Digital twinning fused power grid fault prediction system and method

Abstract

The invention relates to the technical field of digital twinning, in particular to a digital twinning fused power grid fault prediction system and method, and the system comprises a state sensing module, a path deduction module, a twinning mapping module and a prediction execution module. According to the method, multi-source heterogeneous sensing data are fused, a space-time alignment algorithm is executed to construct a structural state tensor, Laplacian operation is combined with a stress wave propagation model to analyze non-rigid deformation characteristics, a load transfer path atlas is generated to reveal recessive mechanical association, Young modulus is dynamically corrected through a finite element grid to form a structural degradation coefficient matrix, and the structural degradation coefficient matrix is obtained. The time sequence difference and the long short-term memory network are combined to capture the equipment degradation trend, the Kolmogov test is used to quantify the fault space-time distribution probability, the power grid equipment jumps from single threshold early warning to multi-physics coupling prediction, the fault prediction space-time resolution and dynamic adaptability are improved, the artificial experience dependence is reduced, and the fault prediction efficiency is improved. And the robustness of the complex working condition system is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of digital twins, and particularly to a power grid fault prediction system and method integrating digital twins. Background Technique

[0002] The technical field of digital twins involves the full-life-cycle dynamic mapping technology of physical entities and virtual models. By using a sensor network to collect entity data in real time to construct a three-dimensional visualization model, and combining simulation algorithms to achieve system state monitoring and trend deduction. Among them, the traditional power grid fault prediction system refers to a technical solution based on the statistical analysis of historical operation data and off-line model calculation. The SCADA system is used to collect voltage and current parameters, and the artificial setting of a threshold interval is used for equipment anomaly warning. It relies on regular maintenance personnel to detect the aging degree of equipment on-site, and uses empirical formulas to estimate the remaining life cycle of equipment.

[0003] The traditional method relies on artificial setting of a fixed threshold interval for anomaly judgment, and cannot adapt to the non-linear attenuation characteristics of material properties during the operation of equipment, resulting in a contradiction between warning sensitivity and false alarm rate. The voltage and current parameters collected by the SCADA system are difficult to reflect the internal stress distribution of the mechanical structure of the equipment. There are monitoring blind spots in the regular maintenance mode. The dynamic changes of the load transfer path are ignored when using empirical formulas to estimate the remaining life, resulting in life prediction deviation. The efficiency of on-site manual detection is low and sudden deformations cannot be captured, and it is easy to miss detections when the equipment is under alternating loads or extreme environments, affecting the safety and stability of power grid operation. Summary of the Invention

[0004] The purpose of the present invention is to solve the deficiencies existing in the prior art, and to propose a power grid fault prediction system and method integrating digital twins.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: A power grid fault prediction system integrating digital twins includes: A state perception module, configured to obtain the strain gradient of the steel structure of the transmission tower through a distributed optical fiber sensor, collect the displacement trajectory of the transmission tower base through a Beidou positioning unit, calculate the angular velocity based on a six-axis inertial unit, perform a spatio-temporal alignment algorithm on multi-source data, combine it into a structure state tensor, and transfer the structure state tensor to the path deduction module; A path deduction module, configured to receive the structure state tensor, perform a Laplace operation on the strain gradient, input the displacement trajectory coordinates into a three-dimensional velocity field correction model, construct a stress wave propagation path in combination with the angular velocity parameter, and solve the wave equation to generate a load transfer path map; Wherein, represents the steel structure of the transmission tower at the spatial position at The displacement vector at a moment, with the unit of meter, represents the time variable, with the unit of second, represents the corrected longitudinal wave propagation speed, with the unit of m / s, represents the Laplace operator, with the dimension of m -2 , represents the three-dimensional space coordinates, with the unit of meter, represents the external excitation force function, with the unit of N / m 3 ; Extract the non-rigid deformation components, generate the load transfer path map, and transfer the load transfer path map to the twin mapping module; The twin mapping module is used to receive the load transfer path map, establish a finite element mesh model, input the stress wave parameters into the damage accumulation model, correct the mesh Young's modulus, generate the structure degradation coefficient matrix, and transfer the structure degradation coefficient matrix to the prediction execution module; The prediction execution module is used to receive the structure degradation coefficient matrix, perform time series differentiation on the degradation coefficient, input the sequence into the long short-term memory network, perform the Kolmogorov test, generate the fault spatio-temporal coordinate encoding, and input the fault spatio-temporal coordinate encoding into the power grid dispatching decision.

[0006] As a further solution of the present invention, the structure state tensor includes the strain gradient distribution, displacement trajectory coordinates, and angular velocity parameters. The load transfer path map is specifically the stress wave attenuation path corrected by the displacement trajectory, the non-rigid deformation region identified based on the angular velocity parameters, and the main load conduction direction obtained through strain gradient analysis. The structure degradation coefficient matrix specifically refers to the material fatigue degree, modulus correction value, and mesh element degradation index. The fault spatio-temporal coordinate encoding includes the prediction period sequence, confidence interval threshold, and fault geographical coordinates.

[0007] As a further solution of the present invention, when the path deduction module executes the stress wave propagation model, the formula: ; wherein, represents the longitudinal wave propagation speed of the stress wave in the steel structure of the transmission tower, with the unit of m / s, represents the reference Young's modulus of the steel structure measured by the distributed optical fiber sensor, with the unit of GPa, represents the Poisson's ratio dimensionless parameter of the transmission tower steel, represents the steel density, with the unit of kg / m 3 ; When calculating the stress wave propagation speed, the strain gradient distribution is used as the initial input parameter, and the velocity field is corrected in combination with the displacement trajectory coordinates.

