A method and system for abnormal on-line monitoring of a cement pole grounding resistance

Abstract

The application provides a cement pole grounding resistance abnormality online monitoring method and system, comprising: collecting a cement pole excitation current, an induced voltage, a pole body temperature and an environmental humidity, solving an instantaneous grounding resistance by using the excitation current and the induced voltage in a time window; combining the excitation current and the instantaneous grounding resistance to do phase space reconstruction to obtain a two-dimensional phase diagram, and generating a characteristic grid atlas by quantization after gridding; constructing a state reference surface of a temperature and humidity coordinate axis based on historical data, storing a standard atlas; calculating a difference between a current atlas and a corresponding temperature and humidity standard atlas matrix, and then multiplying the difference with a diagnosis focusing matrix element by element to generate a weighted difference grid matrix; establishing a cause knowledge base, storing a cause prototype containing a basic cause vector and a cause space mask; converting the weighted difference grid matrix into a deviation state vector, calculating a projection similarity and non-negativizing to obtain a cause correlation degree sequence, identifying an abnormal cause, and combining a mask to update a next period diagnosis focusing matrix.

Description

Technical Field

[0001] This application belongs to the field of online monitoring, and in particular relates to an online monitoring method and system for abnormal grounding resistance of cement poles. Background Technology

[0002] The grounding resistance of cement poles is influenced by a combination of external factors, including soil structure, pole material degradation, and ambient temperature and humidity. Changes in ambient temperature and humidity, in particular, cause fluctuations in soil resistivity and pole conductivity, leading to variations in monitored grounding resistance values. When abnormal fluctuations occur in grounding resistance, existing data comparison mechanisms can only determine if the resistance exceeds limits, failing to remove overlapping characteristics caused by ambient temperature and humidity interference and substantial degradation. This results in early, subtle resistance distortions caused by different factors being easily masked by conventional environmental fluctuation noise, causing the tracing of the abnormal state's cause to lag behind the actual development of the damage. Current data processing models typically rely on static matching of global features, lacking sufficient focus and weight adjustment for specific abnormal evolution directions, leading to blurred diagnostic boundaries when multiple abnormal causes coexist or alternate. Therefore, extracting the correlation evolution patterns encompassing environmental parameters and electrical characteristics from multidimensional time-series features, constructing a monitoring mechanism capable of focusing on abnormal feature regions and quantifying the identification of different degradation causes, tracking the evolution trajectory of intermediate grounding states, and determining the causes of anomalies are problems that need to be solved in the current field of power facility operation and maintenance. Summary of the Invention

[0003] To improve the accuracy of long-term online monitoring of grounding resistance of cement poles and the reliability of cause diagnosis, this invention provides a method and system for online monitoring of abnormal grounding resistance of cement poles.

[0004] According to one aspect of the present invention, a method for online monitoring of abnormal grounding resistance of cement poles is provided, the method comprising: The excitation current, induced voltage, pole temperature, and ambient humidity of the cement pole are collected. The instantaneous grounding resistance is calculated based on the excitation current and induced voltage within the time window. A two-dimensional phase diagram is generated by reconstructing the phase space based on the excitation current and instantaneous grounding resistance, and the characteristic grid map is obtained by gridding and quantization processing. Based on historical operating data, a state reference surface is constructed with temperature and humidity as coordinate axes. The state reference surface stores standard feature raster maps. The current feature raster map is matrix subtracted from the standard feature raster map corresponding to the current temperature and humidity in the state reference surface. The subtraction result is multiplied element by element with the diagnostic focusing matrix to generate a weighted difference raster matrix. A causal knowledge base is constructed, which stores multiple causal templates containing basic causal vectors and causal space masks. The weighted difference raster matrix is ​​converted into an offset state vector. The projection similarity of the offset state vector on each basic causal vector is calculated and nonnegated to obtain a causal correlation degree sequence. The abnormal cause is determined based on the causal correlation degree sequence. The causal space mask corresponding to the causal correlation degree sequence is combined and added to a preset small positive basis matrix. After normalization, a diagnostic focusing matrix for the next period is generated.

[0005] According to another aspect of the present invention, an online monitoring system for abnormal grounding resistance of cement poles is provided, the system comprising the following modules: The reconstruction module is used to collect the excitation current, induced voltage, pole temperature and ambient humidity of the cement pole, calculate the instantaneous grounding resistance based on the excitation current and induced voltage within the time window, generate a two-dimensional phase diagram through phase space reconstruction based on the excitation current and instantaneous grounding resistance, and obtain the characteristic grid map through gridding and quantization processing. The generation module is used to construct a state reference surface with temperature and humidity as coordinate axes based on historical operating data. The state reference surface stores standard feature raster maps. The current feature raster map is matrix subtracted from the standard feature raster map corresponding to the current temperature and humidity in the state reference surface. The subtraction result is multiplied element by element with the diagnostic focusing matrix to generate a weighted difference raster matrix. The update module is used to construct a causal knowledge base, which stores multiple causal templates containing basic causal vectors and causal space masks. The weighted difference raster matrix is ​​converted into an offset state vector. The projection similarity of the offset state vector on each basic causal vector is calculated and nonnegated to obtain a causal correlation degree sequence. The abnormal cause is determined based on the causal correlation degree sequence. The causal space mask corresponding to the causal correlation degree sequence is combined and added to a preset small positive basis matrix. After normalization, the diagnostic focusing matrix for the next cycle is generated.

[0006] This invention extracts grounding characteristics of cement poles by comprehensively collecting electrical and environmental parameters and generating feature grid maps through phase space reconstruction and gridded quantization. Combining temperature and humidity to construct a state reference surface for matrix subtraction reduces interference from normal resistance fluctuations caused by environmental factors. A weighted differential grid matrix is ​​generated using a diagnostic focusing matrix, suppressing background noise and highlighting key anomaly areas. A causal knowledge base is used to calculate projection similarity and generate a causal correlation sequence, enabling quantitative diagnosis of anomaly causes. By combining causal space masks to update the diagnostic focusing matrix for the next cycle, the accuracy of long-term monitoring of cement pole grounding resistance and the reliability of causal diagnosis are improved. Attached Figure Description

[0007] Figure 1A flowchart for an online monitoring method for abnormal grounding resistance of cement poles; Figure 2 A schematic diagram of the two-dimensional reference surface for temperature and humidity and node interpolation; Figure 3 This is a schematic diagram illustrating the principle of feature raster spectrum difference and focus weighting calculation. Detailed Implementation

[0008] In this application, the terms "exemplarily" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplarily" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplarily" or "for example" is intended to present the relevant concepts in a specific manner.

