Water affair operation data decision support system

Abstract

The invention relates to the technical field of decision support, in particular to a water affair operation data decision support system, which comprises a joint analysis unit used for acquiring historical full-dimension operation data and historical water affair decision knowledge reference data of a water affair system in a target decision period; performing data normalization on the historical full-dimension operation data to obtain a pipe network parameter time sequence and a capacity data time sequence, and performing joint analysis on the historical full-dimension operation data and the historical water decision knowledge reference data to obtain a plurality of corresponding historical decision triples; and the feature extraction unit is used for carrying out associated feature extraction on the pipe network parameter time sequence and the capacity data time sequence. According to the method, the historical decision triple is associated with knowledge nodes such as the pipe network and the productivity, so that the system can construct a comprehensive decision knowledge graph, the knowledge graph integrates the mutual relation between the pipe network and water plant equipment, and a decision maker is helped to better understand and manage a complex network of a water affair system.

Description

Technical Field

[0001] This invention relates to the field of decision support technology, specifically a water operations data decision support system. Background Technology

[0002] In recent years, the management and allocation of water resources have become increasingly complex. For example, some regions have experienced prolonged droughts, while others may face torrential rains and floods. These changes have placed higher demands on water management systems, and traditional systems are unable to cope with this ever-changing environment.

[0003] Currently, traditional systems often struggle to effectively integrate historical and real-time data, leading to decisions based on a single data dimension and lacking a holistic perspective. They tend to rely on independent operational data for decision-making, lacking systematic and comprehensive knowledge support. They are relatively weak in extracting key features and analyzing correlations, often relying on manual rules or simple algorithms for analysis. This can easily overlook potential system problems and depend on human intervention to determine whether operational standards are met, resulting in slow response times and a high risk of errors. Furthermore, their analysis is typically simplistic, lacking the ability to make multi-faceted judgments from different levels and dimensions. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: a water affairs operation data decision support system, comprising: The joint analysis unit is used to acquire historical full-dimensional operational data and historical water decision-making knowledge reference data of the water system within the target decision-making period; to perform data normalization on the historical full-dimensional operational data to obtain the time series sequence of pipeline parameters and the time series sequence of production capacity data; and to perform joint analysis on the historical full-dimensional operational data and the historical water decision-making knowledge reference data to obtain multiple corresponding historical decision triplets. The feature extraction unit is used to extract associated features from the pipeline network parameter time series and the production capacity data time series to obtain a first associated feature vector corresponding to the pipeline network parameter time series and a second associated feature vector corresponding to the production capacity data time series. The graph modeling unit is used to input the first associated feature vector and the second associated feature vector into the knowledge node generator to obtain the pipeline network knowledge node set and the capacity knowledge node set. Based on the pipeline network knowledge node set, the capacity knowledge node set and multiple historical decision triples, a decision knowledge graph is constructed. The decision fusion unit is used to acquire real-time full-dimensional operational data and real-time water affairs decision-making knowledge reference data, construct multiple real-time decision triples based on the real-time full-dimensional operational data and the real-time water affairs decision-making knowledge reference data, construct decision subgraphs for each real-time decision triple based on the decision correlations in the multiple historical decision triples, and perform subgraph fusion on the decision subgraphs corresponding to each real-time decision triple to obtain a fused decision subgraph for each real-time decision triple. The decision support unit is used to calculate the fit between the fusion decision subgraph of each of the real-time decision triples and the decision knowledge graph, determine whether each of the real-time decision triples is adapted to the decision knowledge graph, and, in response to the real-time decision triples being adapted to the decision knowledge graph, search in the decision knowledge graph for the water flow links and / or scheduling impact range of the pipeline nodes in the real-time decision triples.

[0005] Preferably, the historical full-dimensional operational data includes time information of the target decision cycle, real-time operating parameters of the pipeline network, usage data of water terminals, and time sequence of water plant capacity data; the historical decision triple includes a subject entity, an object entity, and the decision association relationship connecting the subject entity and the object entity, the subject entity includes pipeline network nodes or water plant equipment, and the object entity includes pipeline network nodes, water plant equipment, or water demand information.

[0006] Preferably, the time series sequences of pipeline parameters and the time series sequences of production capacity data are subjected to correlation feature extraction to obtain a first correlation feature vector corresponding to the time series sequences of pipeline parameters and a second correlation feature vector corresponding to the time series sequences of production capacity data, including: The operating status categories corresponding to the time series sequences of the pipeline parameters and the time series sequences of the production capacity data are obtained based on the density clustering algorithm. Extract time-series sequence fragments of pipeline parameters and production capacity data corresponding to each operating status category; perform time-series-frequency joint analysis on the time-series sequence fragments of pipeline parameters and production capacity data respectively to generate two types of time-series frequency feature distribution maps. In the time-series frequency feature distribution maps, time-series sampling points are mapped to the horizontal axis and parameter fluctuation frequencies are mapped to the vertical axis; normalize the parameter feature amplitudes to feature intensity values ​​to obtain two types of two-dimensional feature heat maps. Frequency domain residual correlation detection is performed on the two types of two-dimensional feature heatmaps to obtain feature correlation saliency maps; Based on the mean and variance of the feature association saliency map, a three-level importance threshold is determined, wherein the three-level importance threshold includes a first importance threshold, a second importance threshold, and a third importance threshold; regions with feature intensity values ​​higher than the first importance threshold are labeled as first-level key elements, regions between the first and second importance thresholds are labeled as second-level key elements, and regions between the second and third importance thresholds are labeled as third-level key elements, thus obtaining a hierarchical set of key elements; The parameter co-occurrence matrix of key feature elements is calculated based on multiple preset association directions. The parameter co-occurrence matrix is ​​used to extract coupled texture features, which include association contrast, parameter correlation, feature energy, and coupling homogeneity. Extract the morphological features of key feature elements, wherein the morphological features include the perimeter of the feature region, the area of ​​the feature region, and the morphological invariant moments; By concatenating the coupled texture features and morphological features, we obtain the first associated feature vector corresponding to the time series sequence of pipeline parameters and the second associated feature vector corresponding to the time series sequence of production capacity data.

