A hydrologic anomaly identification method and system for hydraulic engineering

Abstract

This invention discloses a method and system for identifying hydrological anomalies in water conservancy projects, belonging to the field of data processing technology. The invention collects hydrological observation data and maps monitoring sections as graph nodes, calculates the cross-sectional specific energy to construct transient graph signal vectors and unsigned topological Laplace matrices; calculates the global scheduling ground-state signal vector based on matrix polynomial expansion and performs differential operations to remove interference from compliant high-flow scheduling, obtaining local disturbance residual signal vectors; calculates the evolution rate by extracting algebraic invariants of characteristic polynomials, dynamically generating filtering boundaries; uses Chebyshev polynomial approximation to construct a bandpass graph filter to filter the residual signals, retaining structural anomaly signal vectors representing local structural damage; calculates the transient energy sequence vector and compares it with the energy alarm threshold to accurately identify abnormal physical nodes and generate alarm information. This invention eliminates compliant scheduling interference and achieves high-precision perception and positioning of physical structural disasters.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for identifying hydrological anomalies in water conservancy projects. Background Technology

[0002] With the continuous development of hydrological monitoring technology, various hydrological sensors have been widely used in water resource management, disaster prevention and mitigation, and ecological environment monitoring, realizing the real-time acquisition of key hydrological variables. In the safe operation of water conservancy projects and flood control command, hydrological anomaly identification is a crucial core task. Timely and accurate detection of abnormal hydrological data can not only identify equipment failures and extreme weather events, but also provide early warning of serious structural damage disasters such as dam piping, underground pipe leakage, and dam landslides. However, hydrological observation data of water conservancy projects in real environment often presents extremely high complexity. The water flow state is not only affected by natural diffusion driven by gravity, high-frequency environmental disturbances such as wind and wave white noise, but also frequently affected by compliant large-flow artificial scheduling. Currently, methods for detecting anomalies in hydrological data mainly rely on time series analysis, statistical methods, and machine learning techniques. Chinese invention patent application CN120104967A discloses a machine learning-based method and system for hydrological data management. This scheme measures the uncertainty of hydrological variables by calculating their fuzzy entropy values ​​and uses Granger causality analysis to construct a causal relationship model between hydrological variables in the monitoring data. During continuous monitoring, this method utilizes the causal model of the hydrological monitoring data combined with a fuzzy entropy optimization strategy to dynamically adjust the anomaly detection threshold to achieve real-time identification of abnormal hydrological monitoring data. After detecting anomalies, it uses the causal relationship model to perform reverse causal analysis to generate anomaly tracing paths. However, the aforementioned existing technologies struggle to effectively distinguish compliant manual scheduling. In contrast to real structural disasters, when flood control systems discharge water normally, the flow velocity and water level across the entire network undergo drastic changes that conform to physical diffusion laws. Existing technologies are prone to misjudging such overall deviations caused by compliant large-flow scheduling as anomalies, and cannot completely eliminate the scheduling impact as a background ground state. Existing solutions lack dimensionality reduction extraction of the macroscopic topological invariance of the entire water conservancy network when extracting features and setting judgment thresholds. When the logical reordering of sensor node numbers within the water network or when facing local high-frequency environmental wind, wave, and white noise interference, the drastic changes in the position of elements within the monitoring matrix can easily lead to a high false alarm rate in the detection system. This makes it difficult for existing technologies to construct stable dynamic filtering boundaries that are immune to local noise and node reordering, and thus cannot accurately intercept and extract structural anomaly signals that truly characterize the damage to the engineering structure in the frequency domain. Summary of the Invention

[0003] The technical problem solved by this invention is that existing technologies have difficulty in effectively distinguishing between compliant manual scheduling and real structural disasters. Existing technologies tend to misjudge the overall deviation caused by high-volume compliant scheduling as anomalies and cannot completely eliminate the scheduling impact as a background ground state. Existing solutions lack dimensionality reduction extraction of the macroscopic topological invariance of the entire water conservancy network when extracting features and setting judgment thresholds. This makes it difficult for existing technologies to construct stable dynamic filtering boundaries that are immune to local noise and node rearrangement, and thus cannot accurately intercept and extract structural anomaly signals that truly characterize the damage to engineering structures in the frequency domain.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for identifying hydrological anomalies in water conservancy projects, comprising the following steps: Step S1: Collect hydrological observation data and map the monitoring section as a graph node. Calculate the cross-sectional specific energy of each graph node based on the hydrological observation data, construct the transient graph signal vector, and construct an unsigned topological Laplace matrix based on the physical connectivity between the nodes. Step S2: Based on the transient graph signal vector and the unsigned topological Laplace matrix, calculate the global scheduling ground state signal vector using matrix polynomial expansion, and perform a difference operation between the transient graph signal vector and the global scheduling ground state signal vector to obtain the local disturbance residual signal vector; Step S3: Calculate the characteristic polynomial based on the unsigned topological Laplace matrix, extract the algebraic invariant values ​​from the characteristic polynomial, calculate the evolution rate value based on the algebraic invariant values ​​of the current time and the previous time, and generate a dynamic filtering boundary based on the evolution rate value. Step S4: Based on the unsigned topological Laplace matrix and the dynamic filtering boundary, a bandpass graph filter is constructed based on Chebyshev polynomial approximation to filter the local disturbance residual signal vector and obtain the structural anomaly signal vector. Step S5: Calculate the transient energy sequence vector based on the structural anomaly signal vector, compare the elements in the transient energy sequence vector with the preset energy alarm threshold, identify abnormal nodes, and generate alarm information.

[0005] Preferably, the process of constructing the transient signal vector in step S1 specifically includes: Hydrological observation data is collected by hydrological telemetry terminals deployed on the physical cross sections of water conservancy projects, and the sensor monitoring sections in the water conservancy project network are mapped to nodes in the graph topology. The hydrological observation data includes real-time water level values ​​at each node, average cross-sectional flow velocity values ​​at each node, and node number. The cross-sectional specific energy of each node is calculated based on the hydrological observation data. The mathematical expression for the cross-sectional specific energy is as follows: ; in, Let be the gravitational acceleration constant, and take the value of . , For the first The cross-sectional specific energy of each node, For the first Real-time water level values ​​at each node ) is the first The average flow velocity of the cross section at each node; Arrange the cross-sectional energy values ​​of all nodes according to the node number to form a one-dimensional column vector, which is the current transient signal vector.

[0006] Preferably, the process of constructing the unsigned topological Laplacian matrix in step S1 specifically includes: Select any two nodes to form a node pair, collect the physical river course of the water conservancy project, and determine whether there is a direct physical water flow connection between the node pairs. If there is a direct physical river channel connection between the node pairs, it is determined that there is a physical water flow connection relationship; If there is no direct physical waterway connection between the node pairs, it is determined that there is no physical water flow connection. The dynamic connection weight of node pairs that do not have the physical water flow connectivity relationship is zero; The mathematical expression for the dynamic connection weight of node pairs with the aforementioned physical water flow connectivity is: ; in, Starting point and the end point At the present moment Dynamic connection weights, For the current moment, Starting point and the end point The absolute value of the difference in cross-sectional specific energy at the current time t. This is a preset energy scaling constant; Construct a dynamic adjacency matrix, where the rows are the starting points, the columns are the ending points, and the elements are the dynamic connection weights between the corresponding pairs of nodes. Summing the elements of each row of the dynamic adjacency matrix, and generating a degree matrix in diagonal form using the sum of each row as the main diagonal element, and performing matrix addition on the degree matrix and the dynamic adjacency matrix to obtain an unsigned topological Laplace matrix.

