Cable fault positioning method and system based on clustering analysis and hybrid neural network model

Abstract

The invention discloses a cable fault positioning method and system based on clustering analysis and a hybrid neural network model, and belongs to the technical field of cable fault positioning. Preprocessing the data; feature extraction and fusion: firstly extracting high-dimensional features, and then fusing features of different scales by adopting a multi-scale feature fusion technology; clustering analysis: processing the fused features by adopting a dynamic clustering algorithm, then identifying a high-density region by adopting a density-based clustering algorithm, avoiding selecting a noise point as an initial clustering center, and outputting an optimized clustering result; and fault diagnosis: firstly performing probability mapping, then estimating the uncertainty of a fault diagnosis result by adopting an uncertainty quantification technology, and outputting a final fault diagnosis result and an uncertainty estimation result. According to the method, efficient, accurate and intelligent positioning of the cable fault is realized, the adaptability is high, credibility evaluation of a fault diagnosis result can be provided, and operation and maintenance personnel can be helped to make a more reliable decision.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of cable fault location, and specifically to a cable fault location method and system based on cluster analysis and a hybrid neural network model. Background Art

[0002] As society's demand for electricity continues to grow, cable fault location has become a critical issue affecting the stability and safety of power systems. Accurately and quickly transmitting power and locating faults are crucial for ensuring stable power system operation. Existing cable fault location methods suffer from low positioning accuracy, poor real-time performance, and insufficient diagnostic and location efficiency. A more efficient and accurate fault location method is needed. Summary of the Invention

[0003] The technical solution of the present invention addresses the technical problem that the existing technical solutions are too single, and provides a solution that is significantly different from the existing technology. It mainly provides a cable fault location method and system based on cluster analysis and hybrid neural network model, which is used to solve the technical problems of low positioning accuracy, poor real-time performance, and insufficient diagnosis and positioning efficiency of the existing cable fault location method proposed in the above background technology.

[0004] The technical solution adopted by the present invention to solve the above technical problems is:

[0005] The cable fault location method based on cluster analysis and hybrid neural network model includes the following steps:

[0006] S1. Data acquisition: collect cable operation data and output original signal data;

[0007] S2, data preprocessing: first filter and reduce noise, then extract the time-frequency characteristics of the signal and output the time-frequency feature data;

[0008] S3. Feature extraction and fusion: First, the deep learning model automatically learns the feature representation of the time-frequency feature data signal and extracts high-dimensional features. Then, multi-scale feature fusion technology is used to fuse features of different scales and output the fused features.

[0009] S4. Cluster analysis: First, a dynamic clustering algorithm is used to process the fused features. Then, a density-based clustering algorithm is used to identify high-density areas, avoid selecting noise points as the initial cluster centers, and output the optimized clustering results.

[0010] S5. Fault diagnosis: First, perform probability mapping, then use uncertainty quantification technology to estimate the uncertainty of the fault diagnosis results, and output the final fault diagnosis results and uncertainty estimation results.

[0011] Furthermore, in step S1, the cable operation data includes a voltage signal, a current signal, and a temperature signal.

[0012] Furthermore, in step S2, noise is removed by a minimum mean square error adaptive filter to extract a valid signal.

[0013] Furthermore, in step S2, short-time Fourier transform or wavelet transform is used to perform time-frequency analysis to extract the time-frequency features of the signal.

[0014] Furthermore, in step S3, the multi-scale feature fusion technology is a cascaded convolution-self-attention module technology.

[0015] Furthermore, in step S3, the fusion strategy includes at least one of weighted fusion, splicing fusion, and attention mechanism.

[0016] Furthermore, in step S4, the dynamic clustering algorithm is a combination of hierarchical clustering and K-means clustering algorithms.

[0017] Furthermore, in step S4, the density-based clustering algorithm calculates the local density and distance of the data points, determines the core points, and assigns the data points to the cluster to which the nearest core point belongs, thereby obtaining an optimized clustering result.

[0018] Furthermore, in step S5, the probability mapping is to map the clustering results into fault probabilities using a Bayesian network or a trust measure to quantify the possibility of a fault occurring;

[0019] And / or, the uncertainty quantification technique is Monte Carlo dropout or Bayesian inference.

