Multi-band microwave sensing electrical equipment on-line fault positioning method and monitoring equipment

Abstract

The invention relates to the technical field of electrical equipment monitoring, and discloses a multi-band microwave sensing electrical equipment on-line fault positioning method, which comprises the following steps of: acquiring microwave reflection / scattering signal data of electrical equipment as original data by adopting a multi-band sensor array; de-noising processing is carried out on the original data of each multi-band sensor to obtain de-noised data; performing data fusion processing on the de-noised data to obtain fused data; and establishing a visual map according to the fused data, extracting feature vectors from the visual map to perform node fault identification, obtaining fault identification results of a plurality of nodes, and predicting a fault development trend of each node of the electrical equipment so as to redeploy the multi-band sensor array. Through multi-band sensor array acquisition, data denoising fusion, visual map feature extraction and fault trend prediction, electrical equipment fault accurate positioning and sensor dynamic deployment are realized, the reliability and operation and maintenance efficiency of a power system are improved, and the fault risk and cost are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical equipment monitoring, and in particular to an electrical equipment online fault locating method and monitoring equipment using multi-band microwave sensing. Background Art

[0002] In modern power systems, the stable operation of electrical equipment is crucial to ensuring the reliability and safety of power supply. Failure of electrical equipment can not only cause widespread power outages, disrupting social production and life, but can also lead to serious safety incidents. Therefore, rapid and accurate detection and location of electrical equipment faults is crucial for improving power system efficiency and reducing operation and maintenance costs.

[0003] Currently, electrical equipment fault detection and location technologies primarily consist of traditional methods based on physical quantity monitoring and intelligent approaches based on data analysis. While traditional methods such as temperature monitoring and vibration detection can detect equipment anomalies to a certain extent, they suffer from limited detection range, low accuracy, and an inability to predict faults in advance. Intelligent approaches based on data analysis, while able to leverage historical data for fault diagnosis, face challenges such as low data quality and difficulty extracting features. Especially in complex electromagnetic environments, the signals generated by electrical equipment are susceptible to noise interference, leading to inaccurate fault location. The emergence of multi-band microwave sensing technology offers a new approach to addressing these issues. It exploits the differences in reflection and scattering characteristics generated by microwave signals of different frequency bands interacting with electrical equipment to comprehensively capture equipment status information. However, further research and innovation are needed to effectively process multi-band microwave raw data to accurately locate faults and predict their development trends. Summary of the Invention

[0004] The present invention provides an online fault location method and monitoring equipment for electrical equipment using multi-band microwave sensing. Through multi-band sensor array acquisition, data denoising and fusion, visual graph feature extraction, and fault trend prediction, it achieves accurate fault location of electrical equipment and dynamic deployment of sensors, thereby improving the reliability and operation and maintenance efficiency of the power system and reducing fault risks and costs.

[0005] The present invention provides an online fault location method for electrical equipment using multi-band microwave sensing, comprising:

[0006] A multi-band sensor array is used to collect microwave reflection / scattering signal data of electrical equipment as raw data;

[0007] Performing denoising on the raw data of each multi-band sensor to obtain denoised data;

[0008] Performing data fusion processing on the denoised data of each multi-band sensor to obtain fused data;

[0009] Establishing a visual map based on the fused data, and extracting feature vectors from the visual map;

[0010] Performing node fault identification on the electrical equipment according to the characteristic vector to obtain fault identification results of multiple nodes;

[0011] The fault development trend of each node of the electrical equipment is predicted according to the fault identification result, so as to redeploy the multi-band sensor array according to the development trend.

[0012] Furthermore, in the step of using a multi-band sensor array to collect microwave reflection / scattering signal data of electrical equipment as raw data, the multi-band sensor array includes multiple three-band sensors, and the three-band sensors are 2.45GHz, 5.8GHz, and 24GHz three-band sensors. The multiple three-band sensors are deployed in a linear array along the axial direction of the electrical equipment, or in a circular array around the main body of the electrical equipment.

[0013] Furthermore, the step of performing denoising on the raw data of each multi-band sensor to obtain denoised data includes:

[0014] Use W t and W t-1 Represent the sensor array state values ​​at the current time t and the previous time t-1 respectively, use Z to represent the state transition matrix from time t-1 to time t, and use Y t Indicates the current measurement value, G represents the measurement matrix, and V t and U t They represent the state noise and observation noise that obey the normal distribution, V t ~N(0,P),U t ~N(0,Q), where P and Q represent covariance, and the system state equation is W t =ZW t-1 +V t , the observation equation is Y t =GW t +U t ;

[0015] In the multi-band sensor array, V t and U t Considered as Gaussian white noise, T t Represents the state control quantity at time t. If the control quantity does not exist, then T t = 0, A represents the transition matrix from the control input to the current state, according to the state W at time t-1 (t-1|t-1) and the corresponding covariance S (t-1|t-1) Predict the state and covariance at time t to obtain the predicted value W of the state at time t(t-1|t-1) and the corresponding covariance S (t-1|t-1) for:

[0016]

[0017] Combine the observation equation and the above formula to establish the optimal equation W of the system state estimate at time t (t|t) for:

[0018]

[0019] Among them, K t is the Kalman matrix; the covariance W corresponding to the current moment t is obtained at each update iteration (t|t) , as shown below:

[0020] W (t|t) =K t G.W. (t|t-1)

[0021] The noise of the original data is suppressed by the above prediction-update iterative mechanism, thereby obtaining a more realistic true value of the original data.