[0007] Preferably, the process of calculating the global scheduling ground state signal vector in step S2 specifically includes: The current transient signal vector is multiplied sequentially with each power of the unsigned topological Laplace matrix to obtain the topological diffusion vectors of each order. The topological diffusion vectors of each order are multiplied and weighted by the diagonal matrix of the corresponding order. The one-dimensional column vectors after weighting all orders are summed to obtain the global scheduling ground state signal vector. Perform a vector subtraction operation between the current transient graph signal vector and the global scheduling ground state signal vector, and use the difference result as the local disturbance residual signal vector; The elements on the diagonal of the diagonal matrix of the corresponding order are diffusion attenuation coefficients, and the elements on the other off-diagonal positions are all 0. The mathematical expression for the diffusion attenuation coefficient is: ; in, For nodes In the The diffusion attenuation coefficient value under topological diffusion. For nodes The physical length of the river channel from the starting point of the water conservancy project. Let be the spatial diffusion scaling constant. The topological order is currently being calculated, with values ​​of 1, 2, and 3. The mathematical expression for the global scheduling ground state signal vector is: ; in, For a moment The output global scheduling ground state signal vector, For a moment Known unsigned topological Laplace matrix, For a moment Given the graph signal vector, This represents the current topological order being calculated. The highest topological order is set as the cutoff value. Representing an unsigned topological Laplace matrix Power matrix For the first A diagonal matrix of order n.

[0008] Preferably, the process of obtaining the numerical value of the algebraic invariant in step S3 specifically includes: Construct a first identity matrix in the form of a square matrix with the same dimensions as the unsigned topological Laplace matrix, and perform scalar multiplication operations between the preset algebraic symbolic variables and the identity matrix to obtain an identity matrix with algebraic variables; The characteristic matrix is ​​obtained by subtracting the unsigned topological Laplace matrix from the identity matrix with algebraic variables. The determinant of the characteristic matrix is ​​calculated using a determinant expansion algorithm to obtain the characteristic polynomial; The numerical coefficients of the predetermined powers of the algebraic symbolic variables are extracted from the characteristic polynomial and used as the values ​​of the algebraic invariants at the current moment.

[0009] Preferably, the process of obtaining the dynamic filtering boundary in step S3 specifically includes: Obtain the algebraic invariant value from the previous time step, subtract the algebraic invariant value from the previous time step from the current time step, and take the absolute value to obtain the evolution rate value. The dynamic lower limit frequency value is calculated based on the evolution rate value, which is the sum of the product of the reference cutoff frequency value, the evolution rate amplification factor, and the evolution rate value. The dynamic lower limit frequency value is combined with the preset fixed upper limit frequency value to form a closed interval as the dynamic filtering boundary. The mathematical expression for the dynamic lower limit frequency value is: ; in, This is the dynamic lower limit frequency value calculated at the current moment. The set reference cutoff frequency value. This is the preset evolution rate amplification factor. This represents the rate of evolution. This is a preset dynamic lower limit cutoff tolerance.

[0010] Preferably, in step S4, the process of constructing the bandpass filter specifically includes: Construct a second identity matrix, and perform scaling operations on the second identity matrix and the unsigned topological Laplacian matrix, specifically including: Multiply the unsigned topological Laplacian matrix by the first scaling constant, and then subtract the second identity matrix to obtain the scaled Laplacian matrix. The spectral domain eigenvalues ​​of the scaled Laplacian matrix are located within a preset interval. The closed interval formed by the dynamic lower limit frequency value and the fixed upper limit frequency value is used as the dynamic filtering boundary. The Chebyshev polynomial expansion coefficients of the ideal bandpass filter function are calculated within the dynamic filtering boundary. The bandpass filter is mathematically equivalent to the scaled Laplace matrix, the Chebyshev polynomial expansion coefficients, and the Chebyshev polynomial matrix.

[0011] Preferably, the process of obtaining the structural anomaly signal vector in step S4 specifically includes: The local perturbation residual signal vector is used as the zero-order iterative column vector for Chebyshev recursive calculation. The scaling Laplace matrix is ​​multiplied by the zero-order iterative column vector to obtain the first-order iterative column vector. Starting from the second order up to the preset approximation truncation order, the Chebyshev recursive equation is used to perform iterative calculations to obtain the iterative column vectors of each order. The Chebyshev recursive equation states that the iterative column vector of the current order is equal to twice the product of the scaling Laplacian matrix and the iterative column vector of the previous order minus the iterative column vector of the order before that. The structural anomaly signal vector is obtained by linearly combining and accumulating the iterative column vectors of each order using the Chebyshev polynomial expansion coefficients. The mathematical expression for the structural anomaly signal vector is: ; in, This is the structural anomaly signal vector output at the instant t of the current synchronous sampling. To approximate the truncation order, This is the index of the current expansion order. For the first The coefficients of the Chebyshev polynomial expansion of order 1, To scale the Laplacian matrix, The first variable is the scaled Laplace matrix. Chebyshev polynomial matrix of order, It is the vector of the local disturbance residual signal.

[0012] Preferably, the process of identifying the abnormal node in step S5 specifically includes: Extract the numerical elements from the structural anomaly signal vector in row order, and perform a self-square calculation for each extracted numerical element; Arrange the calculated square values ​​in order of their original node index numbers to generate a transient energy sequence vector; The elements in the transient energy sequence vector are compared with the energy alarm threshold. If an element in the transient energy sequence vector is greater than the energy alarm threshold, the value of the element is taken as the target value, and the row number corresponding to the element in the transient energy sequence vector is extracted, where the row number is the abnormal source node number. According to the preset sensor coordinate mapping table, obtain the latitude and longitude coordinates corresponding to the abnormal source node number, and package the latitude and longitude coordinate values, the target value and the current timestamp into a system alarm data packet; The system alarm data packets are sent to the flood control command center via a wired communication network.

[0013] A hydrological anomaly identification system for water conservancy projects includes a topology construction module, a residual separation module, a feature extraction module, an anomaly extraction module, and an early warning module; The topology construction module is used to collect hydrological observation data, map the monitoring section to graph nodes, calculate the cross-sectional specific energy of each graph node based on the hydrological observation data, construct the transient graph signal vector, and construct an unsigned topological Laplace matrix based on the physical connectivity between each node. The residual separation module is used to calculate the global scheduling ground state signal vector based on matrix polynomial expansion according to the transient graph signal vector and the unsigned topological Laplace matrix, and to perform a difference operation between the transient graph signal vector and the global scheduling ground state signal vector to obtain the local disturbance residual signal vector. The feature extraction module is used to calculate the characteristic polynomial based on the unsigned topological Laplacian matrix, extract the algebraic invariant values ​​from the characteristic polynomial, calculate the evolution rate value based on the algebraic invariant values ​​at the current time and the previous time, and generate a dynamic filtering boundary based on the evolution rate value. The anomaly extraction module is used to construct a bandpass graph filter based on Chebyshev polynomial approximation according to the unsigned topological Laplace matrix and the dynamic filtering boundary, and to filter the local disturbance residual signal vector to obtain the structural anomaly signal vector. The early warning module is used to calculate a transient energy sequence vector based on the structural anomaly signal vector, compare the elements in the transient energy sequence vector with a preset energy alarm threshold, identify abnormal nodes, and generate alarm information.