[0009] This invention proposes an online monitoring method for abnormal grounding resistance of cement poles, such as... Figure 1 As shown, it includes: S1. Collect the excitation current, induced voltage, pole temperature and ambient humidity of the cement pole. Calculate the instantaneous grounding resistance based on the excitation current and induced voltage within the time window. Based on the excitation current and instantaneous grounding resistance, generate a two-dimensional phase diagram through phase space reconstruction, and obtain a characteristic grid map through gridding and quantization processing.

[0010] An excitation injection terminal is installed at the grounding down conductor of the cement pole. The first output terminal of the excitation source is connected to the excitation injection terminal, and the second output terminal of the excitation source is connected to the return auxiliary electrode buried in the soil. This allows the excitation current to sequentially pass through the cement pole grounding down conductor, the grounding electrode under test, the soil, and the return auxiliary electrode to form a closed injection loop. A high-frequency through-hole current transformer is installed on the grounding down conductor to collect the excitation current signal in the closed injection loop. One end of a high-impedance voltage sensor is connected to a test point near the grounding electrode under test, and the other end is connected to a potential reference auxiliary electrode to collect the induced voltage signal of the grounding electrode under test relative to the potential reference auxiliary electrode, wherein the potential reference auxiliary electrode and the return auxiliary electrode are spaced apart. A contact temperature sensor is attached to the shady side of the cement pole to collect the pole temperature. A capacitive humidity sensor is installed around the cement pole to collect the ambient humidity.

[0011] All acquired analog signals are input into a signal conditioning module containing an active low-pass filter circuit for noise reduction and anti-aliasing processing. The signals are then converted into discrete-time digital signals via an analog-to-digital converter at a preset high-frequency sampling rate. Within each monitoring period, the excitation source outputs excitation current according to a preset rated amplitude, preset frequency, and preset waveform, ensuring the comparability of excitation current coordinates across different monitoring periods. When the detected excitation current amplitude deviation does not exceed the preset allowable deviation, the excitation current sequence is proportionally calibrated according to the rated amplitude before participating in phase space reconstruction. When the excitation current amplitude, frequency, or waveform deviation exceeds the preset allowable range, the monitoring period is marked as an excitation abnormal period, and re-acquisition is triggered.

[0012] A sliding time window with a time span encompassing an integer number of power frequency cycles is defined, and the discrete sequences of excitation current and induced voltage within this sliding time window are extracted. Using the phase of the excitation current sequence as a reference, the induced voltage sequence is synchronously detected to extract the non-negative voltage amplitude sequence in phase with the excitation current. For sampling points where the excitation current amplitude is less than or equal to a preset minimum current threshold, no division operation is performed, and the amplitude is filled using interpolation of adjacent valid sampling points or by preserving the previous valid value. For the remaining valid sampling points, the non-negative voltage amplitude is divided by the corresponding excitation current amplitude to calculate the non-negative finite instantaneous grounding resistance sequence.

[0013] The excitation current sequence within the time window is used as the horizontal axis data, and the corresponding instantaneous grounding resistance sequence is used as the vertical axis data. In a two-dimensional Cartesian coordinate system, each pair of current and resistance values ​​at the same instant are mapped to a scatter point in the coordinate system according to the time sampling sequence, thus completing the phase space reconstruction and generating a two-dimensional phase diagram.

[0014] Historical statistical extreme values ​​of excitation current and instantaneous grounding resistance are read to determine the upper and lower boundaries of the horizontal and vertical coordinate axes of the two-dimensional phase diagram. When a trajectory point in the real-time two-dimensional phase diagram exceeds the upper and lower boundaries of the horizontal and vertical coordinate axes, the coordinate values ​​exceeding the upper or lower boundaries are truncated and mapped to the corresponding outermost boundary grid, and the number of out-of-bounds trajectory points and the out-of-bounds ratio are recorded simultaneously. When the out-of-bounds ratio exceeds a preset out-of-bounds ratio threshold, the monitoring period is determined as a coordinate out-of-bounds anomaly period, not used for grounding anomaly cause determination, and data re-acquisition for the monitoring period is triggered. When the out-of-bounds ratio does not exceed the preset out-of-bounds ratio threshold, the truncated and mapped trajectory points are retained to participate in the generation of the feature grid map. The horizontal and vertical coordinate axes of the two-dimensional phase diagram are each divided into a preset M interval, thereby dividing the continuous two-dimensional phase diagram space into a discrete grid of M rows and M columns. The total number of scattered points falling into each independent grid is counted, and the number of scattered points in each grid is divided by the total number of scattered points collected within the sliding time window for numerical normalization. A two-dimensional numerical matrix of size M rows and M columns is generated according to the spatial arrangement of the grid, with all internal elements having values ​​between 0 and 1, which serves as the feature raster map.

[0015] As one possible implementation, the step of generating a two-dimensional phase diagram based on the excitation current and the instantaneous grounding resistance through phase space reconstruction includes: Based on the excitation current sequence and induced voltage sequence within the time window, the induced voltage sequence is synchronously detected to extract the non-negative voltage amplitude sequence in phase with the excitation current. At sampling points where the excitation current amplitude is greater than a preset minimum current threshold, the ratio of the non-negative voltage amplitude to the excitation current amplitude is calculated to obtain the instantaneous grounding resistance sequence. With the excitation current as the abscissa and the instantaneous grounding resistance as the ordinate, the excitation current and instantaneous grounding resistance values ​​at the same extraction time are used to form two-dimensional coordinate points. All times within the time window are traversed in chronological order, and the generated multiple two-dimensional coordinate points are connected sequentially in a two-dimensional plane to obtain the two-dimensional phase diagram with phase space trajectory characteristics.