[0008] As a further solution of the present invention, when the twin mapping module corrects the grid Young's modulus, a damage accumulation model is adopted: ; wherein, represents the corrected grid element Young's modulus with the unit of GPa, represents the initial Young's modulus nominal at the steel factory with the unit of GPa, represents the dimensionless damage factor generated in the q-th load cycle, is the load cycle serial number, represents the total number of load cycles, is obtained by calculating through , represents the amplitude attenuation rate of the stress wave on the propagation path in the q-th cycle, and the calculation formula is (peak amplitude - trough amplitude) / peak amplitude, represents the energy dissipation rate in the main load conduction direction in the q-th cycle, and the calculation formula is 1-(output energy / input energy).

[0009] As a further solution of the present invention, the long short-term memory network in the prediction execution module includes = 128 hidden units, and its gating mechanism adopts the sigmoid activation function; The significance level of the Kolmogorov test is set to 0.05, which is used to verify the distribution consistency between the predicted cycle sequence and the historical fault data in the fault spatio-temporal coordinate coding; wherein, represents the number of neurons in the hidden layer and is optimized and determined on the training set by the gradient descent method, and 0.05 represents the upper limit of the probability of the first type of error allowed in the statistical test.

[0010] As a further solution of the present invention, the spatio-temporal alignment algorithm adopts the dynamic time warping method to unify the sampling frequencies of the strain gradient, displacement trajectory, and angular velocity parameters to 100Hz; When extracting the non-rigid deformation component, the curvature change threshold is set to 0.15rad / m 2 ; wherein, 100Hz represents the sampling frequency of collecting 100 data points per second and is set according to the maximum sampling ability of the distributed optical fiber sensor, and 0.15rad / m 2 represents the critical determination value of the surface curvature change of the transmission tower structure and is obtained by calibrating through the yield test of Q235 steel.

[0011] As a further solution of the present invention, the element size of the finite element grid model is adaptively adjusted according to the main load conduction direction; Adopt a 5mm×5mm grid in the area with intensive load conduction and a 10mm×10mm grid in the area with sparse conduction; Among them, 5mm represents the side length dimension of the square grid unit and is applicable to the area where the stress value in the main load conduction direction exceeds 50MPa, and 10mm represents the side length dimension of the square grid unit and is applicable to the area where the stress value in the main load conduction direction is lower than 50MPa.

[0012] As a further solution of the present invention, the update period of the structure deterioration coefficient matrix is set to 30 minutes; When the deterioration index of the grid unit exceeds 0.7, trigger the real-time warning mechanism of the prediction execution module; Among them, 30 minutes represents the minimum time interval for the system to update the structure deterioration state, and 0.7 represents the critical threshold of the grid unit safety factor and is determined according to the failure critical value of the 85% confidence interval in the tensile test of metallic materials - Part 1: Tensile testing at ambient temperature GB / T228.1 - 2021.

[0013] A power grid fault prediction method integrating digital twin, which is executed based on the above - mentioned power grid fault prediction system integrating digital twin, and includes the following steps: S1: Obtain the strain gradient of the steel structure of the transmission tower through a distributed optical fiber sensor, collect the displacement trajectory of the transmission tower base through a Beidou positioning unit, calculate the angular velocity based on a six - axis inertial unit, perform a spatio - temporal alignment algorithm on multi - source data, and combine it into a structure state tensor; S2: Perform a Laplace operation on the strain gradient in the structure state tensor, input the result into a stress wave propagation model, extract the non - rigid deformation component, and generate a load transfer path map; S3: Establish a finite - element grid model based on the load transfer path map, input the stress wave parameters into a damage accumulation model, correct the Young's modulus of the grid, and generate a structure deterioration coefficient matrix; S4: Perform a temporal difference on the deterioration coefficients in the structure deterioration coefficient matrix, input the sequence into a long short - term memory network, perform a Kolmogorov test, generate a fault spatio - temporal coordinate code and record it into the power grid dispatching decision.

[0014] As a further solution of the present invention, the establishment of the stress wave propagation model in step S2 includes: S201: Calculate the reference propagation speed according to the material property parameters ; S202: Construct a three - dimensional velocity field distribution in combination with the displacement trajectory coordinates; S203: Solve the partial differential equation of stress wave propagation by the finite - difference method; The Young's modulus correction in step S3 includes: S301: Calculate the cumulative damage factor of each grid unit ; S302. Perform modulus reduction according to the damage factor; S303. Update the stiffness matrix of the finite element model.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, by fusing multi-source heterogeneous sensing data and implementing a spatio-temporal alignment algorithm, a structural state tensor is constructed to achieve full-dimensional feature coupling of physical entities. The Laplace operation is combined with the stress wave propagation model to analyze non-rigid deformation characteristics, and a load transfer path map is generated to reveal hidden mechanical correlations. Based on the dynamic correction of Young's modulus for finite element meshes, a structural degradation coefficient matrix is formed. Combining time series difference and long short-term memory network to capture the degradation trend of equipment, and quantifying the probability of fault spatio-temporal distribution through the Kolmogorov test, the power grid equipment realizes the leap from single-parameter threshold warning to multi-physical field coupling prediction, improves the spatio-temporal resolution and dynamic adaptability of fault prediction, reduces the dependence on artificial experience, and enhances the system robustness under complex working conditions. Description of the Drawings

[0016] Figure 1 It is the flowchart of the power grid fault prediction system integrating digital twin of the present invention; Figure 2 It is the flowchart of the state perception module of the present invention; Figure 3 It is the process of the path deduction module of the present invention; Figure 4 It is the flowchart of the twin mapping module of the present invention; Figure 5 It is the flowchart of the prediction execution module of the present invention. Detailed Embodiments

[0017] To make the purpose, technical solution and advantages of the present invention clearer, the following will detail the technical solution implemented based on software in combination with the system architecture diagram and embodiments. It should be understood that the specific embodiments described here are only used to explain the technical solution of the present invention and do not constitute a limitation on the protection scope.