[0007] Preferably, the operating status categories corresponding to the time series sequences of the pipeline parameters and the time series sequences of the production capacity data are obtained based on a density clustering algorithm, including: Calculate the dynamic distance matrix between data samples in the time series sequence of pipeline parameters and the time series sequence of production capacity data to obtain the adaptive truncation distance; Candidate state cluster centers are selected based on the local fluctuation density and relative deviation of the data in the pipeline network parameter time series and the capacity data time series. The number of final state clusters is determined by the density distribution heatmap based on the center of the candidate state clusters. A hierarchical clustering tree is constructed based on the number of final state clusters. After cluster fusion, the stable state cluster structure is retained, and the operating state category is output. The operating state category includes high load state, stable operating state, and low load state.

[0008] Preferably, the first associated feature vector and the second associated feature vector are respectively input into the knowledge node generator to obtain the pipeline network knowledge node set and the capacity knowledge node set, including: The first associated feature vector is matched with the node attributes of the pipeline feature nodes in the pre-constructed pipeline feature label system to obtain the first matching result; wherein, the pipeline feature label system includes multiple pipeline feature nodes; The second associated feature vector is matched with the node attributes of the capacity feature nodes in the pre-constructed capacity feature label system to obtain the second matching result; wherein, the capacity feature label system includes multiple capacity feature nodes; Based on the first matching result, at least one pipeline feature node that is successfully matched by the first associated feature vector is determined, and at least one pipeline feature node is used as a set of pipeline knowledge nodes. Based on the second matching result, at least one capacity feature node that is successfully matched by the second associated feature vector is determined, and at least one capacity feature node is used as a set of capacity knowledge nodes.

[0009] Preferably, the first associated feature vector is matched with the node attributes of pipeline feature nodes in the pre-constructed pipeline feature label system to obtain a first matching result, including: The first associated feature vector is matched with the node attributes of the first-level pipeline feature nodes in the pipeline feature label system; After identifying the first-level pipeline feature node that has been successfully matched, the first associated feature vector is matched with the node attributes of the next-level pipeline feature node belonging to the first-level node in descending order of level. The matching degree between the first associated feature vector and the feature nodes of each level of the pipeline network is calculated. The pipeline feature node with the highest matching degree and located at the lowest level is identified as the successfully matched pipeline feature node. The second associated feature vector is matched with the node attributes of the capacity feature nodes in the pre-constructed capacity feature label system to obtain the second matching result, including: The second associated feature vector is matched with the node attributes of the first-level capacity feature nodes in the capacity feature label system; After identifying the first-level capacity feature node that has been successfully matched, the second associated feature vector is matched with the node attributes of the next-level capacity feature node belonging to the first-level node in descending order of level. The matching degree between the second associated feature vector and the capacity feature nodes at each level is calculated. The capacity feature node with the highest matching degree and located at the lowest level is identified as the successfully matched capacity feature node.

[0010] Preferably, a decision knowledge graph is constructed based on the pipeline network knowledge node set, the capacity knowledge node set, and multiple historical decision triples, including: The pipeline knowledge node set and the production capacity knowledge node set are respectively input into the graph construction module based on the relation inference engine to obtain the first association set and the second association set; Calculate the intersection of the first set of associations and the second set of associations to obtain the core set of associations; Based on the decision-related relationships in the multiple historical decision triples, the decision knowledge graph is obtained by connecting the pipeline knowledge node set and the production capacity knowledge node set based on the core relationship set.

[0011] Preferably, the decision subgraph includes a decision subgraph centered on a pipeline node and / or a decision subgraph centered on a water plant equipment. Based on the decision relationships in the multiple historical decision triples, a decision subgraph is constructed for each of the real-time decision triples, including: In response to the real-time decision triplet including a subject entity of type network node or water plant equipment and an object entity of type network node or water plant equipment, based on the decision association provided by the multiple historical decision triplets, a decision subgraph with the subject entity as the central node and a decision subgraph with the object entity as the central node are constructed for the real-time decision triplet. In response to the real-time decision triple including a subject entity of water plant equipment and an object entity of water demand information, a decision subgraph with the subject entity as the central node is constructed for the real-time decision triple based on the decision association provided by the multiple historical decision triples. Among them, the decision association types in the decision subgraph with pipeline nodes as the central nodes include water flow connectivity between pipeline nodes, pressure transmission between pipeline nodes, and supply and demand correspondence between pipeline nodes and water demand. Among them, the decision-making relationships in the decision-making subgraph with water plant equipment as the central node include the collaborative operation relationship between water plant equipment, the supply relationship between water plant equipment and water demand, and the water transmission relationship between water plant equipment and pipeline network nodes.

[0012] Preferably, subgraph fusion is performed on the decision subgraphs corresponding to each of the real-time decision triples to obtain a fused decision subgraph for each of the real-time decision triples, including: In response to the decision subgraph centered on the subject entity and the decision subgraph centered on the object entity, the two types of decision subgraphs are concatenated by features to obtain a fused decision subgraph. In response to a decision subgraph centered on the main entity, the decision subgraph is used as the fused decision subgraph.

[0013] Preferably, calculating the fit between the fused decision subgraph of each of the real-time decision triples and the decision knowledge graph includes: Calculate the feature similarity between the fused decision subgraph of each of the real-time decision triples and the decision knowledge graph.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention combines historical full-dimensional operational data with real-time full-dimensional operational data to generate real-time decision triples, which are then effectively fused through a decision fusion unit. This fusion not only improves the accuracy of decision-making but also provides real-time and reliable decision support for water systems in a dynamically changing environment. By establishing associations between historical decision triples and knowledge nodes such as pipelines and production capacity, the system can construct a comprehensive decision knowledge graph. This knowledge graph integrates the interrelationships between pipelines and water plant equipment, helping decision-makers better understand and manage the complex network of water systems, thereby improving decision-making efficiency and the effectiveness of water resource utilization. This invention extracts correlation features between pipeline parameters and production capacity data through density clustering algorithm and time-series-frequency joint analysis. This analysis helps identify key elements and potential problems in the system and provides targeted optimization solutions based on these key features, improving the stability and predictability of system operation. Through the real-time decision triplet fitness calculation, the system can quickly determine whether the real-time decision conforms to the existing decision knowledge graph and provide real-time support based on the correlation of historical decisions. This automated decision support can reduce human intervention and improve the automation level and response speed of water management. When constructing decision subgraphs, this invention can perform multi-level analysis from pipeline nodes, water plant equipment to water demand, ensuring accurate judgments on water operations from different dimensions and perspectives. The fusion of each decision subgraph further refines the decision-making process, making water management more precise. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall system architecture in one embodiment of the present invention.