[0022] Furthermore, the step of performing data fusion processing on the denoised data of each multi-band sensor to obtain fused data includes:

[0023] Assume that the multi-band sensor array contains m sensors in total, and establish n fusion nodes in the sensor array to fuse the denoised data of m sensors. Let a represent the input parameter, and the dimension is d, that is, a∈R d , b represents the fusion result, b∈R, based on j independent distribution observation samples (a1, b1),…, (a j ,b j ) Establish the optimal function f(a) to describe the dependency between a and b. Use the nonlinear function φ to map the domain a to a high-dimensional feature space and perform linear regression in it to obtain the nonlinear regression result of the original space. Use v to represent a vector and c to represent a scalar. Then f(a) is as follows:

[0024] f(a)=v·φ(a)+c

[0025] According to the above formula, the problem is to solve v and c based on the known j observation samples; the ε-insensit order function is used as the regression measure function. According to the principle of support vector machine, the regression problem is regarded as a problem that satisfies the structural risk minimization. ak and bk represent the kth observation sample, k∈[1,j], and the corresponding minimization form is As shown below:

[0026]

[0027] Introducing the relaxation factor τ into the ε-insensit order function k and Complete the linear fitting of all training samples based on the accuracy ε, as shown below:

[0028]

[0029] According to the above formula, The minimization regression problem is expressed as follows:

[0030]

[0031] Among them, α is a constant set in advance, and the Lagrangian function is introduced to solve the regression optimization problem, and μ is used k 、 σ k 、 represents the Lagrange coefficient, μ k 、 σ k 、 Lagrangian function In the equation, L is related to μ k 、 σ k 、 Should be maximized, for v, c, τ k 、 To minimize L, the following conditions should be met when obtaining the extreme value of L: Introducing a suitable inner product function ξ(a i ,a k ), i, k∈[1, j], establish μ k 、 Maximize the objective function for:

[0032]

[0033] The above formula must satisfy and 0≤μ k , According to the constraints, the interior point algorithm can be used to solve μ k 、 Combined with the partial derivative of v and μ k 、 get

[0034] According to the above conditions, the fusion data is obtained

[0035] Furthermore, the step of establishing a visual map based on the fused data and extracting feature vectors from the visual map includes:

[0036] Based on the fused data, use B T and D T They represent the structural transfer matrix when there is electrical fault and no fault at time T, y T Indicates the data transmission status target when there is an electrical fault at time T, z T Indicates the output target in the case of no fault, and the output matrix y in the case of fault at the next moment T+1 T+1 Expressed as:

[0037] y T+1 =B T y T +D T z T

[0038] Get the operating structure matrix x under electrical fault conditions kT =K kT y T+1 +w kT , where K kT and w kT Represent different electrical fault operation measurement matrices, k is a positive integer; extract the sample set features after m sensors are fused, detect the number of samples, identify sensor information in the p-th operation system and establish an electrical fault information network, further analyze the electrical operation data and reconstruct the feature model by extracting the electrical fault information features, obtain the eigenvalue, and use the visual map to calculate the node coordinate related parameters in the construction of the electrical fault information network; the electrical fault information feature quantity S iT All satisfy the following formula:

[0039]

[0040] Among them, e -αβγ and e -kβγ is the Euler formula form of trigonometric function, α, β, γ are the trigonometric function parameters, To receive the model coefficients; assume that there are arbitrary electrical nodes i and j. If the vector from i to j is a directed vector, it is represented by the matrix [ij] = 1, otherwise it is represented by [ij] = 0; after obtaining the node coordinates and characteristic distribution maps of all electrical equipment, combining the eigenvalues ​​and operating data, an electrical visualization map is established, and the eigenvector is extracted from the image to identify the electrical fault point based on the eigenvector.

[0041] Furthermore, the step of performing node fault identification on the electrical equipment according to the characteristic vector to obtain fault identification results of multiple nodes includes:

[0042] According to the eigenvector extracted from the electrical visualization map, R is used to represent the eigenvalue of the region where the node is located. Represents the characteristic mean of the region, ΔR represents the characteristic difference, if Then there is no fault point in the area; if R, If ΔR is not equal, it means there is a fault point in the area and the location of the fault point needs to be further determined. The specific method is as follows:

[0043] when When , the current node has a fault tendency. And when ΔR>R, the node is a fault point;

[0044] The final identification of the fault point is determined by the assignment method, and R, ΔR, first compare R, like Then skip the assignment and continue with the next step of detection. Otherwise, add 1 to the assignment κ=0, and then compare R and ΔR. If R=ΔR, directly output the assignment result κ=0, otherwise continue to add 1 to the assignment. After the assignment, directly determine the fault situation based on the assignment. If κ=0, there is no fault point in the electrical system. If κ=1, there is a fault tendency in the electrical system, and attention needs to be paid to the identification node. If κ=2, the node has failed and needs to be repaired immediately to avoid electrical equipment collapse.