[0018] In the description of the present invention, the system architecture relationships or data processing flows indicated by terms such as "hierarchy", "module", "interface", "data stream", "client", "server", etc. are all defined based on the architecture diagram or flowchart corresponding to the embodiments. This expression method is only used to clearly explain the logical relationships of the elements in the technical solution, rather than limiting the physical deployment form. The "multiple" includes two or more technical units, including but not limited to expandable elements such as multiple data nodes, processing threads, service instances or functional components. The specific quantity needs to be determined according to the actual business scenario and is subject to special instructions.

[0019] Please refer to Figure 1 and Figure 2, the present invention provides a technical solution: A power grid fault prediction system integrating digital twin includes: A state perception module, which is used to obtain the strain gradient of the steel structure of the transmission tower through a distributed optical fiber sensor, collect the displacement trajectory of the transmission tower base through a Beidou positioning unit, calculate the angular velocity based on a six-axis inertial unit, execute a spatio-temporal alignment algorithm on multi-source data, combine it into a structure state tensor, and transfer the structure state tensor to the path deduction module; The structure state tensor includes strain gradient distribution, displacement trajectory coordinates, and angular velocity parameters; The spatio-temporal alignment algorithm adopts the dynamic time warping method to unify the sampling frequencies of the strain gradient, displacement trajectory, and angular velocity parameters to 100 Hz; 100 Hz represents the sampling frequency of collecting 100 data points per second and is set according to the maximum sampling ability of the distributed optical fiber sensor.

[0020] First, on a 220 kV "ZBC-35" type straight tower located in Zhangjiakou City, Hebei Province, along the windward and leeward sides of its four main material legs, starting from 5 meters above the tower foot, distributed optical fiber sensors are arranged vertically every 10 meters until 55 meters above the tower top. At each arrangement point, 4 sensing units are evenly installed along the circumferential direction of the rod, and a total of sensing units are installed on the tower body. When the timestamp is , the strain collected by the sensor at point A (located on the windward side) at a tower height of 15 meters is 152.3 microstrains ( ), while the strain collected by the sensor at point B, which is 0.2 meters away from it along the axial direction of the rod, is 153.5 . At this time, the strain gradient value at this position is obtained through the operation of , and the result is 6.0 . This calculation is synchronously applied to all adjacent pairs of sensing units to form a real-time strain gradient distribution matrix covering the entire tower. The original data acquisition frequency of this distributed optical fiber sensor is 50 Hz.

[0021] At the same time, a monitoring terminal integrating a Beidou high-precision positioning module and a six-axis inertial measurement unit (IMU) is installed at the center position of the tower top. The Beidou positioning unit continuously collects the three-dimensional geographical coordinates of the tower top at a frequency of 10 Hz. For example, at , the collected coordinates are (114.88251° E, 40.75162° N, altitude 1535.62 m), and at , the coordinates are (114.88252° E, 40.75161° N, altitude 1535.61 m). This series of continuous coordinate points constitutes the displacement trajectory of the tower top.

[0022] The six-axis IMU collects the three-axis acceleration and three-axis angular velocity at the top of the tower at a frequency of 200 Hz. The gyroscopes inside the IMU directly output the rotation rates around its internal x, y, and z axes. At the output angular velocity vector is (0.03 rad / s, -0.01 rad / s, 0.08 rad / s). This data reflects the real-time torsion and swing rates of the transmission tower under wind loads.

[0023] After obtaining the three groups of heterogeneous data of strain gradient, displacement trajectory, and angular velocity, spatio-temporal alignment processing is performed. The goal is to unify the sampling frequencies of all data streams to 100 Hz. The setting of this frequency is based on the following experiment: A Q235 steel component with the same material as the transmission tower is installed on a vibration table, and distributed fiber optic sensors are pasted on the component. A sine sweep excitation with a frequency ranging from 1 Hz to 150 Hz is applied through the vibration table, and a laser Doppler vibrometer is used as the true value reference. The response signals of the fiber optic sensors and the vibrometer signals are recorded and compared.

[0024] Table 1: Data table of the sampling frequency verification experiment of fiber optic sensors

[0025] As shown in Table 1, the experimental data shows that when the excitation frequency exceeds 100 Hz, the signal-to-noise ratio of the sensor signal is lower than 40 dB, and the signal distortion degree increases sharply. Therefore, 100 Hz is set as the maximum effective sampling frequency. The alignment process is as follows: A reference time axis with an interval of 0.01 seconds (i.e., 100 Hz) is established. For the strain data with an original frequency of 50 Hz, between two original data points, the linear interpolation method is used to calculate the new data at the reference time points. For example, the strain is 152.3 , the strain is 152.9 , then the interpolated strain at is . For the displacement data with a frequency of 10 Hz, the same method is used for interpolation to generate 100 Hz data points. For the angular velocity data with a frequency of 200 Hz, at each 100 Hz reference time point, the two original data points before and after it are taken (for example, take the data of and ), and it is downsampled to 100 Hz by taking the arithmetic mean. For example, the angular velocity at is 0.08 rad / s, the angular velocity at is 0.09 rad / s, then the angular velocity value at

[0026] Finally, at each 0.01-second time stamp, the aligned strain gradient distribution matrix, a three-dimensional displacement trajectory coordinate vector, and a three-dimensional angular velocity parameter vector are combined into a structural state tensor. This tensor at The data slice instance is: {strain gradient matrix [96x1], displacement coordinates (114.882515°, 40.751615°, 1535.615 m), angular velocity (0.035 rad / s, -0.012 rad / s, 0.085 rad / s)}. This structural state tensor is passed as a whole to the path deduction module.

