Commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data

Abstract

The invention discloses a commercial space intelligent optimization method based on a comprehensive knowledge graph and tourist data, and the method comprises the steps: collecting the three-dimensional data of an urban building and an open space through a high-precision unmanned plane and laser equipment, integrating the urban business state distribution and traffic network information, and constructing an urban three-dimensional space data set. And according to the building physical position, the function association and the traffic network topology, constructing, matching and fusing space, function and traffic network relationships, and forming an urban three-dimensional space business state knowledge graph. And performing graph matching and optimization pedigree construction on the to-be-optimized region by using a sub-graph isomorphism algorithm, and screening an optimization scheme based on homogeneous competition and passenger flow data. And periodically predicting and iteratively optimizing the tourist amount of the optimized area through a graph network prediction algorithm, planning shopping and entertainment paths of tourists in combination with a path algorithm, and displaying an optimization result and an adjustment scheme by using virtual reality and mixed reality equipment.

Description

Technical Field

[0001] The present invention relates to the field of urban planning, and specifically to a commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data. Background Art

[0002] Traditional urban business renewal methods usually rely on manual measurement, data collection, manual update and modeling based on static data, and the source of collected data is single, which leads to problems such as low efficiency, long and delayed data update cycle, lack of real-time and accuracy of static urban models, small data volume, limited coverage and lack of relevance. It is difficult to cope with the rapidly changing urban environment and fully reflect the urban business forms and connections. At the same time, the data island phenomenon is serious, making it difficult to carry out intelligent reasoning and decision support, and provide high-level decision support and optimization suggestions. However, with the acceleration of urbanization, the spatial distribution and functional types of the original urban business forms are highly dynamic and complex. The rapid update of urban business forms can improve urban management efficiency through real-time monitoring and management, optimize urban resource allocation, improve the quality of life of citizens and promote the development of urban economy. The intelligent update method and equipment of urban three-dimensional spatial business forms based on knowledge graphs can help solve the above problems of traditional urban business form renewal. Through knowledge graph technology and three-dimensional city modeling technology, data from different sources can be integrated, the relevance and comprehensibility of data can be enhanced through semantic technology, and automatic logical reasoning can be performed based on existing data. Geographic location data, building attribute data, real-time sensor data, etc. can be integrated into the knowledge graph to build a dynamic, semantically rich urban three-dimensional spatial model, and predict the business development trend of a certain area based on the knowledge graph or automatically update the urban three-dimensional spatial business model based on real-time data. In addition, through edge computing and cloud computing technology, efficient processing and analysis of large-scale urban business data can be achieved to support real-time updating and reasoning of the knowledge graph. Summary of the invention

[0003] In order to solve the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide a commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data. The present invention can realize large-scale, high-precision and short-cycle intelligent optimization of urban three-dimensional space business formats.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data includes the following steps:

[0006] Step S1: Data collection. Using a backpack drone with GPS+GLONASS dual-mode positioning and a maximum flight altitude greater than 500m, equipped with a laser module with a measurement range of 3-1200m and a high-definition camera with a resolution greater than 20MP, collect building three-dimensional data (including building plan, building outline, building height and three-dimensional layer data) and urban open space data (including open space plan and open space outline), collect business distribution data and Internet celebrity facility data (including business type and location) through the API interface of the open map platform, and collect the latest urban transportation road network information and rail station data through the open transportation information platform. The data collected in the above manner is converted and integrated in a unified format to form a complete urban three-dimensional space data set.

[0007] Step S2: Classification construction and matching fusion of three-dimensional spatial knowledge graph. According to the physical proximity of buildings, the number of layers of functional formats, and the topological connection of traffic networks, the relationship graphs of space, function, and traffic are constructed respectively. The graph data is cleaned and the space-time network is aligned. Through the Blossom graph matching fusion algorithm, the augmented path and the contraction odd ring are recursively searched to achieve the maximum similarity matching of the relationship graph and form the urban three-dimensional spatial format knowledge graph.

[0008] Step S3: Sample graph matching and optimization spectrum construction of the area to be optimized. Define the graph of the area to be optimized, initialize the candidate matrix through the Ullmann subgraph isomorphism algorithm, and use recursive backtracking and pruning rules to find the subgraph isomorphism relationship between the two graphs, identify similar subgraph areas in the urban three-dimensional space business knowledge graph, and perform k-means clustering to form the business optimization spectrum. Calculate the score and screen the optimization spectrum based on the homogeneous competition distance and the number of passenger visits. Use a ground laser scanner to collect three-dimensional data of the business stores and Internet celebrity facilities in the optimization spectrum.

[0009] Step S4: Periodic prediction and iterative optimization of the number of tourists. Construct the GraphSAGE graph network prediction algorithm, train the case set model for 10,000 iterations, and predict the number of visitors for the optimized area. Every quarter, measure the number of visits to the business format through an infrared counter with a 95% accuracy within a detection range of 8m and a video crowd analysis device equipped with the Faster R-CNN target detection algorithm to determine whether the actual number is close to the predicted number. If not, repeat S3.

[0010] Step S5: Tourist tour route organization planning. Input tourist needs, use the Dijkstra shortest path algorithm to plan shopping destination-oriented streamlines to achieve the shortest tour time; use the Floyd-Warshall multi-hop path algorithm to plan leisure and entertainment roaming routes to achieve maximum spatial tour coverage.

[0011] Step S6: Virtual reality digital screen and mixed reality equipment display. The virtual reality digital screen is used to display the business layout and visitor flow planning, and Apple Vision Pro is used to display the layout of Internet celebrity facilities and store adjustments.

[0012] Furthermore, constructing the relationship map of space, function and transportation in S2 refers to importing the complete data set formed in S1 into the software ArcGis or ArcGisPro and generating models and data sets of urban space, business format and transportation respectively. The physical proximity relationship of buildings, urban open spaces and Internet celebrity facilities is determined by calculating the boundary distance between surface elements through the neighborhood analysis tool of ArcGis, including: proximity (boundary distance is 0-100m), close (boundary distance is 100m-500m) and far away (boundary distance is greater than 500m). Generate the plane and business format distribution position of each floor of the building, associate the distribution coordinates of each business format and the number of floors, view the element count and summary statistical data through ArcGis to analyze the number and type of business format distribution on each floor, and calculate the density distribution of the business format through kernel density analysis.

[0013] Furthermore, the atlas data in S2 is cleaned and aligned with the spatiotemporal network, which means that the data sets from three different sources, namely urban space, business format and transportation, are mapped to the WGS_1984_UTM_Zone_50N projection coordinate system using the ArcGis projection tool, and the converted data are spatially aligned to ensure that the geographical locations of all data sources on the map are consistent. The ArcGis map visualization tool is used to ensure that all geographical locations are correct, and if errors are found, they are corrected. Time data from different data sources are collected, including timestamps and time formats. The ArcGis conversion time field tool is used to convert the time formats of the three data sources, namely urban space, business format and transportation, into a supported unified format, and the ArcGis time zone (environment setting) tool is used to convert the timestamps of different data sources to a unified time scale of UTC+8 to ensure time consistency. For data collected at different times, the linear difference method is used to synchronize the exact time points in the data set. Finally, the time alignment results are checked to ensure that the timestamps are consistent, and if there are errors, they are corrected.