[0020] The present invention also provides a cable fault location system based on cluster analysis and hybrid neural network model, which is applied to the above-mentioned cable fault location method based on cluster analysis and hybrid neural network model; the cable fault location system based on cluster analysis and hybrid neural network model includes a data acquisition module, a data preprocessing module, a feature extraction and fusion module, a cluster analysis module and a fault diagnosis module;

[0021] The data acquisition module is used to collect cable operation data;

[0022] The data preprocessing module includes an adaptive filter and a time-frequency analysis unit. The adaptive filter is used to remove noise and extract effective signals. The time-frequency analysis unit uses short-time Fourier transform or wavelet transform to perform time-frequency analysis and extract the time-frequency characteristics of the signal.

[0023] The feature extraction and fusion module includes a feature self-learning unit and a multi-scale feature fusion unit. The feature self-learning unit uses a convolutional neural network or autoencoder in deep learning to automatically learn the feature representation of the signal and extract high-dimensional features. The multi-scale feature fusion unit uses a multi-scale feature fusion technology to perform multi-scale feature fusion.

[0024] The cluster analysis module includes a dynamic clustering algorithm and an improved density clustering algorithm. The dynamic clustering algorithm dynamically adjusts the cluster center; the improved density clustering algorithm uses a density-based clustering algorithm to identify high-density areas by calculating the local density and distance of data points;

[0025] The fault diagnosis module includes a fault probability mapping unit and an uncertain information quantification unit. The fault probability mapping unit uses a Bayesian network or a trust measure to map the clustering results into fault probabilities to quantify the possibility of fault occurrence; the uncertain information quantification unit uses uncertainty quantification technology to estimate the uncertainty of the fault diagnosis results.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] (1) This invention achieves efficient and intelligent cable fault location through technologies such as adaptive filters, time-frequency analysis, feature self-learning and multi-scale feature fusion, improved clustering algorithms, probability mapping, and uncertainty quantification. The system can provide credibility assessments of fault diagnosis results, helping operators make more reliable decisions. Furthermore, it can adaptively adjust filters and clustering algorithms based on different cable operating environments and noise conditions, ensuring stability and reliability in different scenarios, demonstrating its strong adaptability.

[0028] (2) Real-time performance. The present invention significantly improves the speed of signal processing through adaptive filtering and time-frequency analysis technology, ensuring that the system can process cable operation data in real time and quickly respond to fault signals. The adaptive filter can dynamically adjust the filter weight, quickly remove noise, and extract effective signals. The time-frequency analysis unit captures the time-frequency characteristics of the signal in real time, ensuring the rapid identification and location of the fault signal. This real-time performance enables the system to quickly locate the fault point in the early stages of a cable fault, reducing the impact of the fault on the power system.

[0029] (3) Robustness. The present invention uses a deep learning model through feature self-learning technology to automatically learn the high-dimensional features of the signal and adapt to different environments and noise conditions. The feature self-learning technology can extract fault-related features from complex cable operation data, avoiding the limitations of manually designed features in traditional methods. The present invention can estimate the uncertainty of fault diagnosis results through uncertain information quantification technology. This technology enhances the system's tolerance to noise and abnormal data, ensuring that high diagnostic accuracy can be maintained in different operating environments. The robustness of the system enables it to operate stably in a complex power network environment and adapt to various noises and interferences.

[0030] (4) High-precision positioning. The present invention can fuse features of different scales through multi-scale feature fusion technology to improve the discrimination and robustness of features; the multi-scale feature fusion unit can extract signal features from different levels to ensure the comprehensiveness and accuracy of fault features. The present invention combines the advantages of dynamic clustering and density clustering through an improved clustering algorithm, and can dynamically adjust the cluster center according to the data distribution to avoid the interference of noise points and improve the accuracy of clustering; the improved density clustering algorithm can effectively identify high-density areas by calculating local density and distance, and avoid misjudging noise points as cluster centers. This high-precision positioning capability enables the system to accurately identify the location of cable faults, reduce false alarms and missed alarms, and improve the accuracy of fault diagnosis.

[0031] (5) Reliability. The present invention can map the clustering results into fault probabilities through fault probability mapping technology, quantifying the possibility of fault occurrence; the fault probability mapping unit can generate a fault probability map, intuitively display the fault location, and help operation and maintenance personnel quickly locate the fault point. Uncertainty quantification technology can estimate the uncertainty of fault diagnosis results through Monte Carlo dropout or Bayesian reasoning, providing more reliable decision support; the uncertainty quantification unit can generate uncertainty maps to help operation and maintenance personnel evaluate the credibility of fault diagnosis results and make more reliable decisions. This reliability enables the system to provide stable and reliable fault diagnosis results in a complex power network environment, helping operation and maintenance personnel to quickly locate and repair faults.

