Graph theory-based hydrological spatio-temporal data model construction and medium and long term runoff prediction method
By using a graph theory-based hydrological spatiotemporal data model and a CNN-LSTM-Attention model, the accuracy and robustness issues of existing hydrological models in multi-scale coupling and extreme event prediction are solved, enabling the representation of complex hydrological processes and medium- to long-term runoff prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING VOCATIONAL UNIV OF IND TECH
- Filing Date
- 2025-12-18
- Publication Date
- 2026-04-10
AI Technical Summary
Existing hydrological models based on physical mechanisms are insufficient in terms of prediction accuracy and robustness to meet the needs of refined production applications when dealing with the multi-scale coupling of the combined effects of climate change and human activities and predicting extreme events. Furthermore, traditional hydrological models have limitations in describing distributed characteristics and networked relationships.
A hydrological spatiotemporal data model is constructed based on graph theory, and medium- and long-term runoff prediction is achieved through a CNN-LSTM-Attention model. This includes constructing conceptual relationships, graphs, paths, subgraphs, and attribute modules. Path and time delay analysis is used to screen key features. Spatial features are extracted by combining a CNN network and temporal features by combining an LSTM network. Automatic weighted average calculation is performed using the Attention mechanism.
It has achieved the characterization and data organization of complex hydrological spatiotemporal changes, completed the screening of key spatial and temporal features affecting medium- and long-term runoff prediction, and improved the effectiveness of medium- and long-term runoff prediction.
Smart Images
Figure CN121834331A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for characterizing hydrological change processes and organizing data, as well as for medium- and long-term runoff prediction, in the field of hydrological big data technology. Specifically, it is a method for constructing a hydrological spatiotemporal data model and predicting medium- and long-term runoff based on graph theory. Background Technology
[0002] Spatiotemporal processes describe the dynamic evolution of geographical phenomena across time and space, encompassing complex processes such as natural disasters (e.g., floods, earthquakes) and human activities (e.g., urbanization, population migration). Among these, hydrological processes, as a typical example of spatiotemporal processes, focus on the phase transitions, movement, and interactions of water bodies with the environment. Their dynamic evolution directly impacts flood disasters, water resource allocation, and ecosystem stability. With the development of the national economy and adjustments to national water management policies, existing physical mechanism-based models, due to their numerous simplification assumptions and high parameter sensitivity, struggle to meet the demands of refined production applications in scenarios involving the combined effects of climate change and human activities across multiple scales and in extreme event prediction.
[0003] In recent years, artificial intelligence (AI) technology has provided a new paradigm for hydrological process simulation due to its powerful nonlinear fitting capabilities, high-dimensional feature extraction, and spatiotemporal dynamic modeling advantages. However, theoretical gaps remain regarding the applicability of AI models in the field of hydrology. On the other hand, hydrological processes exhibit high spatial heterogeneity and dynamic complexity, and traditional hydrological models have limitations in describing distributed characteristics and networked relationships. Summary of the Invention
[0004] Purpose of the invention: To address the problems and shortcomings of existing technologies, this invention provides a method for constructing a hydrological spatiotemporal data model and predicting medium- and long-term runoff based on graph theory.
[0005] Technical Solution: A graph-based method for constructing a hydrological spatiotemporal data model and predicting medium- and long-term runoff. First, it classifies hydrological entities within a region based on conceptual relationships. Then, it constructs a graph-based spatiotemporal data model, including four modules: graph, path, subgraph, and attributes, to represent and organize hydrological changes. Next, based on the constructed model, it employs a feature selection method based on path and time-delay analysis to identify key features influencing medium- and long-term runoff prediction. Finally, it constructs a CNN-LSTM-Attention model to achieve medium- and long-term runoff prediction. The method includes the following steps: S1-1 Constructing conceptual relationships.
[0006] S1-2 Determine the types of hydrological entities within the region based on conceptual relationships.
[0007] S1-3 Based on step S1-2, construct a hydrological spatiotemporal data model, including four modules: graph, path, subgraph, and attributes.
[0008] S1-4 Based on step S1-3, form the motion change path process of all entities within the region, including the entity motion change process attribute set and path map.
[0009] Based on the movement and change processes of all entities within the aforementioned region, S1-5 employs a feature screening method based on path and time delay analysis to screen key features that influence medium- and long-term runoff forecasts.