[0014] The beneficial effects of this invention are as follows: By introducing cross-sectional specific energy as a fundamental physical quantity and constructing an unsigned topological Laplace matrix in conjunction with the physical connectivity of the water network, this invention can characterize the overall energy accumulation state and hydrophysical topological framework of water flow under dynamic conditions. Based on this, this invention calculates the global scheduling ground state signal vector based on matrix polynomial expansion and removes it from the transient graph signal by subtraction, thereby accurately separating the local disturbance residual signal. This invention uses the characteristic polynomial of the unsigned topological Laplace matrix to extract algebraic invariants, and uses the invariance of the macroscopic topological structure as a benchmark for measuring the evolution rate of the water network. Since the algebraic invariants obtained by the characteristic polynomial have absolute robustness to the rearrangement of the logical numbering of sensor nodes inside the water conservancy network, this method reduces the risk of false alarms caused by changes in the sensor network and provides an extremely stable and high signal-to-noise ratio evolution rate basis for the generation of dynamic filtering boundaries. The bandpass graph filter constructed by combining the dynamic lower limit frequency and Chebyshev polynomial approximation can intercept the mid-to-high frequency topological distortion signal representing local structural damage, while filtering out residual low-frequency natural fluctuations and extremely high frequency environmental wind and wave white noise. Attached Figure Description

[0015] Figure 1 The flowchart illustrates the steps of a method for identifying hydrological anomalies in water conservancy projects, as provided in one embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example 1, referring to Figure 1 A method for identifying hydrological anomalies in water conservancy projects is provided, comprising the following steps: Step S1: Collect hydrological observation data and map the monitoring section to map nodes. Calculate the cross-sectional specific energy of each map node based on the hydrological observation data, construct the transient map signal vector, and construct an unsigned topological Laplace matrix based on the physical connectivity between each node. Step S2: Based on the transient diagram signal vector and the unsigned topological Laplace matrix, calculate the global scheduling ground state signal vector based on matrix polynomial expansion, and perform a difference operation between the transient diagram signal vector and the global scheduling ground state signal vector to obtain the local disturbance residual signal vector. Step S3: Calculate the characteristic polynomial based on the unsigned topological Laplacian matrix, extract the algebraic invariant values ​​from the characteristic polynomial, calculate the evolution rate values ​​based on the algebraic invariant values ​​of the current time and the previous time, and generate the dynamic filtering boundary based on the evolution rate values. Step S4: Based on the unsigned topological Laplace matrix and dynamic filtering boundary, a bandpass graph filter is constructed based on Chebyshev polynomial approximation to filter the local disturbance residual signal vector and obtain the structural anomaly signal vector. Step S5: Calculate the transient energy sequence vector based on the structural anomaly signal vector, compare the elements in the transient energy sequence vector with the preset energy alarm threshold, identify abnormal nodes, and generate alarm information.

[0018] This invention maps monitoring sections of water conservancy projects to graph nodes and, by combining cross-sectional energy density and unsigned topological Laplace matrix, constructs a graph signal model that can truly reflect the transient physical topological skeleton of water flow. Based on this, matrix polynomial expansion is used to extract and remove the global scheduling ground state, thus immunizing against interference from compliant high-flow scheduling for anomaly identification. At the same time, algebraic invariants are used to calculate the evolution rate to generate a dynamic filtering boundary, and a bandpass graph filter is constructed by combining Chebyshev polynomial approximation. This filter can adaptively and accurately intercept abnormal signals representing local structural damage. Finally, alarms for abnormal nodes are achieved through energy sequence comparison.

[0019] Step S1, the process of constructing the transient signal vector, specifically includes: Hydrological observation data is collected by hydrological telemetry terminals deployed on the physical cross sections of water conservancy projects, and the sensor monitoring sections in the water conservancy project network are mapped to nodes in the graph topology. Hydrological observation data includes real-time water level values ​​at each node, average cross-sectional flow velocity values ​​at each node, and node number. The cross-sectional specific energy of each node is calculated based on hydrological observation data. The mathematical expression for the cross-sectional specific energy is as follows: ; in, Let be the gravitational acceleration constant, and take the value of . , For the first The cross-sectional specific energy of each node, For the first Real-time water level values ​​at each node ) is the first The average flow velocity of the cross section at each node; Arrange the cross-sectional energy values ​​of all nodes according to the node number to form a one-dimensional column vector, which is the current transient signal vector.

[0020] In a specific embodiment of the present invention, hydrological observation data is acquired by hydrological telemetry terminals arranged on the physical cross sections of the water conservancy project. The entire water conservancy project network contains N sensor monitoring sections, and these N sections are mapped as nodes in a graph topology. The reason for calculating the cross-sectional specific energy is that a single water level data cannot truly reflect the overall energy state of flowing water under dynamic conditions. Cross-sectional specific energy is the most fundamental energy conservation benchmark in open channel flow dynamics, which can accurately characterize the physical potential state of the current water flow. Will The cross-sectional specific energy values ​​of each node are arranged sequentially according to the geographic number of the sensor, resulting in a dimension of... The one-dimensional column vector is the transient signal vector.

[0021] When constructing the transient signal vector, this invention introduces cross-sectional specific energy. Since single water level data cannot truly reflect the overall energy state of water flow under dynamic conditions, this scheme can characterize the true physical potential and hydrodynamic characteristics of the current water flow by fusing real-time water level and cross-sectional average flow velocity.

[0022] Step S1, the process of constructing the unsigned topological Laplacian matrix, specifically includes: Select any two nodes to form a node pair, collect the physical river course of the water conservancy project, and determine whether there is a direct physical water flow connection between the node pairs. If there is a direct physical river channel connection between node pairs, it is determined that there is a physical water flow connection relationship. If there is no direct physical river channel connection between node pairs, it is determined that there is no physical water flow connection relationship. The dynamic connection weight of node pairs that do not have a physical water flow connection relationship is zero; The mathematical expression for the dynamic connection weight of node pairs with physical water flow connectivity is: ; in, Starting point and the end point At the present moment Dynamic connection weights, For the current moment, Starting point and the end point The absolute value of the difference in cross-sectional specific energy at the current time t. This is a preset energy scaling constant; Construct a dynamic adjacency matrix, where the rows are the starting points, the columns are the ending points, and the elements are the dynamic connection weights between the corresponding pairs of nodes. Summing the elements of each row of the dynamic adjacency matrix, and using the sum of each row as the main diagonal element, a degree matrix in diagonal form is generated. The degree matrix is ​​then added to the dynamic adjacency matrix to obtain an unsigned topological Laplacian matrix.

[0023] In one specific embodiment of the present invention, the topological connectivity between various sensors is predetermined based on the physical river channel orientation of the water conservancy project. In this embodiment, The value is 0.5 meters; Perform a row summation operation on the dynamic adjacency matrix. Specifically, sum the rows of the dynamic adjacency matrix... The sum of all N columns of a row is the value of the node. Overall connectivity strength in the current water network; Construct a degree matrix, the degree matrix having dimensions of . A diagonal matrix, where the elements on the diagonal correspond to the overall connectivity strength of the nodes; By directly adding the degree matrix and the dynamic adjacency matrix, a matrix with dimension is constructed. The unsigned topological Laplace matrix; The reason for using addition to construct an unsigned matrix instead of the traditional subtraction method to construct a standard Laplace matrix is ​​that the standard Laplace matrix describes the spatial diffusion differences of physical quantities. However, when disasters such as breaches or blockages occur in water conservancy networks, the core manifestation is the illegal accumulation or loss of water volume in a certain locality. The unique algebraic properties of the unsigned topological Laplace matrix can extremely sensitively measure the summation and concentration of the entire network flow. When the engineering structure is intact, the algebraic structure of the matrix remains stable; once leakage or dam failure occurs, the eigenvalues ​​at the corresponding positions of the unsigned topological Laplace matrix will undergo instantaneous algebraic distortion. The resulting unsigned topological Laplace matrix completely encapsulates the transient hydrophysical topological framework of the entire water conservancy project at the current moment.

