Water and electricity immigrant placement and site selection optimization method, system and device based on knowledge graph

Abstract

The invention provides a water and electricity immigrant resettlement site selection optimization method, system and device based on a knowledge graph, and belongs to the technical field of immigrant resettlement. According to the method, multi-source heterogeneous data indexes are integrated, a spatial suitability index, an economic cost index, a livelihood recovery index and a sociality index are introduced, the indexes are mapped into a structured entity-relation network, and node feature propagation and aggregation of candidate placement points are performed through a HeteroData heterogeneous graph neural network model; potential association between candidate settlement points and multiple types of entities is mined, complex condition filtering and path reasoning are supported, and a scientific, explainable and easy-to-expand immigrant settlement point site selection scheme can be generated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of immigrant resettlement, and particularly relates to a hydropower immigrant resettlement site selection optimization method, system and device based on a knowledge graph. BACKGROUND

[0002] While large and medium-sized hydropower projects continue to expand to alleviate energy shortages, they have also generated a large demand for resettlement.

[0003] Traditional commercial site selection methods usually consider spatial suitability for multi-index superposition analysis and rely on expert experience to determine the weights of each index and comprehensive evaluation. Hydropower resettlement needs to consider balancing spatial suitability and the integrity of social networks to ensure the sustainable recovery of the livelihoods of the immigrant population. At present, there is a lack of consideration of social indicators and livelihood recovery indicators, as well as quantitative analysis of the willingness to move, and the site selection evaluation cannot form a traceable and interpretable evaluation logic, resulting in a lack of transparent decision-making basis for site selection results. SUMMARY

[0004] In view of the deficiencies in the prior art, the present application provides a hydropower immigrant resettlement site selection optimization method, system and device based on a knowledge graph, which integrates multi-source heterogeneous data indicators, introduces spatial suitability indicators, economic cost indicators, livelihood recovery indicators and social indicators, maps them into a structured entity-relation network, propagates and aggregates the node features of candidate resettlement points through a HeteroData heterogeneous graph neural network model, and mines the potential associations between candidate resettlement points and multiple entities, thereby generating a scientific, interpretable and easily expandable immigrant resettlement site selection scheme.

[0005] To achieve the above technical purposes, the present application adopts the following technical solutions: The present application provides a hydropower immigrant resettlement site selection optimization method based on a knowledge graph, comprising the following steps: S1, constructing a multi-source heterogeneous data indicator system, wherein the indicator system comprises spatial suitability indicators, economic cost indicators, livelihood recovery indicators and social indicators; S2, collecting and preprocessing data related to the resettlement point and the indicator system, arranging the indicator system into triples of entities, attributes and relationships, and generating attribute tables and relationship tables of resettlement point nodes corresponding to the entities; S3, importing the entities, attribute tables and relationship tables into a graph database to construct a knowledge graph; S4, deriving entity and relationship information from the knowledge graph to construct HeteroData heterogeneous graph data; S5. Design a HeteroData heterogeneous graph neural network model based on HeteroData heterogeneous graph data. Use the relevant data of the indicator system of the established resettlement sites and the candidate resettlement site labels to train the neural network model, perform feature propagation and aggregation of candidate resettlement site nodes, explore the potential relationship between candidate resettlement sites and multiple types of entities, obtain the weight matrix and the fusion feature of candidate resettlement site nodes, and then obtain the trained HeteroData heterogeneous graph neural network model. S6. Load the trained HeteroData heterogeneous graph neural network model and the collected index system data of the candidate resettlement points, predict all candidate resettlement points, and output the comprehensive score of the candidate resettlement points based on the weight matrix and the fusion features of the candidate resettlement point nodes. S7. Generate a list of priority recommended land parcels based on the comprehensive score.

[0006] Preferably, the HeteroData heterogeneous graph neural network model includes a two-layer HeteroConv structure, wherein each sub-convolutional network uses the GCNConv module to learn and aggregate the features of candidate placement point nodes for a specific relation type, thereby obtaining the fused features of the candidate placement point nodes. A ReLU activation function and Dropout operation are added between the two HeteroConv layers, and finally, the comprehensive score of the entity node corresponding to the candidate placement point is output through a linear mapping layer.