[0014] Furthermore, the S2 uses the Blossom graph matching fusion algorithm: 1. Graph structure construction: The data collected in S1 and S2 are modeled as a graph structure, where nodes represent different spatial entities (surface elements), including buildings, urban open spaces, Internet celebrity facilities, urban road networks and rail stations, and edges represent the physical proximity between nodes (surface elements). Each node and edge has additional area and functional type attributes. 2. Construct initial graph matching: Use the Blossom graph matching fusion algorithm to match and fuse data from different sources, construct the graph data into a subgraph and perform initial matching based on node attributes. 3. Augmenting paths and contracting odd rings: In the initial matching, find an augmenting path from an unmatched node to another unmatched node. If an odd ring is encountered when searching for an augmenting path, it is reduced to a single point, and the augmenting path is continued on the way after the reduction. 4. Optimize matching: After finding the augmenting path in the contraction graph, match and expand it in the original graph through the reverse operation, repeat the augmenting path and the contraction steps of the flower until the augmenting path is no longer found, and the maximum match is obtained.

[0015] Furthermore, in S3, the Ullmann subgraph isomorphism algorithm is used to initialize the candidate matrix and use recursive backtracking and pruning rules to find the subgraph isomorphism relationship between the two graphs. It refers to constructing a matching matrix M of size m×n, where m is the number of vertices of the pattern graph and n is the number of vertices of the target graph. The pattern graph vertex is i and the target graph vertex is j. If M[i][j]=1, it means that the pattern graph vertex i can be matched with the target graph vertex j. If M[i][j]=0, ​​it means that the pattern graph vertex i cannot be matched with the target graph vertex j. During initialization, all M[i][j] are set to 1, and impossible vertex pairings are screened out according to vertex attributes and edge attributes, and the matching matrix M is updated, and the impossible pairings are set to 0. The isomorphic mapping between the pattern graph and the target graph is gradually constructed by a recursive method. Specifically, for each unmatched pattern graph vertex i, the candidate matrix is ​​searched for its possible matching target graph vertex j, that is, the pairing that satisfies M[i][j]=1. Whenever a vertex pairing is determined, it is added to the current partial isomorphism mapping, and the next pattern graph vertex is processed recursively. If it is found at a certain step that the current pairing cannot meet the subgraph isomorphism condition (for example, the matching relationship of some edges does not hold in the target graph), it is backtracked to the previous step, the current pairing is canceled, and other possible pairing paths are tried. The pruning rules are applied to prune impossible matching paths, and whether to continue recursion is determined by checking in advance whether the current partial isomorphism mapping violates the necessary conditions for subgraph isomorphism.

[0016] Furthermore, the homogeneous competition distance and passenger flow visit quantity calculation score in S3 are as follows: 1) the functional relationship map in S2 that has been cleaned and aligned with the spatiotemporal network is converted into point feature data using the feature conversion tool in the ArcGIS data management tool to convert the business format store and Internet celebrity facility surface feature data, and the passenger flow visit quantity n is assigned to each point feature through the field calculation tool; 2) the business format nodes and Internet celebrity facilities of the same functional type are selected through the ArcGIS attribute selection tool and the distance between the feature points is calculated using the ArcGIS neighborhood analysis tool, that is, the homogeneous competition distance d; 3) the homogeneous competition distance d and the passenger flow visit quantity n are normalized by the formula to obtain the normalized homogeneous competition distance d norm and normalized passenger flow n norm , the calculation formula is as follows:

[0017] (1)

[0018] Where: d i : The original value of the i-th homogeneous competition distance in the data set; d min : The minimum value of all homogeneous competition distances in the data set; d max : The maximum value of all homogeneous competition distances in the data set; d norm,i : The normalized value of the i-th homogeneous competition distance; n i : The original value of the i-th passenger flow visit volume in the data set; n min : The minimum value of all passenger flow visits in the data set; n max : the maximum value of all passenger flow visits in the data set; n norm,i : The normalized value of the i-th passenger flow visit volume;

[0019] Use the normalized d norm and n norm Calculate the score a of each store using the following formula:

[0020] (2)a i =d norm,i ×n norm,i

[0021] Among them: a i : The rating of the i-th store;

[0022] And sum up the scores of all stores to get the total score A total and maximum score A max , the calculation formula is as follows:

[0023] (3)

[0024] (4)A max =max(a1,a2,…,a m);

[0025] Among them: A total : The total score of all stores; A total : The maximum value of the ratings among all stores; m: The total number of stores.

[0026] Furthermore, the GraphSAGE graph network prediction algorithm in S4 refers to constructing a multi-layer sampling aggregation (SAGEConv) neural network layer, each layer is defined by an input feature dimension, a hidden layer feature dimension and an output feature dimension, and the output features of each layer are used as the input of the next layer by sampling and aggregating the neighbor nodes in the graph. In the model initialization stage, the feature dimension of each layer is defined, the input feature dimension is X, the hidden layer feature dimension is H, the output feature dimension is Y, and the mean aggregation operation is used for each layer. The graph data set is divided into a training set, a validation set and a test set. The training set is used for model training, the validation set is used for model parameter adjustment, and the test set is used for model performance evaluation. The method for training the model includes defining a loss function and an optimizer, calculating the predicted value and loss function value of each node by forward propagation, and updating the model parameters by back propagation until a preset number of training rounds is reached or the early termination condition is met. The optimized pedigree and its feature data screened in S3 are added to the graph structure, and the adjacency matrix and feature matrix of the graph are updated. The trained GraphSAGE model is used to propagate the new node forward, calculate the representation vector of the new node, and input it into the regression layer to output the predicted visit volume of the new node. The regression layer calculates the dot product of the new node representation vector and the preset weight matrix to obtain the predicted visit volume value d of the new node. 预测i , sum up the visits of all new nodes to get the total amount d 预测总 .

[0027] Furthermore, the determination of whether the actual volume is close to the predicted volume in S4 refers to the total flow rate d of the entrance and exit of the area measured by an infrared counter with a detection range of 8m and a 95% accuracy every quarter after the business optimization. 实测总 , using the video flow analysis device equipped with the Faster R-CNN target detection algorithm to identify the daily passenger flow of the business node 实测i ; Compare the measured value with the predicted value. If the following conditions cannot be met at the same time, repeat step S3;

[0028] (5)

[0029] (6)0.6*d 预测总 -d 实测总 >0;

[0030] Where: d 实测i : the actual passenger flow of the ith business node; d 预计i: predicted passenger flow of the ith business node; m: the total number of business nodes; d 预测总 : The total predicted passenger flow of all business nodes; d 实测总 : The sum of actual passenger flow of all business nodes.

