A power distribution line fault diagnosis method, system, device and medium

Abstract

This invention belongs to the field of power line fault detection technology and discloses a method, system, equipment, and medium for diagnosing power distribution line faults. The method includes: acquiring initial fault signal data corresponding to the power distribution line, preprocessing the initial fault signal data, and outputting fault signal data; decomposing the fault signal data using joint decomposition technology to obtain intrinsic mode components (IMCs); introducing information mining technology to extract the distribution change feature vector of the fault signal and fusing it with the IMCs to obtain a change image; extracting local features of the change image through a convolutional neural network and combining it with a strategy function optimized by a Huffman tree for fault classification, and outputting the fault diagnosis result. This invention applies the Softmax strategy of Huffman trees, improving the efficiency and accuracy of judgment, enabling early identification and warning of fault signals in power distribution lines, and improving the efficiency of power distribution line fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of line fault detection technology, and in particular to a method, system, equipment and medium for diagnosing power distribution line faults. Background Technology

[0002] Extracting early fault characteristics of power distribution lines and related equipment is a key technology for improving the operation and maintenance level and power supply reliability of power distribution systems. Accurate identification and early warning of early faults are an important foundation for preventing the expansion of faults and ensuring power supply reliability.

[0003] Early defects in power distribution equipment, such as insulator damage, overheating, or mechanical damage, will develop into permanent faults if not addressed in a timely manner. Based on existing research and practice in the field of power distribution equipment condition monitoring, online monitoring devices deployed at line nodes, switchgear, and transformers can continuously collect multi-dimensional physical quantities reflecting the health status of the equipment and transmit them to the monitoring platform through data communication networks, providing data support for early fault warning.

[0004] Therefore, most current fault diagnosis methods are based on data such as temperature, current, and vibration collected by monitoring terminals. They use feature engineering algorithms to identify potential hazards. Specifically, by performing time-domain, frequency-domain, or time-frequency-domain analysis on the monitoring signals, indicators such as temperature rise trend, harmonic distortion rate, discharge pulse frequency, and vibration spectrum characteristics are extracted to construct equipment status feature vectors. Classification or regression models are then used to determine whether there are potential hazards and their severity.

[0005] However, the features extracted by this method often rely on expert experience design, which has limited adaptability to complex operating conditions and new fault modes. Moreover, single features are easily affected by on-site interference, resulting in insufficient early warning accuracy and robustness. Another approach is based on deep learning end-to-end feature learning, which automatically mines sensitive representations directly from the original monitoring data. This method can avoid the limitations of manual feature design, but it has high requirements for the number and quality of samples, and the model interpretability is weak. In practical applications, its generalization ability and reliability still need further verification. Overall, early fault hazard feature extraction technology is developing towards a combination of multi-source data fusion, adaptive feature learning, and highly interpretable diagnosis to meet the need for accurate detection of hidden defects in the complex operating environment of the power distribution network. In general, as an important part of the power system, power distribution lines are exposed outdoors for a long time and are easily affected by environmental factors, equipment aging, and other factors, resulting in frequent line defects. Traditional power distribution line fault diagnosis methods mainly rely on manual inspection, which is inefficient and poses safety hazards.

[0006] Therefore, how to provide a method, system, equipment, and medium for diagnosing power distribution line faults is an urgent problem to be solved. Summary of the Invention

[0007] This invention provides a method, system, device, and medium for diagnosing power distribution line faults to solve the problems mentioned above in the prior art.

[0008] According to a first aspect of the present invention, a method for diagnosing faults in power distribution lines is provided.

[0009] In one embodiment, the power distribution line fault diagnosis method includes:

[0010] Based on the simulation operation results of the power distribution line, the initial fault signal data corresponding to the power distribution line is collected, and the initial fault signal data is preprocessed and the fault signal data is output.

[0011] The fault signal data is decomposed using joint decomposition technology to obtain intrinsic mode components; information mining technology is introduced to extract the distribution change feature vector of the fault signal, and then fused with the intrinsic mode components to obtain the change image.

[0012] The system extracts local features from the changing image using a convolutional neural network, and combines this with a policy function optimized by a Huffman tree to classify faults and output fault diagnosis results.

[0013] In one embodiment, initial fault signal data corresponding to the power distribution line is collected based on the simulation operation results of the power distribution line, and the initial fault signal data is preprocessed to output fault signal data including:

[0014] The power simulation technology is used to simulate the actual operating conditions of power distribution lines, such as early faults and sudden load changes. The voltage and current data of each node of the power distribution line are monitored in real time during the simulation.