[0010] S1-6 Constructing a CNN-LSTM-Attention model to achieve medium- and long-term runoff prediction, namely, extracting spatial features of the water movement change process path map through a CNN network, using LSTM with input gate, forget gate and output gate control mechanism to effectively extract the temporal feature information of hydrological sequence, and then using the Attention mechanism to automatically calculate the weighted average of the output of the LSTM network. Finally, the best trained CNN-LSTM-Attention model is obtained through the training set.
[0011] S1-7 uses a trained CNN-LSTM-Attention model to predict medium- to long-term runoff in a given region.
[0012] In step S1-1, conceptual relationships are constructed. Specifically, conceptual semantic relationships refer to the semantic relationships between different entities belonging to the same concept within a conceptual classification system, including belonging, co-occupation, and hierarchical relationships. The conceptual relationship model is defined as follows: (1) In the formula, It represents a set of domain concepts, in a multi-level tree structure. Represents a set of concept relations, including relations within the concept set. , Represents an entity set, which is the result of instantiation of a concept.
[0013] In step S1-2, the types of hydrological entities in the region are determined according to the conceptual relationships. The entity types mainly include multiple entities such as unit water, water body and water carrier.
[0014] The graphs in steps S1-3 are defined as follows: (2) in: It is a collection of entities consisting of water units, water bodies, and water carriers; It is a set that has the same related properties.
[0015] gather It can also be divided into a set of water body transfer paths. and the collection of mutual influences between water carriers .definition , That is to say, it is believed This refers to the part of the hydrological process that directly involves movement, transfer, and impact. This refers to the part of the relationship that is not directly affected.
[0016] Assumption middle The quantities are respectively ,if There is a direct correlation between them, that is, it is believed that... yes The neighbor is defined as ,in As the starting entity, The endpoint entity.
[0017] The out-degree of an entity is defined as the number of directed edges originating from that entity, and the in-degree as the number of directed edges ending at that entity. Associations belong to... hour, The main issue is a single direction with a defined orientation, meaning the out-degree is not equal to the in-degree: The association belongs to hour, The main interaction is a correlation-symmetric relationship, where the out-degree equals the in-degree: .
[0018] make All neighbors are a set ,Right now: (3) by The number of associated ends can be expressed as The degree, denoted as : (4) Within the selected range, the sum of the degrees of all entities and the number of associated ends are related as follows: (5) The paths in steps S1-3 are used to define the path representations and calculation methods between different entities, specifically: In the figure In the middle, if from an entity Starting from a point, following some connections, passing through some entities, and arriving at another entity. This forms a path with Path of the link : (6) The length of the path can be defined by formula 6. The shortest path distance is determined by the number of entities traversed in the path. If there is at least one path between two entities, then the shortest path distance between them can be defined. : (7) In this case, the distance from an entity to itself is 0, that is... .like Then it is believed yes of Neighbors.
[0019] The subgraphs in steps S1-3 are used to describe different entities. k The method for representing order subgraphs and calculating information entropy is as follows: Hypothesis diagram middle, and Then the diagram It is a picture A subgraph. A certain entity. of A subgraph can be represented as: (8) in: (9) Therefore, based on the association diversity and distribution uniformity in the subgraph, The information entropy it possesses can be considered as distance. of The sum of uncertainties in all associations among the order neighbors.
[0020] If each entity Transfer to neighbor The probability is: (10) in yes arrive The weighting coefficients, and and It is related to the entity attributes.
[0021] Calculate each path Each edge on Information entropy: (11) in: Distance weights reflect the distance weights in the path. Distance from the starting point The closer the distance, the higher the weight, indicating that the impact of the process is more intuitive and its information entropy is greater.
[0022] Information entropy weights reflect The amount of information about the entity itself. The greater the information entropy of the entity, the higher its weight.
[0023] When entities have indirect influences, it constitutes a single-step influence. ,Right now: (12) Other characteristics that constitute an influence Similarly, its information entropy can be obtained: (13) in: Distance weights reflect the influence Distance from the starting point The degree of attenuation.
[0024] Information entropy weights reflect Influence The amount of information. The greater the information entropy of an entity, the higher its weight.
[0025] This represents the probability of each association type appearing in the total association set.
