Targeted partitioning method and system for water network system

Abstract

The invention discloses a targeted partitioning method and system for a water network system, and the method comprises the steps: collecting water network data of a research region, recognizing water network nodes, extracting the data of each water network node, and constructing a feature value matrix of each water network node; calculating a composite distance between every two nodes, setting a neighbor quantity value based on the quantity of the characteristic values, screening water network nodes which are nearest to each other, and constructing a hierarchical graph by taking the nodes as vertexes and the nearest neighbor relation as edges to obtain a water network characteristic neighborhood graph; identifying a cutting edge based on a neighborhood relationship among the nodes, and dividing the water network feature domain graph into a plurality of water network feature sparse graphs; determining the central point of each water network feature sparse graph based on the local density of the water network nodes, calculating the similarity between the water network feature sparse graphs, clustering the water network feature sparse graphs to obtain a plurality of clustering centers, and dividing the research region into a plurality of water network system targeted partitions. According to the method, multi-source data is considered, and the refinement and scientization level of water network management is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water network partitioning, in particular to a water network system targeted partitioning method. BACKGROUND

[0002] As an important infrastructure for ensuring the rational allocation of urban and regional water resources, flood control and drought resistance, and ecological maintenance, the scientific planning and management of the water network system is crucial. With the acceleration of urbanization and the impact of climate change, the water network system is facing many challenges such as increased flood disaster risk, uneven distribution of water resources, and increased difficulty in water pollution control. In order to effectively respond to these challenges and improve the operational efficiency and problem-solving capabilities of the water network system, it is necessary to carry out reasonable partitioning management of the water network system.

[0003] Currently, the existing water network partitioning methods have many shortcomings. Some methods only rely on geographical regions or administrative boundaries for partitioning, ignoring key factors such as the topological structure of the water network, hydraulic characteristics, and water regime changes in different regions. This simple partitioning method cannot accurately reflect the complex relationships within the water network system, leading to low resource allocation efficiency and difficulty in precise policy implementation in actual water resource allocation, flood control and disaster reduction, and water pollution control. For example, when a flood arrives, it is difficult to quickly and accurately determine the flood propagation path and potentially affected areas, making it impossible to take effective flood control measures in a timely manner. In periods of water scarcity, it is difficult to achieve optimal allocation of water resources to meet the water demand of different regions. Therefore, there is an urgent need for a water network system targeted partitioning method that can consider multiple factors to improve the refinement and scientific level of water network management.

[0004] The present application proposes a water network system targeted partitioning method and system to solve the above problems and improve the refinement and scientific level of water network management. SUMMARY

[0005] The present application provides a water network system targeted partitioning method to solve the above problems in the prior art. On the other hand, a water network system targeted partitioning system is provided.

[0006] Technical solution: A water network system targeted partitioning method, comprising the following steps: Step S1, collect water network data of the study area, identify water network nodes and extract data of each water network node, and reduce the dimension of the water network node data to obtain a feature value matrix of each water network node; Step S2, calculate the complex distance between each two nodes based on the feature value matrix of each water network node, set the number of nearest neighbors based on the number of feature values, and screen water network nodes that are nearest neighbors of each other, to construct a hierarchical graph with nodes as vertices and nearest neighbor relationships as edges, and obtain a water network feature neighborhood graph; Step S3, cutting edges are identified based on the neighborhood relationship between each node, and the water network feature field graph is divided into several water network feature sparse graphs; Step S4, the center points of each water network feature sparse graph are determined based on the local density of the water network nodes, the similarity between each water network feature sparse graph is calculated, and the water network feature sparse graphs are clustered to obtain several clustering centers, and the study area is divided into several water network system targeting partitions.

[0007] According to an aspect of the present application, the step S1 is further: Step S11, collecting water network data of the study area, including: water network topology data, hydraulic data, geographic data, historical drought data and historical flood data; Step S12, identifying water network nodes and extracting data of each water network node, the water network nodes including: basin nodes, engineering nodes, monitoring nodes and geographic feature nodes; Step S13, sequentially reducing the dimension of data of all water network nodes to obtain the eigenvalue matrix of each water network node.

[0008] According to an aspect of the present application, the step S13 is further: Step S13a, extracting several water network node attributes based on the data of the water network nodes, and calculating the correlation coefficient between all water network node attributes, classifying the water network node attributes with a correlation coefficient greater than a threshold value into the same water network node feature attribute to obtain a water network node feature attribute set; Step S13b, calculating the numerical value of each water network node feature attribute by using the entropy weight method, and constructing the eigenvalue matrix of each water network node.

[0009] According to an aspect of the present application, the step S2 is further: Step S21, calculating the composite distance between each two nodes based on the eigenvalue matrix of each water network node; Step S22, setting an initial neighbor number value based on the number of eigenvalues, and screening the neighbor nodes of each water network node, gradually increasing the size of the neighbor number value, and screening the neighbor nodes of each water network node based on the current neighbor number value, until all water network nodes meet the condition of having at least one neighbor node, at this time the neighbor number value is the effective neighbor number value; Step S23, screening the neighbor nodes of each water network node based on the effective neighbor number value, and constructing a hierarchical graph with nodes as vertices and nearest neighbor relationships as edges to obtain a water network feature neighborhood graph.

[0010] According to an aspect of the present application, the step S21 is further: Step S21a, identify the river basin in the study area as a river basin node, the water conservancy hub, the flood drainage channel and the storage project as an engineering node, the hydrological station as a monitoring node, the river junction, the drought area and the high flood area as a geographical feature node; Step S21b, calculate the geographical distance, the hydraulic distance, the climate distance and the management distance between each two nodes in turn; Step S21c, determine the weight of each distance and calculate the composite distance between each two nodes.

[0011] According to an aspect of the present application, the step S23 is further: Step S23a, construct a river basin node, an engineering node and a geographical feature node hierarchical neighborhood graph with nodes as vertices and near-neighbor relationships as edges; Step S23b, represent the water conservancy association, the water transfer association and the climate association with blue edges, red edges and yellow edges respectively, represent the high flood risk area node with red, the medium flood risk area node with orange and the drought area node with yellow, and represent the climate correlation with the thickness of the edge to obtain a water network feature neighborhood graph.

[0012] According to an aspect of the present application, the step S3 is further: Step S31, set a cutting threshold based on the number of water network node feature values, calculate the number of near neighbors of each water network node, mark the water network nodes with a number of near neighbors greater than the cutting threshold, and reserve the edges between the marked water network nodes and all their near neighbor nodes as cutting edges, divide the water network feature neighborhood graph into M subgraphs and a number of scattered water network nodes based on the cutting edges, and M is a positive integer greater than 10; Step S32, calculate the distance between each scattered water network node and the M subgraphs in turn, and divide the scattered water network nodes into the nearest sub-region to obtain M water network feature sparse graphs.

[0013] According to an aspect of the present application, the step S4 is further: Step S41, calculate the local density of each water network node, and determine the center point of each water network feature sparse graph based on the local density of the water network node; Step S42, calculate the distance between the center points of each two water network feature sparse graphs in turn, which is the distance between the two water network feature sparse graphs; Step S43, set a shared nearest neighbor number, and take the ratio of the number of shared natural nearest neighbors to the distance between the two water network feature sparse graphs as the similarity between the two water network feature sparse graphs, and calculate the similarity between all water network feature sparse graphs in turn; Step S44, screening out two water network feature sparse graphs with the highest similarity, clustering them into the same water network partition, obtaining a clustering label, and taking it as a water network feature sparse graph to calculate the similarity between all water network feature sparse graphs again, and repeatedly iterating until all water network feature sparse graphs belong to a certain clustering label, obtaining N clustering labels, dividing the study area into N water network system targeted partitions, and N is a positive integer less than M.

[0014] According to another aspect of the present application, a water network system targeted partition system is provided, comprising: at least one processor; and a memory in communication connection with the at least one processor; wherein, the memory stores instructions executable by the processor, and the instructions are executed by the processor to implement the water network system targeted partition method of any one of the above technical solutions.

[0015] Beneficial effects: the water network system targeted partition method is adopted to realize the targeted partition of the study area, and differentiated water saving and water supply guarantee strategies are formulated according to the water resource conditions and water demand of different partitions. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a flowchart of the present application.

[0017] Figure 2 is a flowchart of step S1 of the present application.

[0018] Figure 3 is a flowchart of step S2 of the present application.

[0019] Figure 4 is a flowchart of step S3 of the present application.

[0020] Figure 5 is a flowchart of step S4 of the present application. DETAILED DESCRIPTION

[0021] As Figure 1 shown, the following technical solutions are proposed. According to one aspect of the present application, a water network system targeted partition method is provided, characterized in that it comprises the following steps: Step S1, collecting water network data of a study area, identifying water network nodes and extracting data of each water network node, and reducing the dimension of the water network node data to obtain a feature value matrix of each water network node; Step S2, calculating the composite distance between each two nodes based on the feature value matrix of each water network node, setting the number of nearest neighbors based on the number of feature values, screening water network nodes that are nearest neighbors of each other, constructing a hierarchical graph with nodes as vertices and nearest neighbor relationships as edges, and obtaining a water network feature neighborhood graph; Step S3, identify the cutting edge based on the neighborhood relationship between each node, and divide the water network feature field map into several water network feature sparse maps; Step S4, determine the center point of each water network feature sparse map based on the local density of the water network node, calculate the similarity between each water network feature sparse map, and cluster the water network feature sparse map to obtain several clustering centers, and divide the study area into several water network system target partitions.