[0007] Preferably, in the two-layer HeteroConv structure, the first layer takes the initial node features of the heterogeneous graph as input and outputs a first-order fusion feature. Each relation corresponds to a separate GCNConv submodule. The second layer takes the first-order fusion feature as input, propagates it again and converges to the entity node corresponding to the candidate placement point, and outputs a higher-order fusion feature of the candidate placement point node. Finally, the higher-order fusion feature of the candidate placement point node is input into the linear layer, and the comprehensive score is output through the linear layer.

[0008] Preferably, the mathematical expression for the HeteroData heterogeneous graph neural network structure model is:

[0009] in, For the first Layer nodes Features For a set of relation types, For relationship Next node Neighbors For relationship The weight matrix, The normalization coefficient is... This is the ReLU activation function.

[0010] Preferably, the mathematical expression of the comprehensive score is:

[0011] Wherein, S is the comprehensive score, W is the weight matrix of the linear layer, b is the bias term, is the high-order fusion feature of the candidate resettlement point node.

[0012] Preferably, in step S4, the CSV data derived from the knowledge graph is used to construct a HeteroData heterogeneous graph, generate an index mapping of each node, and correspond the node feature vector thereto.

[0013] Preferably, the entity includes a candidate land plot, a clan, a second industrial park, a third industrial park, an e-commerce node, a traffic node, and a water source node.

[0014] Preferably, the relationship includes being located, having, covering, being close to, and belonging to.

[0015] Preferably, the spatial suitability index includes area, slope, elevation, engineering geological condition, site suitability, and water source condition.

[0016] Preferably, the spatial suitability index data is collected from the 30-meter resolution DEM provided by the State Bureau of Surveying and Mapping, land use planning data, and geological disaster investigation report, wherein the area, slope, and elevation are calculated and standardized in GIS; the engineering geological condition, site suitability, and water source condition are standardized after being assigned values according to the grade.

[0017] Preferably, the social index includes clan coverage, cultural homogeneity index, and willingness to move index.

[0018] Preferably, the clan coverage is calculated by the ratio of the number of people that the clan can settle in the candidate resettlement point to the total number of people in the clan; the cultural homogeneity index is calculated by the square sum of the proportion of the ethnic population of the people to be resettled in the candidate resettlement point; and the willingness to move index is obtained by weighting and normalizing the scores of immigrants on the type of candidate resettlement point.

[0019] Preferably, the economic cost index includes land expropriation cost, external transportation distance, water supply network distance, and power laying distance.

[0020] Preferably, the land expropriation cost is calculated and inversely normalized in the comprehensive area; the external transportation distance, water supply network distance, and power laying distance are calculated according to the distance from the candidate resettlement point to the nearest station and inversely normalized.

[0021] Preferably, the livelihood recovery index includes surrounding cultivated land area, developable forest and grass area, service industry density, and e-commerce logistics accessibility.

[0022] Preferably, the peripheral cultivated area and the developable forest and grassland area are collected and normalized by the land use planning and forest land protection and utilization planning report within 5km of the candidate resettlement site; the service industry density is collected and normalized by the number of catering, agricultural supplies, repair and logistics commercial stores within a radius of 10km; and the e-commerce logistics accessibility is collected and normalized by measuring the distance from the candidate resettlement site to the nearest e-commerce node.