[0032] The present invention will be explained in detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a structural block diagram of the cable fault location system based on cluster analysis and hybrid neural network model in the present invention;

[0034] Figure 2 This is a flow chart of the cable fault location method based on cluster analysis and hybrid neural network model in the present invention. DETAILED DESCRIPTION

[0035] To facilitate understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in different forms and is not limited to the embodiments described in the text. On the contrary, these embodiments are provided to make the content disclosed in the present invention more thorough and comprehensive.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly associated with those skilled in the art to which the present invention pertains. The terminology used herein in the specification of the present invention is for the purpose of describing specific embodiments and is not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0037] Example: Please refer to the attached Figure 1 , a cable fault location system based on cluster analysis and hybrid neural network model, including:

[0038] 1. Data acquisition module

[0039] Function: Collect cable operation data, including voltage, current, temperature and other signals.

[0040] Output: raw signal data.

[0041] 2. Data preprocessing module

[0042] Includes adaptive filter and time-frequency analysis unit.

[0043] ① Adaptive filter:

[0044] Function: Remove noise and extract valid signals through the least mean square error (LMS) adaptive filter.

[0045] Input: original signal data (output by data acquisition module).

[0046] Output: filtered signal.

[0047] Specifically, the structure of the adaptive filter includes:

[0048] a. Input signal area

[0049] Function: Receive raw signal data, which may contain noise and useful signals.

[0050] Output: original signal x(n).

[0051] b. Adaptive filtering area

[0052] Function: Minimize the error signal e(n) by adjusting the filter weight vector w(n).

[0053] Input: original signal x(n) and error signal e(n).

[0054] Output: The filtered signal y(n) and the updated weight vector w(n+1) [updated according to the original signal x(n) and the error signal e(n)].

[0055] The calculation formula is: y(n) = w T (n)·x(n), where T represents the transpose operation of the vector.

[0056] Specifically: y(n) is the output signal at time n; w(n) is the filter weight vector at time n; x(n) is the input vector at time n.

[0057] To calculate the dot product of two vectors, one of the vectors needs to be transposed first. Assume that w(n) and x(n) are both two-dimensional column vectors:

[0058]

[0059] The transpose w of w(n) T (n) is:

[0060] w T (n) = [w1(n) w2(n)]

[0061] Will w T Multiply x(n) by x(n):

[0062]

[0063] This is exactly the definition of a dot product. Therefore, T represents the transpose of the vector so that w(n) can be dot-producted with x(n).

[0064] c. Error calculation area

[0065] Function: Calculate the error between the filter output signal y(n) and the expected signal d(n).

[0066] Input: filter output y(n) and desired signal d(n), where the desired signal d(n) is usually extracted from the reference signal.

[0067] Output: error signal e(n).

[0068] The calculation formula is: e(n)=d(n)-y(n).

[0069] d. Weight update area

[0070] Function: Update the filter weight vector according to the error signal e(n) and the input signal x(n).

[0071] Input: error signal e(n) and input signal x(n).

[0072] Output: Updated weight w(n+1).

[0073] The calculation formula is: w(n+1)=w(n)-μ·e(n)·x(n), where μ is the learning rate parameter that controls the step size of weight update.

[0074] ②Time-frequency analysis unit:

[0075] Function: Use short-time Fourier transform (STFT) or wavelet transform to perform time-frequency analysis and extract the time-frequency characteristics of the signal.

[0076] Input: Filtered signal.

[0077] Output: time-frequency feature data.

[0078] Specifically, the structure of the time-frequency analysis unit includes:

[0079] a. Input signal area

[0080] Function: Receive the filtered signal y(n), which has been partially noise-removed by the adaptive filter.

[0081] Output: filtered signal y(n).

[0082] b. Time-frequency analysis area

[0083] Function: Convert the signal from the time domain to the time-frequency domain and extract the time-frequency characteristics of the signal. Time-frequency analysis can use short-time Fourier transform (STFT) or wavelet transform (WT).

[0084] Input: Filtered signal y(n).

[0085] Output: time-frequency feature map.