[0022] The current water network system partition management faces three key problems: first, the data dimension is single, and the traditional method mainly depends on geographical topology or water power data, ignores key information such as climate characteristics, historical disasters, and management ownership, resulting in that the partition result cannot reflect the comprehensive function of the water network (such as splitting the related nodes of the inter-basin water transfer project); second, the correlation description is one-sided, only using geographical distance to measure the relationship between nodes, which cannot reflect the core attributes such as water power connectivity (such as the flow influence of upstream on downstream) and climate correlation (such as the response of nodes in the same rainfall zone), and the neighborhood relationship is disconnected from the actual function; third, the partition logic is extensive, and the traditional clustering method (such as K-means) is easy to forcibly divide the densely connected core area (such as reservoir group) or miss the isolated nodes (such as small irrigation canal), resulting in that the partition is disconnected from the management demand, and it is difficult to support the targeted decision of flood control and drought resistance.

[0023] In actual implementation, the traditional water network partition method faces three implementation obstacles: first, it is difficult to process high-dimensional data, the node attribute of the water network involves more than 20 items (such as flow, drought frequency, elevation, etc.), direct calculation is easy to produce data redundancy, and subjective weighting will cause feature deviation; second, it is difficult to define the node correlation, the water network system has geographical, water power, climate and other multi-dimensional correlations, a single distance index cannot accurately distinguish key correlation and secondary correlation, and it is easy to misjudge the near neighbor relationship (such as excluding the nodes in the irrigation area due to geographical distance); third, it is difficult to guarantee the integrity of the partition, and it is difficult to balance the attribution of the core area and the isolated node, if forcibly clustering, the control relationship of the water conservancy project will be split, and if ignoring the isolated node, the management blind area will be caused.

[0024] The present application firstly reduces the dimension through multi-source data fusion and entropy weight method, integrates topological, water power, geographical, historical disaster and other data, extracts core features (such as flood risk, water power connectivity), constructs an objective feature value matrix, solves the problems of high-dimensional data redundancy and subjective deviation; secondly, the composite distance model is innovated, the geographical, water power, climate and management four types of distance are weighted and fused, the multi-dimensional neighborhood graph is constructed combined with dynamic near neighbor adjustment, and the node correlation is accurately described; thirdly, the core correlation (such as the connection between large water conservancy projects) is reserved by cutting edge identification, and the isolated node is distributed to the nearest subgraph, and the integrity of the partition is guaranteed; finally, the center point is determined based on the local density, the similarity is calculated by sharing the near neighbor + distance, and the function-oriented targeted partition is obtained by iterative clustering, which realizes the partition to adapt to the management demand.

[0025] According to an aspect of the present application, the step S1 is further: Step S11, collecting water network data of the study area, including: water network topology data, hydraulic data, geographic data, historical drought data and historical flood data; Step S12, identifying water network nodes and extracting data of each water network node, the water network nodes including: basin nodes, engineering nodes, monitoring nodes and geographic feature nodes; Step S13, sequentially reducing the dimensionality of the data of all water network nodes to obtain the eigenvalue matrix of each water network node.

[0026] According to an aspect of the present application, the step S13 is further: Step S13a, extracting a plurality of water network node attributes based on the data of the water network nodes, and calculating the correlation coefficients between all water network node attributes, and classifying the water network node attributes with correlation coefficients greater than a threshold value into the same water network node characteristic attribute to obtain a water network node characteristic attribute set; From the collected topology, hydraulic, geographic, historical disaster, management data, screening the quantitative indicators reflecting the core function and characteristics of the nodes, specifically: Identify node types (basin / engineering / monitoring / geographic feature nodes) and determine attribute extraction direction (e.g. engineering nodes need to highlight the regulation capacity, and geographic feature nodes need to highlight the disaster sensitivity); Screening quantifiable indicators from the corresponding data sources (excluding qualitative descriptions, such as converting river width to river average width); Unify data units (e.g. drought days unit is d, and flow unit is m3 / s).

[0027] The water network node attributes specifically include: Hydraulic function attributes: multi-year average flow, water level amplitude, water transmission efficiency, river flow capacity; Geographical environment attributes: node elevation, soil permeability, vegetation coverage, river average width; Disaster risk attributes: annual average flood days, flood inundation area, annual average drought days, drought impact range; Engineering regulation attributes: total reservoir capacity, gate regulation accuracy, pumping capacity of pumping station; Monitoring feedback attributes: water level monitoring error, water quality compliance rate, flow monitoring frequency; Management coordination attributes: administrative division attribution, cross-regional dispatch response time.

[0028] The Pearson correlation coefficient method is used to calculate the linear correlation degree of two attributes, and the specific steps are: Determine the attribute pair to be calculated (e.g. annual average flood days and node elevation, multi-year average flow and water transmission efficiency); Collect two attribute data of all nodes (assuming 100 nodes, get two groups of 100 data: X=[x_1,x_2,...,x_{100}], Y=[y_1,y_2,...,y_100]; Substitute the Pearson correlation coefficient formula to calculate the correlation coefficient; Determine the correlation: set the threshold value, if the correlation coefficient is greater than the threshold value, it is considered that the two attributes are strongly correlated, and can be classified into the same characteristic attribute.

[0029] In an embodiment, 22 original attributes are extracted, and after correlation calculation, a set of 8 characteristic attributes is finally formed, specifically: Flood risk characteristics: average annual flood days, flood inundation area, node elevation, correlation: "average annual flood days-flood inundation area" r=0.85, "flood inundation area-node elevation" r=-0.78 (negative correlation is still strong correlation); Drought risk characteristics: average annual drought days, drought impact range, soil permeability, correlation: "average annual drought days-drought impact range" r=0.91, "drought impact range-soil permeability" r=-0.73; Hydraulic connectivity characteristics: multi-year average flow, water delivery efficiency, river flow capacity, correlation: "multi-year average flow-water delivery efficiency" r=0.82, "multi-year average flow-river flow capacity" r=0.87; Engineering regulation characteristics: total reservoir capacity, gate regulation accuracy, pumping station water lifting capacity, only engineering node attribute, correlation>0.75; Geographical location characteristics: node elevation, river average width, distance from main stream, correlation: "node elevation-distance from main stream" r=0.72; Ecological environment characteristics: vegetation coverage, water quality compliance rate, correlation: r=0.71; Monitoring accuracy characteristics: water level monitoring error, flow monitoring frequency, only monitoring node attribute, correlation r=0.76; Management coordination characteristics: cross-regional scheduling response time, administrative partition attribution code, correlation: r=0.74.

[0030] In particular, in the characteristic attribute set, some original attributes (such as node elevation) may be repeatedly classified because they are associated with multiple core characteristics (such as flood risk and geographical location), and need to be prioritized in the core associated characteristics (such as node elevation prioritized in the flood risk characteristic). Step S13b, calculate the numerical value of each water network node characteristic attribute using the entropy weight method, and construct the characteristic value matrix of each water network node.

[0031] The entropy weight method is used to calculate the numerical values ​​of the characteristic attributes of each water network node, specifically as follows: Standardize the original attribute data, where the positive attribute is: x' ij = (x ij -x jmin ) / (x jmax -x jmin ); Negative attribute: x' ij = (x jmax -x ij ) / (x jmax -x jmin ); Where, x ij Let x be the j-th original attribute value of the i-th node. jmax / x jmin x' is the maximum / minimum value of the j-th attribute. ij The standardized value (range 0-1); Calculate the information entropy of an attribute. Information entropy reflects the degree of dispersion of the attribute: the smaller the entropy value, the higher the dispersion of the attribute, the more information it contains, and the greater its weight. Calculate the weight of each original attribute in the feature attributes based on information entropy: The standardized original attribute values ​​are summed with their corresponding weights to obtain the final value (range 0-1) for that node under that feature attribute: Repeat the above steps for other feature attributes to obtain the value of each node under all feature attributes.

[0032] In one embodiment, specifically: Selecting three nodes in a watershed: A, B, and C, the flood risk characteristics include three original attributes: A: The average number of flood days per year is 12, the flood-inundated area is 80 km2, and the node elevation is 25 m. B: Average number of flood days per year: 8 days; flood inundation area: 50 km2; node elevation: 35 m. C: Average number of flood days per year: 15 days; flood inundation area: 100 km2; node elevation: 18 m. After standardization: A: 0.57, 0.6, 0.59; B: 0.0, 0.0, 0.0; C: 1.0, 1.0, 1.0; Calculate information entropy: A: (0.57+10^{-6}) / (0.57+0+1+3×10^{-6})≈0.36; B: ≈0; C: ≈0.64; e_1 = -1 / ln3 * (0.36 * ln0.36 + 0 * ln0 + 0.64 * ln0.64) = 0.89; e_2 = 0.89; e_3 = 0.89; wherein e_1 represents flood days, e_2 represents inundated area, and e_3 represents elevation; Calculate weights: \sum_{j=1}^{3} (1 - e_j) = (1 - 0.89) + (1 - 0.89) + (1 - 0.89) = 0.33; w_1 = w_2 = w_3 = 0.11 / 0.33 = 0.33; Calculate feature attribute values: A: 0.33 * 0.57 + 0.33 * 0.6 + 0.33 * 0.59 = 0.59; B: 0.33 * 0 + 0.33 * 0 + 0.33 * 0 = 0; C: 0.33 * 1 + 0.33 * 1 + 0.33 * 1 = 1.0; After calculating the values of all 8 feature attributes, a feature value matrix of 3 nodes is constructed.

[0033] In another embodiment of the present application, step S13 can also be: The node function mutation recognition in step S13a includes: constructing a function state vector containing five dimensions of drainage, water storage, regulation and storage, flood discharge, and water transport, each dimension representing the activation degree of the corresponding function; calculating the rainfall impact through the hyperbolic tangent function of the difference between the rainfall intensity and the critical value, and obtaining the function change amount combined with the state transition matrix; decomposing the state transition matrix into a basic matrix and a sum of correction matrices of soil water content, pipe network load, and terrain gradient; decomposing the function change into minute-level fast variables, hour-level medium variables, and day-level slow variables, and obtaining the comprehensive function state by superposition.