[0027] Please refer to Figure 1 and Figure 3 for the path deduction module, which is used to receive the structural state tensor, perform the Laplace operation on the strain gradient, input the displacement trajectory coordinates into the three-dimensional velocity field correction model, construct the stress wave propagation path in combination with the angular velocity parameters, and solve the wave equation to generate the load transfer path map; where represents the displacement vector of the steel structure of the transmission tower at the spatial position at the moment in meters, represents the time variable in seconds, represents the corrected longitudinal wave propagation velocity in m / s, represents the Laplace operator with the dimension of m -2 , represents the three-dimensional space coordinates in meters, represents the external excitation force function in N / m 3 ; Extract the non-rigid deformation components, generate the load transfer path map, and transfer the load transfer path map to the twin mapping module; The load transfer path map specifically includes the stress wave attenuation path corrected by the displacement trajectory, the non-rigid deformation region identified based on the angular velocity parameters, and the main load conduction direction obtained through strain gradient analysis; The stress wave propagation model uses the formula: ; where represents the longitudinal wave propagation velocity of the stress wave in the steel structure of the transmission tower in m / s, represents the reference Young's modulus of the steel structure measured by the distributed optical fiber sensor in GPa, represents the Poisson's ratio dimensionless parameter of the transmission tower steel, represents the steel density in kg / m 3 ; The stress wave propagation model innovatively introduces a displacement trajectory feedback mechanism to correct the velocity field parameters in real time through Beidou positioning data. The specific implementation is as follows: where is the displacement influence factor, is the base seat displacement, is the maximum allowable displacement threshold; When calculating the stress wave propagation speed, the strain gradient distribution is used as the initial input parameter, and the velocity field is corrected in combination with the displacement trajectory coordinates; When extracting the non-rigid deformation component, the curvature change threshold is set to 0.15 rad / m 2 ; 0.15 rad / m 2 represents the critical determination value of the surface curvature change of the transmission tower structure and is obtained by calibrating through the yield test of Q235 steel.

[0028] After the path deduction module receives the structure state tensor transmitted by the state perception module, it first performs the Laplace operation on the strain gradient distribution data therein. The specific execution method of this operation on the discrete data field is as follows: For a certain sensing unit on the tower structure , its strain gradient value is , and the strain gradient values of the four adjacent sensing units on the structure are respectively , , , , and the physical distance between adjacent units is . Then the Laplace value of the point is . Taking a certain node at 15 meters of the tower body as an example, its strain gradient value is 6.0 , and the gradient values of its four adjacent nodes above, below, left, and right are 5.8, 6.4, 6.1, and 5.9 respectively , and the adjacent physical distance is 0.2 m, then the Laplace value of this node is . This operation traverses all nodes to generate a Laplace value distribution map, and the magnitude of its value reflects the local curvature of the strain field. The area with a higher value is marked as the stress concentration area.

[0029] Subsequently, the above operation results are input into the stress wave propagation model to calculate the propagation speed of the stress wave in the tower structure. The stress wave propagation model uses the formula: . This formula is used to calculate the propagation speed of longitudinal waves (P-waves) in an ideal elastic medium. The logic of the formula is that the wave speed is jointly determined by the elasticity (ability to resist deformation) and inertia (density) of the material. Among them, represents the longitudinal wave propagation speed of the stress wave in the steel structure of the transmission tower, with the unit of m / s; is the reference Young's modulus of the steel structure measured by a distributed fiber optic sensor, with the unit of GPa; is the Poisson's ratio of the steel for the transmission tower, which is a dimensionless parameter; is the steel density, with the unit of kg / m 3 . The numerator term represents the longitudinal stiffness of the material under constrained conditions, and the denominator term incorporates the density of the material and the influence of the transverse deformation caused by the Poisson effect. The square root operation converts the ratio of stiffness to density into a velocity dimension. The advantage of the formula is that by using the reference Young's modulus obtained from on-site measurement to replace the nominal Young's modulus of the material when it leaves the factory, the calculation result is closer to the actual service state of the transmission tower.

[0030] The assignment process of each parameter in the formula is as follows: (Steel density): Directly adopt the national standard density value of Q235 steel used for the transmission tower, = 7850 kg / m 3 . (Poisson's ratio): Obtained by querying the Q235 steel manual, dimensionless parameter = 0.28. (Reference Young's modulus): Calibrated during the initial deployment stage of the system. At a point 5.0 meters away from the fiber optic sensor A on a main member at the tower foot, apply an instantaneous impact load with a force hammer, with an impact force of 10 kN and a duration of 1 ms. Record the exact arrival time of the stress wave peak through the fiber optic sensors at point A and point B which is 10.0 meters away from point A. If the time recorded at point A is , and the time recorded at point B is , then the measured wave velocity is . Substitute into the formula to back-calculate , and obtain the reference Young's modulus of 198.5 GPa. This value is used as the reference for subsequent calculations.