[0086] b1. Short-time Fourier transform (STFT) region

[0087] Function: Decomposes a signal into a representation in the time-frequency domain using the Short-Time Fourier Transform (STFT). The STFT performs a Fourier transform on the signal at each time point, obtaining the frequency components of the signal at different time points.

[0088] Input: Filtered signal y(n).

[0089] Output: time-frequency graph S(t,f), where t represents time and f represents frequency.

[0090] Calculation formula:

[0091]

[0092] Where w(t) is the window function, usually a Hanning window or a Gaussian window.

[0093] b2. Wavelet transform (optional)

[0094] Function: Decomposes a signal into multi-scale time-frequency representations through wavelet transform, which is applicable to non-stationary signals. Wavelet transform can provide more flexible time and frequency resolution.

[0095] Input: Filtered signal y(n).

[0096] Output: Wavelet time-frequency diagram W(t,s), where t represents time and s represents scale.

[0097] Calculation formula:

[0098]

[0099] Among them, ψ(t) is the wavelet basis function and s is the scale parameter.

[0100] 3. Feature extraction and fusion module

[0101] It includes feature self-learning unit and multi-scale feature fusion unit.

[0102] ①Feature self-learning unit:

[0103] Function: Use convolutional neural networks (CNNs) or autoencoders in deep learning to automatically learn signal feature representations and extract high-dimensional features.

[0104] Input: time-frequency feature data (output by the data preprocessing module).

[0105] Output: high-dimensional features.

[0106] ②Multi-scale feature fusion unit:

[0107] Function: Use multi-scale feature fusion technology, such as the cascaded convolution-self-attention module (FIF-UNet), to perform multi-scale feature fusion and improve feature robustness.

[0108] Input: high-dimensional features.

[0109] Output: fused features.

[0110] Specifically, the multi-scale feature fusion unit includes:

[0111] a. Input signal area

[0112] Function: Receive high-dimensional features output from the feature extraction module. These features may contain information of different scales.

[0113] Output: high-dimensional features F.

[0114] b. Multi-scale feature extraction

[0115] Function: Extract features of different scales through multiple convolutional layers. Each convolutional layer extracts feature maps of different scales.

[0116] Input: high-dimensional features F.

[0117] Output: multi-scale features F1, F2, ...F n .

[0118] Calculation formula:

[0119] F i =Conv(F,W i )

[0120] Among them, W i is the weight of the i-th convolutional layer.

[0121] c. Feature fusion module

[0122] Function: Fuse features of different scales to improve the robustness and discrimination of features.

[0123] Input: multi-scale features F1, F2, ...F n .

[0124] Output: fused feature F 融合 .

[0125] c1. Upsampling and downsampling

[0126] Function: Adjust the size of features of different scales through upsampling and downsampling operations for fusion.

[0127] Input: multi-scale features F1, F2, ...F n .

[0128] Output: Resized feature F1 ＇ , F2 ＇ ,...,F n ＇ .

[0129] Calculation formula:

[0130] F i ＇ =Upsample(F i ) or F i ＇ =Downsample(F i )

[0131] c2. Fusion strategy

[0132] Function: Use strategies such as weighted fusion, splicing fusion, or attention mechanism to fuse the resized features.

[0133] Input: Resized feature F1 ＇ , F2 ＇ ,...,F n ＇ .

[0134] Output: fused feature F 融合 .

[0135] Calculation formula:

[0136] Weighted fusion:

[0137]

[0138] Among them, α i is the weight coefficient.

[0139] Splicing and fusion:

[0140] F 融合 =Concat(F1 ＇ , F2 ＇ ,...,F n ＇ )

[0141] Attention Mechanism:

[0142]

[0143] 4. Cluster analysis module

[0144] Dynamic clustering algorithm and improved density clustering algorithm are used.

[0145] ①Dynamic clustering algorithm:

[0146] Function: Combines hierarchical clustering and K-means clustering algorithms to dynamically adjust cluster centers and improve clustering accuracy.

[0147] Input: fused features.

[0148] Output: clustering results.

[0149] ② Improved density clustering algorithm:

[0150] Function: Use density-based clustering algorithms (such as DBSCAN) to identify high-density areas by calculating the local density and distance of data points, avoiding selecting noise points as initial cluster centers.

[0151] Input: Clustering results.

[0152] Output: optimized clustering results.

[0153] Specifically, the improved density clustering algorithm includes:

[0154] a. Input data

[0155] Function: Receive the data points to be clustered, which can be multidimensional feature vectors.