[0016] With the acquisition time window size set to T=20ms and the sampling frequency set to fs=10kHz, each time window contains N=200 synchronous sampling points. The discrete sequence of the excitation current within the time window is extracted. and induced voltage discrete sequence For the k-th sampling time in the sequence, k iterates from 1 to 200. First, the induced voltage is synchronously detected using the phase of the excitation current sequence as a reference to obtain the in-phase non-negative voltage amplitude. Then read the corresponding excitation current amplitude. ;when When the current exceeds the preset minimum threshold, according to Calculate the instantaneous grounding resistance at this sampling moment; when When the current is less than or equal to the preset minimum current threshold, division is not performed, and the result is padded using interpolation of adjacent valid sampling points or by holding the previous valid value. This refers to the in-phase non-negative voltage amplitude obtained after synchronous detection. For example, the in-phase non-negative voltage amplitude after synchronous detection at the 50th sampling point. =15.0V, excitation current amplitude When the current is 1.5A, calculate the instantaneous grounding resistance at that moment. .

[0017] A two-dimensional Cartesian coordinate system is established with the excitation current as the X-axis (range -5A to +5A) and the instantaneous grounding resistance as the Y-axis (range 0Ω to 50Ω). The instantaneous grounding resistance on the Y-axis is obtained by the ratio of the amplitude of the in-phase non-negative voltage to the amplitude of the excitation current, without directly dividing the signed AC instantaneous voltage by the signed AC instantaneous current. Each pair of extracted values... and Numerical combinations form a spatial point , generate point set Following the sampling time series 1 to 200, adjacent points are... and Connect the beginning and end of a line segment or spline curve.

[0018] Through this continuous connection operation, the originally independent one-dimensional time-domain signal is mapped into a continuous trajectory curve with a definite evolution direction on a two-dimensional plane, thereby generating a two-dimensional phase diagram of the phase space trajectory that reflects the hysteresis characteristics of the grounding resistance.

[0019] As one possible implementation, the gridding and quantization process to obtain a feature raster map includes: Based on the preset excitation current coordinate range and the preset instantaneous grounding resistance coordinate range, the fixed plane region where the two-dimensional phase diagram is located is determined. The plane region where the two-dimensional phase diagram is located is divided into a two-dimensional grid matrix of preset fixed size. The number of trajectory points of the two-dimensional phase diagram contained within the boundary of each single grid is counted as the first basic parameter. The first basic parameter is divided by the total number of trajectory points of the two-dimensional phase diagram in the plane region to calculate the trajectory point distribution probability corresponding to each grid. The trajectory point distribution probability is used as the pixel value to update the spatial state value of the corresponding grid, thereby generating the feature grid map in which the structural features are represented by density distribution.

[0020] Define the fixed boundaries of the current two-dimensional phase diagram plane, setting the X-axis range to [-5A, 5A] and the Y-axis range to [0Ω, 50Ω], and uniformly subdivide it into a two-dimensional grid matrix of size 64×64. The horizontal grid step size is calculated as follows. The vertical grid step size is calculated as follows: .

[0021] Traverse the total N=200 phase diagram trajectory points within the current time window, and determine the grid position corresponding to each trajectory point according to the above boundary truncation rule. For a specific grid cell in the i-th row and j-th column, i,j∈[1,64], count the number of cells falling into the boundary range. The actual number of trajectory points within the range is assigned to the first basic parameter. For example, if after traversal, 12 coordinate points are found to fall within the grid at row 32 and column 32, then it is determined that... .

[0022] After completing the full grid statistics, a normalization operation is performed on each grid cell, and the first basic parameter is... Divide by the total number of trajectory points N within the time window to calculate the probability distribution of trajectory points corresponding to that grid. In the example above, the grid distribution probability... The calculated probability distribution of all trajectory points is directly used as the spatial state value of the corresponding grid, and filled into the matrix according to the original 64×64 row and column structure to generate a feature raster map with internal element values ​​ranging from 0 to 1, which is used for subsequent standard feature raster map construction, matrix difference and causal similarity calculation.

[0023] S2. Based on historical operating data, construct a state reference surface with temperature and humidity as coordinate axes. The state reference surface stores standard feature raster maps. Perform matrix subtraction between the current feature raster map and the standard feature raster map corresponding to the current temperature and humidity in the state reference surface. Multiply the subtraction result element-wise with the diagnostic focusing matrix to generate a weighted difference raster matrix.

[0024] Retrieve historical operational monitoring data of the cement poles, which were confirmed to be healthy and without any abnormalities during the initial stage of operation, for at least one full calendar year. The total temperature and humidity distribution ranges in historical data are divided into multiple discrete temperature and humidity sub-intervals with fixed step sizes. All historical feature raster maps falling within both specific temperature and humidity sub-intervals are selected. When the number of historical feature raster maps within a certain temperature and humidity grid area reaches a preset minimum sample size, the matrix elements at the same row and column positions in these maps are summed and averaged to obtain the average matrix of that temperature and humidity grid area. When the number of historical feature raster maps within a certain temperature and humidity grid area is less than the preset minimum sample size, multiple effective temperature and humidity grid areas adjacent to that area and meeting the sample size requirement are extracted. The standard feature raster maps corresponding to these multiple effective temperature and humidity grid areas are weighted according to the reciprocal of the temperature and humidity distance, and a weighted average is calculated to generate a supplementary standard feature raster map for the insufficient sample area. If adjacent areas still do not meet the supplementary conditions, the average matrix of the normal historical feature raster maps across the entire area is used as a backup standard feature raster map.

[0025] The average value matrix, the completed standard feature raster map, or the spare standard feature raster map are used as the standard feature raster maps for the corresponding temperature and humidity grid regions. A two-dimensional state reference surface is established with the discretized temperature as the horizontal axis and humidity as the vertical axis, and all calculated standard feature raster maps are stored at the center representative point or region index position of the corresponding temperature and humidity grid region.

[0026] When the current pole temperature or ambient humidity exceeds the full temperature and humidity range covered by the state reference surface, the excess portion is truncated to the nearest boundary value of the state reference surface, and the standard feature raster map corresponding to the nearest boundary node or the nearest effective temperature and humidity grid area is extracted as the reference.