[0034] The authenticity discrimination of the function mutation includes: calculating the time integral of the function state relative to the reference state as a persistence index, calculating the complement of the correlation coefficient of the current state and the historical state as an irreversibility index, and calculating the average value of the function change of the adjacent nodes as a cascading index; combining the three indexes by weighting and mapping through an S-shaped function to obtain a mutation authenticity score; when the authenticity score is greater than a preset threshold, it is determined that there is a real function mutation, otherwise it is determined that there is a false mutation.

[0035] Optionally, further comprising: correcting the detected function state according to the authenticity score, retaining the detection result of the real mutation, and restoring the pseudo mutation to the benchmark state; predicting the mutation duration based on a logarithmic function of a difference between the corrected function state and the benchmark state; and using the corrected function state vector as a node feature attribute for subsequent water network partitioning.

[0036] Specifically, the method comprises the following steps: Step S13a1, constructing a node function state vector; In this embodiment, a five-dimensional function vector is defined to represent the multi-function characteristics of the water network node. Specifically, the function vector is represented as: F(t)=[f_drain, f_store, f_regulate, f_discharge, f_transfer]; Wherein, F(t) represents the node function state vector at time t; f_drain represents the drainage function activation degree, with a value range of 0 to 1; f_store represents the water storage function activation degree, with a value range of 0 to 1; f_regulate represents the regulation function activation degree, with a value range of 0 to 1; f_discharge represents the flood discharge function activation degree, with a value range of 0 to 1; f_transfer represents the water transfer function activation degree, with a value range of 0 to 1.

[0037] The initial values of each function activation degree are obtained based on the statistical data of the historical 3-year operation data. For example, the initial function vector of a certain drainage pump station is [0.8, 0.1, 0.05, 0.03, 0.02], indicating that the node mainly undertakes the drainage function.

[0038] Step S13a2, calculating the state transition driven by rainfall; In this embodiment, a rainfall impact function is introduced to simulate the influence of extreme rainfall on node function. Specifically, the state transition calculation formula is: ΔF(t)=tanh(α×(R(t)-R_critical))×W×F(t-1); Wherein, ΔF(t) represents the change amount of the function vector; R(t) represents the rainfall intensity at time t, with a unit of millimeters / hour; R_critical represents the critical rainfall intensity, with a value of 50-100 millimeters / hour; α represents the response sensitivity, with a value of 0.01-0.05; W represents a 5x5 state transition matrix; tanh represents the hyperbolic tangent function, which is used to limit the response to between -1 and 1.

[0039] Step S13a3, extracting the early influence factor; In this embodiment, the fixed transition matrix is improved to a dynamic matrix considering multiple environmental factors. Specifically, the dynamic transition matrix is calculated as: W(t) = W_base + AW_soil(0) + AW_load(L) + AW_topo(G); where W_base represents the base transfer matrix; AW_soil(0) represents the soil water content correction matrix, 0 is the soil volumetric water content, ranging from 0 to 0.4; AW_load(L) represents the pipe network load correction matrix, L is the pipe network load rate, ranging from 0 to 1; AW_topo(G) represents the topographic gradient correction matrix, G is the topographic gradient, unit: degree.

[0040] The calculation formula of the soil water content correction matrix is: AW_soil(0) = k_soil x (0 - 0_field) x M_soil where k_soil represents the soil influence coefficient, taking a value of 0.5-2.0; 0_field represents the field water holding capacity, a typical value of 0.25; M_soil represents the soil influence template matrix.

[0041] Step S13a4, performing multi-time scale fusion; In this embodiment, the functional state change is decomposed into three time scales, specifically: F(t) = F_fast(t) + F_medium(t) + F_slow(t); The calculation formula of the fast variable (minute level) is: F_fast(t) = φ1 x exp(-t / τ1) x AF_rain; where φ1 represents the fast variable weight, taking a value of 0.4-0.6; τ1 represents the fast variable time constant, taking a value of 0.1-0.5 hours; AF_rain represents the functional change caused by rainfall.

[0042] The calculation formula of the medium variable (hour level) is: F_medium(t) = φ2 x (1-exp(-t / τ2)) x AF_load; where φ2 represents the medium variable weight, taking a value of 0.2-0.4; τ2 represents the medium variable time constant, taking a value of 1-5 hours; AF_load represents the functional change caused by load.

[0043] The calculation formula of the slow variable (day level) is: F_slow(t) = φ3 x (1 / (1+exp(-(t-τ3) / s))) x AF_seasonal; Wherein, φ3 represents the slow variable weight, the value is 0.1-0.2; τ3 represents the slow variable time center, the value is 12-24 hours; s represents the conversion width, the value is 2-6 hours; ΔF_seasonal represents the seasonal function change.

[0044] Step S13a5, extracting mutation feature fingerprint; In this embodiment, a three-dimensional mutation feature vector is defined to distinguish real mutations and false mutations, specifically, the feature vector is calculated as: Persistence index: P=(∫(F(t)-F_baseline)dt) / T; Wherein, the integral interval is T hours after the mutation starts, T takes the value of 2-6 hours; F_baseline represents the baseline function state.

[0045] Irreversibility index: I=1-corr(F(t),F(t-T)); Wherein, corr represents the correlation coefficient calculation function; T represents the observation time window.

[0046] Cascade index: C=(Σneighbor_change) / N_neighbors; Wherein, neighbor_change represents the function change of the adjacent node; N_neighbors represents the number of adjacent nodes.

[0047] Step S13a6, constructing real and false mutation discriminator; In this embodiment, the authenticity of the mutation is determined by weighted combination of the three feature indexes, specifically, the discrimination function is: S_real=1 / (1+exp(-(w1×P+w2×I+w3×C-θ_truth))); Wherein, S_real represents the mutation authenticity score, ranging from 0 to 1; w1, w2, w3 represent the feature weight, which is obtained by training historical data, and the typical values are 0.4, 0.3, 0.3 respectively; θ_truth represents the authenticity threshold, the value is 0.5-0.7.

[0048] The weight is obtained by least square method optimization: minΣ(S_real-S_observed)²; Step S13a7, performing function state correction and prediction.

[0049] In this embodiment, the detected function state is corrected based on the authenticity score, specifically: F_corrected(t) = S_real x F_detected(t) + (1 - S_real) x F_baseline(t); The mutation duration prediction formula is: T_duration = kappa x ln(1 + ||F_corrected - F_baseline||) x (1 / decay_rate); Wherein, kappa represents the duration coefficient, the value is 10-30; ||·|| represents the vector norm; decay_rate represents the recovery rate, the value is 0.05-0.2 per hour.

[0050] For example, when the 1-hour rainfall of a certain drainage pump station reaches 120 mm, the function vector changes from [0.8, 0.1, 0.05, 0.03, 0.02] to [0.2, 0.6, 0.15, 0.03, 0.02] quickly, the drainage function decreases, and the water storage function increases rapidly, through mutation feature extraction, the persistence P=0.8, the irreversibility I=0.7, and the cascading C=0.6 are obtained, the discrimination score S_real=0.82 is obtained, and the real mutation is confirmed, and it is predicted that the mutation will last for 8.5 hours.

[0051] Through the above steps, the node function mutation recognition accuracy reaches more than 95%, and the response time is controlled within 20 seconds.

[0052] According to one aspect of the application, the step S2 is further: Step S21, calculating the composite distance between each two nodes based on the eigenvalue matrix of each water network node; Step S22, setting an initial neighbor number value based on the number of eigenvalues, and screening the neighbor nodes of each water network node, gradually increasing the size of the neighbor number value, and screening the neighbor nodes of each water network node based on the current neighbor number value, until all water network nodes meet at least one neighbor node, and the neighbor number value at this time is the effective neighbor number value; Step S23, screening the neighbor nodes of each water network node based on the effective neighbor number value, and constructing a hierarchical graph with nodes as vertices and nearest neighbor relationships as edges to obtain a water network feature neighborhood graph.

[0053] Water network targeted partitioning is the core basis of water network planning layout and system optimization, and its scientific demarcation needs to consider multiple dimensional influence factors, the present application comprehensively considers the distribution characteristics of regional topography, water system, social and economic development level, etc., combines rainfall, evaporation and aridity, annual runoff spatial distribution law, arid region division, floodwater risk elements, comprehensive risk degree and floodwater partitioning, and adopts a water network targeted partitioning method based on a water network feature neighborhood graph.

[0054] If each data point has at least one natural neighbor, the entire water network data set can be considered to be in a stable distribution state, and the definition of the existence of a natural neighbor relationship between two data points will have an edge between the corresponding nodes in the water network feature neighborhood graph. Further, to construct the water network natural neighborhood graph, the following four variables need to be determined: Nearest neighbor determination: for any data point, by searching the nearest neighbor function, the search times of the data point are determined, and the nearest neighbor set is obtained; Stable search state: for each data point in the data set, there is at least one data point such that the two are nearest neighbors; Natural neighbor: if data point 1 is one of the nearest neighbors of data point 2, and data point 2 is also one of the nearest neighbors of data point 1, then data point 2 is considered to be a natural neighbor of data point 1; Natural feature value set: is a set of positive integers that satisfy the condition that for each data point in the data set, there is at least one data point such that the two points are nearest neighbors.