[0156] Output: data point set D.

[0157] b. Calculate local density

[0158] Function: Calculate the local density of each data point, which reflects the number of neighbors around the data point.

[0159] Input: A set of data points D.

[0160] Output: local density ρ i .

[0161] Calculation formula:

[0162]

[0163] Where θ(x) is the step function, ∈ is the cutoff distance, and d ij is the Euclidean distance between data points i and j.

[0164] c. Calculate distance

[0165] Function: Calculate the distance from each data point to other data points, usually using Euclidean distance.

[0166] Input: A set of data points D.

[0167] Output: distance matrix δ ij .

[0168] Calculation formula:

[0169]

[0170] Among them, x ik and x jk are the coordinates of data points i and j in the kth dimension, and M is the dimension of the feature.

[0171] d. Determine the core points

[0172] Function: Determine core points based on local density and distance. Core points are points with high local density and far away from other high-density points.

[0173] Input: local density ρ i and the distance matrix δij .

[0174] Output: core point set C.

[0175] Calculation formula:

[0176] γ i =ρ i ·δ i

[0177] Among them, δ i The data point i has a local density greater than ρ i The core point is γ i Points with larger values.

[0178] e. Cluster assignment

[0179] Function: Assign data points to the cluster to which the nearest core point belongs to form the final clustering result.

[0180] Input: core point set C and data point set D.

[0181] Output: optimized clustering result Cluster.

[0182] Calculation formula:

[0183]

[0184] Among them, Cluster(i) is the cluster to which data point i belongs, d ic is the distance from data point i to core point c.

[0185] 5. Fault diagnosis module

[0186] It includes a fault probability mapping unit and an uncertain information quantification unit.

[0187] ①Fault probability mapping unit:

[0188] Function: Use Bayesian networks or trust measures to map clustering results into failure probabilities and quantify the likelihood of failures.

[0189] Input: Optimized clustering results (output by the clustering analysis module).

[0190] Output: Failure probability.

[0191] Specifically, the fault probability mapping unit includes:

[0192] a. Input data

[0193] Function: Receive the optimized clustering results output by the cluster analysis module, which contain the cluster labels of the data points.

[0194] Output: optimized clustering result Cluster.

[0195] b. Fault feature extraction

[0196] Function: Extract fault-related features from clustering results. These features are used to calculate the fault probability.

[0197] Input: Optimized clustering result Cluster.

[0198] Output: Fault features.

[0199] Calculation formula:

[0200] Features={Feature1, Feature2,..., Feature n}

[0201] Among them, each feature i It can be cluster labels, local density of data points, distance, etc.

[0202] c. Failure probability calculation

[0203] Function: Use Bayesian networks or trust measures to map fault characteristics to fault probabilities and quantify the likelihood of fault occurrence.

[0204] Input: Fault features.

[0205] Output: Fault probability P(Fault).

[0206] Calculation formula:

[0207]

[0208] Among them, P(Features Fault) is the probability of the feature appearing under fault conditions, P(Fault) is the prior probability of fault, and P(Features) is the marginal probability of the feature.

[0209] d. Probability Mapping

[0210] Function: Map the fault probability to the specific fault location or area to generate a fault probability map.

[0211] Input: Fault probability P(Fault) and location information of data points.

[0212] Output: Fault Map.

[0213] Calculation formula:

[0214] Fault Map(x,y)=P(Fault|Features(x,y))

[0215] Among them, (x,y) is the location coordinate of the data point, and Features(x,y) is the fault feature at that location.

[0216] ② Uncertain information quantification unit:

[0217] Function: Use uncertainty quantification techniques, such as Monte Carlo dropout or Bayesian inference, to estimate the uncertainty of fault diagnosis results and provide more reliable decision support.

[0218] Input: Failure probability.

[0219] Output: final fault diagnosis results and uncertainty estimation results.

[0220] Specifically, the uncertain information quantification unit includes:

[0221] a. Input data

[0222] Function: Receive the fault probability map output by the fault probability mapping unit. This data contains the fault probability of each location.

[0223] Output: Fault Map.

[0224] b. Uncertainty estimation

[0225] Function: Use Monte Carlo dropout or Bayesian inference to estimate the uncertainty of fault diagnosis results.

[0226] Input: Fault Map.

[0227] Output: Uncertainty Estimate.