[0031] Substitute the above parameters into the formula for calculation: . The result shows that the theoretical propagation velocity of the stress wave under the current structural state is 5685 m / s.

[0032] Next, correct the velocity field in combination with the displacement trajectory coordinates. According to the Beidou data, within the time period from to , the displacement of the tower top in the vertical direction is -0.01 m, then the vertical vibration velocity of the tower top during this time period is (the negative sign indicates downward movement). When the stress wave propagates from the tower top downward, its corrected velocity is .

[0033] When extracting non-rigid deformation components, a curvature change threshold needs to be set. The threshold is 0.15 rad / m 2 Calibrated through the following experiment: Select 20 standard specimens of Q235 steel. Use an MTS universal material testing machine to conduct unidirectional tension according to the GB / T 228.1-2021 standard. During the test, use three-dimensional digital image correlation (3D-DIC) technology to monitor and calculate the curvature change on the surface of the specimen in real time. Record the critical curvature value of each specimen at the moment when it enters the plastic yield stage from the elastic stage. After removing the highest and lowest values from the 20 groups of data and taking the average, 0.151 rad / m is obtained 2 , with a standard deviation of 0.009 rad / m 2 . Therefore, the engineering threshold is set to 0.15 rad / m 2 . In the system, convert the numerical value of the stress concentration area obtained from the aforementioned Laplace operation into an equivalent surface curvature. Any area with a curvature value exceeding 0.15 rad / m 2 is determined to have undergone non-rigid deformation

[0034] Finally, based on the above analysis results: According to the difference between the corrected wave velocity and the measured wave velocity, draw the attenuation path of the stress wave in the tower structure; Mark all non-rigid deformation areas with excessive curvature in the three-dimensional model; Determine the conduction direction of the main load according to the path with the highest strain gradient value and the smallest attenuation. These three together constitute the load transfer path map and are transmitted to the twin mapping module

[0035] Please refer to Figure 1 and Figure 4 , the twin mapping module is used to receive the load transfer path map, establish a finite element mesh model, input the stress wave parameters into the damage accumulation model, correct the Young's modulus of the mesh, generate a structural degradation coefficient matrix, and transmit the structural degradation coefficient matrix to the prediction execution module The structural degradation coefficient matrix specifically refers to material fatigue degree, modulus correction value, and mesh element degradation index The damage accumulation model uses: ; Among them, represents the corrected Young's modulus of the mesh element and the unit is GPa, represents the initial Young's modulus nominal at the time of steel factory shipment and the unit is GPa, represents the dimensionless damage factor generated in the q-th load cycle, is the load cycle serial number, represents the total number of load cycles, Through Calculated and obtained, represents the amplitude attenuation rate of the stress wave on the propagation path in the q-th cycle and the calculation formula is (peak amplitude - trough amplitude) / peak amplitude It represents the energy dissipation rate in the main load conduction direction of the q-th cycle, and the calculation formula is 1 - (output energy / input energy); The element size of the finite element mesh model is adaptively adjusted according to the main load conduction direction; A 5mm×5mm mesh is adopted in the area with intensive load conduction, and a 10mm×10mm mesh is adopted in the area with sparse conduction; 5mm represents the side length dimension of the square mesh element and is applicable to the area where the stress value in the main load conduction direction exceeds 50MPa, and 10mm represents the side length dimension of the square mesh element and is applicable to the area where the stress value in the main load conduction direction is lower than 50MPa; The update period of the structural deterioration coefficient matrix is set to 30 minutes; When the deterioration index of the mesh element exceeds 0.7, the real-time warning mechanism of the prediction execution module is triggered; 30 minutes represents the minimum time interval for the system to update the structural deterioration state, and 0.7 represents the critical threshold of the safety factor of the mesh element and is determined according to the failure critical value of the 85% confidence interval in the tensile test of metallic materials - Part 1: Tensile test at room temperature (GB / T 228.1 - 2021).

[0036] After receiving the load transfer path map, the twin mapping module immediately starts to establish the finite element mesh model. This process is adaptive: first, the reference 3D CAD model of the transmission tower is retrieved, and then mesh division is performed according to the load transfer path map. For the areas marked as the main load conduction direction in the map, such as the windward side of the tower leg and the key cross-arm connection, the stress value after conversion of the strain monitoring value (for example, through stress = strain , ) If it is lower than 50MPa, a mesh size of 10mm×10mm is adopted; if the stress value of 65MPa is monitored at another place, exceeding the threshold of 50MPa, the mesh is automatically refined in this area, and a unit size of 5mm×5mm is adopted. The setting of this threshold of 50MPa is based on the fatigue S - N curve of Q235 steel. This value is in the transition zone between high - cycle fatigue and low - cycle fatigue. Stress cycles below this value usually do not cause significant damage accumulation in the short term, while stresses above this value require more refined calculations to capture their damage effects.