[0055] According to one aspect of the application, the step S21 is further: Step S21a, identifying the basins in the study area as basin nodes, the water conservancy hubs, flood drainage channels and storage projects as engineering nodes, the hydrological stations as monitoring nodes, the river junctions, arid regions and flood-prone areas as geographic feature nodes; Step S21b, calculating the geographic distance, hydraulic distance, climate distance and management distance between each two nodes in turn; Geographical distance is the straight-line distance of the spatial position of the node, which is calculated by the Haversine formula based on latitude and longitude coordinates; Hydraulic distance represents the transmission time of water flow from upstream node to downstream node; Climate distance represents the difference in climate characteristics between nodes, which is quantified by the standard deviation of multi-year average rainfall; Management distance is the degree of difficulty of water conservancy management coordination between nodes, which is quantified by a 0-1 non-dimensional value, and a management unit attribution + response time double factor comprehensive assignment is adopted: Management unit attribution: 0.2 for attribution to the same level of basin administration, 0.5 for attribution to the same level of water conservancy department, and 0.8 for attribution across levels; Dispatch response time: the average response time of cross-regional dispatch is standardized to 0-0.2; Final management distance: D_{man}=management unit attribution value+response time standardized value.

[0056] Step S21c, determine the weight of each distance and calculate the composite distance between each two nodes.

[0057] The importance of the four types of distances is scored, a judgment matrix is constructed, the maximum eigenvalue, random consistency index, and consistency ratio of the judgment matrix are calculated, and finally the weight vector is determined; The composite distance is calculated as: D AB =w1D geo，AB +w2D hyd，AB +w3D cli，AB +w4D man，AB ; Wherein D AB is the composite distance between nodes A and B.

[0058] According to an aspect of the present application, the step S23 is further: Step S23a, taking nodes as vertices and near-neighbor relationships as edges, respectively constructing the watershed node, engineering node and geographical feature node hierarchical neighborhood graph; Step S23b, respectively taking blue edges, red edges and yellow edges to represent water-related, water transfer-related and climate-related, taking red to represent high flood risk area nodes, orange to represent medium flood risk area nodes, and yellow to represent arid area nodes, and taking the thickness of the edge to represent climate correlation, to obtain a water network feature neighborhood graph.

[0059] The sub-graphs are constructed according to the node types, including: watershed node hierarchical graph, engineering node hierarchical graph and geographical feature node hierarchical graph, and multi-dimensional visual variables of color-shape-thickness are used to intuitively display node types, risk levels and correlation characteristics, specifically: node color represents disaster risk level, wherein red: flood risk characteristic value > 0.8 (high risk); orange: 0.5 ≤ flood risk characteristic value ≤ 0.8 (medium risk); yellow: drought risk characteristic value > 0.7 (arid area); green: other (low risk); Node shape represents node type, wherein circle: watershed node; square: engineering node; triangle: monitoring node; rhombus: geographical feature node; Edge color represents correlation type, wherein blue: water-related between watershed nodes (water collection / water transfer); red: water transfer-related between engineering nodes; yellow: climate-related between geographical feature nodes; Edge thickness represents climate correlation, wherein thick edge (line width 3pt): climate distance < 10mm (high correlation); medium edge (2pt): 10mm ≤ climate distance ≤ 30mm (medium correlation); thin edge (1pt): climate distance > 30mm (low correlation).

[0060] According to an aspect of the present application, the step S21 of calculating the hydraulic distance can also be: The hydraulic conduction delay is calculated, including: The basic time delay is calculated according to the river length between nodes, the basic wave speed and the real-time flow, and is asymmetrically modified through the exponential decay relationship between the potential energy gradient and the flow; the hydraulic state is divided into steady state, gradual change state and sudden change state according to the flow change rate, and a smooth transition function is used to realize the continuous transition between different states; the modified basic time delay is used for the steady state, the time delay adjusted by a logarithmic function is used for the gradual change state, and the time delay of exponential decay is used for the sudden change state; the comprehensive time delay is obtained by weighted combination of the time delays of the three states, and the historical memory mechanism is introduced to smooth the output time delay, and the smoothed time delay is used as the hydraulic distance between nodes.

[0061] The specific implementation of the hydraulic state division is as follows: When the flow change rate is less than a first threshold value, it is determined to be a steady state, when the flow change rate is between the first threshold value and a second threshold value, it is determined to be a gradual change state, and when the flow change rate is greater than or equal to the second threshold value, it is determined to be a sudden change state; the value range of the first threshold value is 0.1 to 0.5 cubic meters per second square, and the value range of the second threshold value is 2.0 to 5.0 cubic meters per second square; the smooth transition function uses an S-shaped function, and the smoothness of state transition is controlled by adjusting the transition steepness parameter.

[0062] The historical memory mechanism includes: The current calculated comprehensive time delay and the output time delay at the previous time are weighted and averaged, wherein the update rate is dynamically adjusted according to the flow change rate, and the more intense the flow change, the higher the update rate; near the state transition boundary, the steepness parameter of the transition function is dynamically adjusted according to the variance of the flow change rate, and the larger the variance, the smaller the steepness, so as to reduce the time delay jump during state switching.

[0063] Specifically, the method comprises the following steps: Step S21b1, constructing a basic time delay matrix; In this embodiment, the time delay difference factor is introduced based on the simplified Saint-Venant equation to construct the basic time delay matrix between nodes, specifically, for any two nodes i and j in the water network, the basic time delay calculation formula is: τ(i,j)=L(i,j) / (c0+α×Q(t)); Wherein, τ(i,j) represents the hydraulic conduction time delay from node i to node j, and the unit is hour; L(i,j) represents the river length between node i and node j, and the unit is kilometer; c0 represents the basic wave speed, and the value range is 0.5-2.0 m / s, which is determined according to the average water depth and the bottom slope of the river; Q(t) represents the real-time flow at time t, and the unit is cubic meter per second; α represents the flow influence coefficient, and the value range is 0.001-0.01, which is used to quantify the influence degree of flow on propagation speed.

[0064] Step S21b2, asymmetrically modifying; In the embodiment, a potential energy gradient correction term is introduced to capture the asymmetric characteristics of hydraulic conduction, and specifically, the calculation formula of the correction term is: Δτ=β×(Hi-Hj)×exp(-γ×|Q(t)| / Qbase); wherein Δτ represents a time delay correction quantity, in hours; Hi represents the water level elevation of node i, in meters; Hj represents the water level elevation of node j, in meters; β represents a potential energy influence coefficient, with a value range of 0.1-0.5, taking a positive value when water flows from a high place to a low place, and taking a negative value otherwise; γ represents a flow decay coefficient, with a value range of 0.5-2.0; and Qbase represents a reference flow, usually taking the annual average flow value, in cubic meters per second.

[0065] The corrected time delay is: τ_corrected(i,j)=τ(i,j)+Δτ; Step S21b3, identifying a hydraulic state; In the embodiment, the current hydraulic state is determined by calculating the flow rate of change, and specifically, three hydraulic states are defined: steady state, gradual change state and sudden change state, and the state determination is based on the following: When dQ / dt<θ1, it is determined to be a steady state; when θ1≤dQ / dt<θ2, it is determined to be a gradual change state; and when dQ / dt≥θ2, it is determined to be a sudden change state, wherein dQ / dt represents the derivative of flow with respect to time, in cubic meters per second 2 ; θ1 represents a steady state threshold, with a value range of 0.1-0.5 cubic meters per second 2 ; and θ2 represents a sudden change threshold, with a value range of 2.0-5.0 cubic meters per second 2 .

[0066] Step S21b4, performing segmented time delay calculation; In the embodiment, different time delay calculation methods are used according to the identified hydraulic state. For the steady state, the corrected time delay of step S21b2 is used; for the gradual change state, the time delay calculation formula is: τ_gradual=τ_base×(1+κ×ln(1+|dQ / dt|)); wherein τ_gradual represents the time delay of the gradual change state; τ_base represents the base time delay; and κ represents a gradual change adjustment coefficient, with a value range of 0.2-0.8.

[0067] For the sudden change state, the time delay calculation formula is: τ_sudden=τ_min+(τ_base-τ_min)×exp(-λ×t); wherein τ_sudden represents the time delay of the sudden change state; τ_min represents the minimum time delay, usually 0.1-0.3 times of τ_base; λ represents an attenuation rate, with a value range of 0.5-2.0; and t represents the time after the sudden change, in hours.

[0068] Step S21b5, constructing a smooth state transition function; In this embodiment, a Sigmoid function is used to realize smooth transition between different states. Specifically, the weight function is defined as: w(x) = 1 / (1+exp(-p x (x-theta))); wherein w(x) represents the weight function value, ranging from 0 to 1; x represents the flow rate change rate dQ / dt; p represents the transition steepness parameter, taking a value ranging from 2.0 to 10.0; and theta represents the transition center point, corresponding to the state transition threshold theta1 or theta2.

[0069] Through the combination of the three weight functions, the final time delay is calculated as: tau_final = w1(dQ / dt) x tau_steady + w2(dQ / dt) x tau_gradual + w3(dQ / dt) x tau_sudden; wherein w1, w2 and w3 respectively correspond to the weights of the steady state, the gradual state and the sudden state, satisfying w1+w2+w3=1.

[0070] Step S21b6, applying a history memory mechanism; In this embodiment, a time delay history buffer is introduced to prevent instantaneous jumps. Specifically, the update formula of the output time delay is: tau_output(t) = eta x tau_final(t) + (1-eta) x tau_output(t-Dt); wherein tau_output(t) represents the output time delay at time t; eta represents the update rate, with a basic value of 0.3; and Dt represents the time step, usually taking 5 minutes.

[0071] The update rate is dynamically adjusted according to the hydraulic state: eta = eta_base x (1+mu x |dQ / dt|); wherein eta_base represents the basic update rate, taking a value of 0.2-0.4; and mu represents the sensitivity coefficient, taking a value of 0.1-0.5.

[0072] Step S21b7, performing boundary adaptive correction.

[0073] In this embodiment, a buffer width adaptive mechanism is introduced near the state transition boundary. Specifically, the transition steepness parameter is dynamically adjusted as: p(t) = p_base x (1-v x Var(dQ / dt)); wherein p(t) represents the transition steepness at time t; p_base represents the basic steepness value, taking a value of 5.0; Var(dQ / dt) represents the variance of the flow rate change rate in the past 1 hour; and v represents the variance sensitivity coefficient, taking a value of 0.1-0.3.