[0024] This invention constructs an unsigned topological Laplace matrix using additive operations, rather than the traditional standard Laplace matrix. The standard Laplace matrix can only characterize the spatial diffusion differences of physical quantities, while the unsigned topological Laplace matrix constructed in this invention and its dynamic connection weights based on the difference in cross-sectional specific energy can measure the summation and aggregation degree of the entire network flow. Once illegal water accumulation and loss phenomena such as leakage, piping, or levee breach occur in a local part of the water conservancy network, the algebraic structure of the matrix and the eigenvalues ​​at the corresponding positions will undergo instantaneous algebraic distortion, thus completely and sensitively preserving the transient hydrophysical topological skeleton of the entire water conservancy project at the current moment.

[0025] Step S2, the process of calculating the global scheduling ground state signal vector, specifically includes: The current transient signal vector is multiplied sequentially with each power of the unsigned topological Laplace matrix to obtain the topological diffusion vectors of each order. The topological diffusion vectors of each order are multiplied and weighted by the diagonal matrix of the corresponding order. The one-dimensional column vectors after weighting all orders are summed to obtain the global scheduling ground state signal vector. Perform a vector subtraction operation between the current transient signal vector and the global scheduling ground state signal vector, and use the difference as the local disturbance residual signal vector; The elements on the diagonal of the diagonal matrix of the corresponding order are the diffusion attenuation coefficients, while the elements on the off-diagonal are all 0. The mathematical expression for the diffusion attenuation coefficient is: ; in, For nodes In the The diffusion attenuation coefficient value under topological diffusion. For nodes The physical length of the river channel from the starting point of the water conservancy project. Let be the spatial diffusion scaling constant. The topological order is currently being calculated, with values ​​of 1, 2, and 3. The mathematical expression for the global scheduling ground state signal vector is: ; in, For a moment The output global scheduling ground state signal vector, For a moment Known unsigned topological Laplace matrix, For a moment Given the graph signal vector, This represents the current topological order being calculated. The highest topological order is set as the cutoff value. Representing an unsigned topological Laplace matrix Power matrix For the first A diagonal matrix of order n.

[0026] In one specific embodiment of the present invention, as water flows downstream from the gate, the nodes closer to the gate are affected by the greater wave intensity, while the nodes farther away are affected by the less, and the attenuation rate is different under different topological orders. In this embodiment, the highest topological order is 3. The highest topological order is the change in water flow energy that is transmitted outward and affects the three connected river segments at a distance of 3 from the source node. For conventional water conservancy networks, the topological connectivity range covered by the highest topological order cutoff value needs to cover the main hydrodynamic diffusion area of ​​a sudden water flow scheduling. when When this occurs, it indicates that the node's original state does not undergo spatial diffusion; in this case, a dimension of is generated. diagonal matrix The matrix has all diagonal elements fixed at 1, and all other off-diagonal elements filled with 0. when Values At that time, for each node Calculate the first The diffusion attenuation coefficient represents the diffusion attenuation coefficient in the topological order. The larger the neighborhood, the greater the physical distance. The resulting attenuation effect will be diluted accordingly; After each level of calculation is completed, these N diffusion attenuation coefficients are sequentially filled into a dimension. The new square matrix is ​​generated by filling the main diagonal position with 0, and filling all other non-diagonal positions with 0, thus generating the [missing information]. Diagonal matrix of order ; Perform matrix polynomial accumulation calculation to obtain the global scheduling ground state signal vector. The mathematical expression of the global scheduling ground state signal vector is: ; in, For a moment The output global scheduling ground state signal vector, The zero-order native state diagonal matrix, It is a first-order diffusion attenuation diagonal matrix. It is a second-order diffusion attenuation diagonal matrix. It is a third-order diffusion attenuation diagonal matrix. For a moment The unsigned topological Laplace matrix, , , These are the first, second, and third powers of the unsigned topological Laplace matrix, respectively. For a moment The current transient signal vector; The power matrix of an unsigned topological Laplace matrix For the nodes in the graph The topological reachability step involves multiplying the power matrix by the transient graph signal vector, representing the water flow energy according to the physical river network topology. Subspace cascade smoothing; Then use a diagonal matrix By performing left multiplication, the scaling weights of the smoothing results across different geographic nodes are adjusted to account for the differential non-uniformity. All The weighted products of order are summed to obtain a final product with dimension . The one-dimensional column vector is the global scheduling ground state signal vector. Each value in the global scheduling ground state signal vector is fitted to a reasonable continuous and gradual water level fluctuation value generated at that node under compliant scheduling, such as the opening of a sluice gate to release water. Extract the transient signal vector obtained in step S1. The transient signal vector internally incorporates compliance scheduling changes, environmental noise, and potential disaster mutations. Perform element-wise vector subtraction with the global scheduling ground state signal vector to calculate the difference column vector, which is also a dimensionless vector. The one-dimensional column vector is used as the local disturbance residual signal vector. The significance of this subtraction operation is that it completely eliminates the drastic changes in the network data caused by compliant high-volume scheduling as a known reasonable background, thereby immune to the interference of high-volume scheduling on anomaly identification. The final obtained local disturbance residual signal vector only retains high-frequency environmental wind and wave white noise and structural abrupt signals caused by engineering structural damage, such as leaks in underground pipes or dam breaks, which do not conform to the continuous and smooth diffusion law of water flow.

[0027] This invention, by introducing a diffusion attenuation coefficient and weighted summation of various powers of the unsigned topological Laplace matrix, fits the continuous and gradual spatial cascade diffusion process of water flow energy in the physical river network topology as it decays with distance, thereby generating a global scheduling ground-state signal vector. The transient signal vector is then subtracted from this ground-state signal vector to extract local disturbance residuals. Its core advantage lies in its ability to completely eliminate drastic changes in the entire network data caused by compliant large-flow scheduling such as sluice gate opening and water release, treating them as a known and reasonable background. This mechanism effectively immunizes against interference from human scheduling in anomaly identification, ensuring that the final extracted residual signal retains only environmental wind and wave noise and structural abrupt signals that do not conform to the smooth diffusion law caused by engineering structure damage.

[0028] Step S3, the process of obtaining the values ​​of algebraic invariants, specifically includes: Construct a first identity matrix in the form of a square matrix with the same dimensions as the unsigned topological Laplace matrix, and perform scalar multiplication operations between the preset algebraic symbolic variables and the identity matrix to obtain an identity matrix with algebraic variables; The characteristic matrix is ​​obtained by subtracting the unsigned topological Laplace matrix from the identity matrix with algebraic variables; The determinant of the characteristic matrix is ​​calculated using a determinant expansion algorithm to obtain the characteristic polynomial; Extract the numerical coefficients of the predetermined powers of the algebraic symbolic variables from the characteristic polynomial, and use them as the values ​​of the algebraic invariants at the current moment.