[0045] Furthermore, the step of predicting the fault development trend of each node of the electrical equipment according to the fault identification result and redeploying the multi-band sensor array according to the development trend includes:

[0046] When κ=1, that is, there is a fault tendency in the electrical system, the multi-band microwave data after Kalman filtering and denoising is extracted, and the corresponding fault level label κ∈{0,1,2};

[0047] For a sample with κ = 1 in the historical data, if it evolves to κ = 2 within the next time t, it is marked as a positive example. If it remains at k = 1 or reverts to k = 0, it is marked as a negative example. A sliding window W is used to generate the input sequence [W, D], where D is the feature dimension. The output is the fault evolution label.

[0048] Extract fault feature S kT and eigenvalue R, ΔR, calculate its time rate of change, the calculation formula is The features were Z-Score normalized to eliminate the difference in frequency band dimensions;

[0049] Build a two-layer LSTM network model with an input layer dimension of [W, D]. The output layer uses Sigmoid activation to output binary classification probabilities. Configure the Adam optimizer and the binary cross entropy loss function, and add Dropout to prevent overfitting.

[0050] The two-layer LSTM network model is trained based on the sample data to obtain a trained two-layer LSTM network model. When κ=1 is detected, the real-time window data is collected and input into the model after feature engineering, and the probability P of κ=2 is output;

[0051] If P is greater than or equal to a preset value, an early warning is triggered, and the double-layer LSTM network model is updated using incremental learning, and the multi-band sensor array is redeployed.

[0052] The present invention also provides an online fault monitoring device for electrical equipment using multi-band microwave sensing, and an online fault location method for electrical equipment using multi-band microwave sensing as described above. The monitoring device includes a multi-band sensor array, a data acquisition module, an edge computing module, a data transmission module, a fault identification module, a visualization map module, and a change prediction module. The multi-band sensor array is connected to the data acquisition module, the data acquisition module is connected to the edge computing module, the edge computing module is connected to the data transmission module, the data transmission module is connected to the visualization map module, the visualization map module is connected to the fault identification module, and the fault identification module is connected to the change prediction module.

[0053] The data acquisition module is used to collect microwave reflection / scattering signal data of electrical equipment as raw data through a multi-band sensor array;

[0054] The edge computing module is used to perform denoising on the raw data of each multi-band sensor and perform data fusion processing on the denoised raw data to obtain fused data;

[0055] The data transmission module is used to transmit the fused data to the fault identification module to identify and locate the fault of the electrical equipment;

[0056] The visualization map module is used to create a visualization map based on the fused data and extract feature vectors from the visualization map;

[0057] The fault identification module is used to perform node fault identification on the electrical equipment according to the feature vector to obtain fault identification results of multiple nodes;

[0058] The change prediction module is used to predict the fault development trend of each node of the electrical equipment according to the fault identification result, so as to redeploy the multi-band sensor array according to the development trend.

[0059] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0060] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0061] The beneficial effects of the present invention are:

[0062] The present invention collects microwave signals from electrical equipment through a multi-band sensor array and performs denoising and fusion processing, effectively improving data quality and anti-interference ability, and solving the problems of limited detection range and insufficient accuracy of traditional methods; constructing a visual map based on the fused data and extracting feature vectors, it realizes the accurate identification and positioning of electrical equipment node faults, and overcomes the defect of difficult feature extraction of existing intelligent methods; through the prediction of fault development trends and the dynamic redeployment of multi-band sensor arrays, potential faults can be warned in advance and the monitoring plan can be optimized, significantly improving the reliability and operation efficiency of the power system, and reducing the risk of equipment failure and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 Schematic diagram of a method flow according to an embodiment of the present invention.

[0064] Figure 2 FIG. 1 is a schematic diagram of the device structure according to an embodiment of the present invention.

[0065] Figure 3 Schematic diagram of the internal structure of a computer device according to an embodiment of the present invention.

[0066] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0067] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0068] like Figure 1 As shown, the present invention provides an online fault location method for electrical equipment using multi-band microwave sensing, comprising:

[0069] S1. Use a multi-band sensor array to collect microwave reflection / scattering signal data of electrical equipment as raw data.

[0070] The multi-band sensor array includes multiple three-band sensors, namely 2.45GHz, 5.8GHz, and 24GHz three-band sensors. The multiple three-band sensors are deployed in a linear array along the axis of the electrical equipment with a spacing of 10-20cm; or in a circular array around the main body of the electrical equipment, with the radius adjusted according to the equipment size (such as the radius of a 1m equipment is set to 0.5m) to achieve three-dimensional spatial scanning; the angle between the sensor antenna normal and the equipment surface is ≤30° to ensure the strength of the reflected signal; the 24GHz sensor is facing the moving parts of the equipment (such as the transformer tap changer), and the 5.8GHz sensor is perpendicular to the electric field direction (to reduce polarization loss).

[0071] Among them, 2.45GHz has strong penetration and is suitable for internal structure detection of equipment; 5.8GHz has good anti-interference performance and is used for surface defect identification; 24GHz has high resolution and can monitor tiny deformations.

[0072] S2. Perform denoising processing on the raw data of each multi-band sensor to obtain denoised data.