[0074] For example, in a certain water network system, node A is located in an upstream reservoir, and node B is located in a downstream urban area. Under normal circumstances, the distance between the two nodes is 30 kilometers, the basic wave speed is 1.5 meters per second, and the calculated basic time delay is 5.56 hours. When the upstream floodgate is opened, the flow increases from 100 cubic meters per second to 500 cubic meters per second within 10 minutes, and dQ / dt reaches 40 cubic meters per second². The system determines that it is in a sudden state, and at this time the time delay is rapidly reduced to 1.2 hours. Through the historical memory mechanism, the output time delay is smoothly transitioned, avoiding system shock caused by sudden changes.

[0075] Through the above steps, the dynamic quantification of hydraulic conduction time delay is realized, the time delay prediction accuracy is improved to more than 85%, and the calculation time is controlled within 50 milliseconds.

[0076] According to one aspect of the present application, the step S3 is further: Step S31, based on the number of water network node characteristic values, set the cutting threshold, calculate the number of near neighbors of each water network node, mark the water network nodes with a number of near neighbors greater than the cutting threshold, and reserve the edges of the marked water network nodes and all their near neighbor nodes as cutting edges. Based on the cutting edges, the water network feature neighborhood graph is divided into M subgraphs and a number of scattered water network nodes, where M is a positive integer greater than 10. Step S32, calculate the distance of each scattered water network node and M subgraphs in turn, and divide the scattered water network nodes into the nearest sub-region to obtain M water network feature sparse graphs.

[0077] Hierarchical clustering is an important branch of clustering algorithms, which constructs a tree-like hierarchical structure through similarity calculation between objects. Hierarchical clustering is divided into two types: agglomerative hierarchical clustering and divisive hierarchical clustering. The latter is a variant based on the former. Agglomerative hierarchical clustering mainly includes the following two steps: first, calculate the similarity between data point pairs; then, iteratively select the data point pair or cluster with the highest similarity for merging; while divisive hierarchical clustering involves three main steps: first, identify and select the cluster with the highest dissimilarity for splitting; second, find the point with the highest separation degree in the cluster to be split and divide it into a new cluster; finally, recalculate the similarity between the data points in the to-be-split cluster and the data points in the new cluster, and according to these similarities, objects are grouped into more similar clusters.

[0078] Hierarchical clustering often improves the performance of hierarchical clustering by combining the advantages of other algorithms, but at the same time it also brings some limitations, such as being sensitive to parameter configuration and being too low in efficiency when dealing with large data.

[0079] In the embodiment, a hierarchical clustering algorithm based on the water feature neighborhood graph is adopted, which includes two processes of water feature neighborhood graph division and neighborhood subgraph merging. Firstly, in the division process, the standard water feature neighborhood graph is cut to obtain the subgraph located in the data core, and the remaining points are distributed based on the shortest distance. Secondly, in the neighborhood subgraph merging process, the two most similar subgraphs are dynamically merged according to the similarity matrix, thereby effectively solving the problem of density difference between clusters.

[0080] The specific process of water feature neighborhood graph division is as follows: Search for the nearest neighbor of each data point; Find the stable search state of natural neighbors; Add edges between data points with natural neighbor relationship; Identify the cut edge and divide the water feature neighborhood graph into a water feature sparse graph; Assign the remaining nodes to the nearest water feature sparse graph; Obtain the final water feature sparse graph.

[0081] The cut edge refers to the edge connecting the nodes, at least one of which has more natural neighbors than the threshold value. These edges constitute the core area of the natural neighborhood graph, specifically: cut_edge(v i ,v j )={e(v i ,v j )∣(nb(i)>(u-1)∣∣(nb(j)>(u-1)),(e(v i ,v j )∈nan_dege)}; When there is a natural neighbor relationship between nodes, an edge will be formed between them. For each pair of natural neighbor nodes, as long as the number of natural neighbors of one of the nodes exceeds the threshold value, the edge will be retained, which is called a cut edge. These cut edges define the core part of the natural neighborhood graph. By using these cut edges and water feature sparse graphs, and distributing the remaining nodes to the nearest subgraph based on the nearest neighbor principle, the division of the natural neighborhood graph is completed.

[0082] According to one aspect of the present application, step S31 further includes water network edge risk assessment, specifically: The hydraulic importance is calculated according to the ratio of the actual flow of the edge to the maximum design flow, the supply guarantee rate and the complement of the redundancy of the alternative path; the recursive formula of the node function integrity is used to simulate the cascade propagation process after the edge failure, wherein the function loss is related to the influence weight and loss proportion of the upstream failed edge; the time-varying propagation rate is calculated according to the power function of the ratio of the current water level to the conventional water level and the exponential function of the reciprocal of the flow; the cascade influence is divided into a direct layer, an indirect layer and a remote layer according to the propagation distance, and the accurate calculation, linear approximation and statistical estimation methods are used respectively.

[0083] The specific implementation of the hierarchical cascade evaluation includes: for the direct layer nodes within one hop range from the failed edge, the direct loss is obtained by accumulating the product of the function integrity and the influence weight of each node; for the indirect layer nodes within two to three hops from the failed edge, the indirect loss is obtained by multiplying the product of the number of nodes and the average loss rate by the attenuation coefficient; for the remote layer nodes more than three hops from the failed edge, the remote loss is estimated by using a negative exponential decay function; and the total loss approximation value is obtained by weighted summation of the three layer losses.

[0084] The identification of the vulnerable edge further includes: the vulnerability evaluation of the edge is dynamically updated by a weighted average model of the calculated value and the historical observed value; the vulnerability at a future time is predicted according to the vulnerability change rate and the seasonal periodic function; and the edges with the vulnerability exceeding a preset threshold are marked as critical vulnerable edges, which are preferentially protected from being cut off during graph cutting.

[0085] In other words, according to an aspect of the present application, in step S31, a water network edge risk assessment is further included, specifically: Identify the vulnerable edge in the system and calculate the influence of its failure on the stability of the partition; When determining the cutting threshold, consider the distribution of the vulnerable edge to avoid using a critical vulnerable edge as the only connection between partitions; For the region containing a high-risk vulnerable edge, appropriately reduce the cutting threshold and increase the redundant connection.

[0086] In other words, step S3A, identifying the vulnerable edge of the water network system: Calculate the hydraulic importance and failure risk of each edge; Simulate the cascade influence of edge failure; Mark the critical vulnerable edge and establish a risk profile; Step S31, adjust the cutting strategy based on the risk assessment: Adjust the region containing the vulnerable edge based on the original cutting threshold; Ensure that each partition has sufficient redundant connections.

[0087] In one embodiment, according to an aspect of the present application, step S31 further includes partition stability verification: Performing vulnerability analysis on each subgraph after cutting; Identifying critical vulnerable edges between and within partitions; Evaluating the impact of vulnerable edge failure on partition function; If it is found that the partition is overly dependent on the vulnerable edge, adjust the partition boundary.

[0088] Specifically, the following steps are included: Step S31a1, constructing a hydraulic dependency graph; In this embodiment, the hydraulic importance of the edge is defined to quantify the criticality of each edge in the water network system. Specifically, the hydraulic importance calculation formula is: I_h(e)=Q(e) / Q_max×P_supply(e)×(1-R_alternative(e)); Where I_h(e) represents the hydraulic importance of edge e, with a value range of 0 to 1; Q(e) represents the actual flow of edge e, with a unit of cubic meters per second; Q_max represents the maximum design flow of the edge, with a unit of cubic meters per second; P_supply(e) represents the water supply guarantee rate of edge e, with a value range of 0.85 to 0.99; R_alternative(e) represents the redundancy of alternative paths, and the calculation formula is: R_alternative(e)=N_alt / (N_alt+1); where N_alt represents the number of alternative paths.

[0089] Step S31a2, simulating cascading failure propagation; In this embodiment, a failure propagation model is established to predict the cascading effect caused by single edge failure. Specifically, the propagation formula of node failure state is: F(v,t+1)=F(v,t) ×(1-Σ(w_e×Loss(e))); Where F(v,t) represents the functional integrity of node v at time t, with a value range of 0 to 1, and the initial value is 1; w_e represents the influence weight of edge e on node v, which is determined according to the topological distance; Loss(e) represents the functional loss ratio caused by edge e failure.

[0090] The calculation formula of loss ratio is: Loss(e)=min(1,Q_lost(e) / Q_demand(v)); Where Q_lost(e) represents the flow loss caused by edge e failure; Q_demand(v) represents the water demand of node v.

[0091] Step S31a3, calculating the dynamic propagation rate; In this embodiment, time-varying propagation rate is introduced to adapt to the cascade speed difference under different water conditions. Specifically, the cascade propagation rate calculation formula is: v_cascade(t)=v_base×(H(t) / H_normal)^β×exp(-γ / Q(t)); Wherein, v_cascade(t) represents the cascade propagation rate at time t, the unit is node / hour; v_base represents the basic propagation rate, the value is 1-5 node / hour; H(t) represents the current water level, the unit is meter; H_normal represents the normal water level, the unit is meter; β represents the water level influence index, the value is 1.5-2.5; γ represents the flow influence coefficient, the value is 10-50; Q(t) represents the current flow, the unit is cubic meter / second.

[0092] Step S31a4, analyze multi-path cascade coupling; In this embodiment, the mutual influence between multiple failure paths is considered. Specifically, the total loss calculation formula is: Loss_total=Loss_direct+Σ(α_ij×Loss_i×Loss_j); Wherein, Loss_total represents the total loss; Loss_direct represents the direct loss; Loss_i and Loss_j represent the loss of path i and j respectively; α_ij represents the path coupling coefficient, the calculation formula is: α_ij=overlap(i,j) / min(length(i),length(j)); Wherein, overlap(i,j) represents the number of overlapping nodes of path i and j; length represents the path length.