[0016] In the diagram: 1. Joint analysis unit; 2. Feature extraction unit; 3. Graph modeling unit; 4. Decision fusion unit; 5. Decision support unit. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1, please refer to Figure 1 This invention provides a technical solution: a water affairs operation data decision support system, comprising: Joint analysis unit 1 is used to acquire historical full-dimensional operational data of the water system and historical water decision-making knowledge reference data within the target decision-making period; data normalization is performed on the historical full-dimensional operational data to obtain the time series sequence of pipeline parameters and the time series sequence of production capacity data; joint analysis is performed on the historical full-dimensional operational data and historical water decision-making knowledge reference data to obtain multiple corresponding historical decision triplets. Feature extraction unit 2 is used to extract associated features from pipeline parameter time series and capacity data time series to obtain the first associated feature vector corresponding to the pipeline parameter time series and the second associated feature vector corresponding to the capacity data time series. The graph modeling unit 3 is used to input the first associated feature vector and the second associated feature vector into the knowledge node generator to obtain the pipeline network knowledge node set and the capacity knowledge node set. Based on the pipeline network knowledge node set, the capacity knowledge node set and multiple historical decision triples, a decision knowledge graph is constructed. The decision fusion unit 4 is used to acquire real-time full-dimensional operational data and real-time water affairs decision-making knowledge reference data. Based on the real-time full-dimensional operational data and real-time water affairs decision-making knowledge reference data, it constructs multiple real-time decision triples. Based on the decision-making relationships in multiple historical decision triples, it constructs decision subgraphs for each real-time decision triple. It then performs subgraph fusion on the decision subgraphs corresponding to each real-time decision triple to obtain the fused decision subgraph of each real-time decision triple. Decision support unit 5 is used to calculate the fit between the fused decision subgraph of each real-time decision triple and the decision knowledge graph, determine whether each real-time decision triple is fit to the decision knowledge graph, and, in response to the real-time decision triple being fit to the decision knowledge graph, search for the water flow links and / or scheduling impact range of the pipeline nodes in the real-time decision triple in the decision knowledge graph.

[0019] It should be noted that historical full-dimensional operational data refers to multi-dimensional operational data of the water system, including pipeline data, capacity data, load data, etc., covering all aspects of the water system; historical water decision-making knowledge reference data refers to past decision-making cases or reference knowledge, such as operational decisions, optimization schemes, and management strategies of the water system under different scenarios; historical decision triples refer to data triples composed of decision-making context, decision-making objective, and decision-making strategy, used to represent decision-making information within a certain decision-making cycle; pipeline parameter time series refers to numerical sequences recording the changes in pipeline operation status over time, such as water flow, pressure, and temperature; capacity data time series refers to data sequences of water system capacity changes over time, such as water plant production and transmission volume; correlation feature vectors represent feature vectors with certain correlations or dependencies in pipeline and capacity data, used to quantify the relationships between systems; knowledge nodes represent important knowledge points or decision-making factors in the water system, such as pipeline nodes, capacity nodes, and scheduling nodes; and decision knowledge graphs. It is a directed graph where nodes represent various decision elements or knowledge points in the system, and edges represent the relationships between them; the graph is used to systematically store decision logic, historical experience, and reasoning paths; real-time decision triples refer to decision information triples constructed from real-time generated decision data, using real-time full-dimensional operational data and real-time decision knowledge reference data; decision subgraphs are local decision graphs formed by extracting a part from the historical decision graph for specific real-time decision triples, used for local decision analysis and reasoning; fit refers to the degree of matching between real-time decision triples and decision knowledge graphs; high fit means that real-time decisions are more in line with historical knowledge and decision patterns, and may be more effective; water flow links refer to the transmission paths of water flow in the pipeline network, including the water delivery path from the water plant to the user end, the distribution of the pipeline network, etc.; the impact range of water plant equipment scheduling refers to the impact range of the operation and scheduling of different equipment (such as water pumps, water treatment equipment, etc.) in the water plant on the entire water system; for example, a failure of a certain equipment may affect the water supply in certain areas.

[0020] In an optional embodiment, the historical full-dimensional operational data includes time information of the target decision cycle, real-time operating parameters of the pipeline network, usage data of water terminals, and time sequence of water plant capacity data; the historical decision triple includes a subject entity, an object entity, and the decision association relationship connecting the subject entity and the object entity. The subject entity includes pipeline network nodes or water plant equipment, and the object entity includes pipeline network nodes, water plant equipment, or water demand information.

[0021] In an optional embodiment, correlation feature extraction is performed on the pipeline parameter time series and the capacity data time series to obtain a first correlation feature vector corresponding to the pipeline parameter time series and a second correlation feature vector corresponding to the capacity data time series, including: Based on density clustering algorithm, the operating status categories corresponding to the time series sequences of pipeline parameters and production capacity data are obtained respectively; Extract time-series sequence fragments of pipeline parameters and production capacity data corresponding to each operating status category; perform time-series-frequency joint analysis on the time-series sequence fragments of pipeline parameters and production capacity data respectively to generate two types of time-series frequency feature distribution maps. In the time-series frequency feature distribution maps, time-series sampling points are mapped to the horizontal axis and parameter fluctuation frequencies are mapped to the vertical axis; normalize the parameter feature amplitudes to feature intensity values ​​to obtain two types of two-dimensional feature heat maps. Frequency domain residual correlation detection was performed on the two types of two-dimensional feature heatmaps to obtain feature correlation saliency maps. The three-level importance thresholds are determined based on the mean and variance of the feature association saliency map. The three-level importance thresholds include the first importance threshold, the second importance threshold, and the third importance threshold. Regions with feature intensity values ​​higher than the first importance threshold are labeled as first-level key elements, regions between the first and second importance thresholds are labeled as second-level key elements, and regions between the second and third importance thresholds are labeled as third-level key elements, thus obtaining a set of hierarchical key elements. The parameter co-occurrence matrix of key feature elements is calculated based on multiple preset association directions. The parameter co-occurrence matrix is ​​used to extract coupled texture features, which include association contrast, parameter correlation, feature energy, and coupling homogeneity. Extract the morphological features of key feature elements, including the perimeter of the feature region, the area of ​​the feature region, and the morphological invariant moments; By concatenating the coupled texture features and morphological features, we obtain the first associated feature vector corresponding to the time series sequence of pipeline parameters and the second associated feature vector corresponding to the time series sequence of production capacity data.