[0031] Furthermore, the Dijkstra shortest path algorithm in S5 plans the shopping destination-oriented streamline, assuming that there are n access nodes in the business knowledge graph, denoted as N = {n1, n2, ..., n n}, the connection relationship between nodes forms a directed graph G = (N, E)G = (N, E)G = (N, E), where E is the edge set, including the weight w of each edge ij , represents node n i To node n j Distance weights. First, initialize the distances of all nodes, set the distance of the starting node to 0, and the distances of other nodes to infinity, and set the predecessor node array. Then, select a node with the shortest current distance from the unvisited node set, remove it from the set, and update the distances and predecessor node information of all its neighbor nodes. Repeat this process until all nodes are visited or the shortest distance of the current node is infinite.

[0032] Furthermore, the Floyd-Warshall multi-hop path algorithm in S5 plans the leisure and entertainment roaming route, which means initializing a distance matrix D, where D[i][j] represents node n i To node n j The path weight of node n i and n j Directly connected, then D[i][j] is equal to the edge weight w ij , otherwise D[i][j]=∞, and for all nodes i, D[i][i]=0. Then, by introducing the intermediate node k, the shortest paths between all node pairs are gradually updated. The specific update formula is D[i][j]=min(D[i][j],D[i][k]+D[k][j]). By performing the above update operation on all nodes, the shortest path matrix is ​​output to ensure that tourists cover the maximum space and business formats during their leisure and entertainment roaming.

[0033] Furthermore, the mixed reality device in S6 displays the layout of Internet celebrity facilities and store adjustments, which refers to the display of three-dimensional data collected by the ground laser scanner in s3, and the visual fusion of spatial position and surface texture with the real scene through Apple Vision Pro, to assist owners in choosing business style, layout and surrounding facilities.

[0034] Beneficial effects:

[0035] 1. The present invention reduces the time and labor costs of urban business renewal schemes. Traditional urban business renewal schemes of 15-20 square kilometers require a team of 6-10 people to spend 4-6 weeks to collaborate and complete the design. The present invention can quickly generate and adjust multiple schemes through intelligent algorithms. Only 2-3 people are required to generate and adjust the scheme within 7 days, which significantly reduces the demand for human resources, improves the design efficiency of urban business renewal schemes, and reduces labor costs.

[0036] 2. The present invention improves the accuracy and predictability of urban business updates. The accuracy of data collected by intelligent collection equipment is 60%-70% higher than that of traditional manual collection. Open platform data can obtain the latest business data and road network information to ensure the comprehensiveness and accuracy of the data. Compared with traditional methods, the accuracy of business data is improved by 40%-50%, and the accuracy of road network data is improved by 50%-60%. The Blossom graph matching fusion algorithm is used to ensure the accuracy of the knowledge graph and the reliability of data analysis, which can be improved by 35%-45% compared with traditional methods. The GraphSAGE graph network algorithm is used to predict the number of visitors to the optimized area and ensure the reliability of the prediction model, which can be improved by 30% compared with traditional methods.

[0037] 3. The present invention expands the design scope and design depth of urban business renewal schemes. Through backpack drones and multi-source data integration, it can efficiently collect urban three-dimensional spatial data, urban business data and urban road network data of 15-20 square kilometers or even larger. High-precision data and short-cycle scheme generation and adjustment capabilities can help planners complete iterative optimization of design schemes more times within a limited time, and improve design depth while efficiently completing large-scale urban business renewal designs. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flow chart of the method of the present invention;

[0039] Figure 2 It is a schematic diagram of a three-dimensional spatial knowledge graph;

[0040] Figure 3 It is a schematic diagram of sample map matching and optimization spectrum of the area to be optimized. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] A commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data, such as Figure 1-3 As shown, the following steps are included:

[0043] Step 1: Data collection. Using a backpack drone with GPS+GLONASS dual-mode positioning and a maximum flight altitude greater than 500m, equipped with a laser module with a measurement range of 3-1200m and a high-definition camera with a resolution greater than 20MP, collect building 3D data (including building plan, building outline, building height and 3D layer data) and urban open space data (including open space plan and open space outline), collect business distribution data and popular facility data (including business type and location) through the API interface of the open map platform, and collect the latest urban transportation road network information and rail station data through the open transportation information platform. The data collected by the above method is converted and integrated in a unified format to form a complete urban 3D space data set.

[0044] Step 2: Classification construction and matching fusion of three-dimensional spatial knowledge graph. According to the physical proximity of buildings, the number of layers of functional formats, and the topological connection of traffic networks, the relationship graphs of space, function, and traffic are constructed respectively. The graph data is cleaned and the space-time network is aligned. Through the Blossom graph matching fusion algorithm, the augmented path and the contraction odd ring are recursively searched to achieve the maximum similarity matching of the relationship graph and form the urban three-dimensional spatial format knowledge graph.

[0045] The construction of the relationship map of space, function and transportation refers to importing the complete data set formed in step one into the software ArcGis or ArcGisPro and generating models and data sets of urban space, business format and transportation respectively. The physical proximity relationship of buildings, urban open spaces and Internet celebrity facilities is determined by calculating the boundary distance between surface elements through the neighborhood analysis tool of ArcGis, including: proximity (boundary distance is 0-100m), close (boundary distance is 100m-500m) and far away (boundary distance is greater than 500m). Generate the plane and business format distribution position of each floor of the building, associate the distribution coordinates of each business format and the number of floors, view the element count and summary statistical data through ArcGis to analyze the number and type of business format distribution on each floor, and calculate the density distribution of the business format through kernel density analysis.

[0046] The atlas data is cleaned and the space-time network is aligned, which means that the data sets from three different sources, urban space, business format and transportation, are mapped to the WGS_1984_UTM_Zone_50N projection coordinate system using the ArcGis projection tool, and the converted data are spatially aligned to ensure that the geographical locations of all data sources on the map are consistent. The ArcGis map visualization tool is used to ensure that all geographical locations are correct, and if errors are found, they are corrected. Time data from different data sources are collected, including timestamps and time formats. The ArcGis conversion time field tool is used to convert the time formats of the three data sources of urban space, business format and transportation into a supported unified format, and the ArcGis time zone (environment setting) tool is used to convert the timestamps of different data sources to a unified time scale of UTC+8 to ensure time consistency. For data collected at different times, the linear difference method is used to synchronize the exact time points in the data set. Finally, the time alignment results are checked to ensure that the timestamps are consistent, and if there are errors, they are corrected.