[0093] Step S31a5, perform critical path pre-screening; In this embodiment, the critical path set is screened based on the hydraulic importance and topological position. Specifically, the screening criteria are: K={p|I_h(p)>θ_importanceANDDepth(p)<D_max}; Wherein, K represents the critical path set; p represents the path; θ_importance represents the importance threshold, the value is 0.6-0.8; Depth(p) represents the depth of the path in the network, that is, the shortest distance to the source node; D_max represents the maximum depth threshold, the value is 3-5.

[0094] Step S31a6, implement hierarchical cascade approximation; In this embodiment, the cascade influence is divided into three layers according to the propagation distance for differential calculation. Specifically: Direct tier (1-hop range) uses exact calculation: Loss_direct =∑(F(v,t) x w_direct(v)); Indirect tier (2-3-hop range) uses linear approximation: Loss_indirect = k1 x N_indirect x avg_loss; Far tier (more than 3-hop) uses statistical estimation: Loss_far = k2 x exp(-distance / λ) x total_nodes; Where w_direct(v) represents direct influence weight; N_indirect represents indirect tier node number; avg_loss represents average loss rate; k1 represents indirect tier attenuation coefficient, taking value 0.3-0.5; k2 represents far tier coefficient, taking value 0.1-0.2; λ represents attenuation length, taking value 5-10.

[0095] Total loss approximation is: Loss_approx = Loss_direct + k1 x Loss_indirect + k2 x Loss_far; Step S31a7, performing dynamic update of vulnerability.

[0096] In this embodiment, the vulnerability assessment of the edge is continuously corrected based on historical failure events, specifically, the update formula is: V_new(e) = λ x V_calculated(e) + (1-λ) x V_observed(e); Where V_new(e) represents the updated vulnerability of edge e; V_calculated(e) represents the vulnerability calculated by the model; V_observed(e) represents the actual observed vulnerability; λ represents learning rate, taking value 0.7-0.9.

[0097] The time series prediction formula of vulnerability is: V(e,t+Δt) = V(e,t) + μ x dV / dt + ν x seasonality(t); Where μ represents trend coefficient, taking value 0.1-0.3; dV / dt represents vulnerability change rate; ν represents seasonal coefficient, taking value 0.05-0.15; seasonality(t) represents seasonal function, using sine function fitting: seasonality(t) = A x sin(2π x t / T + φ); Where A represents amplitude; T represents period, usually 365 days; φ represents phase.

[0098] For example, in a certain urban water network system, a water pipeline connecting the main water plant and the urban area is identified, with a hydraulic importance I_h=0.85. Through cascade analysis, it is found that the failure of the pipeline will cause 15 downstream nodes to lose water within 2 hours, affecting 300,000 residents. Using the hierarchical approximation algorithm, the calculation time is reduced from 45 minutes to 30 seconds, while the prediction accuracy remains above 92%.

[0099] Through the above steps, the accuracy of critical vulnerable edge identification reaches 93%, the prediction error of cascade range is controlled within 15%, and the calculation efficiency is improved by more than 50 times.

[0100] According to an aspect of the present application, the step S4 is further: Step S41, calculate the local density of each water network node, and determine the center point of each water network feature sparse graph based on the local density of the water network node; The traditional k-nearest neighbor algorithm is often used to calculate the density, but the selection of k value is difficult to achieve the optimal, therefore, in the embodiment, the natural neighbor method is adopted to search the nearest neighbor information, which is different from the k-nearest neighbor algorithm, the natural neighbor method can adaptively find the nearest neighbor according to the structure characteristics of the data itself, and the whole process does not need any parameter, the natural neighbor set and the natural feature value are obtained through the natural neighbor method, and then the local density is calculated based on the natural neighbor.

[0101] Step S42, calculate the distance between the center points of each two water network feature sparse graphs in turn, that is, the distance of the two water network feature sparse graphs; Step S43, set the shared nearest neighbor number, take the ratio of the number of shared natural nearest neighbors and the distance of the two water network feature sparse graphs as the similarity of the two water network feature sparse graphs, and calculate the similarity between all water network feature sparse graphs in turn; Water network neighborhood subgraph merging is a step for merging water network feature sparse graphs to achieve more accurate clustering results. The input of the algorithm is the water network feature sparse graph and the expected cluster number, and the output is the clustering label, which is specifically: Initialize the number of subgraphs to be the number of water network feature sparse graphs; According to the shared natural nearest neighbor, the center distance and the similarity between subgraphs, calculate the similarity between each pair of subgraphs; When the number of subgraphs is greater than the expected cluster number, perform the following operations: Find the two subgraphs with the highest similarity; Update the similarity between the subgraph and other subgraphs; Subtract 1 from the number of subgraphs; For each data point belonging to the subgraph, assign a clustering label.

[0102] Return the final clustering label.

[0103] The shared nearest neighbor determines the number of shared nearest neighbors between two subgraphs, which is the intersection of the nearest neighbor sets of two points, for evaluating the similarity between two subgraphs; The calculation of the center point of the subgraph is specifically: C(m)=argmaxp i , v i ∈G NS (m), dc=|c(m)-c(n)|; Where p i is the local density, and v i is the ith eigenvalue.

[0104] The center distance dc is defined as the Euclidean distance between the two center points |c(m)-c(n)|.

[0105] The ratio of the number of shared natural nearest neighbors between two subgraphs and the center distance between subgraphs is the similarity between two subgraphs.

[0106] Step S44, screening out two water network feature sparse graphs with the highest similarity, and clustering into the same water network partition to obtain a clustering label, and taking the clustering label as a water network feature sparse graph to calculate the similarity between all water network feature sparse graphs again, and repeating the iteration until all water network feature sparse graphs belong to a certain clustering label, obtaining N clustering labels, and dividing the study area into N water network system targeting partitions, and N is a positive integer less than M.

[0107] In another embodiment of the present application, in step S44, when the water exchange oscillation of the adjacent partition boundary occurs, a suppression process is further included, which is specifically: The number of times of changing the direction of the boundary flow in a unit time is counted as an oscillation index; the oscillation risk is predicted according to the water level difference on both sides of the boundary, the flow rate of change, and the number of adjacent nodes through an S-shaped function; an energy function including the product of oscillation intensity and water exchange amount and the distance change of the boundary position is constructed; the boundary position is iteratively optimized through the gradient descent method combined with the momentum term to minimize the total oscillation energy.

[0108] According to the water level and the flow rate of change, a day is divided into multiple water regime stable periods, and the boundary optimization is independently performed for each period; the second-order Taylor expansion approximation of the energy function is adopted, and the optimal boundary adjustment amount is directly obtained by solving a linear equation set; a buffer zone with a width dynamically changing with the flow standard deviation and predictability is set at the boundary.

[0109] The buffer mechanism comprises that a buffer width is composed of a basic width, a flow fluctuation term and a predictability term; nodes in the buffer are not fixedly attributed to a certain partition, but are dynamically allocated according to real-time loads of each partition and node joining cost; temporary attribution of the nodes is determined by minimizing the sum of the load and the cost, so as to realize load balancing and oscillation suppression among the partitions.

[0110] Specifically, the method comprises the following steps: Step S44a1, extracting boundary oscillation features; In the embodiment, an oscillation index is defined to quantify the state switching frequency of the boundary node, and specifically, the oscillation index calculation formula is O(b) = (Σ|sign(Q(t))-sign(Q(t-1))|) / T. Wherein, O(b) represents the oscillation index of the boundary b; Q(t) represents the flow through the boundary at t, a positive value represents water supply, and a negative value represents water receiving; sign represents a sign function; T represents an observation time window, usually 1 hour.

[0111] The oscillation risk prediction formula is Risk(b) = 1 / (1+exp(-(a1 x DeltaH + a2 x dQ / dt + a3 x N_neighbors))); Wherein, Risk(b) represents the oscillation risk of the boundary b, ranging from 0 to 1; DeltaH represents the water level difference on both sides of the boundary, in meters; a1 represents a water level difference coefficient, taking a value of 0.5-1.5; a2 represents a flow rate change coefficient, taking a value of 0.1-0.3; a3 represents a neighbor influence coefficient, taking a value of 0.2-0.4; N_neighbors represents the number of adjacent nodes.

[0112] Step S44a2, performing preventive boundary adjustment; In the embodiment, the boundary position is adjusted in advance when the oscillation risk exceeds a threshold value, and specifically, the boundary adjustment formula is b_new = b_old + delta x grad(Risk) / ||grad(Risk)||. Wherein, b_new represents the new boundary position coordinate; b_old represents the original boundary position coordinate; delta represents an adjustment step, taking a value of 0.5-2.0 kilometers; grad(Risk) represents a risk gradient vector; ||·|| represents a vector module.

[0113] Step S44a3, defining an oscillation energy function; In the embodiment, a global oscillation energy function is constructed to evaluate the overall oscillation level, and specifically, E_total = Sigma(O(b)2 x V(b)) + Lambda x Sigma(Dist(b_new, b_old)). Wherein, E_total represents the total oscillation energy; V(b) represents the daily water exchange of boundary b, unit: ten thousand cubic meters; λ represents the boundary stability weight, value: 0.1-0.5; Dist represents the Euclidean distance function.

[0114] Step S44a4, gradient descent boundary optimization is performed; In the embodiment, the oscillation energy is minimized by iterative optimization, specifically, the iterative formula is: B(k+1)=B(k)-η×∇E_total+β×(B(k)-B(k-1)); Wherein, B(k) represents the boundary set of the kth iteration; η represents the learning rate, value: 0.01-0.1; ∇E_total represents the energy gradient; β represents the momentum coefficient, value: 0.5-0.9.