[0026] Summing the weighted entropy of all paths and their effects yields the entropy of the entire subgraph: (14) As can be seen from Formula 14, the information content of hydrological processes Its path characteristics have the following features: (1) In a specific theme constitutes a certain Information entropy of hydrological processes within the sub-graph It is determined by the information transmitted along the path and the information that influences the interactions between entities, thus directly contributing to the overall information entropy of the water body. Because entities must be controlled within... Within the order range, therefore, in calculating the single-step effect... At that time, hydrological entities must remain in relation to the subject entity. Within the neighborhood.
[0027] (2) Hydrological processes The sub-diagram shows Direct path associations between hydrological entities and Single-step influence relationships of hydrological entities. Selected The larger the order, the more possible paths, and the more entities on each path, the more entities and relationships can be included. Potential information interaction and changes can increase the complexity and uncertainty of the system, thereby increasing the overall information entropy.
[0028] (3) Under the influence of distance weight, the information content of an entity will decrease as the path and the distance to the corresponding subject entity increase. Therefore, when determining the selection of hydrological features, it is necessary to consider the information content of the entity itself and its neighbor order with the subject entity to achieve the optimal "entity-information content" configuration.
[0029] Therefore, the selection of features for all elements of a hydrological process is mainly based on the map. This is done in the form of a graph. When the number of entities and relationships is insufficient, a subgraph can be created and configured. The threshold, or directly select Extraction is performed using the order of steps.
[0030] The attributes in steps S1-3 are used to add additional attribute information to the graph. Specifically, for an attribute graph, it mainly includes entity metadata, secondary data, and derived or combined data. Metadata mainly includes entity data and relationship data. Secondary data mainly includes entity attribute data and relationship attribute data. High-dimensional and combined data (derived data) mainly includes multi-point combination, multi-edge combination, path (single path, multi-path, loop, etc.), subgraph, forest, cross-subgraph combination data, etc.
[0031] The aforementioned feature screening method for selecting key features affecting medium- and long-term runoff forecasting involves, firstly, using the spatial transformation method to extract path features of the entity's movement and change process based on the movement and change process of hydrological entities within the aforementioned region, and then using the partial mutual information method to obtain time-delay analysis features. Finally, the feature screening method that couples path and time-delay analysis is used to select key features affecting medium- and long-term runoff forecasting.
[0032] The CNN-LSTM-Attention model specifically involves extracting spatial features of the path map of water movement changes through a CNN network, effectively extracting the temporal feature information of the hydrological sequence through the control mechanism of LSTM with input gate, forget gate and output gate, and then using the Attention mechanism to automatically calculate the weighted average of the output of the LSTM network. Finally, the optimal CNN-LSTM-Attention model is obtained through the training set.
[0033] The medium- and long-term runoff prediction specifically involves using a trained CNN-LSTM-Attention model to predict the medium- and long-term runoff in a given area.
[0034] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of constructing a graph theory-based hydrological spatiotemporal data model and a medium- to long-term runoff prediction method as described above.
[0035] A computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implements the steps described above for constructing a graph theory-based hydrological spatiotemporal data model and a medium- to long-term runoff prediction method.
[0036] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: The present invention can represent and organize the complex spatiotemporal changes of hydrology, and can complete the screening of key spatial and temporal features that affect medium and long-term runoff prediction. Finally, the CNN-LSTM-Attention model is constructed to improve the medium and long-term runoff prediction effect. Attached Figure Description
[0037] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 Conceptual relationship diagram of an embodiment of the present invention. Detailed Implementation
[0038] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading the present invention, any modifications of the present invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.
[0039] A graph theory-based method for constructing a spatiotemporal hydrological data model and predicting medium- to long-term runoff includes the following steps: S1-1 constructs conceptual relationships based on the semantic relationships between different entities belonging to the same concept, including belonging, primacy, and hierarchical relationships, specifically as follows: Figure 2 As shown. The conceptual relationship model is defined as follows. (1) In the formula, It represents a set of domain concepts, in a multi-level tree structure. Represents a set of concept relations, including relations within the concept set. , Represents an entity set, which is the result of instantiation of a concept.
[0040] S1-2 Based on conceptual relationships, the types of hydrological entities within the region are determined, mainly including multiple entities such as unit water, water body and water carrier.
[0041] S1-3 Based on step S1-2, construct a hydrological spatiotemporal data model, including four modules: graph, path, subgraph, and attributes.
[0042] According to the definition of graph, based on the interaction characteristics between entity behavior and its environment, the correlation in hydrological processes can be summarized into two main categories: water body transfer paths and mutual influence between water carriers. These two categories describe the possible flow of characteristic dimensional information between water carriers from different perspectives, as shown in Table 1.