[0029] In a specific embodiment of the present invention, a dimension is also constructed. The first identity matrix is ​​in the form of a square matrix, with all diagonal elements being the number 1 and all other off-diagonal elements being the number 0. Introducing algebraic symbolic variables Perform a scalar multiplication of the algebraic symbolic variables with the identity matrix, such that all elements on the main diagonal of the identity matrix become algebraic symbolic variables. Subtract the known unsigned topological Laplace matrix from the identity matrix with algebraic variables to obtain a matrix of dimension . The feature matrix; A purely algebraic determinant expansion algorithm, such as the Leibniz formula, is used to calculate the determinant of this characteristic matrix. The result after expansion is a univariate high-order polynomial with a highest power of N. It is the identity matrix with the same dimensions as the unsigned topological Laplacian matrix. Let the unsigned topological Laplace matrix be the matrix at the current time. The mathematical expression for a univariate high-degree polynomial is: ; in, This represents a univariate high-degree polynomial obtained after expansion through determinant calculation, with algebraic variables as unknowns. This represents a purely algebraic symbolic variable introduced before the operation. The dimension of the input feature matrix is ​​equal to the total number of sensor nodes in the hydraulic engineering network. This represents the highest power of the algebraic variable in the polynomial, and its order is equal to the total number of sensor nodes. Representing algebraic variables The real coefficient value attached to the exponentiation term. Representing algebraic variables The real coefficient value attached to the exponentiation term. The real coefficients of the first-degree terms of the algebraic variable. This represents the constant term after the polynomial expansion; In complex dynamic pipe networks of water conservancy projects, directly monitoring the threshold values ​​of specific elements in dynamic adjacency matrices or Laplace matrices often faces a very high risk of false alarms. This is because when the sensor node numbers inside the flood control system are logically rearranged, the positions of the elements inside the matrix will change drastically. By constructing the feature matrix and calculating its characteristic polynomial, we are essentially performing dimensionality reduction feature extraction on the entire hydraulic map signal topology. The characteristic polynomial of the unsigned topological Laplace matrix completely encapsulates the macroscopic topological invariance of the graph, that is, no matter how the arrangement order of the nodes changes (mathematical similarity transformation), the coefficients of the polynomial always remain absolutely constant. Therefore, it is used as a benchmark to measure the overall structural stability, providing a stable and high signal-to-noise ratio evolution rate benchmark for the subsequent dynamic filtering boundary generation.

[0030] This invention achieves dimensionality reduction feature extraction of the macroscopic topological invariance of the entire hydraulic map signal by constructing a feature matrix and using a determinant expansion algorithm to calculate the feature polynomial. In complex and dynamic hydraulic networks, the logical rearrangement of sensor nodes can lead to drastic changes in the positions of elements within the dynamic adjacency matrix. Directly monitoring the matrix elements for thresholds faces a high risk of false alarms. The algebraic invariants of the feature polynomial extracted by this invention maintain absolute constant coefficients under node similarity transformations, avoiding interference caused by sensor number rearrangement. This provides an extremely stable and high signal-to-noise ratio evolution rate benchmark for the subsequent generation of dynamic filtering boundaries.

[0031] Step S3, the process of obtaining the dynamic filtering boundary, specifically includes: Obtain the algebraic invariant value from the previous time step, subtract the algebraic invariant value from the previous time step from the current time step, and take the absolute value to obtain the evolution rate value. The dynamic lower limit frequency value is calculated based on the evolution rate value. The dynamic lower limit frequency value is the sum of the reference cutoff frequency value and the product of the evolution rate amplification factor and the evolution rate value. The dynamic lower limit frequency value is combined with the preset fixed upper limit frequency value to form a closed interval as the dynamic filtering boundary; The mathematical expression for the dynamic lower limit frequency value is: ; in, This is the dynamic lower limit frequency value calculated at the current moment. The set reference cutoff frequency value. This is the preset evolution rate amplification factor. This represents the rate of evolution. The preset dynamic lower limit cutoff tolerance is 9.0 in this embodiment; In a specific embodiment of the present invention, the algebraic invariant value of the previous moment is extracted, and the difference between the algebraic invariant value of the previous moment and the absolute value is taken. The absolute value of the difference is used as the evolution rate value, which reflects the degree of damage and distortion of the water network topology at two adjacent sampling moments. This evolution rate value is used to dynamically generate the dynamic lower limit frequency value of the bandpass filter; In this embodiment, The value is fixed at 0.1 because in the spectrum of water conservancy map signals, the low-frequency part between 0 and 0.1 corresponds to the slow and compliant gradual fluctuation of water bodies across the entire network due to gravity. Setting the baseline to 0.1 is to ensure that safe low-frequency fluctuations are blocked under normal conditions. In this embodiment, the evolution rate amplification factor is fixed at 5.0. Since the changes in algebraic invariants caused by micro-piping or microcracks in the physical dam body are usually small, the evolution rate amplification factor is used to amplify them, causing the filter's interception threshold to shift to the right, thereby framing high-frequency abrupt signals. Calculate the dynamic lower limit frequency value Then, it is compared with the preset fixed upper limit frequency value. Combining them, they form a closed interval. In this embodiment, The fixed value is 10.0. This represents the theoretical extreme value of the highest frequency oscillation energy generated by the water flow. Closed interval This is the final output dynamic filtering boundary, which determines the frequency range that the bandpass filter is allowed to pass through in the subsequent step S4.

[0032] This invention utilizes the evolution rate values ​​calculated from algebraic invariants at adjacent time points to dynamically generate the lower limit frequency boundary of the filter, achieving adaptive and precise targeting of anomalous signals. By setting a reference cutoff frequency, this invention can safely intercept slow, compliant global gradual fluctuations caused by gravity under normal conditions. When minor piping or microcracks occur in the physical dam body, causing subtle changes in algebraic invariants, the introduced evolution rate amplification factor can significantly amplify this distortion, causing the filter's interception threshold to shift rapidly to the right. This dynamic boundary mechanism can extremely sensitively and adaptively lock onto high-frequency abrupt signals, improving the ability to detect hidden structural disasters.

[0033] Step S4, the process of constructing the bandpass graph filter, specifically includes: Construct a second identity matrix, and perform scaling operations on the second identity matrix and the unsigned topological Laplacian matrix, specifically including: Multiply the unsigned topological Laplacian matrix by the first scaling constant, and then subtract the second identity matrix to obtain the scaled Laplacian matrix. The spectral domain eigenvalues ​​of the scaled Laplacian matrix are located within a preset interval. The closed interval formed by the dynamic lower limit frequency value and the fixed upper limit frequency value is used as the dynamic filtering boundary. The Chebyshev polynomial expansion coefficients of the ideal bandpass filter function are calculated within the dynamic filtering boundary. The bandpass filter is mathematically equivalent to the scaled Laplace matrix, the Chebyshev polynomial expansion coefficients, and the Chebyshev polynomial matrix.

[0034] In a specific embodiment of the present invention, the construction dimension is... The second identity matrix is ​​then subjected to scaling operations. The scaling operation specifically includes: Set the first scaling constant to Set the second scaling constant to 1; Multiplying all elements of the unsigned topological Laplacian matrix by the first scaling constant, and then subtracting the identity matrix, yields a matrix with dimension 1. The Laplace matrix is ​​scaled, with its rows and columns representing hydraulic sensor nodes. Due to the orthogonal domain constraint of the Chebyshev polynomial, this scaling operation shifts and scales the spectral domain eigenvalues ​​of the original Laplace matrix to a new value. Within the effective approximation interval; Obtain the dynamic filtering boundary interval An ideal bandpass filter is constructed in the frequency domain, that is, the signal passes through completely within the dynamic filtering boundary interval with a gain of 1, and the signal is completely blocked outside the interval with a gain of 0. In this embodiment, the approximation truncation order is fixed at 10. A Chebyshev polynomial with an approximation truncation order of 10 can fit the steep bandpass filter boundary with a very small approximation error. A higher order would increase meaningless cyclic multiplication operations and cause system delay. In response to arrive For each order of the ideal bandpass filter, the scalar coefficients of the approximation expansion are calculated using the Chebyshev numerical integral formula, thus obtaining 11 definite constants, denoted as the Chebyshev polynomial expansion coefficients.