[0073] The multi-sensor characteristics and the complexity of the electrical working environment lead to inevitable errors in the data collected by the sensor array. In order to improve the data accuracy and maximize the authenticity of the electrical data, the Kalman filter is introduced to remove the data noise, that is, through the state equation W t =ZW t-1 +V t and the observation equation Y t =GW t +U t Dynamically optimize signal characteristics to eliminate the impact of environmental electromagnetic interference on multi-band data. Specifically including:

[0074] Use W t and W t-1 Represent the sensor array state values ​​at the current time t and the previous time t-1 respectively, use Z to represent the state transition matrix from time t-1 to time t, and use Y t Indicates the current measurement value, G represents the measurement matrix, and V t and U t They represent the state noise and observation noise that obey the normal distribution, V t ~N(0,P),U t ~N(0,Q), where P and Q represent covariance, and the system state equation is W t =ZW t-1 +V t , the observation equation is Y t =GW t +U t .

[0075] In the multi-band sensor array, V t and U t Considered as Gaussian white noise, T t Represents the state control quantity at time t. If the control quantity does not exist, then T t = 0, A represents the transition matrix from the control input to the current state, according to the state W at time t-1 (t-1|t-1) and the corresponding covariance S (t-1|t-1) Predict the state and covariance at time t to obtain the predicted value W of the state at time t (t-1|t-1) and the corresponding covariance S (t-1|t-1) for:

[0076]

[0077] Combine the observation equation and the above formula to establish the optimal equation W of the system state estimate at time t (t|t) for:

[0078]

[0079] Among them, K t is the Kalman matrix; the covariance W corresponding to the current moment t is obtained at each update iteration (t|t) , as shown below:

[0080] W (t|t) =K t G.W. (t|t-1 )

[0081] The above prediction-update iterative mechanism achieves noise suppression on the raw data, obtaining a more realistic true value of the raw data. The Kalman filter algorithm can effectively remove the system noise and sensor measurement noise in the sensor array data, obtaining a more realistic true value of the electrical data.

[0082] S3. Perform data fusion processing on the denoised data from each multi-band sensor to obtain fused data. Specifically, based on the SVM data fusion framework, the amplitude, phase, frequency offset, and other parameters of the multi-band microwave signal are mapped to a high-dimensional feature space. A regression model f(a) = v·φ(a) + c is constructed using the nonlinear function φ(a), thereby achieving weighted fusion of features from different frequency bands.

[0083] The sensor array contains multiple sensors. By fusing data from multiple sensor information sources, it can not only improve the accuracy of subsequent identification of electrical fault points, but also improve overall performance.

[0084] Assume that the multi-band sensor array contains m sensors in total, and establish n fusion nodes in the sensor array to fuse the denoised data of m sensors. Let a represent the input parameter, and the dimension is d, that is, a∈R d, b represents the fusion result, b∈R, based on j independent distribution observation samples (a1, b1),…, (a j ,b j ) Establish the optimal function f(a) to describe the dependency between a and b. Use the nonlinear function φ to map the domain a to a high-dimensional feature space and perform linear regression in it to obtain the nonlinear regression result of the original space. Use v to represent a vector and c to represent a scalar. Then f(a) is as follows:

[0085] f(a)=v·φ(a)+c

[0086] According to the above formula, the problem is to solve v and c based on the known j observation samples; the ε-insensit order function is used as the regression measure function. According to the principle of support vector machine, the regression problem is regarded as a problem that satisfies the structural risk minimization. ak and bk represent the kth observation sample, k∈[1,j], and the corresponding minimization form is As shown below:

[0087]

[0088] Introducing the relaxation factor τ into the ε-insensit order function k and Complete the linear fitting of all training samples based on the accuracy ε, as shown below:

[0089]

[0090] According to the above formula, The minimization regression problem is expressed as follows:

[0091]

[0092] Among them, α is a constant set in advance, and the Lagrangian function is introduced to solve the regression optimization problem, and μ is used k 、 σ k 、 represents the Lagrange coefficient, μ k 、 σ k 、 Lagrangian function In the equation, L is related to μ k 、 σ k 、 Should be maximized, for v, c, τ k 、 To minimize L, the following conditions should be met when obtaining the extreme value of L: Introducing a suitable inner product function ξ(a i ,ak ), i, k∈[1, j], establish μ k 、 Maximize the objective function for:

[0093]

[0094] The above formula must satisfy and 0≤μ k , According to the constraints, the interior point algorithm can be used to solve μ k 、 Combined with the partial derivative of v and μ k 、 get

[0095] According to the above conditions, the fusion data is obtained

[0096] In the following steps S4 and S5, based on the multi-sensor fusion data, the structural transfer matrix B of the electrical fault and no-fault state at time T is defined. T and D T , establish the output matrix equation y T+1 =B T y T +D T Z T , describing the fault state transition law. Extract the fault feature S from the sample set after the fusion of m sensors kT , combined with Euler's formula form e -αβγ Eigenvalues ​​are calculated to form a eigenvector containing parameters such as amplitude and phase. At the sensor nodes, a directed connection matrix is ​​constructed based on the similarity of the eigenvectors. If a fault exists between nodes i and j, the value is set to 1, and otherwise to 0, forming a fault propagation topology. Based on the fault information network, a graph layout algorithm is used to calculate node coordinates. Eigenvalues ​​are mapped to graph color / shape parameters, generating a visual graph containing the feature distribution to intuitively display the fault area. From the visual graph, the eigenvectors of each node are extracted, and the eigenvalue R, regional mean Rˉ, and difference ΔR are calculated. The fault level is determined using an assignment method.