[0037] Next, the stress wave parameters are input into the damage accumulation model to correct the Young's modulus of the mesh element. The damage accumulation model is: . The logic of this formula is to dynamically adjust its mechanical property parameters by simulating the fatigue damage process of the material under cyclic loads. Among them, represents the corrected Young's modulus of the mesh element, in GPa; is the initial Young's modulus nominal at the time of steel factory shipment, in GPa; is the dimensionless damage factor generated by the qth load cycle, q is the load cycle number, and n is the total number of load cycles. Represents the total damage caused by all load cycles within a period of time (30 minutes in this example). This factor directly reflects the degree of degradation of material properties. By multiplying this degradation factor with the initial modulus, we can get the corrected Young's modulus that is closer to the actual state. . Damage Factor pass Calculated, where represents the amplitude attenuation rate of the stress wave on the propagation path in the qth period, represents the energy dissipation rate in the direction of the main load conduction in the qth cycle. The benefit of the formula is that it converts the damage factor and two physical quantities that can be directly monitored: stress wave amplitude attenuation rate and energy dissipation rate The correlation enables damage assessment to be driven directly by real-time sensor data.

[0038] The parameter assignment and calculation process in the formula are as follows: (Initial Young's modulus): The nominal value of Q235 steel on the factory certificate is used, set to 206GPa. In a 30-minute update cycle, the system uses the rainflow counting method to identify the strain data of a key node. A complete load-unload cycle. Take a loop as an example to calculate: (Amplitude decay rate): In this cycle, the stress wave propagates from upstream point A to downstream point B. The peak value of the stress wave monitored at point A is 70MPa and the trough is -60MPa; the peak value monitored at point B is 65MPa and the trough is -55MPa. The calculation calls the peak and trough amplitudes of point A, =(70MPa-65MPa) / 70MPa=0.0714. (Energy dissipation rate): The stress wave energy is proportional to the square of the stress amplitude. The input energy is proportional to , the output energy is proportional to . = . (Damage factor for a single load cycle): Assuming that in 150 cycles, the calculated damage factor The average value of is 0.00009, so the total damage accumulation . (Corrected Young's modulus): This result indicates that after 30 minutes of load cycling, the Young's modulus of this grid element has decreased from 206 GPa to 203.22 GPa.

[0039] The above calculations are applied to each grid element in the finite element model. Subsequently, the system generates a structural degradation coefficient matrix, and the update period of this matrix is 30 minutes. For the grid element calculated above, the corresponding degradation coefficient entries are: {material fatigue degree: 0.0135, modulus correction value: 203.22 GPa, grid element degradation index: 0.0135}. Among them, the grid element degradation index is calculated by Calculated.

[0040] The system will continuously monitor the degradation indices of all grid elements. When the degradation index of any element exceeds 0.7, an early warning is triggered. The determination of this threshold of 0.7 is based on the following experimental analysis: Table 2: Association Table between Degradation Index and Failure State in Tensile Test of Q235 Steel

[0041] As shown in Table 2, the Q235 steel is subjected to a tensile test in accordance with the GB / T 228.1-2021 standard, and the ratio of its tangent modulus to the initial modulus is calculated in real time to obtain the degradation index. The experimental results show that when the degradation index reaches 0.7, the material has entered the necking stage, and its load-bearing capacity is about to or has already started to decline. At this time, the lower limit of the 85% confidence interval of the structural failure probability has exceeded the critical value. Therefore, 0.7 is set as the critical threshold for early warning. In this example, the calculated degradation index of 0.0135 is much lower than 0.7. The structural degradation coefficient matrix is transmitted to the prediction execution module as a whole.

[0042] Please refer to Figure 1 and Figure 5 , the prediction execution module is used to receive the structural degradation coefficient matrix, perform time series differentiation on the degradation coefficients, input the sequence into a long short-term memory network, perform the Kolmogorov test, generate a fault spatio-temporal coordinate encoding, and enter the fault spatio-temporal coordinate encoding into the power grid dispatching decision-making; The fault spatio-temporal coordinate encoding includes a prediction period sequence, a confidence interval threshold, and a fault geographical coordinate; The long short-term memory network contains = 128 hidden units, and its gating mechanism uses the sigmoid activation function; The significance level of the Kolmogorov test is set to 0.05, which is used to verify the distribution consistency between the prediction period sequence in the fault spatio-temporal coordinate encoding and the historical fault data; denotes the number of neurons in the hidden layer and is determined by optimizing on the training set using the gradient descent method, and 0.05 represents the upper limit of the type I error probability allowed in the statistical test.

[0043] After the prediction execution module receives the structure deterioration coefficient matrix updated every 30 minutes, it performs a temporal difference operation on the key indicator therein, namely the "grid cell deterioration index". This operation aims to shift from the absolute value of the deterioration state to focusing on the rate and acceleration of its change. Taking the grid cells in a certain high-stress area as an example, the recorded deterioration indices for three consecutive cycles are: , , . The temporal difference calculation process is to calculate the difference between adjacent cycles: Difference 1 = 0.0135 - 0.0130 = 0.0005; Difference 2 = 0.0142 - 0.0135 = 0.0007. Thus, a time series [0.0005, 0.0007] representing the deterioration rate is obtained.

[0044] This differenced sequence is input into a pre-trained long short-term memory network (LSTM). The network structure contains 128 hidden units ( ). The determination process of the number of hidden units is as follows: In the model development stage, historical monitoring data is used as the training set, and LSTM models with 32, 64, 128, and 256 hidden units are respectively constructed. These models are evaluated on an independent validation set, and the evaluation metric is the root mean square (RMSE) of the prediction error. Tests found that the RMSE of the model with 128 hidden units on the validation set is 0.00008, while the RMSE of the 64-unit model is 0.00011, and the RMSE of the 256-unit model is 0.000085, but its training time increases significantly. Therefore, 128 is selected as the number of units that balances both accuracy and efficiency. The gating mechanisms of the network, such as the input gate, forget gate, and output gate, all adopt the Sigmoid activation function, which maps the internal calculated value to between 0 and 1.