[0023] The application also provides a knowledge graph-based hydropower resettlement site selection optimization system, comprising: A multi-source heterogeneous data index system construction module is configured to construct a multi-source heterogeneous data index system, wherein the index system comprises spatial suitability indexes, economic cost indexes, livelihood recovery indexes and social indexes. A data collection and preprocessing module is configured to collect and preprocess data related to entities associated with resettlement sites according to the index system, to organize the index system into triples of entities, attributes and relationships, and to generate attribute tables and relationship tables of resettlement site nodes corresponding to the entities. A knowledge graph construction module is configured to import the entities, the attribute tables and the relationship tables into a graph database to construct a knowledge graph. A HeteroData heterogeneous graph data construction module is configured to derive entity and relationship information from the knowledge graph to construct HeteroData heterogeneous graph data. A heterogeneous graph neural network analysis module is configured to train a HeteroData heterogeneous graph neural network model using index system-related data of resettlement sites and candidate resettlement site labels, to propagate and aggregate features of candidate resettlement site nodes, to mine potential correlations between candidate resettlement sites and multiple types of entities, to obtain a weight matrix and a fusion feature of candidate resettlement site nodes, to further obtain the trained HeteroData heterogeneous graph neural network model, and to output comprehensive scores of candidate resettlement sites by predicting all candidate resettlement sites according to the weight matrix and the fusion feature of candidate resettlement site nodes, and to generate a priority land plot list according to the comprehensive scores.

[0024] The application also provides a knowledge graph-based hydropower resettlement site selection optimization device, comprising a processor capable of executing a computer program, wherein the computer program can implement the knowledge graph-based hydropower resettlement site selection optimization method described above.

[0025] Compared with the prior art, the application has the following beneficial effects: 1. The application associates spatial suitability, economic cost, livelihood recovery and social indexes by constructing a knowledge graph, converts scattered data into a unified knowledge base, and realizes unified management and semantic query of multidimensional data.

[0026] 2. The present application first introduces clan coverage, cultural homogeneity, and willingness to move into the index system, fills the gap of traditional site selection method to the internal social needs of immigrants, significantly reduces the social stability risk after resettlement, and promotes sustainable development.

[0027] 3. The present application automatically mines the potential association between the candidate resettlement site and the multiple entities through the heterogeneous graph neural network model, and outputs the comprehensive score in an end-to-end manner, avoiding the deviation of subjective weighting, and significantly increasing the reliability of decision-making.

[0028] 4. The index system and model structure of the present application can be flexibly adjusted according to project requirements, and adapt to different regional immigrant resettlement scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 A flowchart of a knowledge graph-based hydropower resettlement site selection optimization method according to an embodiment of the present application.

[0030] Figure 2 A knowledge graph diagram according to an embodiment of the present application.

[0031] Figure 3 A heterogeneous graph neural network overall architecture diagram according to an embodiment of the present application.

[0032] Figure 4 A HeteroConv layer internal working principle diagram according to an embodiment of the present application. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0034] As shown in the drawings, Figures 1-4 The present application provides a knowledge graph-based hydropower resettlement site selection optimization method, comprising the following steps: S1, constructing a multi-source heterogeneous data index system, the index system comprising spatial suitability index, economic cost index, livelihood recovery index and social index; S2, collecting and preprocessing the data related to the index system related to the entity of the resettlement site, arranging the index system into triples of entity, attribute and relationship, and generating the attribute table and relationship table of the resettlement site node corresponding to the entity; the attribute is the parameter or characteristic of the entity, the area can be xx mu; the geological condition can be: suitable, unsuitable, etc. S3, treat the attributes of each entity as a numerical node feature vector, which is used for subsequent model calculation; the entity is a thing in the objective world, such as a candidate land plot, an industrial park, etc.; the attribute is a characteristic or parameter possessed by the entity, such as the slope and area of a candidate land plot; the relationship describes the connection between entities, such as a candidate land plot being close to a certain traffic node.

[0035] S3, import the entity, attribute table and relationship table into the graph database to construct a knowledge graph; a unique index mapping, i.e., a unique integer ID, is generated for each entity node in the knowledge graph, which facilitates efficient indexing of the model. At the same time, the node feature vector generated by S3 is associated with the index mapping to form input data available to the model.