[0047] The Blossom graph matching fusion algorithm refers to 1. Graph structure construction: modeling the data collected in step one and step two into a graph structure, where nodes represent different spatial entities (surface elements), including buildings, urban open spaces, Internet celebrity facilities, urban road networks and rail stations, and edges represent the physical proximity between nodes (surface elements). Each node and edge has additional area and functional type attributes. 2. Constructing initial graph matching: using the Blossom graph matching fusion algorithm to match and fuse data from different sources, constructing the graph data into a subgraph and performing initial matching based on node attributes. 3. Augmenting paths and contracting odd rings: searching for an augmenting path from an unmatched node to another unmatched node in the initial matching. If an odd ring is encountered when searching for an augmenting path, it is reduced to a single point, and the augmenting path is continued on the way after the reduction. 4. Optimizing matching: After finding the augmenting path in the contraction graph, the matching expansion is performed in the original graph through the reverse operation, and the augmenting path and the contraction steps of the flower are repeated until the augmenting path is no longer found, and the maximum match is obtained.

[0048] Step 3: Sample graph matching and optimization spectrum construction of the area to be optimized. Define the graph of the area to be optimized, initialize the candidate matrix through the Ullmann subgraph isomorphism algorithm, and use recursive backtracking and pruning rules to find the subgraph isomorphism relationship between the two graphs, identify similar subgraph areas in the urban three-dimensional space business knowledge graph, and perform k-means clustering to form the business optimization spectrum. Calculate the score and screen the optimization spectrum based on the homogeneous competition distance and the number of passenger visits. Use a ground laser scanner to collect three-dimensional data of the business stores and Internet celebrity facilities in the optimization spectrum.

[0049] The method of using the Ullmann subgraph isomorphism algorithm to initialize the candidate matrix and use recursive backtracking and pruning rules to find the subgraph isomorphism relationship between the two graphs refers to constructing a matching matrix M of size m×n, where m is the number of vertices of the pattern graph and n is the number of vertices of the target graph. The pattern graph vertex is i and the target graph vertex is j. If M[i][j]=1, it means that the pattern graph vertex i can be matched with the target graph vertex j. If M[i][j]=0, it means that the pattern graph vertex i cannot be matched with the target graph vertex j. During initialization, all M[i][j] are set to 1, and impossible vertex pairings are screened out according to vertex attributes and edge attributes, and the matching matrix M is updated, and the impossible pairings are set to 0. The isomorphic mapping between the pattern graph and the target graph is gradually constructed by a recursive method. Specifically, for each unmatched pattern graph vertex i, the candidate matrix is ​​searched for its possible matching target graph vertex j, that is, the pairing that satisfies M[i][j]=1. Whenever a vertex pairing is determined, it is added to the current partial isomorphism mapping, and the next pattern graph vertex is processed recursively. If it is found at a certain step that the current pairing cannot meet the subgraph isomorphism condition (for example, the matching relationship of some edges does not hold in the target graph), it is backtracked to the previous step, the current pairing is canceled, and other possible pairing paths are tried. The pruning rules are applied to prune impossible matching paths, and whether to continue recursion is determined by checking in advance whether the current partial isomorphism mapping violates the necessary conditions for subgraph isomorphism.

[0050] The calculation score of homogeneous competition distance and passenger flow visit quantity refers to 1. Using the feature conversion tool in the ArcGIS data management tool to convert the functional relationship map of the business format stores and Internet celebrity facilities into point feature data after data cleaning and spatiotemporal network alignment in step 2, and assigning the passenger flow visit quantity n to each point feature through the field calculation tool. 2. Using the ArcGIS attribute screening tool to select business format nodes and Internet celebrity facilities of the same functional type and using the ArcGIS neighborhood analysis tool to calculate the distance between feature points, that is, the homogeneous competition distance d. 3. Normalizing the homogeneous competition distance d and the passenger flow visit quantity n through the formula to obtain d norm and n norm , the calculation formula is as follows:

[0051]

[0052] Where: d i : The original value of the i-th homogeneous competition distance in the data set; d min : The minimum value of all homogeneous competition distances in the data set; d max : The maximum value of all homogeneous competition distances in the data set; d norm,i : The normalized value of the i-th homogeneous competition distance; n i : The original value of the i-th passenger flow visit volume in the data set; nmin : The minimum value of all passenger flow visits in the data set; n max : the maximum value of all passenger flow visits in the data set; n norm,i : The normalized value of the i-th passenger flow visit volume.

[0053] Use the normalized d norm and n norm Calculate the score a of each store using the following formula:

[0054] (2)a i =d norm,i ×n norm,i

[0055] Among them: a i : The rating of the i-th store.

[0056] And sum up the scores of all stores to get the total score A total and maximum score A max , the calculation formula is as follows:

[0057] (3)

[0058] (4)A max =max(a1,a2,…,a m ).

[0059] Among them: A total : The total score of all stores; A total : The maximum value of the ratings among all stores; m: The total number of stores.

[0060] Step 4: Periodic prediction and iterative optimization of the number of tourists. Construct the GraphSAGE graph network prediction algorithm, train the case set model for 10,000 iterations, and predict the number of visitors for the optimized area. Every quarter, measure the number of visits to the business format through an infrared counter with a 95% accuracy within a detection range of 8m and a video crowd analysis device equipped with the Faster R-CNN target detection algorithm to determine whether the actual number is close to the predicted number. If not, repeat the third step.

[0061] The GraphSAGE graph network prediction algorithm refers to constructing a multi-layer sampling aggregation (SAGEConv) neural network layer, each layer is defined by the input feature dimension, the hidden layer feature dimension and the output feature dimension, and the output features of each layer are used as the input of the next layer by sampling and aggregating the neighbor nodes in the graph. In the model initialization stage, the feature dimension of each layer is defined, the input feature dimension is X, the hidden layer feature dimension is H, the output feature dimension is Y, and the mean aggregation operation is used for each layer. The graph data set is divided into a training set, a validation set and a test set. The training set is used for model training, the validation set is used for model parameter adjustment, and the test set is used for model performance evaluation. The method for training the model includes defining a loss function and an optimizer, calculating the predicted value and loss function value of each node by forward propagation, and backpropagating to update the model parameters until the preset number of training rounds is reached or the early termination condition is met. The optimized pedigree and its feature data selected in step three are added to the graph structure, and the adjacency matrix and feature matrix of the graph are updated. The trained GraphSAGE model is used to propagate the new node forward, calculate the representation vector of the new node, and input it into the regression layer to output the predicted visit volume of the new node. The regression layer calculates the dot product of the new node representation vector and the preset weight matrix to obtain the predicted visit volume value d of the new node. 预测i , sum up the visits of all new nodes to get the total amount d 预测总 .

[0062] The judgment of whether the actual volume is close to the predicted volume refers to the total flow rate d at the entrance and exit of the area measured every quarter after the business optimization by an infrared counter with a 95% accuracy within a detection range of 8m. 实测总 , using the video flow analysis device equipped with the Faster R-CNN target detection algorithm to identify the daily passenger flow of the business node 实测i . Compare the measured value with the predicted value. If the following conditions cannot be met at the same time, repeat the third step.