[0045] Currently, the sequence [0.0005, 0.0007] is fed into the LSTM network. The network predicts the future deterioration rate based on the historical fault evolution patterns it has learned. The predicted rates for the next three cycles (90 minutes) output by the network are [0.0010, 0.0015, 0.0022]. By accumulating these predicted rates, the future deterioration indices can be deduced: ; ; . This sequence of [0.0152, 0.0167, 0.0189] is the "prediction cycle sequence".

[0046] Subsequently, a Kolmogorov-Smirnov test is performed on the prediction results to verify their distribution consistency with the true failure modes. The significance level is set to 0.05. This value setting is a common standard in statistics and engineering practice, meaning tolerating a 5% probability of misjudging a prediction that is actually consistent with historical failure modes as inconsistent (Type I error). When performing the test, the predicted cycle sequence output by the LSTM is used as sample one, and the degradation evolution data of 10 cases that are similar to the current working conditions (such as wind speed, temperature) and ultimately lead to failures are retrieved from the historical database as sample two. A p-value is calculated through the test. If the p-value is 0.42, since , the null hypothesis is accepted, that is, it is considered that there is no significant difference in the statistical distribution between the currently predicted degradation trend and the true failure evolution trend in history.

[0047] Finally, the system integrates the above information to generate a failure spatio-temporal coordinate encoding. This encoding is a structured data object, and the specific content is: {predicted cycle sequence: [0.0152, 0.0167, 0.0189], confidence interval threshold: 0.95, fault geographical coordinates: {transmission tower number: 'ZBC-35', structural part:'main material leg on the south side, height 25m', geographical coordinates: (114.88251° E, 40.75162° N)}}. This encoding is immediately generated and entered into the decision support system of the power grid dispatching center in the standard data interface format.

[0048] The power grid fault prediction method integrating digital twin includes the following steps: S1: Obtain the strain gradient of the transmission tower steel structure through a distributed optical fiber sensor, collect the displacement trajectory of the transmission tower base through a Beidou positioning unit, calculate the angular velocity based on a six-axis inertial unit, and perform a spatio-temporal alignment algorithm on the multi-source data to combine it into a structural state tensor; S2: Perform a Laplace operation on the strain gradient in the structural state tensor, input the result into the stress wave propagation model, extract the non-rigid deformation component, and generate a load transfer path map; Establishing the stress wave propagation model includes: S201. Calculate the reference propagation speed according to the material property parameters ; S202. Combine the displacement trajectory coordinates to construct a three-dimensional velocity field distribution; S203. Use the finite difference method to solve the partial differential equation of stress wave propagation; S3: Establish a finite element mesh model based on the load transfer path map, input the stress wave parameters into the damage accumulation model, correct the Young's modulus of the mesh, and generate a structural degradation coefficient matrix; The correction of Young's modulus includes: S301. Calculate the cumulative damage factor of each mesh element ; S302. Perform modulus reduction according to the damage factor; S303. Update the stiffness matrix of the finite element model; S4: Perform time series difference on the deterioration coefficients in the structural deterioration coefficient matrix, input the sequence into the long short-term memory network, perform the Kolmogorov test, generate the fault space-time coordinate encoding and record it into the power grid dispatching decision-making.

[0049] The above embodiments show the preferred implementation manners of the present invention. Any equivalent adjustment of the technical solution based on the software engineering method falls within the protection scope, including but not limited to: implementing the algorithm logic in different programming languages, performing service-oriented reconstruction on the functional modules, adjusting the data interaction protocol, optimizing the resource scheduling strategy and other technical improvements. Any implementation scheme derived from reasonable modifications of the data processing flow, service call link or system architecture level without departing from the technical core of the present invention shall be regarded as within the protection scope defined by the claims of the present invention.

Claims

1. A power grid fault prediction system integrating digital twins, characterized in that, The system includes: A state perception module, which is used to obtain the strain gradient of the steel structure of the transmission tower through a distributed optical fiber sensor, collect the displacement trajectory of the transmission tower base through a Beidou positioning unit, calculate the angular velocity based on a six-axis inertial unit, execute a spatio-temporal alignment algorithm on multi-source data, combine them into a structure state tensor, and transfer the structure state tensor to the path deduction module; A path deduction module, which is used to receive the structure state tensor, execute a Laplace operation on the strain gradient, input the result into a stress wave propagation model, extract non-rigid deformation components, generate a load transfer path map, and transfer the load transfer path map to the twin mapping module; A twin mapping module, which is used to receive the load transfer path map, establish a finite element mesh model, input stress wave parameters into a damage accumulation model, correct the Young's modulus of the mesh, generate a structure deterioration coefficient matrix, and transfer the structure deterioration coefficient matrix to the prediction execution module; A prediction execution module, which is used to receive the structure deterioration coefficient matrix, perform a time series difference on the deterioration coefficient, input the sequence into a long short-term memory network, execute a Kolmogorov test, generate a fault spatio-temporal coordinate code, and enter the fault spatio-temporal coordinate code into the power grid dispatching decision.

2. The power grid fault prediction system integrating digital twins according to claim 1, wherein The structure state tensor includes strain gradient distribution, displacement trajectory coordinates, and angular velocity parameters. The load transfer path map is specifically the stress wave attenuation path corrected by the displacement trajectory, the non-rigid deformation region identified based on the angular velocity parameters, and the main load conduction direction obtained through strain gradient analysis. The structure deterioration coefficient matrix specifically refers to material fatigue degree, modulus correction value, and mesh unit deterioration index. The fault spatio-temporal coordinate code includes a prediction period sequence, a confidence interval threshold, and a fault geographical coordinate.