[0022] It should be noted that classifying pipeline network parameters and production capacity data time series using density clustering first requires determining a suitable distance metric to extract meaningful clustering results from the time series data; each category represents an operating state; extracting corresponding time series segments from each state category is for in-depth analysis of system behavior under different states; this can be accomplished using sliding window techniques or other time series segmentation methods; performing time-series-frequency analysis on each segment (using techniques such as Fast Fourier Transform (FFT)) to generate a time-series frequency feature distribution map; the goal here is to identify the main frequency components in the signal to understand the dynamic characteristics under different operating states; normalizing the characteristic amplitudes in the frequency domain analysis results to generate a two-dimensional heatmap can help visualize the intensity of system behavior under different states; for frequency domain data By calculating frequency domain residuals, the correlation between parameters under different states can be detected, thereby identifying key change patterns. Based on the saliency of features, different thresholds are set to determine the level of each feature, helping to classify and prioritize features with different levels of importance. Co-occurrence matrices are used to describe the correlation between features, and coupled texture features are extracted. This is a method for analyzing the interaction between parameters, especially in complex systems where strong coupling relationships may exist between features. Morphological analysis of key feature regions, such as calculating perimeter, area, and invariant moments, helps to geometrically describe the structural characteristics of the features. Finally, coupled texture features and morphological features are concatenated to form feature vectors for further analysis. These vectors can be used to build models for prediction or optimization of pipeline network operation.

[0023] In an optional embodiment, obtaining the operating status categories corresponding to the time series sequences of pipeline parameters and the time series sequences of production capacity data based on a density clustering algorithm includes: Calculate the dynamic distance matrix between data samples in the time series of pipeline parameters and the time series of production capacity data to obtain the adaptive cutoff distance; Candidate state cluster centers are selected based on the local fluctuation density and relative deviation of data in the time series of pipeline parameters and production capacity data. The number of final state clusters is determined by the density distribution heatmap based on the center of the candidate state clusters. A hierarchical clustering tree is constructed based on the number of final state clusters. After cluster fusion, the stable state cluster structure is retained, and the operating state category is output. The operating state categories include high load state, stable operating state, and low load state.

[0024] It should be noted that the distance between each pair of samples in the time series of pipeline parameters and production capacity data is calculated. This "distance" can be based on different metrics (such as Euclidean distance, Dynamic Time Warping (DTW), etc.), depending on the nature of the data and the analysis requirements. This distance matrix reflects the similarity or difference between data samples and can reveal the dynamic trends between samples. After obtaining the distance matrix, a distance threshold (cutoff distance) can be adaptively set based on certain criteria to distinguish between similar and dissimilar samples. The setting of this cutoff distance is usually adjusted based on the distribution or clustering results of the data to adapt to the characteristics of different datasets; for example, it can be achieved through the quantiles of the distance matrix or... Certain statistical properties are used to define local fluctuation density: this means calculating the volatility of data within a local region; areas of high volatility usually indicate rapid changes in the system's state, while areas of low volatility may indicate a more stable system; local fluctuation density can be used to filter out data segments with large (or small) fluctuations, which may represent different operating states; relative deviation refers to the degree of deviation of each data point relative to other data points in its neighborhood; by calculating the deviation of each data point, it can further help identify points that may belong to different state clusters; a large deviation may indicate that the point is a potential cluster center, or that the system has experienced significant fluctuations; using the above two characteristics, suitable data segments can be filtered out. Clusters can be clusters representing different operating states, indicating the system's behavior patterns under specific conditions. A density heatmap can plot the data's density distribution, with each point representing the data density and color intensity indicating density magnitude. This visually shows data aggregation, helping to determine the number of clusters in different states. Areas with higher density may represent different operating states, while areas with lower density may represent noisy or unimportant states. The final state cluster count is determined by analyzing the density heatmap to identify the number of distinct clusters. Each cluster represents an operating state; higher-density clusters represent more stable system states, while lower-density clusters may represent less stable states. The table represents transitional or abnormal states. Once candidate cluster centers are determined, the next step is to construct a hierarchical clustering tree, which is an algorithm that progressively merges or splits clusters. The hierarchical clustering tree can progressively build a hierarchical structure of clusters based on the similarity of the data. During hierarchical clustering, some small clusters may appear, and merging them can produce a larger, more stable cluster structure. Cluster fusion reduces noise and outliers by merging clusters with high similarity, ultimately resulting in more stable operating state clusters. Stable state cluster structure: In the final cluster fusion process, those clusters that have shown strong stability throughout the process are retained. These stable clusters represent the main operating states of the system, such as "high load state," "smooth operating state," and "low load state."High load state: In this state, the system may be under high load, and the parameters or capacity data of the pipeline network may fluctuate significantly, indicating that the system is under great pressure. Stable operation state: In this state, the system may exhibit small fluctuations, indicating that the system is operating smoothly and the load is relatively balanced. Low load state: When the system is under low load, the fluctuations are small, the system load is low, and it may be in an idle or inefficient state.

[0025] In an optional embodiment, the first associated feature vector and the second associated feature vector are respectively input into the knowledge node generator to obtain a pipeline network knowledge node set and a capacity knowledge node set, including: The first associated feature vector is matched with the node attributes of the pipeline feature nodes in the pre-constructed pipeline feature label system to obtain the first matching result; wherein, the pipeline feature label system includes multiple pipeline feature nodes; The second associated feature vector is matched with the node attributes of the capacity feature nodes in the pre-constructed capacity feature label system to obtain the second matching result; wherein, the capacity feature label system includes multiple capacity feature nodes; Based on the first matching result, at least one pipeline feature node that is successfully matched by the first associated feature vector is determined, and at least one pipeline feature node is used as the pipeline knowledge node set. Based on the second matching result, at least one capacity feature node that is successfully matched by the second associated feature vector is determined, and at least one capacity feature node is taken as the set of capacity knowledge nodes.