[0097] S4. Establish a visualization map based on the fused data, and extract feature vectors from the visualization map.

[0098] Based on the fusion data, the electrical structure is analyzed and B T and D T They represent the structural transfer matrix when there is electrical fault and no fault at time T, y T Indicates the data transmission status target when there is an electrical fault at time T, z TIndicates the output target in the case of no fault, and the output matrix y in the case of fault at the next moment T+1 T+1 Expressed as:

[0099] y T+1 =B T y T +D T z T

[0100] Get the operating structure matrix x under electrical fault conditions kT =K kT y T+1 +w kT , where K kT and w kT Represent different electrical fault operation measurement matrices, k is a positive integer; extract the sample set features after m sensors are fused, detect the number of samples, identify sensor information in the p-th operation system and establish an electrical fault information network, further analyze the electrical operation data and reconstruct the feature model by extracting the electrical fault information features, obtain the eigenvalue, and use the visual map to calculate the node coordinate related parameters in the construction of the electrical fault information network; the electrical fault information feature quantity S iT All satisfy the following formula:

[0101]

[0102] Among them, e -αβγ and e -kβγ is the Euler formula form of trigonometric function, α, β, γ are the trigonometric function parameters, To receive the model coefficients; assume that there are arbitrary electrical nodes i and j. If the vector from i to j is a directed vector, it is represented by the matrix [ij] = 1, otherwise it is represented by [ij] = 0; after obtaining the node coordinates and characteristic distribution maps of all electrical equipment, combining the eigenvalues ​​and operating data, an electrical visualization map is established, and the eigenvector is extracted from the image to identify the electrical fault point based on the eigenvector.

[0103] S5. Perform node fault identification on the electrical equipment according to the characteristic vector to obtain fault identification results of multiple nodes.

[0104] Fault feature quantity Mapped to the node coordinates of the visualization graph; where the real part e -αβγ and the imaginary part e -kβγ Corresponding to the horizontal and vertical coordinates of the two-dimensional map, the module value |S kT | is mapped to the node color depth, such as red represents high amplitude. For each node, extract its feature vector S = [S1, S2, ..., Sn] in the visualization graph, where S iThe characteristic parameters (such as reflection coefficient and phase offset) after fusion of multiple frequency bands (2.45 GHz, 5.8 GHz, and 24 GHz) are obtained. The eigenvalue R is calculated by weighted summation: where w i is the frequency band weight (e.g. 24GHz weight 0.5, 5.8GHz weight 0.3, 2.45GHz weight 0.2), n represents the dimension of the eigenvector, i.e. the number of eigenvalues ​​involved in the weighted summation. The fault state transfer matrix is ​​dynamically adjusted according to the formula in step S4. The directed connection matrix of the graph divides the adjacent nodes (connection value 1) in the graph into the same area and calculates Where m is the number of nodes in the region, R j is the eigenvalue R of each node, the difference Display R distribution in the form of a heat map, where red areas indicate R is higher than the mean Marked with a dotted box Click on the map node to display the R, The specific value of ΔR.

[0105] R is used to represent the characteristic value of the region where the node is located. Represents the characteristic mean of the region, ΔR represents the characteristic difference, if Then there is no fault point in the area; if R, If ΔR is not equal, it means there is a fault point in the area and the location of the fault point needs to be further determined. The specific method is as follows:

[0106] when When , the current node has a fault tendency. And when ΔR>R, the node is a fault point;

[0107] The final identification of the fault point is determined by the assignment method, and R, ΔR, first compare R, like Then skip the assignment and continue with the next step of detection. Otherwise, add 1 to the assignment κ=0, and then compare R and ΔR. If R=ΔR, directly output the assignment result κ=0, otherwise continue to add 1 to the assignment. After the assignment, directly determine the fault situation based on the assignment. If κ=0, there is no fault point in the electrical system. If κ=1, there is a fault tendency in the electrical system, and attention needs to be paid to the identification node. If κ=2, the node has failed and needs to be repaired immediately to avoid electrical equipment collapse.

[0108] S6. Predicting a fault development trend of each node of the electrical equipment according to the fault identification result, so as to redeploy the multi-band sensor array according to the development trend.

[0109] S501. When κ=1, that is, there is a fault tendency in the electrical system, extract the multi-band microwave data after Kalman filtering and denoising, including the amplitude, phase, and frequency offset characteristics of the 2.45GHz, 5.8GHz, 24GHz and other frequency bands, and the corresponding fault level label κ∈{0,1,2}; provide a clean time series data basis for step S502, and improve the signal-to-noise ratio of the signal after filtering to ensure the validity of the features.

[0110] S502: For a sample with κ=1 in the historical data, if it evolves to κ=2 within the next time t, it is marked as a positive example; if it remains at k=1 or reverts to k=0, it is marked as a negative example. A sliding window W is used to generate the input sequence [W, D], where D is the feature dimension; the output is the fault evolution label.

[0111] A fixed time window W is used to extract continuous features. For example, the input window consists of a multi-band feature vector containing the current time and the previous W-1 time points, with a dimension of [W, D] (D is the feature dimension). The output label is the fault evolution label at time points t after the current window's corresponding time (whether it has evolved or not, for k = 2). The window length W must match the temporal dynamics of the state transition matrix. For example, W = 12 corresponds to 12 10-minute data points.