[0035] This invention performs scaling operations on the unsigned topological Laplace matrix, shifting its spectral domain eigenvalues ​​as a whole and strictly limiting them to the effective domain interval required for Chebyshev polynomial approximation. Based on this, an ideal bandpass filter is constructed in the graph frequency domain using the Chebyshev polynomial expansion coefficients. This allows for the fitting of steep bandpass boundaries with minimal approximation error, enabling target anomalous signals within the dynamic filtering boundary to pass completely, while interference signals outside the interval are completely blocked. This invention effectively avoids the delay caused by meaningless high-order cyclic multiplication operations while ensuring isolation accuracy in the ultra-high frequency domain.

[0036] Step S4, the process of obtaining the structural anomaly signal vector, specifically includes: The local perturbation residual signal vector is used as the zero-order iterative column vector for Chebyshev recursive calculation. The scaling Laplacian matrix is ​​multiplied by the zero-order iterative column vector to obtain the first-order iterative column vector. Starting from the second order up to the preset approximation truncation order, the Chebyshev recursive equation is used to perform loop calculations to obtain the iterative column vectors of each order. The Chebyshev recursive equation is that the iterative column vector of the current order is equal to twice the product of the scaled Laplace matrix and the iterative column vector of the previous order minus the iterative column vector of the previous order. By using the Chebyshev polynomial expansion coefficients to linearly combine and accumulate the iterative column vectors of each order, the structural anomaly signal vector is obtained. The mathematical expression for the structural anomaly signal vector is: ; in, This is the structural anomaly signal vector output at the instant t of the current synchronous sampling. To approximate the truncation order, This is the index of the current expansion order. For the first The coefficients of the Chebyshev polynomial expansion of order 1, To scale the Laplacian matrix, The first variable is the scaled Laplace matrix. Chebyshev polynomial matrix of order, It is the vector of the local disturbance residual signal.

[0037] In one specific embodiment of the present invention, in order to avoid calculating multiple high powers of the scaled Laplacian matrix itself, matrix multiplication is transformed into a recursive iterative operation of continuous matrix multiplication vectors. Specifically, the local perturbation residual signal vector generated in step S2 is obtained as the zero-order iterative column vector. ; Perform a standard matrix-vector multiplication between the scaled Laplacian matrix and the zero-order iterative column vector to obtain the first-order iterative column vector. from Start until By strictly performing iterative calculations using the Chebyshev recursive equation, the matrix is ​​scaled with the Laplace matrix and the previous iteration column vector. For multiplication, multiply the resulting column vector by 2, and then subtract the vector of the previous order. This yields the iterative column vector of the current order. ; In obtaining from arrive After a total of 11 iterative column vectors, the iterative column vectors are linearly combined and accumulated using the Chebyshev polynomial expansion coefficients. The mathematical expression for this linear combination accumulation is: ; in, Let be the structural anomaly signal vector output at the current synchronous sampling instant t, with dimension . ; The index is the current expansion order. In this embodiment, the value ranges from 0 to M, where M is the set approximation truncation order, and in this embodiment, it is 10. These are the coefficients of the Chebyshev polynomial expansion; It is a mathematical theoretical expression, and in the physical calculation steps, it is completely equivalent to the third stage recursively calculated first... Iterative column vectors of order ; The output structural anomaly signal vector is obtained by summing the elements one by one as described above. The low-frequency global gradual water flow, i.e. the low-frequency residue that could not be removed by step S2, and the extremely high-frequency natural wind and wave white noise, are eliminated; each element in this column vector corresponds to the mid-to-high frequency topological abnormal peaks induced by transient water flow suction caused by engineering structure damage, such as piping inside the dam, at the physical location of each sensor.

[0038] This invention transforms the highly complex matrix power multiplication into a continuous Chebyshev matrix-vector recursive iteration operation when acquiring structural anomaly signal vectors. This reduces hardware computing power consumption and improves the execution efficiency of the algorithm in real-time hydrological monitoring systems. By linearly combining and accumulating the iterative column vectors of each order through Chebyshev expansion coefficients, it can deeply clean up low-frequency residual scheduling fluctuations that cannot be completely eliminated and extremely high-frequency natural wind and wave white noise. The final output structural anomaly signal accurately corresponds to the mid-to-high frequency topological anomaly peaks induced by engineering physical damage such as transient water flow suction.

[0039] Step S5, the process of identifying abnormal nodes, specifically includes: Extract the numerical elements from the structural anomaly signal vector in row order, and perform a self-square calculation on each extracted numerical element; Arrange the calculated square values ​​in order of their original node index numbers to generate a transient energy sequence vector; Compare the elements in the transient energy sequence vector with the energy alarm threshold; If an element in the transient energy sequence vector is greater than the energy alarm threshold, the value of the element is taken as the target value, and the row number corresponding to the element in the transient energy sequence vector is extracted. The row number is the abnormal source node number. Based on the preset sensor coordinate mapping table, obtain the latitude and longitude coordinates corresponding to the abnormal source node number, and package the latitude and longitude coordinate values, target values ​​and current timestamp into a system alarm data packet. The system alarm data packets are sent to the flood control command center via a wired communication network.

[0040] In a specific embodiment of the present invention, in order to identify the node where a structural disaster actually occurs from the transient energy sequence vector, an energy alarm threshold is preset. In this embodiment, the value of the energy alarm threshold is set to 0.04. If, after steps S2 and S4 have completely eliminated the normal surge in water level, a sudden drop or rise in the cross-sectional energy of an isolated section still exceeds the square root of the hard threshold and violates the continuity of the context topology, this is extremely abnormal in hydrodynamics. This indicates that an abnormal loss or obstruction of water has occurred in the area where the node is located, such as piping at the bottom of the dike causing the water flow to be suddenly pumped out, or a sudden large landslide blocking the river channel. The hard threshold is set to prevent engineering-level damage. Compare the elements in the transient energy sequence vector with the energy alarm threshold; When an element in the transient energy sequence vector is greater than the energy alarm threshold, the value of that element is taken as the target value, and the row number corresponding to that element in the column vector is determined. Extracted, line number The source node number where the physical anomaly occurred; After obtaining the node number of the disaster, the latitude and longitude of the abnormal node are obtained according to the preset sensor coordinate mapping table. The preset sensor coordinate mapping table contains three columns of data: the first column is the integer of the node number, the second column is the actual longitude coordinate value of the sensor deployment location, and the third column is the actual latitude coordinate value. The extracted exact physical latitude and longitude coordinates, target values, representing the intensity of disaster damage, and the current timestamp are packaged together and encapsulated into the system alarm data packet according to the hydrological bureau's communication protocol; The data packet was sent directly to the control console of the flood control command center via a wired communication network.

[0041] This invention generates a transient energy sequence by performing self-square calculation on the purified structural anomaly signal vector elements, further amplifying the anomalous hydrological physical energy fluctuation characteristics. By using a set physical energy alarm threshold for rigorous comparison, it can identify drastic drops or rises that violate the continuity of the contextual topology of water flow. From the perspective of fluid dynamics, it can accurately determine engineering-level damage such as levee piping or landslide blockage. Combined with a preset sensor coordinate mapping table, it can directly extract the location of the anomaly source node.