[0026] It should be noted that the system includes a pipeline network feature labeling system, which consists of multiple pipeline network feature nodes. Each pipeline network feature node represents a specific attribute or feature of the pipeline network data, such as pressure, flow rate, temperature, etc. These nodes help describe the state and behavior of the pipeline network system. The first associated feature vector contains features extracted from the time-series of pipeline network parameters, usually a summary or representation of certain time-series data, reflecting the dynamic changes of the pipeline network. It is used for matching with the pipeline network feature labeling system. The matching process compares the first associated feature vector with the pipeline network feature nodes in the pipeline network feature labeling system to find the most matching pipeline network feature node. The matching result may be multiple nodes, indicating a high similarity between the feature vector and the pipeline network feature. The first matching result is obtained through this matching, resulting in one or more pipeline network feature nodes that match the first associated feature vector. These nodes represent certain states or behaviors of the pipeline network system. This matching result helps in understanding the operating status of the pipeline network. Similar to the pipeline network system, the system also has a capacity feature labeling system, which consists of multiple capacity feature nodes. Each capacity feature node represents... The first matching vector represents specific attributes or behaviors of production capacity data, such as capacity utilization and output fluctuations. The second matching vector is extracted from the time-series data of production capacity, reflecting dynamic changes in production capacity. It is used to match nodes in the production capacity feature labeling system. The matching process compares the second matching vector with production capacity feature nodes in the system to find the most suitable matching node, thus identifying different states or behaviors of production capacity data. The second matching result yields one or more production capacity feature nodes that match the second matching vector, representing certain states or characteristics of the production capacity system. The pipeline network knowledge node set is determined based on the first matching result, identifying which pipeline network feature nodes successfully match the first matching vector. These nodes reflect the main operating characteristics of the pipeline network system, and thus, these successfully matched pipeline network feature nodes form the pipeline network knowledge node set. Similarly, based on the second matching result, determining which production capacity feature nodes successfully match the second matching vector, these successfully matched production capacity feature nodes represent key characteristics of the production capacity system, and they form the production capacity knowledge node set.

[0027] In an optional embodiment, the first associated feature vector is matched with the node attributes of pipeline feature nodes in a pre-constructed pipeline feature label system to obtain a first matching result, including: The first associated feature vector is matched with the node attributes of the first-level pipeline feature nodes in the pipeline feature label system; After identifying the first-level pipeline feature node that has been successfully matched, the first associated feature vector is matched with the node attributes of the next-level pipeline feature node belonging to the first-level node in descending order of level. The matching degree between the first associated feature vector and the feature nodes of each level of the pipeline network is calculated. The pipeline feature node with the highest matching degree and located at the lowest level is identified as the successfully matched pipeline feature node. The second associated feature vector is matched with the node attributes of the capacity feature nodes in the pre-constructed capacity feature label system to obtain the second matching result, including: The second associated feature vector is matched with the node attributes of the first-level capacity feature nodes in the capacity feature label system; After identifying the first-level capacity feature node that has been successfully matched, the second associated feature vector is matched with the node attributes of the next-level capacity feature node belonging to the first-level node in descending order of level. The matching degree between the second associated feature vector and the capacity feature nodes at each level is calculated. The capacity feature node with the highest matching degree and located at the lowest level is identified as the successfully matched capacity feature node.

[0028] It should be noted that the pipeline network feature labeling system is designed with a hierarchical structure, with each level representing a different level of pipeline network characteristics. Each feature node has a corresponding level, with the description of the pipeline system's attributes and behaviors gradually becoming more detailed from higher to lower levels. The first associated feature vector is a feature vector containing pipeline parameter information, possibly including data summaries of features such as pressure, flow rate, and temperature. This first associated feature vector is matched with the first-level nodes in the pipeline network feature labeling system. The first-level nodes are typically the most basic or broadest attribute nodes in the pipeline system. If a first-level node matches the first associated feature vector, the process moves to the second-level nodes, and so on, comparing and matching feature vectors with the attributes of pipeline network feature label nodes layer by layer. This layer-by-layer matching method helps to accurately match feature vectors with more detailed and specific nodes. The matching results at each level are statistically analyzed, and the matching degree of each matching node is calculated. Nodes with high matching degrees are considered more valuable. The node matching is more precise; once all levels of matching are completed, the node with the highest matching degree and located at the lowest level will be selected as the final successfully matched pipeline feature node. Similarly, the capacity feature label system is also a hierarchical structure, with each layer representing different capacity data features, possibly refining from overall capacity data to more specific attributes. The second associated feature vector contains features extracted from the capacity time series data, reflecting the dynamic changes in capacity; it will be matched with the first-level nodes in the capacity feature label system. If the first-level node matches the second associated feature vector successfully, the matching will proceed to more refined level nodes. If a successfully matched node is found at a certain level, the second associated feature vector will continue to be matched with the feature nodes of the next level corresponding to that node. Similarly, the matching results of all levels are statistically analyzed and calculated, and the node with the highest matching degree and located at the lowest level is selected as the final successfully matched capacity feature node.

[0029] In an optional embodiment, a decision knowledge graph is constructed based on a set of pipeline knowledge nodes, a set of production capacity knowledge nodes, and multiple historical decision triples, including: Input the set of pipeline knowledge nodes and the set of production capacity knowledge nodes into the graph construction module based on the relation inference engine, respectively, to obtain the first set of association relationships and the second set of association relationships; Calculate the intersection of the first set of associations and the second set of associations to obtain the core set of associations; Based on the decision-making relationships in multiple historical decision triples, a decision knowledge graph is obtained by connecting the pipeline knowledge node set and the production capacity knowledge node set based on the core relationship set.