[0112] The window division is based on the timestamp consistency of step S501, and the label definition is based on the fault registration data of step S501, providing a model training supervision signal for step S503.

[0113] S503, extracting fault feature S kT and eigenvalue R, ΔR, calculate its time rate of change, the calculation formula is The features are Z-Score standardized to eliminate the dimensional differences in the frequency bands (for example, the units of the 2.45GHz amplitude and the 24GHz phase are unified). To address the problem of insufficient sample size when k=1, the dataset is expanded by sliding window translation and adding Gaussian noise to improve the generalization ability of the model.

[0114] Feature extraction relies on the time series data in step S502, and the standardized result is used as the input of the two-layer LSTM in step S504 to ensure the model convergence speed.

[0115] S504. Build a two-layer LSTM network model with an input layer dimension of [W, D]. The output layer uses Sigmoid activation to output binary classification probabilities. Configure the Adam optimizer and binary cross entropy loss function, and add Dropout to prevent overfitting.

[0116] In the two-layer LSTM architecture, the first layer captures short-term dependencies, while the second layer captures long-term trends, referencing the temporal dynamics of state transitions. The Dropout layer randomly drops 20% of neurons to mitigate overfitting of multi-band features. The model structure design refers to the feature dimension (D) in step S503, and the parameter configuration is based on the sample distribution (such as the ratio of positive and negative examples) in step S502, providing the computational framework for training in step S505.

[0117] S505. The two-layer LSTM network model is trained based on the sample data to obtain a trained two-layer LSTM network model. When κ=1 is detected, the real-time window data is collected and input into the model after feature engineering, and the probability P of κ=2 is output.

[0118] The training, validation, and test sets were divided into a ratio of 8:1:1, ensuring a balanced distribution of evolving labels for k = 1 samples in each set. The batch size was set to 32, the number of epochs was set to 50, and early stopping was performed after each round of validation set evaluation (stopping if the validation loss did not decrease for five consecutive rounds). The training data depended on the dataset partitioning results from step S502, and the early stopping strategy was based on the feature validity of step S503. The trained model was used to predict the probability P of κ = 2.

[0119] S506. If P is greater than or equal to a preset value, an early warning is triggered, and incremental learning is used to update the two-layer LSTM network model (based on the time series data characteristics of steps S502-S503), and the multi-band sensor array is redeployed, such as adding multi-band sensors to nodes that are greater than or equal to the preset value.

[0120] In addition, after detecting that electrical equipment has a tendency to fail, the data collection density and accuracy can be improved. By increasing the scanning frequency of microwave sensors (for example, from 1 time / 10 minutes to 1 time / minute), combined with the collaborative collection of the 24GHz high-frequency band (high resolution) and the 2.45GHz low-frequency band (penetration), the dynamic evolution characteristics of the failure tendency can be captured. Microwave sensors are temporarily added near the failure-prone nodes to build a local monitoring network, and the physical location of the potential fault source is located through array signal processing (such as beamforming). Infrared thermal images and ultrasonic signals of the failure-prone nodes are collected simultaneously, and the microwave sensing results are verified through multimodal data fusion.

[0121] like Figure 2As shown, the present invention also provides an online fault monitoring device for electrical equipment with multi-band microwave sensing, and an online fault location method for electrical equipment with multi-band microwave sensing as described above. The monitoring device includes a multi-band sensor array, a data acquisition module, an edge computing module, a data transmission module, a fault identification module, a visualization map module and a change prediction module. The multi-band sensor array is connected to the data acquisition module, the data acquisition module is connected to the edge computing module, the edge computing module is connected to the data transmission module, the data transmission module is connected to the visualization map module, the visualization map module is connected to the fault identification module, and the fault identification module is connected to the change prediction module.

[0122] The data acquisition module is used to collect microwave reflection / scattering signal data of electrical equipment as raw data through a multi-band sensor array;

[0123] The edge computing module is used to perform denoising on the raw data of each multi-band sensor and perform data fusion processing on the denoised raw data to obtain fused data;

[0124] The data transmission module is used to transmit the fused data to the fault identification module to identify and locate the fault of the electrical equipment;

[0125] The visualization map module is used to create a visualization map based on the fused data and extract feature vectors from the visualization map;

[0126] The fault identification module is used to perform node fault identification on the electrical equipment according to the feature vector to obtain fault identification results of multiple nodes;

[0127] The change prediction module is used to predict the fault development trend of each node of the electrical equipment according to the fault identification result, so as to redeploy the multi-band sensor array according to the development trend.

[0128] Each of the above modules is used to execute the corresponding steps in the above multi-band microwave sensing method for online fault location of electrical equipment. The specific implementation method thereof is described in the above method embodiment and will not be repeated here.

[0129] like Figure 3 As shown, the present invention also provides a computer device, which can be a server, and its internal structure can be as follows Figure 3As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor of the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store all data required for the process of the online fault location method for electrical equipment using multi-band microwave sensing. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the online fault location method for electrical equipment using multi-band microwave sensing is implemented.