[0063] (5)

[0064] (6)0.6*d 预测总 -d 实测总 >0.

[0065] Where: d 实测i : the actual passenger flow of the ith business node; d 预计i : predicted passenger flow of the ith business node; m: the total number of business nodes; d 预测总 : The total predicted passenger flow of all business nodes; d 实测总 : The sum of actual passenger flow of all business nodes.

[0066] Step 5: Tourist tour route organization planning. Input tourist needs, use the Dijkstra shortest path algorithm to plan shopping destination-oriented streamlines to achieve the shortest tour time; use the Floyd-Warshall multi-hop path algorithm to plan leisure and entertainment roaming routes to achieve maximum spatial tour coverage.

[0067] The Dijkstra shortest path algorithm plans the shopping destination-oriented streamline, assuming that there are n access nodes in the business knowledge graph, denoted as N = {n1, n2, ..., n n}, the connection relationship between nodes forms a directed graph G = (N, E)G = (N, E)G = (N, E), where E is the edge set, including the weight w of each edge ij , represents node n i To node n j Distance weights. First, initialize the distances of all nodes, set the distance of the starting node to 0, and the distances of other nodes to infinity, and set the predecessor node array. Then, select a node with the shortest current distance from the unvisited node set, remove it from the set, and update the distances and predecessor node information of all its neighbor nodes. Repeat this process until all nodes are visited or the shortest distance of the current node is infinite.

[0068] The Floyd-Warshall multi-hop path algorithm plans the leisure and entertainment roaming route, which is to initialize a distance matrix D, where D[i][j] represents the node n i To node n j The path weight of node n i and n j Directly connected, then D[i][j] is equal to the edge weight w ij , otherwise D[i][j]=∞, and for all nodes i, D[i][i]=0. Then, by introducing the intermediate node k, the shortest paths between all node pairs are gradually updated. The specific update formula is D[i][j]=min(D[i][j],D[i][k]+D[k][j]). By performing the above update operation on all nodes, the shortest path matrix is ​​output to ensure that tourists cover the maximum space and business formats during their leisure and entertainment roaming.

[0069] Step 6: Virtual reality digital screen and mixed reality equipment display. Use virtual reality digital screen to display business layout and visitor flow planning, and use Apple Vision Pro to display the layout of Internet celebrity facilities and store adjustments.

[0070] The mixed reality device displays the layout of Internet celebrity facilities and store adjustments, which refers to the display of three-dimensional data collected by the ground laser scanner in the s3, and the visual fusion of spatial position and surface texture with the real scene through Apple Vision Pro, to assist owners in choosing business style, layout and surrounding facilities.

[0071] The technical solution of the present invention will be described in detail below by taking the urban design of a certain area in a certain city as an example.

[0072] The specific implementation is as follows:

[0073] (1) Data collection. Taking a certain area of ​​a city as the target city, a backpack drone with GPS+GLONASS dual-mode positioning and a maximum flight altitude of more than 500m is used, equipped with a laser module with a measurement range of 3-1200m and a high-definition camera with a resolution of more than 20MP, to collect three-dimensional building data (including building plan, building outline, building height and three-dimensional number of floors) and urban open space data (including open space plan and open space outline). Business distribution data and Internet celebrity facility data (including business type and location) are collected through the API interface of the open map platform, and the latest urban transportation road network information and rail station data are collected through the open transportation information platform. The data collected in the above way are converted and integrated through a unified format to form a complete urban three-dimensional space data set. The three types of online consumption data and urban population data are obtained to form an actual online consumption data set of the target city, and the digital space sandbox model is constructed using the data set.

[0074] (2) Classification construction and matching fusion of three-dimensional spatial knowledge graphs. According to the physical proximity of buildings, the number of layers of functional formats, and the topological connection of traffic networks, spatial, functional, and traffic relationship graphs are constructed respectively. The graph data is cleaned and the spatiotemporal network is aligned. Through the Blossom graph matching fusion algorithm, the augmented paths and contracted odd rings are recursively searched to achieve the maximum similarity matching of the relationship graph and form a knowledge graph of urban three-dimensional spatial formats. Specifically, it includes:

[0075] (2.1) Import the complete data set generated in step 1 into the software ArcGis or ArcGisPro and generate models and data sets for urban space, business types and transportation respectively. The boundary distance between surface elements is calculated through the neighborhood analysis tool of ArcGis to determine the physical proximity relationship between buildings, urban open spaces and popular facilities, including: proximity (boundary distance is 0-100m), close (boundary distance is 100m-500m) and far away (boundary distance is greater than 500m). Generate the plane and business type distribution position of each floor of the building, associate the coordinates of each business type distribution and the number of floors, view the element count and summary statistical data through ArcGis to analyze the number and type of business type distribution on each floor, and calculate the density distribution of business types through kernel density analysis.

[0076] (2.2) Use the ArcGis projection tool to map the datasets from three different sources, namely urban space, business and transportation, to the WGS_1984_UTM_Zone_50N projection coordinate system, align the converted data spatially, ensure that the geographical locations of all data sources on the map are consistent, and use the ArcGis map visualization tool to ensure that all geographical locations are correct. If errors are found, correct them. Collect time data from different data sources, including timestamps and time formats. Use the ArcGis Convert Time Field tool to convert the time formats of the three data sources, namely urban space, business and transportation, to a supported unified format, and use the ArcGis Time Zone (Environment Settings) tool to convert the timestamps of different data sources to a unified time scale of UTC+8 to ensure time consistency. For data collected at different times, use the linear difference method to synchronize the exact time points in the dataset. Finally, check the time alignment results to ensure that the timestamps are consistent and correct them if errors exist.

[0077] (2.3) Through 1. Graph structure construction: The data collected in steps 1 and 2 are modeled as a graph structure, where nodes represent different spatial entities (surface elements), including buildings, urban open spaces, popular facilities, urban road networks and rail stations, and edges represent the physical proximity between nodes (surface elements). Each node and edge has additional area and functional type attributes. 2. Construct initial graph matching: Use the Blossom graph matching fusion algorithm to match and fuse data from different sources, construct the graph data into a subgraph and perform initial matching based on node attributes. 3. Augmenting paths and contracting odd rings: In the initial matching, find an augmenting path from an unmatched node to another unmatched node. If an odd ring is encountered when searching for an augmenting path, it is reduced to a single point and the augmenting path is continued on the way after the reduction. 4. Optimize matching: After finding the augmenting path in the contraction graph, match and expand it in the original graph through the reverse operation, repeat the augmenting path and the contraction steps of the flower until the augmenting path is no longer found, and the maximum match is obtained. Finally, a knowledge graph of urban three-dimensional space formats is formed.