[0030] It should be noted that the pipeline network knowledge node set includes all feature nodes related to the pipeline network system, representing different attributes of the pipeline network system (such as pressure, flow rate, temperature, etc.); the capacity knowledge node set includes all feature nodes related to the capacity system, representing different attributes of the capacity system (such as capacity utilization rate, output fluctuation, etc.). These two sets will be input into the graph construction module based on the relational inference engine, respectively. The role of the graph construction module is to generate potential connections between the pipeline network and capacity through inference and relational connections. The first set of association relationships: During the relational inference process between the pipeline network knowledge node set and the graph construction module, the system identifies the relationships between pipeline network nodes through inference, forming the first set of association relationships. For example, there may be correlations between pipeline network nodes such as flow rate, pressure, and temperature; the inferred relationships are the association relationships between these nodes. The second set of association relationships: Similarly, the capacity knowledge node set will also identify the association relationships between nodes in the capacity system through inference by the graph construction module; for example, there may be relationships such as output fluctuation and utilization rate changes between capacity nodes; these relationships will form the second set of association relationships. Intersection calculation: The first set of association relationships and the second set of association relationships each... The system represents the internal relationships between the pipeline network and the capacity system, which may have overlap. The core set of relationships is obtained by calculating the intersection of these two sets, representing the common and significant relationships between the pipeline network and capacity. For example, changes in pipeline flow may affect capacity utilization, and these overlapping relationships are part of the core set of relationships. Decision-making relationships: Through historical decision triplets, the system can understand how different states of the pipeline network and the capacity system influence each other under different historical decision-making scenarios, and thus infer the relationship between the pipeline network and capacity during the decision-making process. This historical decision data helps reveal the potential impact and linkage between the pipeline network and the capacity system. Connecting the pipeline network and capacity knowledge node sets: The core set of relationships provides the common relationships between the pipeline network and capacity. Based on this, the system integrates the knowledge and relationships of the two by connecting the pipeline network knowledge node set and the capacity knowledge node set, forming a unified decision knowledge graph. This graph not only represents the relationship between the pipeline network and capacity, but can also contain historical decision data to reflect the correlation patterns under different decision-making conditions, helping to understand how to optimize capacity performance by adjusting the pipeline network system.

[0031] In an optional embodiment, the decision subgraph includes a decision subgraph centered on a pipeline node and / or a decision subgraph centered on a water plant equipment. Based on the decision relationships in multiple historical decision triples, decision subgraphs are constructed for each real-time decision triple, including: In response to the real-time decision triplet, which includes both the subject entity (pipeline node or water plant equipment) and the object entity (pipeline node or water plant equipment), a decision subgraph with the subject entity as the central node and a decision subgraph with the object entity as the central node are constructed based on the decision association provided by multiple historical decision triplets. In response to the real-time decision triplet, which includes a subject entity of water plant equipment and an object entity of water demand information, a decision subgraph with the subject entity as the central node is constructed for the real-time decision triplet based on the decision association provided by multiple historical decision triplets. Among them, the decision association types in the decision subgraph with pipeline nodes as the central nodes include water flow connectivity between pipeline nodes, pressure transmission between pipeline nodes, and supply and demand correspondence between pipeline nodes and water demand. Among them, the decision-making relationships in the decision-making subgraph with water plant equipment as the central node include the collaborative operation relationship between water plant equipment, the supply relationship between water plant equipment and water demand, and the water transmission relationship between water plant equipment and pipeline network nodes.

[0032] It should be noted that in real-time decision-making, the system needs to construct a corresponding decision subgraph for each real-time decision triplet based on historical decision data (decision relationships provided by multiple historical decision triplets). The decision subgraph is centered on a specific entity, reflecting the relationships and influences between that entity and other relevant entities. The main entity is a network node: in a real-time decision triplet, if the main entity is a network node, that node will become the center of the decision subgraph. Water flow connectivity describes the flow path and direction between network nodes; this relationship helps understand the process of water flow from one node to another. Pressure transmission describes how pressure changes in the network system affect the relationships between different nodes, helping to optimize the pressure distribution of water flow. Supply and demand correspondence reflects the relationship between network nodes and water demand, indicating whether the water volume at the network node can meet the water demand of a specific area. The main entity is water plant equipment: in a real-time decision triplet, if the main entity is water plant equipment, that equipment will become the center of the decision subgraph. Relationship types include: Collaborative operation relationship: different equipment within the water plant... The system analyzes the collaborative working relationships between equipment, such as how pumps, valves, and filters coordinate to ensure the normal operation of the water plant; supply relationships: the relationship between water plant equipment and water demand, reflecting how the equipment meets water demand or regulates water supply; water transmission relationships: the relationship between water plant equipment and pipeline nodes, describing how the equipment delivers water to the pipeline network to support the flow of water to end users; constructing decision subgraphs: for each real-time decision triple, the system constructs a decision subgraph centered on the main entity (pipeline node or water plant equipment) based on historical decision relationships; each decision subgraph reflects the relationship between the entity and other related entities (such as pipeline nodes, water demand, etc.) during the decision-making process; optimizing the decision-making process: by constructing these decision subgraphs, the system can more accurately capture the interaction between the pipeline network and water plant equipment, and optimize the current decision-making process by analyzing these subgraphs; for example, in some cases, it may be necessary to adjust the pipeline pressure to meet the water plant's water supply capacity, or to adjust the collaborative working state of the water plant equipment to better respond to water demand.

[0033] In an optional embodiment, subgraph fusion is performed on the decision subgraphs corresponding to each real-time decision triple to obtain a fused decision subgraph for each real-time decision triple, including: In response to decision subgraphs centered on subject entities and decision subgraphs centered on object entities, feature concatenation is performed on the two types of decision subgraphs to obtain a fused decision subgraph. Responding to the decision subgraph centered on the main entity, the decision subgraph is used as the fused decision subgraph.

[0034] It's important to clarify that there are two types of decision subgraphs: 1) **Subgraph centered on the subject entity:** This is a subgraph built around a specific entity (such as a pipeline node or water plant equipment). The nodes and edges in this subgraph represent the relationships between this entity and other related entities (such as water flow connectivity or collaborative operation). 2) **Subgraph centered on the object entity:** This is similar to the subject entity subgraph, but here, the central node is the object entity in the decision triple (e.g., the object of the pipeline node, water demand information, etc.). It also includes the relationships between this entity and other entities, but the focus is different. **Feature concatenation:** This refers to integrating the information from the two types of decision subgraphs to form a new fused decision subgraph. Specifically, this means… By linking the features (such as nodes, edges, and relationship types) of the decision subgraphs centered on the subject entity and the object entity, a new graph structure containing more information is formed. The subject entity subgraph provides features about the relationship between the subject and other nodes (such as pipeline nodes and water demand). The object entity subgraph reflects the decision association between the object and other entities (such as water plant equipment and pipeline nodes). Feature linking combines the node and edge information of these two types of graphs to generate a new graph (i.e., a fused decision subgraph). The fused decision subgraph can provide more comprehensive information, helping to better understand the relationships and decision impacts between entities. The graph is formed by concatenating features from two types of decision subgraphs; this means that by integrating and aggregating the relationships between different entities, a composite graph containing more decision-making information is obtained, which can provide more accurate support for actual decision-making. Application scenarios: The construction of fused decision subgraphs can be used to improve the intelligence and optimization of decision-making; for example, in the optimization of pipeline management and water plant operations, fused decision subgraphs can help decision-makers fully understand the interactive relationships between various pipeline nodes, water plant equipment, and water demand, thereby making more accurate and efficient decisions. Handling decision problems: When facing complex multi-objective, multi-factor decision problems, fused decision subgraphs provide a comprehensive perspective, enabling... It effectively considers the dependencies between multiple levels and entities; it helps analyze the mutual influence between decisions, ensuring the balance and efficiency of system operation; in some cases, when considering a decision subgraph with the subject entity as the central node, the decision subgraph itself may already contain sufficient information to support the current decision-making process; therefore, if it is not necessary to fuse it with the decision subgraph of the object entity through feature concatenation, the decision subgraph of the subject entity can be used directly as the fused decision subgraph; the decision subgraph with the subject entity as the central node is sufficient to reflect the relationship between the subject entity and other related entities in the entire system, and can be directly applied in actual decision-making without further graph structure fusion.