[0130] Those skilled in the art will understand that Figure 3 The structure shown in is merely a block diagram of a portion of the structure related to the present application solution and does not constitute a limitation on the computer device to which the present application solution is applied.

[0131] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, any of the above-mentioned methods for online fault location of electrical equipment using multi-band microwave sensing is implemented.

[0132] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0133] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0134] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for online fault location of electrical equipment using multi-band microwave sensing, characterized in that: include: A multi-band sensor array is used to collect microwave reflection / scattering signal data of electrical equipment as raw data; Performing denoising on the raw data of each multi-band sensor to obtain denoised data; Performing data fusion processing on the denoised data of each multi-band sensor to obtain fused data; Establishing a visual map based on the fused data, and extracting feature vectors from the visual map; Performing node fault identification on the electrical equipment according to the characteristic vector to obtain fault identification results of multiple nodes; The fault development trend of each node of the electrical equipment is predicted according to the fault identification result, so as to redeploy the multi-band sensor array according to the development trend.

2. The method for online fault location of electrical equipment using multi-band microwave sensing according to claim 1, characterized in that: In the step of using a multi-band sensor array to collect microwave reflection / scattering signal data of electrical equipment as raw data, the multi-band sensor array includes multiple three-band sensors, and the three-band sensors are 2.45GHz, 5.8GHz, and 24GHz three-band sensors. The multiple three-band sensors are deployed in a linear array along the axial direction of the electrical equipment, or in a circular array surrounding the main body of the electrical equipment.

3. The method for online fault location of electrical equipment using multi-band microwave sensing according to claim 1, characterized in that: The step of performing denoising on the raw data of each multi-band sensor to obtain denoised data includes: Use W t and W t-1 Represent the sensor array state values ​​at the current time t and the previous time t-1 respectively, use Z to represent the state transition matrix from time t-1 to time t, and use Y t Indicates the current measurement value, G represents the measurement matrix, and V t and U t They represent the state noise and observation noise that obey the normal distribution, V t ~N(0,P),U t ~N(0,Q), where P and Q represent covariance, and the system state equation is W t =ZW t-1 +V t , the observation equation is Y t =GW t +U t ; In the multi-band sensor array, V t and U t Considered as Gaussian white noise, T t Represents the state control quantity at time t. If the control quantity does not exist, then T t = 0, A represents the transition matrix from the control input to the current state, according to the state W at time t-1 (t-1|t-1) and the corresponding covariance S (t-1|t-1) Predict the state and covariance at time t to obtain the predicted value W of the state at time t (t-1|t-1) and the corresponding covariance S (t-1|t-1) for: Combine the observation equation and the above formula to establish the optimal equation W of the system state estimate at time t (t|t) for: Among them, K t is the Kalman matrix; the covariance W corresponding to the current moment t is obtained at each update iteration (t|t) , as shown below: W (t|t) =K t G·W (t|t-1) The noise of the original data is suppressed by the above prediction-update iterative mechanism, thereby obtaining a more realistic true value of the original data.

4. The method for online fault location of electrical equipment using multi-band microwave sensing according to claim 1, characterized in that: The step of performing data fusion processing on the denoised data of each multi-band sensor to obtain fused data includes: Assume that the multi-band sensor array contains m sensors in total, and establish n fusion nodes in the sensor array to fuse the denoised data of m sensors. Let a represent the input parameter, and the dimension is d, that is, a∈R d , b represents the fusion result, b∈R, based on j independent distribution observation samples (a1,b1),…,(a j ,b j ) Establish the optimal function f(a) to describe the dependency between a and b. Use the nonlinear function φ to map the domain a to a high-dimensional feature space and perform linear regression in it to obtain the nonlinear regression result of the original space. Use v to represent a vector and c to represent a scalar. Then f(a) is as follows: f(a)=v·φ(a)+c According to the above formula, the problem is to solve v and c based on the known j observation samples; the ε-insensit order function is used as the regression measure function. According to the principle of support vector machine, the regression problem is regarded as a problem that satisfies the structural risk minimization. ak and bk represent the kth observation sample, k∈[1,j], and the corresponding minimization form is As shown below: Introducing the relaxation factor τ into the ε-insensit order function k and Complete the linear fitting of all training samples based on the accuracy ε, as shown below: According to the above formula, The minimization regression problem is expressed as follows: Among them, α is a constant set in advance, and the Lagrangian function is introduced to solve the regression optimization problem. represents the Lagrange coefficient, Lagrangian function In the equation, L is related to μ k 、 Should be maximized, for v, c, τ k 、 To minimize L, the following conditions should be met when obtaining the extreme value of L: Introducing a suitable inner product function ξ(a i ,a k ), i, k∈[1, j], establish μ k 、 Maximize the objective function for: The above formula must satisfy and 0≤μ k , According to the constraints, the interior point algorithm can be used to solve μ k 、 Combined with the partial derivative of v and μ k 、 get According to the above conditions, the fusion data is obtained 5. The method for online fault location of electrical equipment using multi-band microwave sensing according to claim 1, characterized in that: The step of establishing a visual map based on the fused data and extracting feature vectors from the visual map includes: Based on the fused data, use B T and D T They represent the structural transfer matrix when there is electrical fault and no fault at time T, y T Indicates the data transmission status target when there is an electrical fault at time T, z T Indicates the output target in the case of no fault, and the output matrix y in the case of fault at the next moment T+1 T+1 Expressed as: y T+1 =B T y T +D T z T Get the operating structure matrix x under electrical fault conditions kT =K kT y T+1 +w kT , where K kT and w kT Represent different electrical fault operation measurement matrices, k is a positive integer; extract the sample set features after m sensors are fused, detect the number of samples, identify sensor information in the p-th operation system and establish an electrical fault information network, further analyze the electrical operation data and reconstruct the feature model by extracting the electrical fault information features, obtain the eigenvalue, and use the visual map to calculate the node coordinate related parameters in the construction of the electrical fault information network; the electrical fault information feature quantity S iT All satisfy the following formula: Among them, e -αβγ and e -kβγ is the Euler formula form of trigonometric function, α, β, γ are the trigonometric function parameters, To receive the model coefficients; assume that there are arbitrary electrical nodes i and j. If the vector from i to j is a directed vector, it is represented by the matrix [ij] = 1, otherwise it is represented by [ij] = 0; after obtaining the node coordinates and characteristic distribution maps of all electrical equipment, combining the eigenvalues ​​and operating data, an electrical visualization map is established, and the eigenvector is extracted from the image to identify the electrical fault point based on the eigenvector.