[0078] (3) Sample graph matching and optimization spectrum construction of the area to be optimized. Define the graph of the area to be optimized, initialize the candidate matrix through the Ullmann subgraph isomorphism algorithm, and use recursive backtracking and pruning rules to find the subgraph isomorphism relationship between the two graphs, identify similar subgraph areas in the urban three-dimensional space business knowledge graph, and perform k-means clustering to form the business optimization spectrum. Calculate the score and screen the optimization spectrum based on the homogeneous competition distance and the number of customer visits. Use a ground laser scanner to collect three-dimensional data of the business stores and Internet celebrity facilities in the optimization spectrum. Specifically include:

[0079] (3.1) Initialize the candidate matrix and construct a matching matrix M of size m×n, where m is the number of vertices in the pattern graph and n is the number of vertices in the target graph. The pattern graph vertex is i and the target graph vertex is j. If M[i][j]=1, it means that the pattern graph vertex i can be matched with the target graph vertex j. If M[i][j]=0, ​​it means that the pattern graph vertex i cannot be matched with the target graph vertex j. During initialization, all M[i][j] are set to 1. According to the vertex attributes and edge attributes, the impossible vertex pairings are screened out, and the matching matrix M is updated, and the impossible pairings are set to 0. The isomorphic mapping between the pattern graph and the target graph is gradually constructed by a recursive method. Specifically, for each unmatched pattern graph vertex i, its possible matching target graph vertex j is searched in the candidate matrix, that is, the pairing that satisfies M[i][j]=1. Whenever a vertex pairing is determined, it is added to the current partial isomorphic mapping, and the next pattern graph vertex is recursively processed. If it is found at a certain step that the current pairing cannot satisfy the subgraph isomorphism condition (for example, the matching relationship of some edges does not hold in the target graph), then go back to the previous step, cancel the current pairing, and try other possible pairing paths. Apply pruning rules to prune impossible matching paths, and decide whether to continue recursion by checking in advance whether the current partial isomorphism mapping violates the necessary conditions for subgraph isomorphism.

[0080] (3.2) Use the Feature to Point tool in the ArcGIS data management tool to convert the functional relationship map of the business stores and Internet celebrity facilities in step 2 into point feature data, and assign the number of customer visits n to each point feature through the field calculation tool. 2. Use the ArcGIS attribute selection tool to select business nodes and Internet celebrity facilities of the same functional type and use the ArcGIS neighborhood analysis tool to calculate the distance between feature points, that is, the homogeneous competition distance d. 3. Use the formula to normalize the homogeneous competition distance d and the number of customer visits n to obtain d norm and n norm , the calculation formula is as follows:

[0081] (1)

[0082] Where: d i : The original value of the i-th homogeneous competition distance in the data set; d min : The minimum value of all homogeneous competition distances in the data set; d max : The maximum value of all homogeneous competition distances in the data set; d norm,i : The normalized value of the i-th homogeneous competition distance; n i : The original value of the i-th passenger flow visit volume in the data set; n min : The minimum value of all passenger flow visits in the data set; n max : the maximum value of all passenger flow visits in the data set; n norm,i : The normalized value of the i-th passenger flow visit volume.

[0083] Use the normalized d norm and n norm Calculate the score a of each store using the following formula:

[0084] (2)a i =d norm,i ×n norm,i

[0085] Among them: a i : The rating of the i-th store.

[0086] And sum up the scores of all stores to get the total score A total and maximum score A max , the calculation formula is as follows:

[0087] (3)

[0088] (4)A max =max(a1,a2,…,a m ).

[0089] Among them: A total : The total score of all stores; A total : The maximum value of the ratings among all stores; m: The total number of stores.

[0090] (4) Periodic prediction and iterative optimization of the number of tourists. Construct the GraphSAGE graph network prediction algorithm, iterate the case set model training 10,000 times, and predict the visitor volume of the optimized area. Every quarter, measure the business visits through an infrared counter with a 95% accuracy within a detection range of 8m and a video crowd analysis device equipped with the FasterR-CNN target detection algorithm to determine whether the actual volume is close to the predicted volume. If not, repeat the third step. Specifically include:

[0091] (4.1) Construct a multi-layer sampling aggregation (SAGEConv) neural network layer. Each layer is defined by the input feature dimension, the hidden layer feature dimension, and the output feature dimension. By sampling and aggregating the neighbor nodes in the graph, the output features of each layer are used as the input of the next layer. In the model initialization stage, the feature dimension of each layer is defined. The input feature dimension is X, the hidden layer feature dimension is H, and the output feature dimension is Y. Each layer uses the mean aggregation operation. The graph dataset is divided into a training set, a validation set, and a test set. The training set is used for model training, the validation set is used for model parameter adjustment, and the test set is used for model performance evaluation. The method for training the model includes defining a loss function and an optimizer, calculating the predicted value and loss function value of each node through forward propagation, and updating the model parameters through back propagation until the preset number of training rounds is reached or the early termination condition is met. The optimized pedigree and its feature data selected in S3 are added to the graph structure, and the adjacency matrix and feature matrix of the graph are updated. The trained GraphSAGE model is used to propagate the new node forward, calculate the representation vector of the new node, and input it into the regression layer to output the predicted visit volume of the new node. The regression layer calculates the dot product of the new node representation vector and the preset weight matrix to obtain the predicted visit volume value d of the new node. 预测i , sum up the visits of all new nodes to get the total amount d 预测总 .

[0092] (4.2) After the business optimization, the total flow rate d at the entrance and exit of the area is calculated every quarter by using an infrared counter with a detection range of 8m and a 95% accuracy. 实测总 , using the video flow analysis device equipped with the Faster R-CNN target detection algorithm to identify the daily passenger flow of the business node 实测i . Compare the measured value with the predicted value. If the following conditions cannot be met at the same time, repeat the third step.

[0093] (5)

[0094] (6)0.6*d 预测总 -d 实测总 >0.

[0095] Where: d 实测i : the actual passenger flow of the ith business node; d 预计i : predicted passenger flow of the ith business node; m: the total number of business nodes; d 预测总 : The total predicted passenger flow of all business nodes; d 实测总 : The sum of actual passenger flow of all business nodes.

[0096] (5) Tourist tour route organization planning. Input tourist needs and use the Dijkstra shortest path algorithm to plan shopping destination-oriented streamlines to achieve the shortest tour time; use the Floyd-Warshall multi-hop path algorithm to plan leisure and entertainment roaming routes to achieve maximum spatial tour coverage. Specifically include:

[0097] (5.1) Assume that there are n access nodes in the business knowledge graph, denoted by N = {n1, n2, …, n n}, the connection relationship between nodes forms a directed graph G = (N, E)G = (N, E)G = (N, E), where E is the edge set, including the weight w of each edge ij , represents node n i To node n j Distance weights. First, initialize the distances of all nodes, set the distance of the starting node to 0, and the distances of other nodes to infinity, and set the predecessor node array. Then, select a node with the shortest current distance from the unvisited node set, remove it from the set, and update the distances and predecessor node information of all its neighbor nodes. Repeat this process until all nodes are visited or the shortest distance of the current node is infinite.