[0043] Table 1. Two types of correlations in hydrological processes
[0044] A path, or water transfer path, refers to the process of migration and flow of a large number of units of water between different water carriers, emphasizing the trajectory of water movement in time and space. These connections typically exist in the form of "paths" or "chain structures" and can be categorized into spatial path-type migration, state-change-type migration, and carrier-transfer-type paths. Spatial path-type migration refers to water transfer primarily driven by dynamics, such as "surface runoff—[inflow] → river—[inflow] → lake—[inflow] → estuary," and "cloud—[precipitation] → soil—[infiltration] → groundwater," etc. State-change-type migration refers to water transfer primarily driven by thermodynamics, such as "cloud—[precipitation] → surface (gas state → liquid state)" and "surface—[evaporation] → atmosphere (liquid state → gas state)," etc. Carrier-transfer-type paths refer to water transfer caused by the movement or change of the carrier, such as "grass—[eaten] → animal" and "atmosphere—[transported] → atmosphere," etc. This type of connection emphasizes the continuity, directionality, and conservation characteristics of the water body itself and is suitable for specific water volume tracking.
[0045] In contrast, the interaction between water carriers focuses on the interrelationships formed between water carriers through structural connections, energy exchange, or process coupling. Such associations may only involve a correlation between a single attribute of the water carrier, but can also evolve into associations between entities. Examples include: "Aquifer (structure) – [influence] → Surface river (recharge intensity)", "Soil (particle composition) – [influence] → Soil (moisture retention capacity)", and "Surface (slope) – [influence] → Surface runoff (path)".
[0046] Subgraphs are used to describe different entities. k The structure of the subgraph and the method for calculating information entropy are as follows: Hypothesis diagram middle, and Then the diagram It is a picture A subgraph. A certain entity. of A subgraph can be represented as: (8) in: (9) Therefore, based on the association diversity and distribution uniformity in the subgraph, The information entropy it possesses can be considered as distance. of The sum of uncertainties in all associations among the order neighbors.
[0047] If each entity Transfer to neighbor The probability is: (10) in yes arrive The weighting coefficients, and and It is related to the entity attributes.
[0048] Calculate each path Each edge on Information entropy: (11) in: Distance weights reflect the distance weights in the path. Distance from the starting point The closer the distance, the higher the weight, indicating that the impact of the process is more intuitive and its information entropy is greater.
[0049] Information entropy weights reflect The amount of information about the entity itself. The greater the information entropy of the entity, the higher its weight.
[0050] When entities have indirect influences, it constitutes a single-step influence. ,Right now: (12) Other characteristics that constitute an influence Similarly, its information entropy can be obtained: (13) in: Distance weights reflect the influence Distance from the starting point The degree of attenuation.
[0051] Information entropy weights reflect Influence The amount of information. The greater the information entropy of an entity, the higher its weight.
[0052] This represents the probability of each association type appearing in the total association set.
[0053] Summing the weighted entropy of all paths and their effects yields the entropy of the entire subgraph: (14) As can be seen from Formula 14, the information content of hydrological processes Its path characteristics have the following features: (1) In a specific theme constitutes a certain Information entropy of hydrological processes within the sub-graph It is determined by the information transmitted along the path and the information that influences the interactions between entities, thus directly contributing to the overall information entropy of the water body. Because entities must be controlled within... Within the order range, therefore, in calculating the single-step effect... At that time, hydrological entities must remain in relation to the subject entity. Within the neighborhood.
[0054] (2) Hydrological processes The sub-diagram shows Direct path associations between hydrological entities and Single-step influence relationships of hydrological entities. Selected The larger the order, the more possible paths, and the more entities on each path, the more entities and relationships can be included. Potential information interaction and changes can increase the complexity and uncertainty of the system, thereby increasing the overall information entropy.
[0055] (3) Under the influence of distance weight, the information content of an entity will decrease as the path and the distance to the corresponding topic entity increase. Therefore, when determining the selection of hydrological features, it is necessary to consider the information content of the entity itself and its neighbor order with the topic entity to achieve the optimal "entity-information content" configuration.
[0056] Therefore, the selection of features for all elements of a hydrological process is mainly based on the map. This is done in the form of a graph. When the number of entities and relationships is insufficient, a subgraph can be created and configured. The threshold, or directly select Extraction is performed using the order of steps.