[0036] S4, derive entity and relationship information from the knowledge graph to construct HeteroData heterogeneous graph data; S5, design a HeteroData heterogeneous graph neural network model based on the HeteroData heterogeneous graph data, train the HeteroData heterogeneous graph neural network model using the index system related data of the built resettlement site in combination with the label of the candidate resettlement site, perform candidate resettlement site node feature propagation and aggregation, mine the potential association between the candidate resettlement site and multiple types of entities, obtain a weight matrix and a fusion feature of the candidate resettlement site node, and then obtain a trained HeteroData heterogeneous graph neural network model; S6, load the trained HeteroData heterogeneous graph neural network model and the collected index system related data of the candidate resettlement site, and predict all candidate resettlement sites, output a comprehensive score of the candidate resettlement site according to the weight matrix and the fusion feature of the candidate resettlement site node; S7, generate a priority recommended land plot list according to the comprehensive score.

[0037] According to one specific embodiment of the present application, the HeteroData heterogeneous graph neural network model includes two layers of HeteroConv structure, wherein each sub-convolutional network adopts a GCNConv module for feature learning and aggregation of the candidate resettlement site node for a specific relationship type, obtains a fusion feature of the candidate resettlement site node, and adds a ReLU activation function and a Dropout operation between the two layers of HeteroConv structure, and finally outputs a comprehensive score of the entity node corresponding to the candidate resettlement site through a linear mapping layer.

[0038] According to one specific embodiment of the present application, in the two-layer HeteroConv structure, the first layer inputs the initial node features of the heterogeneous graph, and outputs first-order fusion features, each relationship corresponds to a separate GCNConv submodule, the second layer inputs the first-order fusion features, and propagates and converges to the entity nodes corresponding to the candidate settlement points again, and outputs high-order fusion features of the candidate settlement point nodes, and finally the high-order fusion features of the candidate settlement point nodes are input into the linear layer, and the comprehensive score is output through the linear layer.

[0039] According to one specific embodiment of the present application, the mathematical expression of the HeteroData heterogeneous graph neural network structure model is:

[0040] Among them, is the feature of the node in the i-th layer, is the feature of the node in the i-th layer, is the set of relationship types, is the neighbor of the relationship, is the weight matrix of the relationship, is the normalization coefficient, is the ReLU activation function. According to one specific embodiment of the present application, the mathematical expression of the comprehensive score is:

[0041] Among them, S is the comprehensive score, W is the weight matrix of the linear layer, b is the bias term, is the high-order fusion feature of the candidate settlement point node. According to one specific embodiment of the present application, in step S4, the HeteroData heterogeneous graph is constructed from the CSV data derived from the knowledge graph, the index mapping of each node is generated, and the node feature vector is corresponded.

[0042] According to one specific embodiment of the present application, the entity includes a candidate land plot, a clan, a second industrial park, a third industrial park, an e-commerce node, a traffic node and a water source node.

[0043] According to one specific embodiment of the present application, the relationship includes being located, having, covering, being close to and belonging to.

[0044] According to one specific embodiment of the present application, the spatial suitability index includes area, slope, elevation, engineering geological condition, site suitability and water source condition.

[0045] According to one specific embodiment of the present application, the spatial suitability index includes area, slope, elevation, engineering geological condition, site suitability and water source condition.

[0046] According to one specific embodiment of the present application, the spatial suitability index includes area, slope, elevation, engineering geological condition, site suitability and water source condition.

[0047] ​​According to one specific embodiment of the present application, the space suitability index data is collected from the 30-meter resolution DEM provided by the State Bureau of Surveying and Mapping, land use planning data, and geological disaster investigation reports, and the area, slope and elevation are calculated and standardized in GIS; the engineering geological conditions, site suitability and water source conditions are standardized after being assigned according to the grade.

[0048] According to one specific embodiment of the present application, the social indicators include: clan coverage, cultural homogeneity index and willingness to move index.

[0049] According to one specific embodiment of the present application, the clan coverage is calculated by the ratio of the number of people who can be resettled in the candidate resettlement site to the total number of people in the clan; the cultural homogeneity index is calculated by the square sum of the proportion of the ethnic population of the people to be resettled in the candidate resettlement site; and the willingness to move index is obtained by weighting and normalizing the scores of the immigrants to the type of the candidate resettlement site.