[0115] Step S44a5, identify the water regime mode and segment; In the embodiment, 24 hours is divided into multiple water regime stable segments, and the segmentation criterion is: |dH / dt|<threshold_H and |dQ / dt|<threshold_Q; Wherein, dH / dt represents the water level change rate, threshold threshold_H takes 0.1 meter / hour; dQ / dt represents the flow rate, threshold threshold_Q takes 10 cubic meters / second 2 .

[0116] Step S44a6, fast approximation optimization is performed; In the embodiment, the second-order Taylor expansion is used to approximate the energy function E_approx=E0+g'×ΔB+0.5×ΔB'×H×ΔB; Wherein, E0 represents the current energy value; g represents the gradient vector; H represents the Hessian matrix; ΔB represents the boundary adjustment amount.

[0117] The optimal adjustment amount is obtained by solving the linear equation set: ΔB*=-H^(-1)×g; Step S44a7, a boundary buffer mechanism is established.

[0118] In the embodiment, a buffer with a dynamic width of the boundary is set, specifically, the buffer width is calculated as: Buffer(b,t)=β0+β1×σ(Q)+β2×predictability(t) wherein Buffer(b, t) represents the buffer width of the boundary b at time t, with the unit of kilometer; β0represents the basic width, with the value of 0.5-1.0 kilometer; β1represents the flow fluctuation coefficient, with the value of 0.1-0.3; σ(Q) represents the flow standard deviation; β2represents the predictability coefficient, with the value of 0.2-0.5; predictability(t) represents the predictability index based on historical data, with the range of 0 to 1.

[0119] The dynamic allocation formula of the nodes in the buffer zone is: Assign(v) = argmin(Load(partition) + Cost(v, partition)); wherein Assign(v) represents the partition ownership of the node v; Load(partition) represents the current load of the partition; Cost(v, partition) represents the cost of the node v joining the partition.

[0120] For example, at the boundary of two partitions of a certain water network system, the original oscillation frequency is 12 times per hour, after applying the method, by setting a dynamic buffer zone with a width of 1.2 kilometers, and dynamically allocating 5 nodes in the buffer zone according to real-time load, the oscillation frequency is reduced to 2 times per hour, and the water quantity statistical error is reduced from 30% to 3%.

[0121] Through the above steps, the boundary oscillation frequency is reduced by more than 85%, the water quantity statistical error is controlled within 5%, and the system stability is significantly improved.

[0122] According to an aspect of the present application, the determination process of the water network feature sparse graph is specifically: The edge density of the graph is defined as the ratio of the actual number of edges to the maximum possible number of edges, and the calculation formula is: ρ_graph = 2E / (V(V-1)); wherein E is the number of edges in the graph, and V is the number of nodes in the graph; When the edge density ρ_graph is less than the sparse threshold θ_sparse, the subgraph is determined as a water network feature sparse graph, wherein the calculation formula of the sparse threshold θ_sparse is: θ_sparse = 1 / (2×√V); The threshold value is adaptively adjusted according to the number of nodes, and the more the number of nodes, the stricter the sparse standard; For the subgraph formed after cutting, if the edge density is between 0.05 and 0.3, it is determined as a reasonable sparse graph; if the edge density is less than 0.05, it is determined as an excessively sparse graph and needs to be merged with adjacent subgraphs; if the edge density is greater than 0.3, it is determined as a dense graph and needs to be further cut.

[0123] In a certain embodiment, a subgraph containing 50 nodes has a maximum possible number of edges of 50*49 / 2=1225, an actual number of edges of 120, an edge density of 120 / 1225≈0.098, and a sparse threshold of 1 / (2*sqrt(50))≈0.071. Since 0.098>0.071 and 0.098∈[0.05, 0.3], it is determined that the subgraph is a reasonable water network feature sparse graph.

[0124] According to an aspect of the present application, the determination process of the cutting threshold in the step S31 is specifically as follows: Based on the number of eigenvalues F and the total number of nodes N, the calculation formula of the cutting threshold τ_cut is τ_cut=┗α*ln(N)*sqrt(F)┛. Wherein, α is an adjustment coefficient, and the value range is 0.8 to 1.2, ┗┛ represents rounding down; When the number of eigenvalues F=8 and the total number of nodes N is in different ranges, the calculation of the cutting threshold is as follows: N∈[50, 100]: τ_cut=┗1.0*ln(75)*sqrt(8)┛=┗12.2┛=12; N∈[100, 500]: τ_cut=┗1.0*ln(300)*sqrt(8)┛=┗16.1┛=16; N∈[500, 1000]: τ_cut=┗0.9*ln(750)*sqrt(8)┛=┗16.9┛=16; N>1000: τ_cut=┗0.8*ln(N)*sqrt(8)┛, and the maximum is not more than 20; The physical meaning of the cutting threshold is that when the number of natural neighbors of a node exceeds the threshold, it means that the node is in the core position of the local high-density area, and the connected edge should be marked as a cutting edge to maintain the integrity of the core area.

[0125] In a certain embodiment, the study area contains 235 water network nodes and 8 feature attributes, and the cutting threshold is calculated as τ_cut=┗1.0*ln(235)*sqrt(8)┛=┗15.4┛=15, that is, when the number of natural neighbors of a node is greater than 15, the connected edge between the node and its neighbor is marked as a cutting edge.

[0126] According to an aspect of the present application, the determination process of the shared nearest neighbor number in the step S43 is specifically as follows: The shared nearest neighbor number k_share is dynamically determined based on the size of the subgraph, and the calculation formula is: k_share=max(3, min(┗sqrt(n_i*n_j)┛, 20)); Wherein, n_i and n_j are the number of nodes of two subgraphs to be compared; For different size subgraph pairs, the number of shared nearest neighbors is: Small subgraph pairs (n_i, n_j < 20): k_share = 3 to 5; Medium subgraph pairs (n_i, n_j between 20 and 100): k_share = 6 to 10; Large subgraph pairs (n_i, n_j > 100): k_share = 10 to 20; Subgraph pairs with large size difference: use a smaller value min(k_i, k_j), where k_i = ┗√n_i┛, k_j = ┗√n_j┛; The identification process of shared nearest neighbors is to find the k_share nearest neighbors of the center points of the two subgraphs respectively, calculate the intersection size of the two nearest neighbor sets, and the number of intersection elements is the number of shared nearest neighbors SNN(i, j).

[0127] In an embodiment, subgraph A contains 36 nodes, subgraph B contains 64 nodes, k_share = ┗√(36×64)┛ = ┗48┛ = 20 (take the upper limit), and the 20 nearest neighbors of the center point of subgraph A coincide with the 20 nearest neighbors of the center point of subgraph B. Then SNN(A, B) = 8.

[0128] According to an aspect of the present application, the calculation process of the distance of the dispersed node to the subgraph in step S32 is as follows: The distance of the dispersed node v to the subgraph G is defined as the weighted minimum value of the three distances: d(v, G) = min(w_1×d_center(v, G), w_2×d_boundary(v, G), w_3×d_average(v, G)); Where d_center(v, G) is the composite distance of node v to the center point of subgraph G; d_boundary(v, G) = min{d(v, u) | u ∈ boundary(G)}, which is the minimum composite distance of node v to the boundary node of subgraph G, and the boundary node is defined as the node connected to other subgraphs or dispersed nodes; d_average(v, G) = (1 / |G|)×Σ_{u∈G}d(v, u), which is the average composite distance of node v to all nodes of subgraph G; The weights are set as: w_1 = 0.5 (preferentially consider the center distance), w_2 = 0.3 (secondarily consider the boundary distance), and w_3 = 0.2 (auxiliary consider the average distance); The assignment rule is: assign(v) = argmin_Gd(v, G), that is, assign the dispersed node v to the subgraph with the minimum distance; In some embodiments, the distance of a dispersed node P to three subgraphs is calculated: to subgraph G1 : d_center = 15.2, d_boundary = 8.5, d_average = 18.3, weighted distance = min(7.6, 2.55, 3.66) = 2.55; to subgraph G2: d_center = 22.1, d_boundary = 12.3, d_average = 25.6, weighted distance = min(11.05, 3.69, 5.12) = 3.69; to subgraph G3: d_center = 18.5, d_boundary = 10.2, d_average = 20.1, weighted distance = min(9.25, 3.06, 4.02) = 3.06; Node P is assigned to G1 (the minimum distance is 2.55).

[0129] According to an aspect of the present application, the determination process of the number of clusters N in step S44 is as follows: The optimal number of clusters N is automatically determined using the silhouette coefficient method, and the calculation process is as follows: For each possible number of clusters n (n from 2 to √M), the average silhouette coefficient is calculated: S(n) = (1 / N_total) x ∑_is(i); where s(i) = (b(i) - a(i)) / max(a(i), b(i)); a(i) is the average distance of node i to other nodes in the same cluster, and b(i) is the average distance of node i to the nearest other cluster; The n that maximizes the silhouette coefficient is selected as the final number of clusters N: N = argmax_n S(n); At the same time, the following constraints are set: minimum number of clusters: N_min = max(2, ┗M / 10┛); maximum number of clusters: N_max = min(┗M / 2┛, 20); silhouette coefficient threshold: S(N) > 0.3, otherwise continue to adjust; When the automatically determined N value does not meet the management requirements, manual intervention can be performed to set the expected number of clusters N_expect, and the algorithm searches for the optimal value within the range [N_expect-2, N_expect+2].

[0130] In some embodiments, the initial division results in M = 45 subgraphs, and the silhouette coefficients for different numbers of clusters are calculated: S(5) = 0.42 when n = 5, S(6) = 0.51 when n = 6, S(7) = 0.58 when n = 7, S(8) = 0.55 when n = 8, and S(9) = 0.48 when n = 9, so the optimal number of clusters N is determined to be 7.