3. The power grid fault prediction system integrating digital twin according to claim 2, wherein When the path deduction module executes the stress wave propagation model, the formula is used: ； wherein, represents the longitudinal wave propagation velocity of stress wave in the steel structure of transmission tower, with the unit of m / s, represents the measured reference Young's modulus of steel structure by distributed optical fiber sensor, with the unit of GPa, represents the Poisson's ratio dimensionless parameter of transmission tower steel, represents the steel density, with the unit of kg / m 3 ; When calculating the stress wave propagation speed, the strain gradient distribution is used as the initial input parameter, and the velocity field is corrected in combination with the displacement trajectory coordinates.

4. The power grid fault prediction system integrating digital twins according to claim 3, wherein, When the twin mapping module corrects the Young's modulus of the mesh, the damage accumulation model is used: ； Among them, represents the Young's modulus of the corrected grid element in GPa, represents the initial Young's modulus nominal at the steel factory in GPa, represents the dimensionless damage factor generated in the q-th load cycle, is the load cycle serial number, represents the total number of load cycles, through calculated and obtained, represents the amplitude attenuation rate of the stress wave in the propagation path during the q-th cycle, and the calculation formula is (peak amplitude - trough amplitude) / peak amplitude, represents the energy dissipation rate in the main load conduction direction during the q-th cycle, and the calculation formula is 1 - (output energy / input energy).

5. The power grid fault prediction system integrating digital twins according to claim 4, wherein The long short-term memory network in the prediction execution module includes = 128 hidden units, and its gating mechanism uses the sigmoid activation function; The significance level of the Kolmogorov test is set to 0.05, which is used to verify the distribution consistency between the prediction period sequence in the fault spatio-temporal coordinate code and the historical fault data; Among them, represents the number of neurons in the hidden layer and is determined by optimizing on the training set through the gradient descent method, and 0.05 represents the upper limit of the probability of the first type of error allowed in the statistical test.

6. The power grid fault prediction system integrating digital twins according to claim 5, characterized in that, The spatio-temporal alignment algorithm uses the dynamic time warping method to unify the sampling frequencies of the strain gradient, displacement trajectory, and angular velocity parameters to 100 Hz; When extracting the non-rigid deformation component, the curvature change threshold is set to 0.15 rad / m 2 ; Among them, 100Hz represents the sampling frequency of collecting 100 data points per second and is set according to the maximum sampling capacity of the distributed optical fiber sensor, and 0.15rad / m 2 represents the critical determination value of the surface curvature change of the transmission tower structure and is obtained by calibration through the yield test of Q235 steel.

7. The power grid fault prediction system integrating digital twins according to claim 6, wherein The element size of the finite element mesh model is adaptively adjusted according to the main load conduction direction; A 5mm×5mm mesh is used in the load conduction dense area, and a 10mm×10mm mesh is used in the conduction sparse area; Among them, 5mm represents the side length dimension of the square mesh unit and is applicable to the area where the stress value in the main load conduction direction exceeds 50 MPa, and 10mm represents the side length dimension of the square mesh unit and is applicable to the area where the stress value in the main load conduction direction is lower than 50 MPa.

8. The power grid fault prediction system integrating digital twins according to claim 7, wherein The update period of the structure deterioration coefficient matrix is set to 30 minutes; When the mesh unit deterioration index exceeds 0.7, the real-time warning mechanism of the prediction execution module is triggered; Among them, 30 minutes represents the minimum time interval for the system to update the structural degradation state, and 0.7 represents the critical threshold of the safety factor of the grid cell and is determined according to the failure critical value of the 85% confidence interval in the tensile test of metallic materials in GB / T 228.1-2021.

9. A power grid fault prediction method integrating digital twins, characterized in that, The method is used to implement the power grid fault prediction system integrating digital twin described in any one of claims 1-8, and includes the following steps: S1: Obtain the strain gradient of the steel structure of the transmission tower through a distributed optical fiber sensor, collect the displacement trajectory of the transmission tower base through a Beidou positioning unit, calculate the angular velocity based on a six-axis inertial unit, perform a spatio-temporal alignment algorithm on the multi-source data, and combine it into a structural state tensor; S2: Perform a Laplace operation on the strain gradient in the structural state tensor, input the result into the stress wave propagation model, extract the non-rigid deformation component, and generate a load transfer path map; S3: Establish a finite element mesh model based on the load transfer path map, input the stress wave parameters into the damage accumulation model, correct the Young's modulus of the mesh, and generate a structural degradation coefficient matrix; S4: Perform a time series difference on the degradation coefficients in the structural degradation coefficient matrix, input the sequence into a long short-term memory network, perform a Kolmogorov test, generate a fault spatio-temporal coordinate code, and record it into the power grid dispatching decision-making.

10. The power grid fault prediction method integrating digital twin according to claim 9, characterized in that The establishment of the stress wave propagation model in S2 includes: S201. Calculating the reference propagation speed according to the material property parameters ; S202: Construct a three-dimensional velocity field distribution in combination with the displacement trajectory coordinates; S203: Solve the partial differential equation of stress wave propagation by the finite difference method; The Young's modulus correction in S3 includes: S301, calculating the cumulative damage factor of each grid cell ; S302: Perform modulus reduction according to the damage factor; S303: Update the stiffness matrix of the finite element model.

Images