[0027] The system acquires the current temperature and humidity data of the pole body. It locates the target temperature and humidity cell within the current actual temperature and humidity area on the state reference surface. It extracts the standard feature raster maps stored on the four adjacent representative points of the temperature and humidity grid surrounding the current temperature and humidity point. Using an interpolation algorithm, the interpolation weights are determined based on the relative positions of the current temperature and humidity within the area enclosed by the four representative points. Spatial interpolation is performed on these four standard feature raster maps to obtain a reference feature raster map matching the current temperature and humidity. If there are missing nodes among the four adjacent representative points of the temperature and humidity grid participating in the interpolation, the completed standard feature raster map is used first. If unusable nodes still exist, distance-weighted interpolation is performed using the nearest valid node to the current temperature and humidity, ensuring that the corresponding reference feature raster map can be obtained under the current temperature and humidity conditions. The principle of the two-dimensional temperature and humidity reference surface and four-node interpolation is as follows: Figure 2 As shown. In one embodiment, after locating the corresponding temperature and humidity grid region based on the current temperature and humidity, the standard feature raster map corresponding to that region is directly extracted and matrix subtraction is performed. In another preferred embodiment, the standard feature raster maps corresponding to the multiple representative points of the temperature and humidity grid closest to the current temperature and humidity are obtained, and interpolation calculation is performed based on the node distance weight to obtain the reference feature raster map that continuously matches the current temperature and humidity. Both of the above methods are used to obtain the standard feature raster map under the current temperature and humidity conditions.

[0028] Subtracting the reference feature raster map matrix from the currently generated feature raster map matrix yields a difference matrix representing the quantized offset of the current grounding state of the cement pole relative to the normal reference state.

[0029] The pre-stored M x M column diagnostic focusing matrix is ​​retrieved, with the initial diagnostic focusing matrix initialized as a constant matrix where all elements are 1. Each element in the difference matrix is ​​multiplied by the corresponding element in the diagnostic focusing matrix. This multiplication amplifies high-risk region features and suppresses irrelevant noise using the diagnostic focusing matrix, resulting in a weighted difference raster matrix of the same dimension (M x M). The complete calculation process for feature raster map difference operations and weighted diagnostic focusing matrix is ​​as follows: Figure 3 As shown.

[0030] As one possible implementation, the construction of a state reference surface based on historical operating data with temperature and humidity as coordinate axes includes: Historical data on pole temperature, ambient humidity, and feature grid patterns are synchronously collected during normal operation of the cement pole when the equipment is undamaged and no alarms occur. The two-dimensional temperature and humidity space is divided into multiple temperature and humidity grid regions according to preset temperature and humidity step sizes. All feature grid patterns falling within the same temperature and humidity grid region are extracted. The corresponding elements of the feature grid patterns are matrix-accumulated, and the average value matrix is ​​calculated. The average value matrix is ​​used as the standard feature grid pattern corresponding to the temperature and humidity grid region. All temperature and humidity grid regions and their corresponding standard feature grid patterns together constitute the state reference surface.

[0031] Obtain historical sensor data of the cement pole under conditions of no damage and zero alarms over the past 12 consecutive months. Each valid data tuple contains... ,in The temperature of the rod at the moment of data collection. For ambient relative humidity, This corresponds to a 64×64 dimension feature raster map matrix. The preset operating environment temperature range is [-20℃, 60℃], and the ambient humidity range is [0%, 100%], with a temperature step size ΔT = 2℃ and a humidity step size ΔH = 5%.

[0032] Following the aforementioned step size setting, the continuous two-dimensional environmental parameter plane is divided into 800 non-overlapping temperature and humidity grid regions. During the reference surface construction phase, for any temperature and humidity grid region numbered (m, n), for example, a temperature range of [20℃, 22℃] and a humidity range of [50%, 55%), all data tuples matching these conditions are queried. Assuming L=50 matching records are found, 50 associated spectral matrices are extracted. to .

[0033] A 64×64 cumulative matrix of all zeros is constructed. Elements at the same position (i,j) in these 50 matrices are summed sequentially to obtain the cumulative value. This cumulative value is then divided by the total sample size L to obtain the average value at that position. If the sum of the 50 spectral elements at a certain position (15,20) in the matrix is ​​2.5, then the average value at that position is set to 0.05. The 64×64 average value matrix obtained in this way serves as the standard feature raster map for that specific environmental interval. All 800 temperature and humidity grid regions and their corresponding standard feature raster maps together constitute the state reference surface.

[0034] As one possible implementation, the step of performing matrix subtraction between the current feature raster map and the standard feature raster map corresponding to the current temperature and humidity in the state reference surface includes: The current pole temperature and ambient humidity are collected on-site and used as positioning parameters. The target temperature and humidity grid region containing the current pole temperature and ambient humidity is found in the state reference surface. The standard feature raster map corresponding to the target temperature and humidity grid region is extracted from the state reference surface. The feature raster map generated at the current time is subtracted from the extracted standard feature raster map element by element at each corresponding position to generate a difference matrix composed of difference data as the result of the matrix difference.

[0035] During each monitoring cycle, environmental condition parameters transmitted from the field are acquired, such as the current instantaneous pole temperature. =21.5℃, ambient humidity =53%. Using these two measured values ​​as positioning parameters, it was determined that 21.5℃ falls within the preset temperature range of [20℃, 22℃), and 53% falls within the preset humidity range of [50%, 55%), thus locating the corresponding target temperature and humidity grid area. .

[0036] After successful location, extract and The corresponding 64×64 dimension standard feature raster map matrix Obtain the real-time feature raster map matrix obtained through synchronous processing at the current moment. Perform element-wise subtraction on these two matrices of the same dimension. The resulting difference matrix is... Then for any row List The element, execute For example, if The real-time observed value of the (10,15)th element is 0.08, while If the baseline mean is 0.05 under the same environment, then the difference is... 0.03; if the current value is 0.02 and the reference value is 0.06, the result retains the sign bit and becomes -0.04. A 64×64 difference matrix containing the spatial global offset data is generated, which is used as the result of the difference of the matrix.