[0050] According to one specific embodiment of the present application, the economic cost indicators include land expropriation cost, external traffic distance, water supply network distance and power laying distance.

[0051] According to one specific embodiment of the present application, the land expropriation cost is calculated and inversely normalized in the comprehensive area; the external traffic distance, water supply network distance and power laying distance are calculated according to the distance from the candidate resettlement site to the nearest site and inversely normalized.

[0052] According to one specific embodiment of the present application, the livelihood recovery indicators include surrounding arable land area, developable forest area, service industry density and e-commerce logistics accessibility.

[0053] According to one specific embodiment of the present application, the surrounding arable land area and the developable forest area are collected by land use planning and forest protection and utilization planning reports within a range of 5km around the candidate resettlement site, and are normalized; the service industry density is collected by counting the number of commercial stores such as catering, agricultural supplies, maintenance and logistics within a range of 10km, and is normalized; and the e-commerce logistics accessibility is collected by measuring the distance from the candidate resettlement site to the nearest e-commerce node, and is normalized.

[0054] The present application also provides a knowledge graph-based hydropower resettlement site selection optimization system, comprising: A multi-source heterogeneous data index system construction module is used to construct a multi-source heterogeneous data index system, and the index system includes space suitability indicators, economic cost indicators, livelihood recovery indicators and social indicators. The data acquisition and preprocessing module acquires and preprocesses data related to the index system of the entity related to the resettlement site, arranges the index system into a triple of entity, attribute and relationship, and generates an attribute table and a relationship table of the resettlement site node corresponding to the entity; The knowledge graph construction module is used for importing the entity, the attribute table and the relationship table into a graph database to construct a knowledge graph. The HeteroData heterogeneous graph data construction module is used for deriving entity and relationship information from the knowledge graph to construct HeteroData heterogeneous graph data. The heterogeneous graph neural network analysis module trains a HeteroData heterogeneous graph neural network model by using the index system related data of the resettlement site combined with the candidate resettlement site label, performs candidate resettlement site node feature propagation and aggregation, mines the potential association between the candidate resettlement site and multiple types of entities, obtains a weight matrix and a fusion feature of the candidate resettlement site node, and then obtains the trained HeteroData heterogeneous graph neural network model. Through prediction on all candidate resettlement sites, the comprehensive score of the candidate resettlement site is output according to the weight matrix and the fusion feature of the candidate resettlement site node, and a priority recommended land list is generated according to the comprehensive score.

[0055] The application also provides a knowledge graph-based hydropower resettlement site selection optimization device, which comprises a processor, the processor can execute a computer program, and the computer program can realize the knowledge graph-based hydropower resettlement site selection optimization method.

[0056] Embodiment 1 The following takes a hydropower resettlement site selection as an example to illustrate the implementation details of the application, and the similar parts adopt the above implementation scheme.

[0057] 1. Index system construction The hydropower project in this embodiment divides the factors affecting the resettlement site selection into four first-level indexes and 17 second-level indexes, and constructs a comprehensive evaluation system containing space suitability indexes, economic cost indexes, livelihood recovery indexes and social indexes.

[0058] Table 1. Resettlement site selection evaluation system of a hydropower project

[0059] 2. Knowledge extraction In the middle of knowledge extraction, define each entity and relationship mode. Entities include candidate plots, clans, second industrial parks, third industrial parks, e-commerce nodes, transportation nodes, and water source nodes. Relationships include being located, having, covering, being close to, belonging to, etc. The second-level indicators in Table 1 are the attributes of the specific entity type "candidate plots". After normalization, they collectively constitute the initial feature vector of the "candidate plot" node.

[0060] 3. Constructing the knowledge graph In the Neo4j graph database, construct the knowledge graph, create nodes with unique constraints for each entity, and identify them by ID. Save the node and relationship data as UTF-8 encoded CSV files in the import directory of Neo4j. Import all entity nodes and relationship files into the Neo4j graph database one by one.