[0228] b1. Monte Carlo dropout

[0229] Function: Through multiple forward propagations, use dropout to randomly discard some neurons and estimate the uncertainty of the model output.

[0230] Input: Fault Map.

[0231] Output: Uncertainty Estimate.

[0232] Calculation formula:

[0233] Uncertainty Estimate

[0234] =Var({P(Fault|Features)1,P(Fault|Features)2,...,P(Fault|Features) t})

[0235] Where t is the number of forward propagation, P(Fault|Features) t is the failure probability of the tth forward propagation.

[0236] b2. Bayesian Inference

[0237] Function: Use Bayesian methods to estimate the uncertainty of model output through prior and posterior distributions.

[0238] Input: Fault Map.

[0239] Output: Uncertainty Estimate.

[0240] Calculation formula:

[0241] Uncertainty Estimate=∫P(Fault|Features,θ)·P(θ|Data)dθ

[0242] Here, θ is the model parameter and P(θ|Data) is the posterior distribution of the parameter.

[0243] c. Uncertainty map generation

[0244] Function: Map the uncertainty estimation results to specific fault locations or areas to generate uncertainty maps.

[0245] Input: Uncertainty Estimate and location information of data points.

[0246] Output: Uncertainty Map.

[0247] Calculation formula:

[0248] Uncertainty Map(x,y)=Uncertainty Estimate(x,y)

[0249] Where (x,y) is the location coordinate of the data point, and Uncertainty Estimate(x,y) is the uncertainty estimate of the location.

[0250] Please refer to the attached Figure 2 A cable fault location method based on cluster analysis and hybrid neural network model using the above system includes the following steps:

[0251] S1. Data acquisition: collect cable operation data, including voltage, current, temperature and other signals, and output original signal data;

[0252] S2. Data preprocessing: For the original signal data, remove noise through LMS adaptive filter, extract the effective signal, and output the filtered signal; then, for the filtered signal, use time-frequency analysis technology to convert the signal from time domain to time-frequency domain, extract the time-frequency characteristics of the signal, and output the time-frequency characteristic data;

[0253] In this step, the LMS criterion is used to find the optimal weight coefficients and minimum mean square error, which can effectively suppress sidelobe effects and adapt to different signal and noise environments. The time-frequency characteristics of the signal are extracted to facilitate better analysis and processing in subsequent steps.

[0254] S3. Feature extraction and fusion: For time-frequency feature data, high-dimensional features are extracted by automatically learning the feature representation of the signal using CNN or autoencoders. Then, for the high-dimensional features, multi-scale feature fusion technology is used to fuse features of different scales and output the fused features.

[0255] In this step, automatically learning the feature representation of the signal can improve the robustness and discrimination of the features; fusing features of different scales can improve the model's ability to capture features of different scales and improve the robustness of the features.

[0256] S4. Cluster analysis: Based on the fused features, hierarchical clustering and K-means clustering algorithms are combined to dynamically adjust the cluster centers and output the clustering results. Then, based on the clustering results, a density-based clustering algorithm is used to identify high-density areas, avoid selecting noise points as the initial cluster centers, and output the optimized clustering results.

[0257] In this step, the cluster centers are adjusted dynamically to improve the accuracy and robustness of clustering.

[0258] S5. Fault diagnosis: Based on the optimized clustering results, use Bayesian networks or trust measures to map the optimized clustering results into fault probabilities, quantify the possibility of fault occurrence, and output the fault probability. Then, based on the fault probability, use uncertainty quantification technology to estimate the uncertainty of the fault diagnosis results, and output the final fault diagnosis results and uncertainty estimation results.

[0259] In this step, the clustering results are mapped to fault probabilities, quantifying the possibility of fault occurrence, which can improve the reliability of diagnosis; and estimating the uncertainty of fault diagnosis results can provide more reliable decision support.

[0260] The cable fault location method and system provided by the present invention have technical characteristics such as real-time, robust, high-precision positioning and reliability, and realize efficient and accurate positioning of cable faults, and have significant technical advantages and wide application value.

[0261] The above description of the present invention is exemplified in conjunction with the accompanying drawings. It is obvious that the specific implementation of the present invention is not limited to the above-mentioned method. As long as such non-substantial improvements are made using the method concept and technical solution of the present invention, or the concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the scope of protection of the present invention.