[0037] As one possible implementation, the step of multiplying the difference result element-wise with the diagnostic focusing matrix to generate a weighted difference raster matrix includes: The difference matrix obtained by subtracting the matrix is ​​read. When it is the initial monitoring period, a one matrix with the same dimension as the difference matrix is ​​generated as the diagnostic focus matrix. When it is not the initial monitoring period, the iterative diagnostic focus matrix saved in the previous period is extracted. Each element in the difference matrix is ​​multiplied item by item with the corresponding element in the same row and column position in the diagnostic focus matrix. All product calculation results are saved and reorganized into a new matrix according to the original row and column order to obtain the weighted difference raster matrix that highlights the potential abnormal area.

[0038] The 64×64 dimension difference matrix obtained by reading the matrix and taking the difference The cycle count value t is checked. When t=1, indicating the initial monitoring cycle, a 64×64 matrix with all elements equal to 1.0 is generated and used as the current diagnostic focus matrix F. When t>1, indicating a non-initial monitoring cycle, the iterative diagnostic focus matrix saved in the previous cycle (t-1) is extracted. This is used as the focusing matrix F for the current period.

[0039] Difference matrix Each value in The values ​​F(i,j) at the same row and column coordinates (i,j) in the diagnostic focusing matrix F are multiplied item by item. For example, if the difference at coordinate point (12,12) in the difference matrix is ​​0.05 in a non-initial period, and the weight of the corresponding coordinate in the diagnostic focusing matrix is ​​1.5, then the product is equal to 0.075. All product calculation results are saved, and the data is reconstructed into a new 64×64 matrix according to the original row 1 to 64 and column 1 to 64 layout format, generating a weighted difference raster matrix.

[0040] S3. Construct a causal knowledge base, which stores multiple causal templates containing basic causal vectors and causal space masks. Convert the weighted difference raster matrix into an offset state vector. Calculate the projection similarity of the offset state vector onto each basic causal vector and perform nonnegation processing to obtain a causal correlation degree sequence. Determine the abnormal cause based on the causal correlation degree sequence. Combine the causal space masks corresponding to the causal correlation degree sequence and add them to a preset small positive basis matrix. After normalization processing, generate the diagnostic focusing matrix for the next cycle.

[0041] Beforehand, simulations were performed on typical anomalies of the same type of cement pole, including grounding electrode corrosion and fracture, soil cracking failure, and loose down conductor connections. The difference matrix after each simulated anomaly stabilized was recorded. The difference matrix representing each anomaly was then concatenated row by row from top to bottom and flattened into a one-dimensional column vector of length M squared, which served as the basic cause vector for that type of anomaly. This basic cause vector was not pre-stacked with diagnostic focusing weights and was used for a one-time weight mapping based on the diagnostic focusing matrix of the current cycle during the monitoring phase.

[0042] Based on the grid region in the two-dimensional phase diagram matrix where each anomaly is most likely to produce drastic changes, an M-row, M-column weight distribution matrix with the same dimensions as the feature raster map is defined as a causal space mask. In this mask, matrix elements corresponding to regions of anomalous change are set to high weight values ​​close to 1, while matrix elements in regions with no significant change are set to low weight values ​​close to zero. For example, matrix elements corresponding to regions of anomalous change can be set to high weight values ​​between 0.8 and 1.0, while matrix elements in regions with no significant change can be set to low weight values ​​between 0.0 and 0.2. The basic causal vector of each typical anomaly and its dedicated causal space mask are paired one-to-one, recorded, and persistently stored in the causal knowledge base, resulting in multiple causal templates.

[0043] The M-row, M-column weighted difference raster matrix calculated in real time for the current period is expanded and concatenated row by row, and converted into a one-dimensional column vector of length M squared, defined as the offset state vector. For each basic causal vector extracted from the causal knowledge base, the basic causal vector is first restored to an M-row, M-column basic causal matrix, and then the basic causal matrix is ​​multiplied element-wise at the same position with the diagnostic focus matrix used in the current period to obtain the basic causal matrix to be matched in the current weight space, and converted into the basic causal vector to be matched in the same expansion order; the dot product value of the current offset state vector and the basic causal vector to be matched is calculated. When the L2 norm of the offset state vector or the L2 norm of the basic causal vector to be matched is zero, the corresponding cosine projection similarity value is set to 0; when the L2 norm of both is not zero, the dot product value is divided by the product of the L2 norm of the offset state vector and the L2 norm of the basic causal vector to be matched to obtain the cosine projection similarity value of the two in the multidimensional space. The calculated projection similarity values ​​are processed using a rectification function, retaining all positive values ​​greater than zero and resetting all values ​​less than or equal to zero to zero, thus completing the non-negativity process. All the non-negativity-processed values ​​are then combined sequentially according to the inherent order of anomaly types in the causal knowledge base to form a one-dimensional causal correlation sequence.

[0044] Traverse the causal correlation sequence and extract all elements whose values ​​exceed the set anomaly judgment threshold as a candidate anomaly cause set. When the candidate anomaly cause set contains only one element, the anomaly type corresponding to that element is judged as a single anomaly cause. When the candidate anomaly cause set contains multiple elements, sort them from high to low according to the causal correlation values, take the anomaly type corresponding to the element with the largest value as the dominant anomaly cause, and take the anomaly types corresponding to the other elements that exceed the anomaly judgment threshold as concurrent anomaly causes. When different anomaly types alternately become the dominant anomaly causes in multiple consecutive monitoring periods, and the corresponding causal correlation values ​​all exceed the anomaly judgment threshold, record the judgment results of different anomaly types alternately becoming the dominant anomaly causes.

