitality index, the social dimension includes employment growth rate, resident satisfaction and public service coverage, the environmental dimension evaluates traffic pressure index, carbon emissions and green coverage, and the implementation feasibility dimension analyzes policy landing difficulty, execution cost and policy synergy; each dimension index is weighted by AHP, with the economic dimension weight being 0.4, the social dimension weight being 0.3, the environmental dimension weight being 0.2, and the implementation feasibility dimension weight being 0.

1. Finally, a comprehensive decision-making score of 0-100 is generated and visualized in the form of a three-dimensional radar chart.

10. The GIS-driven based commercial plot spatial correlation visualized analysis method according to claim 1, wherein: In step S9, three different types of commercial areas, including urban central business districts, emerging development zones and community commercial circles, are selected for field application verification, and 3 experimental groups and 1 control group are set up in each area. Actual operation data and policy implementation effect feedback are collected for 6 consecutive months, and the deviation rate of the model prediction results is calculated by comparing the two; when the deviation exceeds 8%, the potential problems in data collection, feature extraction and model training are traced back, and the model is iteratively optimized by adjusting the abnormal value judgment threshold in the data preprocessing stage, optimizing the penalty coefficient in the feature selection algorithm and updating the model training samples; the model performance is evaluated every quarter to ensure that the long-term prediction accuracy is stable above 90% and form an optimization report to guide subsequent business decisions.

Images