[0015] During the simulation, the environment of the power distribution line is monitored to obtain temperature and humidity, and the initial fault signal data is obtained by combining voltage and current data.

[0016] The initial operating parameters are described using the minimum, first quartile, median, third quartile, and maximum value. The description results are arranged from smallest to largest, and abnormal data in the initial fault signal data are removed by combining the outlier detection formula.

[0017] After outlier removal, the initial fault signal data is processed by moving average filtering to eliminate noise and output fault signal data.

[0018] In one embodiment, the fault signal data is decomposed using joint decomposition technology to obtain intrinsic mode components; information mining technology is introduced to extract the distribution variation feature vector of the fault signal, and fused with the intrinsic mode components to obtain a variation image, including:

[0019] The fault signal data is reconstructed into the modal components of the current order and the residual signal. Based on the decomposition results, three joint constraint criteria are established to constrain the spectral uncorrelation between the modal components of the current order and the historically determined modes, so as to ensure that the decomposed intrinsic modal components are mutually non-interfering in the frequency domain.

[0020] The extracted modal components are transformed into a constrained minimization function by combining the equilibrium parameters of the three joint constraint criteria. An augmented Lagrange function is constructed by introducing a Lagrange operator. The alternating multiplier algorithm is used to solve the minimization function and iteratively update the modal components, center frequency and Lagrange multipliers.

[0021] Based on the variance and convergence parameters of the fault signal data, the iteration termination condition is set, and when the iteration update result meets the iteration termination condition, the extraction of the current order intrinsic mode components is completed. The updated residual signal is used as a new processing object to extract all valid intrinsic mode components in the fault signal data step by step.

[0022] The fault signal data is coarsely processed, and a feature vector for extracting the distribution change of the fault signal data is formed based on the processing results. This vector is then fused with the intrinsic mode components to output an entropy change image.

[0023] In one embodiment, the extracted modal components are transformed into a constrained minimization function by combining the equilibrium parameters of three joint constraint criteria, and an augmented Lagrange function is constructed by introducing a Lagrange operator. The alternating multiplier algorithm is used to solve the minimization function, and the modal components, center frequency, and Lagrange multipliers are iteratively updated, including:

[0024] A balance factor based on three joint constraint criteria is introduced to establish a minimization objective function, and the fault signal data reconstruction equation is used as the only constraint condition to transform the extracted modal components into a constrained minimization function.

[0025] Expand the residual signal in the minimization function into the objective form to obtain the equality-constrained residual term, and construct the augmented Lagrange function based on the equality-constrained residual term, which includes the dual term of the introduced Lagrange multipliers and the quadratic penalty term.

[0026] Based on the partial derivatives of the modal components in the augmented Lagrange function with respect to the Lagrange multipliers and the previous modal components, the frequency domain closed-form solution of the modal components of the target wheel iteration is derived, and the center frequency is derived by taking the partial derivative with respect to the center frequency in the augmented Lagrange function.

[0027] The Lagrange multipliers are updated with a fixed step size along the gradient ascent direction of the dual function. The frequency domain closed-form update solution of the Lagrange multipliers is obtained by combining the residuals, thus completing the iterative update of the modal components, center frequency and Lagrange multipliers.

[0028] In one embodiment, coarse-grained processing of fault signal data is performed, and a distribution change feature vector for extracting fault signal data is formed based on the processing result. This vector is then fused with intrinsic mode components to output an entropy change image.

[0029] Based on the length of the fault signal data, the embedding dimension and similarity tolerance are set as parameters. The fault signal data is coarsened on a scale-by-scale, and the fault signal data at each scale is divided according to the length to construct composite coarsened sequences at different scales.

[0030] Based on the composite coarse-grained sequence at the same scale, composite coarse-grained sequence parameters with different offsets are generated. The mean of the composite coarse-grained sequence parameters is calculated. The composite multi-scale entropy value is defined according to the logarithm of the mean ratio. The composite multi-scale entropy values ​​corresponding to all scales are arranged in the target order to form a distribution change feature vector for extracting fault signal data.

[0031] The distribution variation characteristics of different intrinsic modal components are fused by direct connection or multiplication combination. The integrity verification technology is applied to judge the loss of modal features during the fusion process. Based on the judgment result, the entropy change image from non-faulty phase to faulty phase is output.

[0032] In one embodiment, local features of the changed image are extracted using a convolutional neural network, and a fault classification is performed using a policy function optimized by a Huffman tree. The output fault diagnosis results include:

[0033] The convolution kernel of a convolutional neural network is slid across an image with entropy changes. Through the convolution operation and the sliding process, a pixel-by-pixel weighted summation is performed on the image with entropy changes, and local features of the image with entropy changes are extracted based on the summation result.