[0098] (5.2) Initialize a distance matrix D, where D[i][j] represents node n i To node n j The path weight of node n i and n j Directly connected, then D[i][j] is equal to the edge weight w ij , otherwise D[i][j]=∞, and for all nodes i, D[i][i]=0. Then, by introducing the intermediate node k, the shortest paths between all node pairs are gradually updated. The specific update formula is D[i][j]=min(D[i][j],D[i][k]+D[k][j]). By performing the above update operation on all nodes, the shortest path matrix is ​​output to ensure that tourists cover the maximum space and business formats during their leisure and entertainment roaming.

[0099] (6) Virtual reality digital screens and mixed reality equipment displays. Virtual reality digital screens are used to display business layout and visitor flow planning, and Apple Vision Pro is used to display the layout of popular facilities and store adjustments.

[0100] (6.1) Display the 3D data collected by the ground laser scanner in step 3, and use Apple VisionPro to visually integrate the spatial position and surface texture with the real scene to assist the owner in choosing the business style, layout and surrounding facilities.

Claims

1. A commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data, characterized by: Step S1: Data collection: using a backpack drone with GPS+GLONASS dual-mode positioning and a maximum flight altitude greater than 500m, equipped with a laser module with a measurement range of 3-1200m and a high-definition camera with a resolution greater than 20MP, to collect 3D building data and urban open space data, collect business distribution data and popular facilities data through the API interface of the open map platform, and collect the latest urban traffic network information and rail station data through the open traffic information platform; the data collected by the above methods are converted and integrated in a unified format to form a complete urban 3D space data set; Step S2: Classification construction and matching fusion of three-dimensional spatial knowledge graphs; construct spatial, functional and traffic relationship graphs respectively according to the physical proximity of buildings, the number of functional business relationships and the topological connection of traffic networks; clean the graph data and align the spatiotemporal network, and recursively search for augmented paths and contracted odd rings through the Blossom graph matching fusion algorithm to achieve the maximum similarity matching of the relationship graph and form a knowledge graph of urban three-dimensional spatial business formats; Step S3: Sample graph matching and optimization spectrum construction of the area to be optimized; delineate the graph of the area to be optimized, initialize the candidate matrix through the Ullmann subgraph isomorphism algorithm, and use recursive backtracking and pruning rules to find the subgraph isomorphism relationship between the two graphs, identify similar subgraph areas in the urban three-dimensional space business knowledge graph, and perform k-means clustering to form the business optimization spectrum; calculate the score and screen the optimization spectrum according to the homogeneous competition distance and the number of passenger visits; use a ground laser scanner to collect the three-dimensional data of the business stores and Internet celebrity facilities in the optimization spectrum; Step S4: Periodic prediction and iterative optimization of the number of tourists; construct the GraphSAGE graph network prediction algorithm, train the case set model for 10,000 iterations, and predict the number of visitors to the optimized area; every quarter, measure the number of business visits through an infrared counter with a 95% accuracy within a detection range of 8m and a video crowd flow analysis device equipped with the Faster R-CNN target detection algorithm to determine whether the actual number is close to the predicted number. If not, repeat the third step; Step S5: Tourist tour route organization planning: input tourist needs, plan shopping destination-oriented streamlines through Dijkstra shortest path algorithm to achieve the shortest tour time; plan leisure and entertainment roaming routes through Floyd-Warshall multi-hop path algorithm to achieve maximum spatial tour coverage; Step S6: Display of virtual reality digital screen and mixed reality equipment; display of business layout and visitor flow planning through virtual reality digital screen, and display of Internet celebrity facility layout and store adjustments using Apple Vision Pro.

2. The commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data according to claim 1 is characterized in that: The construction of the relationship map of space, function and transportation in S2 refers to importing the complete data set formed in S1 into the software ArcGis or ArcGisPro and generating models and data sets of urban space, business format and transportation respectively; calculating the boundary distance between surface elements through the neighborhood analysis tool of ArcGis to determine the physical proximity relationship of buildings, urban open spaces and popular facilities, including: proximity, closeness and distance; generating the plane and business format distribution position of each floor of the building, associating the distribution coordinates of each business format and the number of floors, analyzing the number and type of business format distribution on each floor through viewing the feature count and summary statistical data of ArcGis, and calculating the density distribution of the business format through kernel density analysis; The atlas data in S2 is cleaned and the space-time network is aligned, which means that the data sets from three different sources, namely urban space, business format and transportation, are mapped to the WGS_1984_UTM_Zone_50N projection coordinate system using the ArcGis projection tool, and the converted data are spatially aligned to ensure that the geographical locations of all data sources on the map are consistent, and the ArcGis map visualization tool is used to ensure that all geographical locations are correct. If errors are found, they are corrected; time data from different data sources are collected, including timestamps and time formats; the ArcGis time field conversion tool is used to convert the time formats of the three data sources, namely urban space, business format and transportation, into a supported unified format, and the ArcGis time zone tool is used to convert the timestamps of different data sources to a unified time scale of UTC+8 to ensure time consistency; for data collected at different times, the linear difference method is used to synchronize the actual time points in the data set; finally, the time alignment results are checked to ensure that the timestamps are consistent, and errors are corrected if any.

3. The commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data according to claim 2 is characterized in that: In S2, the Blossom graph matching fusion algorithm is used: 1) Graph structure construction: the data collected in S1 and S2 are modeled as a graph structure, in which nodes represent different spatial entities, including buildings, urban open spaces, popular facilities, urban road networks and rail stations, and edges represent the physical proximity between nodes. Each node and edge is attached with area and functional type attributes; 2) Initial graph matching is constructed: Blossom is used to m The graph matching fusion algorithm matches and fuses data from different sources, constructs the graph data into subgraphs and performs initial matching based on node attributes; 3) Augmenting paths and shrinking odd rings: In the initial matching, an augmenting path from an unmatched node to another unmatched node is searched. If an odd ring is encountered when searching for an augmenting path, it is reduced to a single point and the augmenting path is continued to be searched on the way after the reduction; 4) Optimized matching: After finding the augmenting path in the shrinking graph, the matching and expansion are performed in the original graph through the reverse operation, and the shrinking steps of the augmenting path and the flower are repeated until the augmenting path is no longer found, and the maximum match is obtained.