[0042] A hydrological anomaly identification system for water conservancy projects includes a topology construction module, a residual separation module, a feature extraction module, an anomaly extraction module, and an early warning module; The topology construction module is used to collect hydrological observation data, map the monitoring sections to graph nodes, calculate the cross-sectional specific energy of each graph node based on the hydrological observation data, construct the transient graph signal vector, and construct an unsigned topological Laplace matrix based on the physical connectivity between each node. The residual separation module is used to calculate the global scheduling ground state signal vector based on the transient diagram signal vector and the unsigned topological Laplace matrix, and to perform a difference operation between the transient diagram signal vector and the global scheduling ground state signal vector to obtain the local disturbance residual signal vector. The feature extraction module is used to calculate the characteristic polynomial based on the unsigned topological Laplacian matrix, extract the algebraic invariant values ​​from the characteristic polynomial, calculate the evolution rate values ​​based on the algebraic invariant values ​​of the current time step and the previous time step, and generate dynamic filtering boundaries based on the evolution rate values. The anomaly extraction module is used to construct a bandpass graph filter based on Chebyshev polynomial approximation using the unsigned topological Laplace matrix and dynamic filtering boundary, and to filter the local disturbance residual signal vector to obtain the structural anomaly signal vector. The early warning module is used to calculate the transient energy sequence vector based on the structural anomaly signal vector, compare the elements in the transient energy sequence vector with the preset energy alarm threshold, identify abnormal nodes, and generate alarm information.

[0043] This invention introduces cross-sectional specific energy as a fundamental physical quantity and constructs an unsigned topological Laplace matrix by combining the physical connectivity of the water network. This matrix can characterize the overall energy accumulation state and hydrophysical topological framework of water flow under dynamic conditions. Based on this, this invention calculates the global scheduling ground state signal vector based on matrix polynomial expansion and removes it from the transient signal by subtraction, thereby accurately separating the local disturbance residual signal. This effectively immunizes against interference from compliant large-flow manual scheduling, such as the opening and releasing of water from sluice gates, and removes the global gradual fluctuations caused by normal hydrological scheduling. This invention utilizes the characteristic polynomial of the unsigned topological Laplace matrix to extract algebraic invariants, taking the invariance of the macroscopic topological structure as a benchmark for measuring the evolution rate of the water network. Since the algebraic invariants obtained by the characteristic polynomial have absolute robustness to the rearrangement of the logical numbers of sensor nodes within the water conservancy network, this method reduces the risk of false alarms caused by changes in the sensor network. It provides an extremely stable and high signal-to-noise ratio evolution rate basis for the generation of dynamic filtering boundaries. Combined with the bandpass graph filter constructed by dynamic lower limit frequency and Chebyshev polynomial approximation, it can adaptively and accurately intercept the mid-to-high frequency topological distortion signals representing local structural damage in the complex graph frequency domain, while filtering out residual low-frequency natural fluctuations and extremely high-frequency environmental wind and wave white noise, thus improving the accuracy and environmental adaptability of graph signal filtering in complex water conservancy environments. This invention, by amplifying the transient energy sequence of filtered pure structural anomaly signals and comparing it with hard thresholds, can identify drastic energy drops or rises that violate the topological continuity of the water network context. On a clean data base stripped of background noise and compliant scheduling, it can directly and clearly pinpoint the abnormal source nodes causing abnormal water loss or obstruction. Furthermore, it links with a pre-set geographic coordinate mapping table to achieve precise latitude and longitude alarms for physical disaster locations. This invention effectively breaks through the limitations of traditional hydrological monitoring that relies solely on statistical thresholds, making it difficult to determine the causes of anomalies. It achieves deep-level perception and precise spatial positioning of engineering-level physical structural damage disasters, improving the intelligent early warning level and response efficiency of flood control and disaster reduction command at water conservancy hubs.

[0044] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0045] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for identifying hydrological anomalies in water conservancy projects, characterized in that, Includes the following steps: Step S1: Collect hydrological observation data and map the monitoring section as a graph node. Calculate the cross-sectional specific energy of each graph node based on the hydrological observation data, construct the transient graph signal vector, and construct an unsigned topological Laplace matrix based on the physical connectivity between the nodes. Step S2: Based on the transient graph signal vector and the unsigned topological Laplace matrix, calculate the global scheduling ground state signal vector using matrix polynomial expansion, and perform a difference operation between the transient graph signal vector and the global scheduling ground state signal vector to obtain the local disturbance residual signal vector; Step S3: Calculate the characteristic polynomial based on the unsigned topological Laplace matrix, extract the algebraic invariant values ​​from the characteristic polynomial, calculate the evolution rate value based on the algebraic invariant values ​​of the current time and the previous time, and generate a dynamic filtering boundary based on the evolution rate value. Step S4: Based on the unsigned topological Laplace matrix and the dynamic filtering boundary, a bandpass graph filter is constructed based on Chebyshev polynomial approximation to filter the local disturbance residual signal vector and obtain the structural anomaly signal vector. Step S5: Calculate the transient energy sequence vector based on the structural anomaly signal vector, compare the elements in the transient energy sequence vector with the preset energy alarm threshold, identify abnormal nodes, and generate alarm information.

2. The method for identifying hydrological anomalies in water conservancy projects as described in claim 1, characterized in that, The process of constructing the transient signal vector in step S1 specifically includes: Hydrological observation data is collected by hydrological telemetry terminals deployed on the physical cross sections of water conservancy projects, and the sensor monitoring sections in the water conservancy project network are mapped to nodes in the graph topology. The hydrological observation data includes real-time water level values ​​at each node, average cross-sectional flow velocity values ​​at each node, and node number. The cross-sectional specific energy of each node is calculated based on the hydrological observation data. The mathematical expression for the cross-sectional specific energy is as follows: ； in, Let be the gravitational acceleration constant, and take the value of . , For the first The cross-sectional specific energy of each node, For the first Real-time water level values ​​at each node ) is the first The average flow velocity of the cross section at each node; Arrange the cross-sectional energy values ​​of all nodes according to their node numbers to form a one-dimensional column vector, which is the current transient signal vector.

3. The method for identifying hydrological anomalies in water conservancy projects as described in claim 2, characterized in that, The process of constructing the unsigned topological Laplacian matrix in step S1 specifically includes: Select any two nodes to form a node pair, collect the physical river course of the water conservancy project, and determine whether there is a direct physical water flow connection between the node pairs. If there is a direct physical river channel connection between the node pairs, it is determined that there is a physical water flow connection relationship; If there is no direct physical waterway connection between the node pairs, it is determined that there is no physical water flow connection. The dynamic connection weight of node pairs that do not have the physical water flow connectivity relationship is zero; The mathematical expression for the dynamic connection weight of node pairs with the aforementioned physical water flow connectivity is: ； in, Starting point and the end point At the present moment Dynamic connection weights, For the current moment, Starting point and the end point The absolute value of the difference in cross-sectional specific energy at the current time t. This is a preset energy scaling constant; Construct a dynamic adjacency matrix, where the rows are the starting points, the columns are the ending points, and the elements are the dynamic connection weights between the corresponding pairs of nodes. Summing the elements of each row of the dynamic adjacency matrix, and generating a degree matrix in diagonal form using the sum of each row as the main diagonal element, and performing matrix addition on the degree matrix and the dynamic adjacency matrix to obtain an unsigned topological Laplace matrix.