6. The method for online fault location of electrical equipment using multi-band microwave sensing according to claim 1, characterized in that: The step of performing node fault identification on the electrical equipment according to the characteristic vector to obtain fault identification results of multiple nodes includes: According to the eigenvector extracted from the electrical visualization map, R is used to represent the eigenvalue of the region where the node is located. Represents the characteristic mean of the region, ΔR represents the characteristic difference, if Then there is no fault point in the area; if R, If ΔR is not equal, it means there is a fault point in the area and the location of the fault point needs to be further determined. The specific method is as follows: when When , the current node has a fault tendency. And when ΔR>R, the node is a fault point; The final identification of the fault point is determined by the assignment method, and R, ΔR, first compare R, like Then skip the assignment and continue with the next step of detection. Otherwise, add 1 to the assignment κ=0, and then compare R and ΔR. If R=ΔR, directly output the assignment result κ=0, otherwise continue to add 1 to the assignment. After the assignment, directly determine the fault situation based on the assignment. If κ=0, there is no fault point in the electrical system. If κ=1, there is a fault tendency in the electrical system, and attention needs to be paid to the identification node. If κ=2, the node has failed and needs to be repaired immediately to avoid electrical equipment collapse.

7. The method for online fault location of electrical equipment using multi-band microwave sensing according to claim 6, characterized in that: The step of predicting the fault development trend of each node of the electrical equipment according to the fault identification result, and redeploying the multi-band sensor array according to the development trend, includes: When κ=1, that is, there is a fault tendency in the electrical system, the multi-band microwave data after Kalman filtering and denoising is extracted, and the corresponding fault level label κ∈{0,1,2}; For a sample with κ = 1 in the historical data, if it evolves to κ = 2 within the next time t, it is marked as a positive example. If it remains at k = 1 or reverts to k = 0, it is marked as a negative example. A sliding window W is used to generate the input sequence [W, D], where D is the feature dimension. The output is the fault evolution label. Extract fault feature S kT and eigenvalue R, ΔR, calculate its time rate of change, the calculation formula is The features were Z-Score normalized to eliminate the difference in frequency band dimensions; Build a two-layer LSTM network model with an input layer dimension of [W, D]. The output layer uses Sigmoid activation to output binary classification probabilities. Configure the Adam optimizer and the binary cross entropy loss function, and add Dropout to prevent overfitting. The two-layer LSTM network model is trained based on the sample data to obtain a trained two-layer LSTM network model. When κ=1 is detected, the real-time window data is collected and input into the model after feature engineering, and the probability P of κ=2 is output; If P is greater than or equal to a preset value, an early warning is triggered, and the double-layer LSTM network model is updated using incremental learning, and the multi-band sensor array is redeployed.

8. An electrical equipment online fault monitoring device with multi-band microwave sensing, characterized in that: An online fault location method for electrical equipment based on multi-band microwave sensing according to any one of claims 1 to 7, wherein the monitoring device comprises a multi-band sensor array, a data acquisition module, an edge computing module, a data transmission module, a fault identification module, a visualization map module, and a change prediction module, wherein the multi-band sensor array is connected to the data acquisition module, the data acquisition module is connected to the edge computing module, the edge computing module is connected to the data transmission module, the data transmission module is connected to the visualization map module, the visualization map module is connected to the fault identification module, and the fault identification module is connected to the change prediction module; The data acquisition module is used to collect microwave reflection / scattering signal data of electrical equipment as raw data through a multi-band sensor array; The edge computing module is used to perform denoising on the raw data of each multi-band sensor and perform data fusion processing on the denoised raw data to obtain fused data; The data transmission module is used to transmit the fused data to the fault identification module to identify and locate the fault of the electrical equipment; The visualization map module is used to create a visualization map based on the fused data and extract feature vectors from the visualization map; The fault identification module is used to perform node fault identification on the electrical equipment according to the feature vector to obtain fault identification results of multiple nodes; The change prediction module is used to predict the fault development trend of each node of the electrical equipment according to the fault identification result, so as to redeploy the multi-band sensor array according to the development trend.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images