[0131] According to an aspect of the present application, the processing procedure of isolated nodes and nodes without natural neighbors, specifically: Identify special nodes: Isolated node: the composite distance to all other nodes is greater than the distance threshold D_isolate=3×D_mean, where D_mean is the average distance of all node pairs; Node without natural neighbor: within the range of the number of effective neighbors, there is no node with mutual nearest neighbor relationship; The processing strategy includes: For isolated nodes, use the extended search radius method: gradually increase the search radius r=r_0×(1+0.2×k), k is the iteration number, until at least one candidate neighbor is found; calculate the weighted distance to all candidate neighbors belonging to the subgraph, considering the subgraph size (high weight for large subgraph) and functional similarity; assign the isolated node to the subgraph with the smallest weighted distance; For nodes without natural neighbors, use the relaxed mutual neighbor condition: relax the mutual nearest neighbor condition to one-way nearest neighbor, that is, if node A is one of the k nearest neighbors of node B (k≤2×effective neighbor number), establish a weak connection from A to B; construct an extended neighborhood graph based on the weak connection; search for natural neighbors in the extended neighborhood graph; If there are still nodes that cannot be assigned after processing, create an "edge zone" subgraph to accommodate these nodes, and when merging the final clusters, give priority to merging the edge zone subgraph with the main subgraph that is most similar in function.

[0132] In an embodiment, the minimum composite distance of node Q to other nodes is 85.3, while the average distance D_mean=22.1, and the threshold D_isolate=66.3, so Q is determined as an isolated node; extended search finds 3 candidate subgraphs, the weighted distance to G1 is 72.5 (subgraph size 15, functional similarity 0.3), the weighted distance to G2 is 68.2 (subgraph size 25, functional similarity 0.4), and the weighted distance to G3 is 70.1 (subgraph size 20, functional similarity 0.35), finally Q is assigned to G2.

[0133] According to an aspect of the present application, the evaluation process of the water network system target partition effect, specifically: Construct a multi-dimensional evaluation index system: Internal compactness index IC=(1 / N)×Σ_k(1 / n_k²)×Σ_{i,j∈C_k}sim(i,j); Where N is the number of clusters, n_k is the number of nodes in the kth partition, and sim(i,j) is the node similarity; The outer separation index ES = min_{k≠l}d(C_k, C_l) / max_k diameter(C_k); wherein d(C_k, C_l) is the distance between partitions, and diameter(C_k) is the diameter of the partition; The functional consistency index FC = (1 / N) x Σ_k var(F_k) / var(F_total); wherein var(F_k) is the variance of the functional features in the kth partition, and var(F_total) is the variance of the global functional features, and the smaller FC is, the more consistent the functions in the partition are; The management feasibility index MA = (1 / N) x Σ_k (1-split(k)); wherein split(k) is the ratio of the number of management units spanned by the kth partition to the number of nodes in the partition; The comprehensive evaluation score Score = w_IC x IC + w_ES x ES + w_FC x (1-FC) + w_MA x MA; wherein the weights w_IC = 0.3, w_ES = 0.2, w_FC = 0.3, and w_MA = 0.2; the evaluation criteria are: Score > 0.7 is excellent, 0.5-0.7 is good, 0.3-0.5 is qualified, and < 0.3 needs to be re-partitioned.

[0134] In a certain embodiment, after a certain watershed is divided into 7 targeted partitions, the IC = 0.72 (the partition is tightly internal), the ES = 0.65 (the separation degree between partitions is good), the FC = 0.28 (the functional consistency is high), and the MA = 0.81 (the management boundary is clear), the comprehensive score Score = 0.3 x 0.72 + 0.2 x 0.65 + 0.3 x 0.72 + 0.2 x 0.81 = 0.724, and the evaluation is excellent, which is 42% higher than the traditional K-means method (Score = 0.51).

[0135] According to another aspect of the present application, a water network system targeted partitioning system is provided, characterized in that it comprises: at least one processor; and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the water network system targeted partitioning method according to any one of the preceding aspects.

[0136] The above describes the preferred embodiments of the present application, but the present application is not limited to the specific details in the above embodiments, and various equivalent transformations can be made to the technical solutions of the present application within the technical concept range of the present application, and these equivalent transformations all belong to the protection range of the present application.

Claims

1. A method for targeting sub-zones of a water network system, characterized in that, The method comprises the following steps: Step S1, collecting water network data of a study area, identifying water network nodes and extracting data of each water network node, and reducing dimension of the data to obtain an eigenvalue matrix of each water network node; Step S2, calculating a composite distance between each two nodes based on the eigenvalue matrix of each water network node, setting a near-neighbor number value based on the number of eigenvalues, screening water network nodes that are nearest neighbors of each other, and constructing a hierarchical graph with nodes as vertices and nearest-neighbor relationships as edges to obtain a water network characteristic neighborhood graph; Step S3, identifying cut edges based on the neighborhood relationship between each node, and dividing the water network characteristic neighborhood graph into a plurality of water network characteristic sparse graphs; Step S4, determining a center point of each water network characteristic sparse graph based on the local density of the water network nodes, calculating the similarity between each water network characteristic sparse graph, and clustering the water network characteristic sparse graphs to obtain a plurality of clustering centers, and dividing the study area into a plurality of water network system targeted partitions.

2. The water network system targeting zonation method of claim 1, wherein, The step S1 further comprises: Step S11, collecting water network data of a study area, including: water network topological structure data, hydraulic data, geographic data, historical drought data and historical flood data; Step S12, identifying water network nodes and extracting data of each water network node, wherein the water network nodes include: basin nodes, engineering nodes, monitoring nodes and geographic feature nodes; Step S13, sequentially reducing dimension of data of all water network nodes to obtain an eigenvalue matrix of each water network node.

3. The water network system targeting zonation method of claim 2, wherein, The step S13 further comprises: Step S13a, extracting a plurality of water network node attributes based on the data of the water network nodes, and calculating correlation coefficients between all water network node attributes, and classifying water network node attributes with correlation coefficients greater than a threshold value into the same water network node characteristic attribute to obtain a water network node characteristic attribute set; Step S13b, calculating values of each water network node characteristic attribute using an entropy weight method, and constructing an eigenvalue matrix of each water network node.

4. The water network system targeting zonation method of claim 1, wherein, The step S2 further comprises: Step S21, calculating a composite distance between each two nodes based on the eigenvalue matrix of each water network node; Step S22, setting an initial near-neighbor number value based on the number of eigenvalues, and screening near-neighbor nodes of each water network node, gradually increasing the size of the near-neighbor number value, and screening near-neighbor nodes of each water network node based on the current near-neighbor number value, until all water network nodes satisfy at least one near-neighbor node, at which time the near-neighbor number value is the effective near-neighbor number value; Step S23, screening near-neighbor nodes of each water network node based on the effective near-neighbor number value, and constructing a hierarchical graph with nodes as vertices and nearest-neighbor relationships as edges to obtain a water network characteristic neighborhood graph.

5. The water network system targeting zonation method of claim 4, wherein, The step S21 further comprises: Step S21a, identifying basins in the study area as basin nodes, identifying water conservancy hubs, flood drainage channels and storage projects as engineering nodes, identifying hydrological sites as monitoring nodes, and identifying river junctions, drought areas and flood-prone areas as geographic feature nodes; Step S21b, sequentially calculating a geographic distance, a hydraulic distance, a climate distance and a management distance between each two nodes; Step S21c, determining weights of each distance and calculating a composite distance between each two nodes.

6. The water network system targeting zonation method of claim 4, wherein, The step S23 further comprises: Step S23a, constructing a hierarchical neighborhood graph of basin nodes, engineering nodes and geographical feature nodes respectively with nodes as vertices and near-neighbor relationships as edges; Step S23b, representing water-related associations, water transfer associations and climate associations with blue edges, red edges and yellow edges respectively, representing high flood risk area nodes with red, orange nodes with medium flood risk area nodes, and yellow nodes with drought area nodes, and representing climate correlation with edge thickness, to obtain a water network feature neighborhood graph.

7. The water network system targeting zonation method of claim 1 wherein, The step S3 further comprises: Step S31, setting a cutting threshold based on the number of water network node feature values, calculating the number of near neighbors of each water network node, marking water network nodes with a number of near neighbors greater than the cutting threshold, and retaining the edges of the marked water network nodes and all their near neighbor nodes as cutting edges, and dividing the water network feature neighborhood graph into M subgraphs and a number of scattered water network nodes based on the cutting edges, M being a positive integer greater than 10; Step S32, calculating the distance of each scattered water network node and M subgraphs in turn, and dividing the scattered water network nodes into the nearest sub-region to obtain M water network feature sparse graphs.

8. The water network system targeting zonation method of claim 1 wherein, The step S4 further comprises: Step S41, calculating the local density of each water network node, and determining the center point of each water network feature sparse graph based on the local density of the water network node; Step S42, calculating the distance between the center points of each two water network feature sparse graphs in turn, which is the distance between the two water network feature sparse graphs; Step S43, setting a shared nearest neighbor number, taking the ratio of the number of shared natural nearest neighbors to the distance between the two water network feature sparse graphs as the similarity of the two water network feature sparse graphs, and calculating the similarity between all water network feature sparse graphs in turn; Step S44, selecting the two water network feature sparse graphs with the highest similarity and clustering them into the same water network division, obtaining a clustering label, and taking it as a water network feature sparse graph to calculate the similarity between all water network feature sparse graphs again, and repeating the iteration until all water network feature sparse graphs belong to a certain clustering label, obtaining N clustering labels, and dividing the study area into N water network system targeted divisions, N being a positive integer less than M.

9. A water network system targeted zoning system characterized by, It comprises: At least one processor; And A memory in communication connection with the at least one processor; wherein The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the water network system targeted division method of any one of claims 1 to 8.

Images