[0045] Each value in the causal correlation sequence is used as a coefficient and multiplied by the corresponding M-row, M-column causal space mask matrix in the causal template to obtain multiple adjustment matrices scaled according to correlation. Elements at the same position in all adjustment matrices are summed to synthesize a comprehensive adjustment matrix. A smooth basis matrix with all elements being preset small positive numbers is added to the comprehensive adjustment matrix to prevent all-zero dead zones. The maximum and minimum elements in the summed matrix are found, and a non-negative adjustment matrix is ​​obtained using a min-max normalization formula with a positive lower bound. This formula maps matrix elements to a closed interval between a preset positive lower bound and a preset upper bound, ensuring the normalized minimum element is not zero. When the maximum and minimum elements are equal, the non-negative adjustment matrix is ​​directly set to a matrix where all elements are equal to the preset positive lower bound. The non-negative adjustment matrix is ​​proportionally adjusted according to a preset amplification factor to maintain the positive weight range of elements in the diagnostic focus matrix, and the adjusted matrix is ​​written into memory as the diagnostic focus matrix for a new period.

[0046] As one possible implementation, the step of converting the weighted difference raster matrix into an offset state vector, calculating the projection similarity of the offset state vector onto each basic causal vector, and performing nonnegation processing to obtain a causal correlation sequence includes: The generated weighted difference raster matrix is ​​element-wise extracted according to a preset row-first or column-first order, and concatenated to convert it into a one-dimensional offset state vector. Each set of basic causal vectors in the causal knowledge base is restored to a basic causal matrix, and multiplied element-wise with the current periodic diagnostic focus matrix and then expanded again to obtain the basic causal vector to be matched in the current weight space. The cosine similarity between the offset state vector and each basic causal vector to be matched is calculated. The values ​​less than zero in the calculated results are set to zero and denegated, and then arranged and combined sequentially to generate a one-dimensional numerical sequence representing the degree of agreement between the current state and each preset causal factor. This sequence is saved as the causal correlation degree sequence. Based on the causal correlation degree sequence, the corresponding causal space mask is combined and then added to a preset small positive basis matrix.

[0047] The generated 64×64 weighted difference raster matrix Using a row-major order, elements are extracted sequentially from the first row and first column of the matrix to the right until the end of the first row. Then, the process switches to the second row and continues from the beginning, until the last value of the 64th row is extracted. All extracted data elements are concatenated to form a one-dimensional offset state vector of length 4096. .

[0048] Read the K pre-set groups from the causal knowledge base, for example, K=5, which represent the basic causal vectors of a specific anomaly. to Their data length is also 4096. Each fundamental causality vector... Restored to a 64×64 fundamental causal matrix, and multiplied element-wise with the current periodic diagnostic focus matrix F, then re-expanded to obtain the fundamental causal vector to be matched in the current weight space. Calculate the offset state vector. With each Cosine similarity between them, when =0 or When =0, let =0; when ≠0 and When ≠0, the formula is expressed as .

[0049] Assuming the calculated original similarity results between the current vector and the first to third basic causes are as follows: , , The calculation results are nonnegated, with the truncation condition as follows: if the value is less than 0.0, it is set to 0; if it is greater than or equal to 0.0, the original calculated value remains unchanged. After the above nonnegation process, the original negative similarity is reduced. Set to 0.0, while positive similarity and Retain the values ​​as 0.72 and 0.45.

[0050] The obtained K non-negative calculation results are arranged and combined in sequence to generate a one-dimensional numerical sequence such as A=[0.72,0.0,0.45,...], which is then saved as a causal correlation sequence.

[0051] This invention also proposes an online monitoring system for abnormal grounding resistance of cement poles, comprising the following modules: The reconstruction module is used to collect the excitation current, induced voltage, pole temperature and ambient humidity of the cement pole, calculate the instantaneous grounding resistance based on the excitation current and induced voltage within the time window, generate a two-dimensional phase diagram through phase space reconstruction based on the excitation current and instantaneous grounding resistance, and obtain the characteristic grid map through gridding and quantization processing. The generation module is used to construct a state reference surface with temperature and humidity as coordinate axes based on historical operating data. The state reference surface stores standard feature raster maps. The current feature raster map is matrix subtracted from the standard feature raster map corresponding to the current temperature and humidity in the state reference surface. The subtraction result is multiplied element by element with the diagnostic focusing matrix to generate a weighted difference raster matrix. The update module is used to construct a causal knowledge base, which stores multiple causal templates containing basic causal vectors and causal space masks. The weighted difference raster matrix is ​​converted into an offset state vector. The projection similarity of the offset state vector on each basic causal vector is calculated and nonnegated to obtain a causal correlation degree sequence. The abnormal cause is determined based on the causal correlation degree sequence. The causal space mask corresponding to the causal correlation degree sequence is combined and added to a preset small positive basis matrix. After normalization, the diagnostic focusing matrix for the next cycle is generated.

[0052] As one possible implementation, the step of generating a two-dimensional phase diagram based on the excitation current and the instantaneous grounding resistance through phase space reconstruction includes: Based on the excitation current sequence and induced voltage sequence within the time window, the induced voltage sequence is synchronously detected to extract the non-negative voltage amplitude sequence in phase with the excitation current. At sampling points where the excitation current amplitude is greater than a preset minimum current threshold, the ratio of the non-negative voltage amplitude to the excitation current amplitude is calculated to obtain the instantaneous grounding resistance sequence. With the excitation current as the abscissa and the instantaneous grounding resistance as the ordinate, the excitation current and instantaneous grounding resistance values ​​at the same extraction time are used to form two-dimensional coordinate points. All times within the time window are traversed in chronological order, and the generated multiple two-dimensional coordinate points are connected sequentially in a two-dimensional plane to obtain the two-dimensional phase diagram with phase space trajectory characteristics.

[0053] As one possible implementation, the gridding and quantization process to obtain a feature raster map includes: Based on the preset excitation current coordinate range and the preset instantaneous grounding resistance coordinate range, the fixed plane region where the two-dimensional phase diagram is located is determined. The plane region where the two-dimensional phase diagram is located is divided into a two-dimensional grid matrix of preset fixed size. The number of trajectory points of the two-dimensional phase diagram contained within the boundary of each single grid is counted as the first basic parameter. The first basic parameter is divided by the total number of trajectory points of the two-dimensional phase diagram in the plane region to calculate the trajectory point distribution probability corresponding to each grid. The trajectory point distribution probability is used as the pixel value to update the spatial state value of the corresponding grid, thereby generating the feature grid map in which the structural features are represented by density distribution.