[0057] Attributes mainly include entity metadata, secondary data, and derived or combined data. Specifically, metadata mainly includes entity data and relationship data; secondary data mainly includes entity attribute data and relationship attribute data; high-dimensional and combined data (derived data) mainly includes multi-point combinations, multilateral combinations, paths (single paths, multiple paths, loops, etc.), subgraphs, forests, and cross-subgraph combined data, etc.
[0058] S1-4 Based on step S1-3, form the motion change path process of all entities within the region, including the entity motion change process attribute set and path map.
[0059] S1-5 Based on the movement and change processes of all entities in the above-mentioned area, a feature screening method based on path and time delay analysis is used to screen key features that affect medium- and long-term runoff forecasts. S1-6 Constructing a CNN-LSTM-Attention prediction model to achieve medium- and long-term runoff prediction, namely, extracting spatial features of the water movement change process path map through a CNN network, using LSTM through the control mechanism of input gate, forget gate and output gate to effectively extract the temporal feature information of hydrological sequence, and then using the Attention mechanism to automatically calculate the weighted average of the output of the LSTM network. Finally, the best CNN-LSTM-Attention prediction model is obtained through the training set.
[0060] S1-7 uses a trained CNN-LSTM-Attention model to predict medium- to long-term runoff in a given region.
[0061] S1-8 End.
[0062] Obviously, those skilled in the art should understand that the steps of the graph theory-based hydrological spatiotemporal data model construction and medium-to-long-term runoff prediction method described in the above embodiments of the present invention can be implemented using general-purpose computing devices. These steps can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, which can then be stored in a storage device for execution by the computing device. Furthermore, in some cases, the steps shown or described can be performed in a different order than presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.
Claims
1. A hydrological spatio-temporal data model construction and medium and long term runoff prediction method based on graph theory, characterized in that, Comprise the following steps: S1-1 constructs the concept relationship; S1-2 determines the entity type related to hydrology in the region according to the concept relationship division; S1-3 according to step S1-2, construct hydrological spatio-temporal data model, including graph, path, subgraph, four modules of attribute; S1-4 according to step S1-3, form the whole entity movement change path process in the region, including entity movement change process attribute set and path graph two parts; S1-5 above the whole entity movement change process in the region as the basis, using the feature screening method based on path and time delay analysis realizes the key feature screening of influence middle and long term runoff prediction; S1-6 construct CNN-LSTM-Attention model to realize middle and long term runoff prediction, that is, through CNN network extracts the spatial features of water body movement change process path graph, adopts LSTM through the control mechanism of input gate, forget gate and output gate, effectively extracts the time feature information of hydrological sequence, then uses Attention mechanism to automatically weighted average calculation on the output of LSTM network, finally through the training set, obtain the best trained CNN-LSTM-Attention model; S1-7 realize the middle and long term runoff prediction of the given region through the trained CNN-LSTM-Attention model.
2. The graph-based hydrological spatio-temporal data model construction and mid-long term runoff prediction method according to claim 1, characterized in that, In step S1-1, the concept relationship is constructed, specifically, the concept semantic relationship refers to the semantic relationship between different entities in the concept classification system, including belonging, same level, superior and subordinate relationship; the concept relationship model is defined as （1） In the formula, represents a set of domain concepts, which is a multi-level tree structure, represents a set of concept relations, including relations within the concept set , represents a set of entities, which is the instantiation result of the concept.
3. The graph-based hydrological spatio-temporal data model construction and mid-long term runoff prediction method according to claim 1, characterized in that, The graph in step S1-3 is defined as follows: （2） wherein: is a collection of entities of unit water, water bodies and water carriers; is a collection with the same property association; set into a set of water body transfer paths and a set of water carrier interactions ; definition , it is considered as the part of the hydrological process that directly occurs in motion transfer and influence, while as the part of the association that is not directly influenced Set number of the number of if a direct association exists between them, it is considered that is a neighbor of , defined as , where is the starting entity, is the end entity; When the association belongs to , the out-degree is not equal to the in-degree; and when the association belongs to , the out-degree is equal to the in-degree. Let all neighbors of i.e.: （3） With the number of associated ends is denoted as in degrees, denoted as : （4） In the selected range, the sum of the degrees of all entities and the associated end number have the following relationship: （5）。 4. The graph-based hydrological spatio-temporal data model construction and mid-long term runoff prediction method according to claim 1, characterized in that, The path in step S1-3 is used to define the path representation between different entities and its calculation method, specifically: In Fig , if one starts from one entity , goes through some entities along some associations and arrives at another entity , a path with steps is formed : （6） Length of a path Determined by the number of entities that are traversed in the path; if there is at least one path between two entities, then the shortest path distance between the two is defined : （7） where the distance of an entity to itself is 0, i.e. ; if , then is considered as 's order neighbor.