[0061] 4. Graph neural network modeling Export CSV data from the knowledge graph to construct a HeteroData heterogeneous graph, generate index mapping for each node, and correspond the node feature vector. Construct a heterogeneous graph neural network model, design two layers of HeteroConv, the first layer inputs the initial node features of the heterogeneous graph, and outputs the first-order fusion features. Each relationship corresponds to a separate GCNConv submodule. The second layer inputs the first-order features, propagates and converges to the CandidatePlot node again, and outputs high-order fusion features. Finally, output the score or classification probability through a linear layer. 5. Data collection and preprocessing of candidate resettlement sites 5.1 Spatial suitability data Spatial suitability index data is collected from 30-meter resolution DEM, land use planning data, and geological disaster investigation reports provided by the National Geomatics Bureau. Area, slope, and elevation are calculated and standardized in GIS. Engineering geological conditions, site suitability, and water source conditions are assigned values according to grades and standardized.

[0062] 5.2 Economic cost data Land expropriation cost is calculated and reverse normalized for the comprehensive area. The cost indicators of external transportation distance, water supply network distance, and power laying distance are collected as the distance from the resettlement site to the nearest station and reverse normalized.

[0063] 5.3 Livelihood recovery data The surrounding farmland area and the developable forest and grassland area are collected within a range of 5 km around the candidate block according to the land use planning and the forest land protection and utilization planning report, and are normalized; the service industry density is collected by counting the number of commercial stores such as catering, agricultural supplies, maintenance and logistics within a radius of 10 km and is normalized; and the e-commerce logistics accessibility is collected by measuring the distance from the block to the nearest e-commerce node and is normalized.

[0064] 5.4 Social data The data of the social indicators are mainly obtained through household questionnaire interviews of the immigrant groups and analysis of historical data, and the data need to be desensitized; the clan coverage degree is calculated according to the assignable population of the block and the total number of households in the block, the proportion of the clan in the block is calculated, and the normalized processing is performed; the cultural homogeneity index is collected by classifying and counting the number of each type of people n g , the total number of people N, calculating , and normalized processing; the willingness to move index needs to be investigated by household, and the weighted sum is calculated according to the proportion of family population and normalized processing.

[0065] 6. Generate training labels Collect the built water and electricity resettlement points as historical samples; collect the index data of these resettlement points before construction and the same dimension as the candidate block for site selection, to constitute the input features of the model; the independent evaluation data of the water and electricity resettlement points after construction is collected in this embodiment, and the independent evaluation comprehensive score is standardized as the training label.

[0066] 7. Model training The feature nodes of a certain historical sample are input into the initialized model, and the high-order fusion feature vector is generated after two layers of HeteroConv network calculation; the weight matrix W and the bias term b of the linear mapping layer are given random initial values, and an initial prediction score is calculated; the prediction score is compared with the training label corresponding to the sample, and the error value between the two is calculated by the mean square error loss function MSE; the contribution of each parameter in the weight matrix W and the bias term b to the error is calculated by using the back propagation algorithm. The above steps are iterated until the prediction error of the model converges to a small enough range, and the final weight matrix W and bias term b are obtained.

[0067] 8. Comprehensive evaluation Load the trained model, including the weight matrix W and the bias term b; load the preprocessed and standardized index data of the candidate block; input the index data of the candidate block into the graph neural network of the trained model for forward propagation and feature fusion calculation. The high-order fusion feature vector , input to the linear mapping layer of the model; perform weighted sum calculation to calculate the comprehensive score S. The calculation formula is as follows:

[0068] Wherein, S is the comprehensive score, W is the weight matrix of the linear layer, b is the bias term, is the high-order fusion feature of the node.

[0069] The above only describes the embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the scope of the application should be included in the protection scope of the present application.