[0054] As one possible implementation, the construction of a state reference surface based on historical operating data with temperature and humidity as coordinate axes includes: Historical data on pole temperature, ambient humidity, and feature grid patterns are synchronously collected during normal operation of the cement pole when the equipment is undamaged and no alarms occur. The two-dimensional temperature and humidity space is divided into multiple temperature and humidity grid regions according to preset temperature and humidity step sizes. All feature grid patterns falling within the same temperature and humidity grid region are extracted. The corresponding elements of the feature grid patterns are matrix-accumulated, and the average value matrix is ​​calculated. The average value matrix is ​​used as the standard feature grid pattern corresponding to the temperature and humidity grid region. All temperature and humidity grid regions and their corresponding standard feature grid patterns together constitute the state reference surface.

[0055] As one possible implementation, the step of performing matrix subtraction between the current feature raster map and the standard feature raster map corresponding to the current temperature and humidity in the state reference surface includes: The current pole temperature and ambient humidity are collected on-site and used as positioning parameters. The target temperature and humidity grid region containing the current pole temperature and ambient humidity is found in the state reference surface. The standard feature raster map corresponding to the target temperature and humidity grid region is extracted from the state reference surface. The feature raster map generated at the current time is subtracted from the extracted standard feature raster map element by element at each corresponding position to generate a difference matrix composed of difference data as the result of the matrix difference.

[0056] As one possible implementation, the step of multiplying the difference result element-wise with the diagnostic focusing matrix to generate a weighted difference raster matrix includes: The difference matrix obtained by subtracting the matrix is ​​read. When it is the initial monitoring period, a one matrix with the same dimension as the difference matrix is ​​generated as the diagnostic focus matrix. When it is not the initial monitoring period, the iterative diagnostic focus matrix saved in the previous period is extracted. Each element in the difference matrix is ​​multiplied item by item with the corresponding element in the same row and column position in the diagnostic focus matrix. All product calculation results are saved and reorganized into a new matrix according to the original row and column order to obtain the weighted difference raster matrix that highlights the potential abnormal area.

[0057] As one possible implementation, the step of converting the weighted difference raster matrix into an offset state vector, calculating the projection similarity of the offset state vector onto each basic causal vector, and performing nonnegation processing to obtain a causal correlation sequence includes: The generated weighted difference raster matrix is ​​element-wise extracted according to a preset row-first or column-first order, and concatenated to convert it into a one-dimensional offset state vector. Each set of basic causal vectors in the causal knowledge base is restored to a basic causal matrix, and multiplied element-wise with the current periodic diagnostic focus matrix and then expanded again to obtain the basic causal vector to be matched in the current weight space. The cosine similarity between the offset state vector and each basic causal vector to be matched is calculated. The values ​​less than zero in the calculated results are set to zero and denegated, and then arranged and combined sequentially to generate a one-dimensional numerical sequence representing the degree of agreement between the current state and each preset causal factor. This sequence is saved as the causal correlation degree sequence. Based on the causal correlation degree sequence, the corresponding causal space mask is combined and then added to a preset small positive basis matrix.

[0058] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for online monitoring of abnormal grounding resistance of cement poles, characterized in that, include: The excitation current, induced voltage, pole temperature, and ambient humidity of the cement pole are collected. The instantaneous grounding resistance is calculated based on the excitation current and induced voltage within the time window. A two-dimensional phase diagram is generated by reconstructing the phase space based on the excitation current and instantaneous grounding resistance, and the characteristic grid map is obtained by gridding and quantization processing. Based on historical operating data, a state reference surface is constructed with temperature and humidity as coordinate axes. The state reference surface stores standard feature raster maps. The current feature raster map is matrix subtracted from the standard feature raster map corresponding to the current temperature and humidity in the state reference surface. The subtraction result is multiplied element by element with the diagnostic focusing matrix to generate a weighted difference raster matrix. A causal knowledge base is constructed, which stores multiple causal templates containing basic causal vectors and causal space masks. The weighted difference raster matrix is ​​converted into an offset state vector. The projection similarity of the offset state vector on each basic causal vector is calculated and nonnegated to obtain a causal correlation degree sequence. The abnormal cause is determined based on the causal correlation degree sequence. The causal space mask corresponding to the causal correlation degree sequence is combined and added to a preset small positive basis matrix. After normalization, a diagnostic focusing matrix for the next period is generated.

2. The method according to claim 1, characterized in that, The process of generating a two-dimensional phase diagram based on the excitation current and instantaneous grounding resistance through phase space reconstruction includes: Based on the excitation current sequence and induced voltage sequence within the time window, the induced voltage sequence is synchronously detected to extract the non-negative voltage amplitude sequence in phase with the excitation current. At sampling points where the excitation current amplitude is greater than a preset minimum current threshold, the ratio of the non-negative voltage amplitude to the excitation current amplitude is calculated to obtain the instantaneous grounding resistance sequence. With the excitation current as the abscissa and the instantaneous grounding resistance as the ordinate, the excitation current and instantaneous grounding resistance values ​​at the same extraction time are used to form two-dimensional coordinate points. All times within the time window are traversed in chronological order, and the generated multiple two-dimensional coordinate points are connected sequentially in a two-dimensional plane to obtain the two-dimensional phase diagram with phase space trajectory characteristics.

3. The method according to claim 1, characterized in that, The gridding and quantization process yields a feature raster map, including: Based on the preset excitation current coordinate range and the preset instantaneous grounding resistance coordinate range, the fixed plane region where the two-dimensional phase diagram is located is determined. The plane region where the two-dimensional phase diagram is located is divided into a two-dimensional grid matrix of preset fixed size. The number of trajectory points of the two-dimensional phase diagram contained within the boundary of each single grid is counted as the first basic parameter. The first basic parameter is divided by the total number of trajectory points of the two-dimensional phase diagram in the plane region to calculate the trajectory point distribution probability corresponding to each grid. The trajectory point distribution probability is used as the pixel value to update the spatial state value of the corresponding grid, thereby generating the feature grid map in which the structural features are represented by density distribution.