[0035] In an optional embodiment, calculating the fit between the fused decision subgraph of each real-time decision triple and the decision knowledge graph includes: Calculate the feature similarity between the fused decision subgraph of each real-time decision triple and the decision knowledge graph.

[0036] It's important to note that feature similarity refers to the similarity between two graphs (such as a fused decision subgraph and a decision knowledge graph). In this scenario, the comparison involves the features of entities and relationships contained in the fused decision subgraph and the existing knowledge features in the decision knowledge graph. This similarity calculation helps assess the degree of matching between real-time data (such as triples) and a predefined decision knowledge graph during the decision-making process, thus providing more accurate background support for real-time decision-making. Calculating similarity typically involves the following steps: comparing the attributes or features of each node in the fused decision subgraph and the decision knowledge graph; nodes can be entities or resources, and their similarity can be measured by comparing node attributes (such as type, state, location, etc.); common node similarity calculation methods include: Euclidean distance: used to quantify the attribute differences between two nodes; cosine similarity: used to compare node attribute vectors. Similarity between graphs is particularly suitable for high-dimensional attribute data; it compares the similarity of edge (i.e., relationship) features; edges typically represent dependencies or interactions between entities, such as "water supply relationship" or "collaborative work relationship"; similarity can be calculated by comparing the type and weight of edges; for example, if edges represent the same type of relationship, similarity can be judged by the weight and direction of the edges; it assesses the overall matching degree between the fused decision subgraph and the decision knowledge graph by calculating the structural similarity of the entire graph; this can be achieved through graph structural similarity measurement methods, such as graph isomorphism, graph edit distance, etc.; methods: graph isomorphism: checks whether two graphs have the same structure, which can be used to compare the overall matching of nodes and edges between the fused decision subgraph and the decision knowledge graph; graph edit distance: quantifies the minimum number of operations (such as adding or deleting nodes or edges) required to transform one graph into another.

[0037] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A water management operation data decision support system, characterized in that, include: The joint analysis unit is used to obtain historical full-dimensional operational data of the water system and historical water decision-making knowledge reference data within the target decision-making cycle; Data normalization is performed on the historical full-dimensional operation data to obtain the time series sequence of pipeline parameters and the time series sequence of production capacity data. The historical full-dimensional operation data and historical water affairs decision-making knowledge reference data are jointly analyzed to obtain multiple corresponding historical decision triples. The feature extraction unit is used to extract associated features from the pipeline network parameter time series and the production capacity data time series to obtain a first associated feature vector corresponding to the pipeline network parameter time series and a second associated feature vector corresponding to the production capacity data time series. The graph modeling unit is used to input the first associated feature vector and the second associated feature vector into the knowledge node generator to obtain the pipeline network knowledge node set and the capacity knowledge node set. Based on the pipeline network knowledge node set, the capacity knowledge node set and multiple historical decision triples, a decision knowledge graph is constructed. The decision fusion unit is used to acquire real-time full-dimensional operational data and real-time water affairs decision knowledge reference data, construct multiple real-time decision triples based on the real-time full-dimensional operational data and the real-time water affairs decision knowledge reference data, and construct decision subgraphs for each real-time decision triple based on the decision association relationships in the multiple historical decision triples. Subgraph fusion is performed on the decision subgraphs corresponding to each of the real-time decision triples to obtain the fused decision subgraph of each of the real-time decision triples; The decision support unit is used to calculate the fit between the fusion decision subgraph of each of the real-time decision triples and the decision knowledge graph, determine whether each of the real-time decision triples is adapted to the decision knowledge graph, and, in response to the real-time decision triples being adapted to the decision knowledge graph, search in the decision knowledge graph for the water flow links and / or scheduling impact range of the pipeline nodes in the real-time decision triples.

2. The water management operation data decision support system according to claim 1, characterized in that, The historical full-dimensional operational data includes time information of the target decision cycle, real-time operating parameters of the pipeline network, usage data of water terminals, and time sequence of water plant capacity data; the historical decision triple includes a subject entity, an object entity, and the decision association relationship connecting the subject entity and the object entity. The subject entity includes pipeline network nodes or water plant equipment, and the object entity includes pipeline network nodes, water plant equipment, or water demand information.

3. The water management operation data decision support system according to claim 2, characterized in that, The pipeline network parameter time series and the production capacity data time series are subjected to correlation feature extraction to obtain a first correlation feature vector corresponding to the pipeline network parameter time series and a second correlation feature vector corresponding to the production capacity data time series, including: The operating status categories corresponding to the time series sequences of the pipeline parameters and the time series sequences of the production capacity data are obtained based on the density clustering algorithm. Extract time-series sequence fragments of pipeline parameters and production capacity data corresponding to each operating status category; perform time-series-frequency joint analysis on the time-series sequence fragments of pipeline parameters and production capacity data respectively to generate two types of time-series frequency feature distribution maps. In the time-series frequency feature distribution maps, time-series sampling points are mapped to the horizontal axis and parameter fluctuation frequencies are mapped to the vertical axis; normalize the parameter feature amplitudes to feature intensity values ​​to obtain two types of two-dimensional feature heat maps. Frequency domain residual correlation detection is performed on the two types of two-dimensional feature heatmaps to obtain feature correlation saliency maps; Based on the mean and variance of the feature association saliency map, a three-level importance threshold is determined, wherein the three-level importance threshold includes a first importance threshold, a second importance threshold, and a third importance threshold; regions with feature intensity values ​​higher than the first importance threshold are labeled as first-level key elements, regions between the first and second importance thresholds are labeled as second-level key elements, and regions between the second and third importance thresholds are labeled as third-level key elements, thus obtaining a hierarchical set of key elements; The parameter co-occurrence matrix of key feature elements is calculated based on multiple preset association directions. The parameter co-occurrence matrix is ​​used to extract coupled texture features, which include association contrast, parameter correlation, feature energy, and coupling homogeneity. Extract the morphological features of key feature elements, wherein the morphological features include the perimeter of the feature region, the area of ​​the feature region, and the morphological invariant moments; By concatenating the coupled texture features and morphological features, we obtain the first associated feature vector corresponding to the time series sequence of pipeline parameters and the second associated feature vector corresponding to the time series sequence of production capacity data.