Claims

1. A cable fault location method based on cluster analysis and hybrid neural network model, characterized by: The steps include: S1. Data acquisition: collect cable operation data and output original signal data; S2, data preprocessing: first filter and reduce noise, then extract the time-frequency characteristics of the signal and output the time-frequency feature data; S3. Feature extraction and fusion: First, we use a deep learning model to automatically learn the feature representation of the time-frequency feature data signal and extract high-dimensional features. Then, we use multi-scale feature fusion technology to fuse features of different scales and output the fused features. S4. Cluster analysis: First, a dynamic clustering algorithm is used to process the fused features. Then, a density-based clustering algorithm is used to identify high-density areas, avoid selecting noise points as the initial cluster centers, and output the optimized clustering results. S5. Fault diagnosis: First, perform probability mapping, then use uncertainty quantification technology to estimate the uncertainty of the fault diagnosis results, and output the final fault diagnosis results and uncertainty estimation results.

2. The cable fault location method based on cluster analysis and hybrid neural network model according to claim 1, characterized in that: In step S1, the cable operation data includes a voltage signal, a current signal, and a temperature signal.

3. The cable fault location method based on cluster analysis and hybrid neural network model according to claim 1, characterized in that: In step S2, the noise is removed by a minimum mean square error adaptive filter to extract the effective signal.

4. The cable fault location method based on cluster analysis and hybrid neural network model according to claim 1, characterized in that: In step S2, short-time Fourier transform or wavelet transform is used to perform time-frequency analysis to extract the time-frequency characteristics of the signal.

5. The cable fault location method based on cluster analysis and hybrid neural network model according to claim 1, characterized in that: In step S3, the multi-scale feature fusion technology is a cascaded convolution-self-attention module technology.

6. The cable fault location method based on cluster analysis and hybrid neural network model according to claim 1, characterized in that: In step S3, the fusion strategy includes at least one of weighted fusion, splicing fusion, and attention mechanism.

7. The cable fault location method based on cluster analysis and hybrid neural network model according to claim 1, characterized in that: In step S4, the dynamic clustering algorithm is a combination of hierarchical clustering and K-means clustering algorithms.

8. The cable fault location method based on cluster analysis and hybrid neural network model according to claim 1, characterized in that: In step S4, the density-based clustering algorithm calculates the local density and distance of the data points, determines the core points, and assigns the data points to the cluster to which the nearest core point belongs, thereby obtaining an optimized clustering result.

9. The cable fault location method based on cluster analysis and hybrid neural network model according to claim 1, characterized in that: In step S5, the probability mapping is to map the clustering results into fault probabilities using a Bayesian network or a trust measure to quantify the possibility of a fault occurring; And / or, the uncertainty quantification technique is Monte Carlo dropout or Bayesian inference.

10. A cable fault location system based on cluster analysis and a hybrid neural network model, applied to the cable fault location method based on cluster analysis and a hybrid neural network model according to any one of claims 1 to 9; characterized in that: The cable fault location system based on cluster analysis and hybrid neural network model includes data acquisition module, data preprocessing module, feature extraction and fusion module, cluster analysis module and fault diagnosis module; The data acquisition module is used to collect cable operation data; The data preprocessing module includes an adaptive filter and a time-frequency analysis unit. The adaptive filter is used to remove noise and extract effective signals. The time-frequency analysis unit uses short-time Fourier transform or wavelet transform to perform time-frequency analysis and extract the time-frequency characteristics of the signal. The feature extraction and fusion module includes a feature self-learning unit and a multi-scale feature fusion unit. The feature self-learning unit uses a convolutional neural network or autoencoder in deep learning to automatically learn the feature representation of the signal and extract high-dimensional features. The multi-scale feature fusion unit uses a multi-scale feature fusion technology to perform multi-scale feature fusion. The cluster analysis module includes a dynamic clustering algorithm and an improved density clustering algorithm. The dynamic clustering algorithm dynamically adjusts the cluster center; the improved density clustering algorithm uses a density-based clustering algorithm to identify high-density areas by calculating the local density and distance of data points; The fault diagnosis module includes a fault probability mapping unit and an uncertain information quantification unit. The fault probability mapping unit uses a Bayesian network or a trust measure to map the clustering results into fault probabilities and quantify the possibility of fault occurrence. The uncertainty information quantization unit uses uncertainty quantification technology to estimate the uncertainty of the fault diagnosis result.

Images