[0034] By introducing an input gate, a forget gate, and an output gate to form a gating mechanism to capture the time-dependent characteristics of fault signal data, a Huffman tree is constructed based on the occurrence probability of fault categories in the power distribution line in the preset training set, and the output layer structure of the policy function is optimized using the Huffman tree.

[0035] Based on the optimized policy function, fault classification is performed by combining local features and temporal dependency features of the entropy change image, and fault diagnosis results are output.

[0036] In one embodiment, constructing a Huffman tree based on the occurrence probability of power line fault categories in a preset training set, and optimizing the output layer structure of the policy function using the Huffman tree includes:

[0037] Based on the occurrence frequency of each fault category in the preset training set, a leaf node is created for each fault category. The data structure of the leaf node contains the fault category tag and the occurrence frequency. All leaf nodes are put into the minimum priority queue for iterative construction of a Huffman tree.

[0038] Starting from the root node of the Huffman tree, perform a depth-first traversal of the Huffman tree to assign a binary code to each leaf node, and specify the marker value for moving from the root node to the left and right leaf nodes to define the Huffman code corresponding to each leaf node, thus determining the binary sequence on the path from the root node to the leaf node.

[0039] The output layer structure of the policy function is optimized based on binary sequence optimization, so that the fault type output by the policy function can be accessed through a unique path from the root node to the leaf node, thus completing the optimization of the policy function.

[0040] According to a second aspect of the present invention, a power distribution line fault diagnosis system is provided.

[0041] In one embodiment, the power distribution line fault diagnosis system includes:

[0042] The data acquisition and processing unit is used to acquire the initial fault signal data corresponding to the power distribution line based on the simulation operation results of the power distribution line, preprocess the initial fault signal data, and output the fault signal data.

[0043] The decomposition and fusion processing unit is used to decompose the fault signal data using joint decomposition technology to obtain the intrinsic mode components; information mining technology is introduced to extract the distribution change feature vector of the fault signal, and then fused with the intrinsic mode components to obtain the change image.

[0044] The diagnostic result output unit is used to extract local features of the changed image through a convolutional neural network, and combine them with a policy function optimized by a Huffman tree to perform fault classification and output fault diagnosis results.

[0045] In one embodiment, the data acquisition and processing unit includes:

[0046] The power simulation technology is used to simulate the actual operating conditions of power distribution lines, such as early faults and sudden load changes. The voltage and current data of each node of the power distribution line are monitored in real time during the simulation.

[0047] During the simulation, the environment of the power distribution line is monitored to obtain temperature and humidity, and the initial fault signal data is obtained by combining voltage and current data.

[0048] The initial operating parameters are described using the minimum, first quartile, median, third quartile, and maximum value. The description results are arranged from smallest to largest, and abnormal data in the initial fault signal data are removed by combining the outlier detection formula.

[0049] After outlier removal, the initial fault signal data is processed by moving average filtering to eliminate noise and output fault signal data.

[0050] In one embodiment, the decomposition and fusion processing unit includes:

[0051] The criteria establishment module is used to reconstruct fault signal data into the modal components of the current order and the residual signal, and establish three joint constraint criteria based on the decomposition results to constrain the spectral uncorrelation between the modal components of the current order and the historically determined modes, so as to ensure that the decomposed intrinsic modal components are mutually non-interfering in the frequency domain.

[0052] The iterative update module is used to combine the equilibrium parameters of the three joint constraint criteria to transform the extracted modal components into a constrained minimization function, and to introduce the Lagrange operator to construct the augmented Lagrange function. The alternating multiplier algorithm is used to solve the minimization function to iteratively update the modal components, center frequency and Lagrange multipliers.

[0053] The intrinsic mode component output module is used to set the iteration termination condition based on the variance and convergence parameters of the fault signal data, and to complete the extraction of the current order intrinsic mode components when the iteration update result meets the iteration termination condition. The updated residual signal is used as a new processing object to extract all valid intrinsic mode components in the fault signal data step by step.

[0054] The change image output module is used to coarsely process the fault signal data, form a distribution change feature vector for extracting the fault signal data based on the processing results, and fuse it with the intrinsic mode components to output an entropy value change image.

[0055] In one embodiment, the extracted modal components are transformed into a constrained minimization function by combining the equilibrium parameters of three joint constraint criteria, and an augmented Lagrange function is constructed by introducing a Lagrange operator. The alternating multiplier algorithm is used to solve the minimization function, and the modal components, center frequency, and Lagrange multipliers are iteratively updated, including:

[0056] A balance factor based on three joint constraint criteria is introduced to establish a minimization objective function, and the fault signal data reconstruction equation is used as the only constraint condition to transform the extracted modal components into a constrained minimization function.