4. The water management operation data decision support system according to claim 3, characterized in that, The operating status categories corresponding to the time series sequences of pipeline parameters and the time series sequences of production capacity data are obtained based on the density clustering algorithm, including: Calculate the dynamic distance matrix between data samples in the time series sequence of pipeline parameters and the time series sequence of production capacity data to obtain the adaptive truncation distance; Candidate state cluster centers are selected based on the local fluctuation density and relative deviation of the data in the pipeline network parameter time series and the capacity data time series. The number of final state clusters is determined by the density distribution heatmap based on the center of the candidate state clusters. A hierarchical clustering tree is constructed based on the number of final state clusters. After cluster fusion, the stable state cluster structure is retained, and the operating state category is output. The operating state category includes high load state, stable operating state, and low load state.

5. A water management operation data decision support system according to claim 4, characterized in that, The first and second associated feature vectors are input into the knowledge node generator to obtain a pipeline network knowledge node set and a capacity knowledge node set, including: The first associated feature vector is matched with the node attributes of the pipeline feature nodes in the pre-constructed pipeline feature label system to obtain the first matching result; wherein, the pipeline feature label system includes multiple pipeline feature nodes; The second associated feature vector is matched with the node attributes of the capacity feature nodes in the pre-constructed capacity feature label system to obtain the second matching result; wherein, the capacity feature label system includes multiple capacity feature nodes; Based on the first matching result, at least one pipeline feature node that is successfully matched by the first associated feature vector is determined, and at least one pipeline feature node is used as a set of pipeline knowledge nodes. Based on the second matching result, at least one capacity feature node that is successfully matched by the second associated feature vector is determined, and at least one capacity feature node is used as a set of capacity knowledge nodes.

6. The water management operation data decision support system according to claim 5, characterized in that, The first associated feature vector is matched with the node attributes of pipeline feature nodes in the pre-constructed pipeline feature label system to obtain the first matching result, including: The first associated feature vector is matched with the node attributes of the first-level pipeline feature nodes in the pipeline feature label system; After identifying the first-level pipeline feature node that has been successfully matched, the first associated feature vector is matched with the node attributes of the next-level pipeline feature node belonging to the first-level node in descending order of level. The matching degree between the first associated feature vector and the feature nodes of each level of the pipeline network is calculated. The pipeline feature node with the highest matching degree and located at the lowest level is identified as the successfully matched pipeline feature node. The second associated feature vector is matched with the node attributes of the capacity feature nodes in the pre-constructed capacity feature label system to obtain the second matching result, including: The second associated feature vector is matched with the node attributes of the first-level capacity feature nodes in the capacity feature label system; After identifying the first-level capacity feature node that has been successfully matched, the second associated feature vector is matched with the node attributes of the next-level capacity feature node belonging to the first-level node in descending order of level. The matching degree between the second associated feature vector and the capacity feature nodes at each level is calculated. The capacity feature node with the highest matching degree and located at the lowest level is identified as the successfully matched capacity feature node.

7. A water management operation data decision support system according to claim 6, characterized in that, Based on the set of pipeline knowledge nodes, the set of production capacity knowledge nodes, and multiple historical decision triples, a decision knowledge graph is constructed, including: The pipeline knowledge node set and the production capacity knowledge node set are respectively input into the graph construction module based on the relation inference engine to obtain the first association set and the second association set; Calculate the intersection of the first set of associations and the second set of associations to obtain the core set of associations; Based on the decision-related relationships in the multiple historical decision triples, the decision knowledge graph is obtained by connecting the pipeline knowledge node set and the production capacity knowledge node set based on the core relationship set.

8. A water management operation data decision support system according to claim 7, characterized in that, The decision subgraph includes decision subgraphs centered on pipeline nodes and / or decision subgraphs centered on water plant equipment. Based on the decision relationships in the multiple historical decision triples, a decision subgraph is constructed for each of the real-time decision triples, including: In response to the real-time decision triplet including a subject entity of type network node or water plant equipment and an object entity of type network node or water plant equipment, based on the decision association provided by the multiple historical decision triplets, a decision subgraph with the subject entity as the central node and a decision subgraph with the object entity as the central node are constructed for the real-time decision triplet. In response to the real-time decision triple including a subject entity of water plant equipment and an object entity of water demand information, a decision subgraph with the subject entity as the central node is constructed for the real-time decision triple based on the decision association provided by the multiple historical decision triples. Among them, the decision association types in the decision subgraph with pipeline nodes as the central nodes include water flow connectivity between pipeline nodes, pressure transmission between pipeline nodes, and supply and demand correspondence between pipeline nodes and water demand. Among them, the decision-making relationships in the decision-making subgraph with water plant equipment as the central node include the collaborative operation relationship between water plant equipment, the supply relationship between water plant equipment and water demand, and the water transmission relationship between water plant equipment and pipeline network nodes.

9. A water management operation data decision support system according to claim 8, characterized in that, Subgraph fusion is performed on the decision subgraphs corresponding to each of the real-time decision triples to obtain a fused decision subgraph for each of the real-time decision triples, including: In response to the decision subgraph centered on the subject entity and the decision subgraph centered on the object entity, the two types of decision subgraphs are concatenated by features to obtain a fused decision subgraph. In response to a decision subgraph centered on the main entity, the decision subgraph is used as the fused decision subgraph.

10. A water management operation data decision support system according to claim 9, characterized in that, Calculating the fit between the fused decision subgraph of each of the real-time decision triples and the decision knowledge graph includes: Calculate the feature similarity between the fused decision subgraph of each of the real-time decision triples and the decision knowledge graph.

Images