Claims

1. A knowledge graph-based method for optimizing the site selection of hydropower resettlement for displaced people, characterized in that, Includes the following steps: S1. Construct a multi-source heterogeneous data indicator system, which includes spatial suitability indicators, economic cost indicators, livelihood recovery indicators, and social indicators; S2. Collect and preprocess data related to the indicator system for entities related to resettlement sites, organize the indicator system into triples of entities, attributes and relationships, and generate attribute tables and relationship tables for resettlement site nodes corresponding to the entities. S3. Import the entities, attribute tables, and relation tables into the graph database to construct a knowledge graph; S4. Derive entity and relationship information from the knowledge graph to construct HeteroData heterogeneous graph data; S5. Design a HeteroData heterogeneous graph neural network model based on HeteroData heterogeneous graph data. Use the relevant data of the indicator system of the established resettlement sites and the candidate resettlement site labels to train the HeteroData heterogeneous graph neural network model, perform feature propagation and aggregation of candidate resettlement site nodes, explore the potential relationship between candidate resettlement sites and multiple types of entities, obtain the weight matrix and the fusion features of candidate resettlement site nodes, and then obtain the trained HeteroData heterogeneous graph neural network model. S6. Load the trained HeteroData heterogeneous graph neural network model and the collected index system data of the candidate resettlement points, predict all candidate resettlement points, and output the comprehensive score of the candidate resettlement points based on the weight matrix and the fusion features of the candidate resettlement point nodes. S7. Generate a list of priority recommended land parcels based on the comprehensive score.

2. The knowledge graph-based site selection optimization method for hydropower resettlement as described in claim 1, characterized in that, The HeteroData heterogeneous graph neural network model consists of a two-layer HeteroConv structure. Each sub-convolutional network uses the GCNConv module to learn and aggregate the features of candidate placement point nodes for a specific relation type, thereby obtaining the fused features of the candidate placement point nodes. A ReLU activation function and Dropout operation are added between the two HeteroConv layers. Finally, the comprehensive score of the entity node corresponding to the candidate placement point is output through a linear mapping layer.

3. The knowledge graph-based site selection optimization method for hydropower resettlement as described in claim 2, characterized in that, In the two-layer HeteroConv structure, the first layer takes the initial node features of the heterogeneous graph as input and outputs the first-order fusion features. Each relation corresponds to a separate GCNConv submodule. The second layer takes the first-order fusion features as input, propagates them again and converges to the entity nodes corresponding to the candidate placement points, and outputs the higher-order fusion features of the candidate placement point nodes. Finally, the higher-order fusion features of the candidate placement point nodes are input into the linear layer, and the comprehensive score is output through the linear layer.

4. The knowledge graph-based site selection optimization method for hydropower resettlement according to any one of claims 2-3, characterized in that, The mathematical expression for the HeteroData heterogeneous graph neural network structure model is: in, For the first Layer nodes Features For a set of relation types, For relationship Next node Neighbors For relationship The weight matrix, The normalization coefficient is... This is the ReLU activation function.

5. The knowledge graph-based site selection optimization method for hydropower resettlement as described in claim 1, characterized in that, In step S4, CSV data is exported from the knowledge graph to construct a HeteroData heterogeneous graph, an index mapping for each node is generated, and the node feature vectors are mapped to it.

6. The knowledge graph-based site selection optimization method for hydropower resettlement as described in claim 1, characterized in that, Entities include candidate land parcels, clans, secondary industrial parks, tertiary industrial parks, e-commerce nodes, transportation nodes, and water source nodes.

7. The knowledge graph-based site selection optimization method for hydropower resettlement as described in claim 1, characterized in that, Relationships include being located at, having, covering, being near, and belonging to.

8. The knowledge graph-based site selection optimization method for hydropower resettlement of displaced persons according to claim 1, characterized in that, Spatial suitability indicators include area, slope, elevation, engineering geological conditions, site suitability, and water source conditions.