5. The graph-based hydrological spatio-temporal data model construction and mid-long term runoff prediction method according to claim 1, characterized in that, The subgraph in the step S1-3 is used to describe different entities k The step S1-3 is used to calculate the information entropy of the subgraph, and the information entropy of the subgraph is calculated according to the following formula: Let G be a graph In, And Then G is a subgraph of G; a certain entity The subgraph of G is denoted by G The subgraph of G is denoted by G The subgraph of G is denoted by G The subgraph of G is denoted by G （8） Wherein: （9） Thus, according to the correlation diversity and distribution uniformity in the subgraph, The information entropy possessed is considered as the distance of the sum of all the uncertainties of the correlation in the k-neighbor If each entity Transfers to neighbor The probability is: （10） wherein is to a weight coefficient related to the entity attributes of and calculating the information entropy of each edge on each path （11） Wherein: Distance weights reflect the distance weights in the path. Distance from the starting point The closer the distance, the higher the weight, indicating that the impact of the process is more intuitive and its information entropy is greater; Information entropy weights reflect The amount of information about the entity itself; the greater the information entropy of the entity, the higher its weight. There is a non-direct influence between entities, i.e. a single step influence i.e.: （12） For other features that constitute an impact , the information entropy thereof can also be acquired: （13） Wherein: is a distance weight, reflecting the degree of attenuation of influence from the starting point of the distance; is the information entropy weight, reflecting influence the amount of information; the greater the information entropy of the entity itself, the higher the weight; the probability of occurrence of each association type in the total association set; Sum the weighted information entropy of all paths and influences to get the information entropy of the whole subgraph: （14）。 6. The graph-based hydrological spatio-temporal data model construction and mid-long term runoff prediction method according to claim 5, characterized in that, Since entities must control in order range, when calculating single-step influence , hydrological entities must be kept within the order neighbors of the subject entity; the order subgraph of hydrological processes exhibits direct path relationships between order hydrological entities and single-step influence relationships; under the action of distance weights, the information amount of entities will present a decay trend with the increase of the distance of the path and its corresponding influence to the subject entity; when determining the selection of hydrological features, the information amount of the entity itself and its neighbor order number with the subject entity need to be considered at the same time to achieve the optimal "entity-information amount" configuration; Feature selection of all elements of hydrological processes based on diagrams This is done in the form of a graph; when the number of entities and relationships cannot meet the requirements, a subgraph is created and configured. The threshold, or directly select Extraction is performed using the order of steps.
7. The hydrological spatio-temporal data model construction and medium and long term runoff prediction method based on graph theory according to claim 1, characterized in that, The attribute in step S1-3 is used to add additional attribute information to the graph; specifically, for an attribute graph, it includes entity metadata, secondary data and derivative data or combined data; metadata includes entity data and association relationship data; secondary data includes entity attribute data and relationship attribute data; High-dimensional, combined data includes multi-point combination, multi-edge combination, path, subgraph, forest and cross-subgraph combination data. 8.The hydrological spatio-temporal data model construction and mid-long term runoff prediction method based on graph theory according to claim 1, characterized in that, The feature screening method realizes the key feature screening of influence middle and long term runoff prediction, specifically, based on the hydrological entity object movement change process in the region, first, the path feature of entity movement change process is extracted by using spatial transformation method, then the time delay analysis feature is obtained by using partial mutual information method, finally, the feature screening method coupling path and time delay analysis is realized to realize the key feature screening of influence middle and long term runoff prediction.
9. A computer device, comprising: The computer device comprises a memory, a processor and a computer program stored on the memory and executable on the processor, and the processor executes the above computer program to realize the steps of the hydrological spatio-temporal data model construction and middle and long term runoff prediction method based on graph theory in any one of claims 1-8.
10. A computer readable storage medium having stored thereon computer programs / instructions, characterized in that: The computer program / instructions, when executed by the processor, implement the steps of the hydrological spatio-temporal data model construction and medium and long term runoff prediction method based on graph theory according to any one of claims 1-8.