4. The method according to claim 1, characterized in that, The state reference surface constructed based on historical operating data, with temperature and humidity as coordinate axes, includes: Historical data on pole temperature, ambient humidity, and feature grid patterns are synchronously collected during normal operation of the cement pole when the equipment is undamaged and no alarms occur. The two-dimensional temperature and humidity space is divided into multiple temperature and humidity grid regions according to preset temperature and humidity step sizes. All feature grid patterns falling within the same temperature and humidity grid region are extracted. The corresponding elements of the feature grid patterns are matrix-accumulated, and the average value matrix is ​​calculated. The average value matrix is ​​used as the standard feature grid pattern corresponding to the temperature and humidity grid region. All temperature and humidity grid regions and their corresponding standard feature grid patterns together constitute the state reference surface.

5. The method according to claim 2, characterized in that, The step of performing matrix subtraction between the current feature raster map and the standard feature raster map corresponding to the current temperature and humidity in the state reference surface includes: The current pole temperature and ambient humidity are collected on-site and used as positioning parameters. The target temperature and humidity grid region containing the current pole temperature and ambient humidity is found in the state reference surface. The standard feature raster map corresponding to the target temperature and humidity grid region is extracted from the state reference surface. The feature raster map generated at the current time is subtracted from the extracted standard feature raster map element by element at each corresponding position to generate a difference matrix composed of difference data as the result of the matrix difference.

6. The method according to claim 1, characterized in that, The step of multiplying the difference result element-wise with the diagnostic focusing matrix to generate a weighted difference raster matrix includes: The difference matrix obtained by subtracting the matrix is ​​read. When it is the initial monitoring period, a one matrix with the same dimension as the difference matrix is ​​generated as the diagnostic focus matrix. When it is not the initial monitoring period, the iterative diagnostic focus matrix saved in the previous period is extracted. Each element in the difference matrix is ​​multiplied item by item with the corresponding element in the same row and column position in the diagnostic focus matrix. All product calculation results are saved and reorganized into a new matrix according to the original row and column order to obtain the weighted difference raster matrix that highlights the potential abnormal area.

7. The method according to claim 1, characterized in that, The step of converting the weighted difference raster matrix into an offset state vector, calculating the projection similarity of the offset state vector onto each basic causal vector, and performing nonnegation processing to obtain the causal correlation sequence includes: The generated weighted difference raster matrix is ​​element-wise extracted according to a preset row-first or column-first order, and concatenated to convert it into a one-dimensional offset state vector. Each set of basic causal vectors in the causal knowledge base is restored to a basic causal matrix, and multiplied element-wise with the current periodic diagnostic focus matrix and then expanded again to obtain the basic causal vector to be matched in the current weight space. The cosine similarity between the offset state vector and each basic causal vector to be matched is calculated. The values ​​less than zero in the calculated results are set to zero and denegated, and then arranged and combined sequentially to generate a one-dimensional numerical sequence representing the degree of agreement between the current state and each preset causal factor. This sequence is saved as the causal correlation degree sequence. Based on the causal correlation degree sequence, the corresponding causal space mask is combined and then added to a preset small positive basis matrix.

8. An online monitoring system for abnormal grounding resistance of cement poles, characterized in that, include: The reconstruction module is used to collect the excitation current, induced voltage, pole temperature and ambient humidity of the cement pole, calculate the instantaneous grounding resistance based on the excitation current and induced voltage within the time window, generate a two-dimensional phase diagram through phase space reconstruction based on the excitation current and instantaneous grounding resistance, and obtain the characteristic grid map through gridding and quantization processing. The generation module is used to construct a state reference surface with temperature and humidity as coordinate axes based on historical operating data. The state reference surface stores standard feature raster maps. The current feature raster map is matrix subtracted from the standard feature raster map corresponding to the current temperature and humidity in the state reference surface. The subtraction result is multiplied element by element with the diagnostic focusing matrix to generate a weighted difference raster matrix. The update module is used to construct a causal knowledge base, which stores multiple causal templates containing basic causal vectors and causal space masks. The weighted difference raster matrix is ​​converted into an offset state vector. The projection similarity of the offset state vector on each basic causal vector is calculated and nonnegated to obtain a causal correlation degree sequence. The abnormal cause is determined based on the causal correlation degree sequence. The causal space mask corresponding to the causal correlation degree sequence is combined and added to a preset small positive basis matrix. After normalization, the diagnostic focusing matrix for the next cycle is generated.

9. The system according to claim 8, characterized in that, The process of generating a two-dimensional phase diagram based on the excitation current and instantaneous grounding resistance through phase space reconstruction includes: Based on the excitation current sequence and induced voltage sequence within the time window, the induced voltage sequence is synchronously detected to extract the non-negative voltage amplitude sequence in phase with the excitation current. At sampling points where the excitation current amplitude is greater than a preset minimum current threshold, the ratio of the non-negative voltage amplitude to the excitation current amplitude is calculated to obtain the instantaneous grounding resistance sequence. With the excitation current as the abscissa and the instantaneous grounding resistance as the ordinate, the excitation current and instantaneous grounding resistance values ​​at the same extraction time are used to form two-dimensional coordinate points. All times within the time window are traversed in chronological order, and the generated multiple two-dimensional coordinate points are connected sequentially in a two-dimensional plane to obtain the two-dimensional phase diagram with phase space trajectory characteristics.

10. The system according to claim 8, characterized in that, The gridding and quantization process yields a feature raster map, including: Based on the preset excitation current coordinate range and the preset instantaneous grounding resistance coordinate range, the fixed plane region where the two-dimensional phase diagram is located is determined. The plane region where the two-dimensional phase diagram is located is divided into a two-dimensional grid matrix of preset fixed size. The number of trajectory points of the two-dimensional phase diagram contained within the boundary of each single grid is counted as the first basic parameter. The first basic parameter is divided by the total number of trajectory points of the two-dimensional phase diagram in the plane region to calculate the trajectory point distribution probability corresponding to each grid. The trajectory point distribution probability is used as the pixel value to update the spatial state value of the corresponding grid, thereby generating the feature grid map in which the structural features are represented by density distribution.

Images