9. The knowledge graph-based site selection optimization method for hydropower resettlement as described in claim 8, characterized in that, Spatial suitability index data were collected from 30-meter resolution DEMs, land use planning data, and geological disaster investigation reports provided by the State Bureau of Surveying and Mapping. Among them, area, slope, and elevation were calculated and standardized in GIS; engineering geological conditions, site suitability, and water source conditions were standardized after being assigned values ​​according to levels.

10. The knowledge graph-based site selection optimization method for hydropower resettlement as described in claim 1, characterized in that, The social indicators include: clan coverage, cultural homogeneity index, and migration intention index.

11. The knowledge graph-based site selection optimization method for hydropower resettlement as described in claim 10, characterized in that, Clan coverage is calculated as the ratio of the number of people that can be resettled by a clan in the candidate resettlement site to the total number of people in the clan; cultural homogeneity index is calculated as the sum of squares of the ethnic population proportions of the people to be resettled in the candidate resettlement site; migration intention index is obtained by scoring the candidate resettlement site type by immigrants, weighting it by population weight, and then normalizing it.

12. The knowledge graph-based site selection optimization method for hydropower resettlement of displaced persons according to claim 1, characterized in that, Economic cost indicators include land acquisition costs, external transportation distances, water supply network distances, and power line laying distances.

13. The knowledge graph-based site selection optimization method for hydropower resettlement as described in claim 12, characterized in that, Land acquisition costs are calculated based on the comprehensive area and region and then reverse-normalized; distances to external transportation, water supply networks, and power lines are calculated based on the distances from candidate resettlement sites to the nearest stations and then reverse-normalized.

14. The knowledge graph-based site selection optimization method for hydropower resettlement of displaced persons according to claim 1, characterized in that, Livelihood recovery indicators include the area of ​​surrounding arable land, the area of ​​exploitable forest and grassland, the density of the service industry, and the accessibility of e-commerce logistics.

15. The knowledge graph-based site selection optimization method for hydropower resettlement as described in claim 14, characterized in that, The area of ​​arable land and exploitable forest and grassland in the surrounding area were collected within 5km of the candidate resettlement site through land use planning and forest land protection and utilization planning reports, and then normalized. The number of catering, agricultural supplies, repair, and logistics businesses within a 10km radius of the service industry density data collection area was statistically analyzed and normalized. E-commerce logistics accessibility data is collected, the distance from candidate resettlement points to the nearest e-commerce node is calculated, and the data is normalized.

16. A knowledge graph-based site selection optimization system for hydropower resettlement, characterized in that, include: A multi-source heterogeneous data indicator system construction module is used to construct a multi-source heterogeneous data indicator system, which includes spatial suitability indicators, economic cost indicators, livelihood recovery indicators, and social indicators. The data acquisition and preprocessing module collects and preprocesses data related to the indicator system for entities related to resettlement sites, organizes the indicator system into triples of entities, attributes, and relationships, and generates attribute tables and relationship tables for resettlement site nodes corresponding to the entities. The knowledge graph construction module is used to import the entities, attribute tables, and relation tables from the graph database to construct a knowledge graph. The HeteroData heterogeneous graph data construction module is used to derive entity and relationship information from the knowledge graph and construct HeteroData heterogeneous graph data. The heterogeneous graph neural network analysis module uses relevant data from the indicator system of completed resettlement sites, combined with candidate resettlement site labels, to train a HeteroData heterogeneous graph neural network model. It performs feature propagation and aggregation of candidate resettlement site nodes, mines potential associations between candidate resettlement sites and multiple types of entities, obtains the weight matrix and the fusion features of candidate resettlement site nodes, and then obtains the trained HeteroData heterogeneous graph neural network model. By predicting all candidate resettlement sites, it outputs a comprehensive score for each candidate resettlement site based on the weight matrix and the fusion features of the candidate resettlement site nodes. Based on this comprehensive score, it generates a list of priority recommended land parcels.

17. A knowledge graph-based site selection optimization device for hydropower resettlement, characterized in that, The device includes a processor capable of executing a computer program that implements the knowledge graph-based hydropower resettlement site optimization method according to any one of claims 1-15.

Images