4. The commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data according to claim 3 is characterized in that: In the S3, the Ullmann subgraph isomorphism algorithm is used to initialize the candidate matrix and use recursive backtracking and pruning rules to find the subgraph isomorphism relationship between the two graphs; it refers to constructing a matching matrix M of size m×n, where m is the number of vertices of the pattern graph and n is the number of vertices of the target graph; the pattern graph vertex is i and the target graph vertex is j. If M[i][j]=1, it means that the pattern graph vertex i is matched with the target graph vertex j, and if M[i][j]=0, it means that the pattern graph vertex i cannot be matched with the target graph vertex j; during initialization, all M[i][j] are set to 1, and impossible vertex pairings are screened out according to vertex attributes and edge attributes, and the matching matrix M is updated, and the impossible pairings are set to 0; the isomorphic mapping between the pattern graph and the target graph is gradually constructed by a recursive method; for each unmatched pattern graph vertex i, the candidate matrix is ​​searched for its possible matching target graph vertex j, that is, the pairing that satisfies M[i][j]=1; whenever a vertex pairing is determined, it is added to the current partial isomorphism mapping, and the next pattern graph vertex is recursively processed; If it is found at a certain step that the current pairing cannot satisfy the subgraph isomorphism condition, then go back to the previous step, cancel the current pairing, and try other possible pairing paths; apply pruning rules to prune impossible matching paths, and decide whether to continue recursion by checking in advance whether the current partial isomorphism mapping violates the necessary conditions for subgraph isomorphism.

5. The commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data according to claim 4 is characterized in that: The homogeneous competition distance and passenger flow visit number calculation score in S3 are as follows: 1) the functional relationship map in S2 that has been cleaned and aligned with the spatiotemporal network is converted into point feature data using the feature conversion tool in the ArcGIS data management tool, and the passenger flow visit number n is assigned to each point feature through the field calculation tool; 2) the business nodes and Internet celebrity facilities of the same functional type are selected through the ArcGIS attribute selection tool, and the distance between the feature points, i.e., the homogeneous competition distance d, is calculated using the ArcGIS neighborhood analysis tool; 3) The homogeneous competition distance d and the passenger flow n are normalized by the formula to obtain the normalized homogeneous competition distance d norm and normalized passenger flow n norm , the calculation formula is as follows: (1) Where: d i : The original value of the i-th homogeneous competition distance in the data set; d min : The minimum value of all homogeneous competition distances in the data set; d max : The maximum value of all homogeneous competition distances in the data set; d norm,i : The normalized value of the i-th homogeneous competition distance; n i : The original value of the i-th passenger flow visit volume in the data set; n min : The minimum value of all passenger flow visits in the data set; n max : the maximum value of all passenger flow visits in the data set; n norm，i : The normalized value of the i-th passenger flow visit volume; Use the normalized d norm and n norm Calculate the score a of each store using the following formula: (2)a i =d norm,i ×n norm,i Among them: a i : The rating of the i-th store; And sum up the scores of all stores to get the total score A total and maximum score A max , the calculation formula is as follows: (3) (4)A max =max(a1,a2,...,a m ); Among them: A total : The total score of all stores; A total : The maximum value of the ratings among all stores; m: The total number of stores.

6. The commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data according to claim 5 is characterized in that: The GraphSAGE graph network prediction algorithm in S4 refers to constructing a multi-layer sampling aggregation neural network layer, each layer is defined by an input feature dimension, a hidden layer feature dimension and an output feature dimension, and the output features of each layer are used as the input of the next layer by sampling and aggregating neighbor nodes in the graph; in the model initialization stage, the feature dimension of each layer is defined, the input feature dimension is X, the hidden layer feature dimension is H, the output feature dimension is Y, and the mean aggregation operation is used for each layer; the graph data set is divided into a training set, a validation set and a test set, the training set is used for model training, the validation set is used for model parameter adjustment, and the test set is used for model performance evaluation; training The method of training the model includes defining the loss function and optimizer, calculating the predicted value and loss function value of each node through forward propagation, and updating the model parameters through back propagation until the preset number of training rounds is reached or the early termination condition is met; adding the optimized pedigree and its feature data selected in S3 to the graph structure, updating the adjacency matrix and feature matrix of the graph; forward propagating the new node through the trained GraphSAGE model, calculating the representation vector of the new node, and inputting it into the regression layer, outputting the predicted visit volume of the new node, and the regression layer calculates the dot product of the new node representation vector and the preset weight matrix to obtain the predicted visit volume value d of the new node. 预测i , sum up the visits of all new nodes to get the total amount d 预测总 .

7. The commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data according to claim 6 is characterized in that: The judgment in S4 whether the actual volume is close to the predicted volume refers to the total flow d of the entrance and exit of the area measured every quarter by an infrared counter with 95% accuracy within a detection range of 8m after the business optimization. 实测总 , using the video flow analysis device equipped with the Faster R-CNN target detection algorithm to identify the daily passenger flow of the business node 实测i ; Compare the measured value with the predicted value. If the following conditions cannot be met at the same time, repeat step S3; (5) (6)0.6*d 预测总 -d 实测总 >0; Where: d 实测i : the actual passenger flow of the ith business node; d 预计i : predicted passenger flow of the ith business node; m: the total number of business nodes; d 预测总 : The total predicted passenger flow of all business nodes; d 实测总 : The sum of actual passenger flow of all business nodes.

8. The commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data according to claim 7 is characterized in that: The Dijkstra shortest path algorithm in S5 plans the shopping destination-oriented streamline, assuming that there are n access nodes in the business knowledge graph, denoted as N = {n1, n2, ..., n n }, the connection relationship between nodes forms a directed graph G = (N, E)G = (N, E)G = (N, E), where E is the edge set, including the weight w of each edge ij , represents node n i To node n j Distance weight; first, initialize the distances of all nodes, set the distance of the starting node to 0, and the distances of other nodes to infinity, and set the predecessor node array; then, select a node with the shortest current distance from the unvisited node set, remove it from the set, and update the distances and predecessor node information of all its neighbor nodes; repeat this process until all nodes are visited or the shortest distance of the current node is infinite.

9. The commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data according to claim 8 is characterized in that: The Floyd-Warshall multi-hop path algorithm in S5 plans the leisure and entertainment roaming route, which means initializing a distance matrix D, where D[i][j] represents node n i To node n j The path weight of node n i and n j Directly connected, then D[i][j] is equal to the edge weight w ij , otherwise D[i][j]=∞, and for all nodes i, D[i][i]=0; by introducing the intermediate node k, the shortest paths between all node pairs are gradually updated. The specific update formula is D[i][j]=min(D[i][j], D[i][k]+D[k][j]); the above update operation is performed on all nodes to obtain the shortest path matrix, ensuring that tourists cover the maximum space and business formats during their leisure and entertainment roaming.

10. The commercial space intelligent optimization method based on comprehensive knowledge graph and tourist data according to claim 9 is characterized in that: The mixed reality device in S6 displays the layout of Internet celebrity facilities and store adjustments, which refers to the display of three-dimensional data collected by the ground laser scanner in S3, and the visual fusion of spatial position and surface texture with the real scene through Apple Vision Pro, to assist owners in choosing business style, layout and surrounding facilities.

Images