4. The method for identifying hydrological anomalies in water conservancy projects as described in claim 3, characterized in that, The process of calculating the global scheduling ground state signal vector in step S2 specifically includes: The current transient signal vector is multiplied sequentially with each power of the unsigned topological Laplace matrix to obtain the topological diffusion vectors of each order. The topological diffusion vectors of each order are multiplied and weighted by the diagonal matrix of the corresponding order. The one-dimensional column vectors after weighting all orders are summed to obtain the global scheduling ground state signal vector. Perform a vector subtraction operation between the current transient graph signal vector and the global scheduling ground state signal vector, and use the difference result as the local disturbance residual signal vector; The elements on the diagonal of the diagonal matrix of the corresponding order are diffusion attenuation coefficients, and the elements on the other off-diagonal positions are all 0. The mathematical expression for the diffusion attenuation coefficient is: ； in, For nodes In the The diffusion attenuation coefficient value under topological diffusion. For nodes The physical length of the river channel from the starting point of the water conservancy project. Let be the spatial diffusion scaling constant. The topological order is currently being calculated, with values ​​of 1, 2, and 3. The mathematical expression for the global scheduling ground state signal vector is: ； in, For a moment The output global scheduling ground state signal vector, For a moment The known unsigned topological Laplace matrix, For a moment Given the graph signal vector, This represents the current topological order being calculated. The highest topological order is set as the cutoff value. Representing an unsigned topological Laplace matrix Power matrix For the first A diagonal matrix of order n.

5. The method for identifying hydrological anomalies in water conservancy projects as described in claim 4, characterized in that, The process of obtaining the numerical value of the algebraic invariant in step S3 specifically includes: Construct a first identity matrix in the form of a square matrix with the same dimensions as the unsigned topological Laplace matrix, and perform scalar multiplication operations between the preset algebraic symbolic variables and the identity matrix to obtain an identity matrix with algebraic variables; The characteristic matrix is ​​obtained by subtracting the unsigned topological Laplace matrix from the identity matrix with algebraic variables. The determinant of the characteristic matrix is ​​calculated using a determinant expansion algorithm to obtain the characteristic polynomial; The numerical coefficients of the predetermined powers of the algebraic symbolic variables are extracted from the characteristic polynomial and used as the values ​​of the algebraic invariants at the current moment.

6. The method for identifying hydrological anomalies in water conservancy projects as described in claim 5, characterized in that, The process of obtaining the dynamic filtering boundary in step S3 specifically includes: Obtain the algebraic invariant value from the previous time step, subtract the algebraic invariant value from the previous time step from the current time step, and take the absolute value to obtain the evolution rate value. The dynamic lower limit frequency value is calculated based on the evolution rate value, which is the sum of the product of the reference cutoff frequency value, the evolution rate amplification factor, and the evolution rate value. The dynamic lower limit frequency value is combined with the preset fixed upper limit frequency value to form a closed interval as the dynamic filtering boundary. The mathematical expression for the dynamic lower limit frequency value is: ； in, This is the dynamic lower limit frequency value calculated at the current moment. The set reference cutoff frequency value. This is the preset evolution rate amplification factor. This represents the rate of evolution. This is a preset dynamic lower limit cutoff tolerance.

7. The method for identifying hydrological anomalies in water conservancy projects as described in claim 6, characterized in that, In step S4, the process of constructing the bandpass filter specifically includes: Construct a second identity matrix, and perform scaling operations on the second identity matrix and the unsigned topological Laplacian matrix, specifically including: Multiply the unsigned topological Laplacian matrix by the first scaling constant, and then subtract the second identity matrix to obtain the scaled Laplacian matrix. The spectral domain eigenvalues ​​of the scaled Laplacian matrix are located within a preset interval. The closed interval formed by the dynamic lower limit frequency value and the fixed upper limit frequency value is used as the dynamic filtering boundary. The Chebyshev polynomial expansion coefficients of the ideal bandpass filter function are calculated within the dynamic filtering boundary. The bandpass filter is mathematically equivalent to the scaled Laplace matrix, the Chebyshev polynomial expansion coefficients, and the Chebyshev polynomial matrix.

8. The method for identifying hydrological anomalies in water conservancy projects as described in claim 7, characterized in that, The process of obtaining the structural anomaly signal vector in step S4 specifically includes: The local perturbation residual signal vector is used as the zero-order iterative column vector for Chebyshev recursive calculation. The scaling Laplace matrix is ​​multiplied by the zero-order iterative column vector to obtain the first-order iterative column vector. Starting from the second order up to the preset approximation truncation order, the Chebyshev recursive equation is used to perform iterative calculations to obtain the iterative column vectors of each order. The Chebyshev recursive equation states that the iterative column vector of the current order is equal to twice the product of the scaling Laplacian matrix and the iterative column vector of the previous order minus the iterative column vector of the order before that. The structural anomaly signal vector is obtained by linearly combining and accumulating the iterative column vectors of each order using the Chebyshev polynomial expansion coefficients. The mathematical expression for the structural anomaly signal vector is: ； in, This is the structural anomaly signal vector output at the instant t of the current synchronous sampling. To approximate the truncation order, This is the index of the current expansion order. For the first The coefficients of the Chebyshev polynomial expansion of order 1, To scale the Laplacian matrix, The first variable is the scaled Laplace matrix. Chebyshev polynomial matrix of order, This is the vector of the local disturbance residual signal.

9. The method for identifying hydrological anomalies in water conservancy projects as described in claim 8, characterized in that, In step S5, the process of identifying the abnormal node specifically includes: Extract the numerical elements from the structural anomaly signal vector in row order, and perform a self-square calculation for each extracted numerical element; Arrange the calculated square values ​​in order of their original node index numbers to generate a transient energy sequence vector; The elements in the transient energy sequence vector are compared with the energy alarm threshold. If an element in the transient energy sequence vector is greater than the energy alarm threshold, the value of the element is taken as the target value, and the row number corresponding to the element in the transient energy sequence vector is extracted, where the row number is the abnormal source node number. According to the preset sensor coordinate mapping table, obtain the latitude and longitude coordinates corresponding to the abnormal source node number, and package the latitude and longitude coordinate values, the target value and the current timestamp into a system alarm data packet; The system alarm data packets are sent to the flood control command center via a wired communication network.

10. A hydrological anomaly identification system for water conservancy projects, applied in a hydrological anomaly identification method for water conservancy projects as described in any one of claims 1-9, characterized in that, It includes a topology construction module, a residual separation module, a feature extraction module, an anomaly extraction module, and an early warning module; The topology construction module is used to collect hydrological observation data, map the monitoring section to graph nodes, calculate the cross-sectional specific energy of each graph node based on the hydrological observation data, construct the transient graph signal vector, and construct an unsigned topological Laplace matrix based on the physical connectivity between each node. The residual separation module is used to calculate the global scheduling ground state signal vector based on matrix polynomial expansion according to the transient graph signal vector and the unsigned topological Laplace matrix, and to perform a difference operation between the transient graph signal vector and the global scheduling ground state signal vector to obtain the local disturbance residual signal vector. The feature extraction module is used to calculate the characteristic polynomial based on the unsigned topological Laplacian matrix, extract the algebraic invariant values ​​from the characteristic polynomial, calculate the evolution rate value based on the algebraic invariant values ​​at the current time and the previous time, and generate a dynamic filtering boundary based on the evolution rate value. The anomaly extraction module is used to construct a bandpass graph filter based on Chebyshev polynomial approximation according to the unsigned topological Laplace matrix and the dynamic filtering boundary, and to filter the local disturbance residual signal vector to obtain the structural anomaly signal vector. The early warning module is used to calculate a transient energy sequence vector based on the structural anomaly signal vector, compare the elements in the transient energy sequence vector with a preset energy alarm threshold, identify abnormal nodes, and generate alarm information.

Images