[0057] Expand the residual signal in the minimization function into the objective form to obtain the equality-constrained residual term, and construct the augmented Lagrange function based on the equality-constrained residual term, which includes the dual term of the introduced Lagrange multipliers and the quadratic penalty term.

[0058] Based on the partial derivatives of the modal components in the augmented Lagrange function with respect to the Lagrange multipliers and the previous modal components, the frequency domain closed-form solution of the modal components of the target wheel iteration is derived, and the center frequency is derived by taking the partial derivative with respect to the center frequency in the augmented Lagrange function.

[0059] The Lagrange multipliers are updated with a fixed step size along the gradient ascent direction of the dual function. The frequency domain closed-form update solution of the Lagrange multipliers is obtained by combining the residuals, thus completing the iterative update of the modal components, center frequency and Lagrange multipliers.

[0060] According to a third aspect of the present invention, a computer device is provided.

[0061] In some embodiments, the computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0062] According to a fourth aspect of the present invention, a computer-readable storage medium is provided.

[0063] In one embodiment, a computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the above method.

[0064] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0065] 1. This invention addresses the issue that fault signals in power distribution lines are significantly affected by operating conditions, resulting in substantial differences in fault signal characteristics under different conditions. In the extraction of early-stage fault characteristics for power distribution lines and related equipment, a model of early-stage fault characteristics is first established using simulation software. This model simulates early-stage fault characteristics of power distribution lines and related equipment under various operating conditions, including load surges and motor starting, to obtain fault data. A machine learning-based SVMD-RCMSE-multimodal joint feature extraction method is then employed. The SVMD algorithm decomposes the three-phase voltage signal to obtain each intrinsic mode component. The RCMSE algorithm is then introduced to characterize the fault signal, extracting the distribution variation feature vectors from low-frequency components to high-frequency components at different scales. The extracted three-phase feature vectors are horizontally concatenated to form a set of data samples, and multimodal data fusion is performed. Finally, a deep learning-based CNN-LSTM-Softmax integrated model is used for fault classification, achieving fault diagnosis.

[0066] 2. Compared with the traditional feature extraction scheme combining Variational Mode Decomposition (VMD) and Composite Multiscale Entropy (CMSE), this invention can effectively solve the mode aliasing problem in the decomposition of complex fault signals, while improving the stability and reliability of entropy features. In SVMD, by introducing center frequency focusing constraints, residual spectrum separation constraints, and inter-mode spectrum de-overlap constraints, the objective function and iterative process of mode decomposition are optimized, ensuring that the spectral boundaries of each order of intrinsic mode components are clear and the center frequency focusing is strong. This avoids the mode crossover phenomenon that is prone to occur in the decomposition of non-stationary fault signals in traditional VMD. At the same time, the RCMSE algorithm significantly reduces the probability of undefined entropy values ​​caused by a single coarse-grained sequence by calculating the mean of multiple sliding windows of composite coarse-grained sequences. This makes the extracted fault entropy features more consistent and distinguishable at different scales, and can accurately capture the subtle time-domain, frequency-domain, and entropy-domain features of early fault hazards in power distribution lines and related equipment. This provides high-dimensional and highly representative feature inputs for subsequent fault classification and diagnosis, and significantly improves the sensitivity of early fault hazard identification.

[0067] 3. This invention, through further optimization of the CNN-LSTM-Softmax model, fully integrates the efficient extraction capability of convolutional neural networks for feature space correlations with the precise capture advantage of long short-term memory networks for temporal dependencies. This improves the diagnostic accuracy of early faults in power distribution lines and related equipment, and eliminates the need for manual intervention in the feature selection process, significantly shortening the fault diagnosis cycle and enhancing the convenience and real-time performance of engineering applications. Furthermore, the application of the Huffman tree Softmax strategy further improves the efficiency and accuracy of judgment, enabling the identification and early warning of early fault signals in power distribution lines, thereby improving the efficiency of line fault diagnosis and ensuring the power supply reliability of power distribution lines and related equipment.

[0068] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0069] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0070] Figure 1 This is a flowchart illustrating a method for diagnosing power distribution line faults according to an exemplary embodiment;

[0071] Figure 2 This is a schematic diagram of a power distribution line fault diagnosis system according to an exemplary embodiment;

[0072] Figure 3 This is a schematic diagram of the structure of a computer device according to an exemplary embodiment. Detailed Implementation

[0073] The following description and accompanying drawings fully illustrate specific embodiments described herein to enable those skilled in the art to practice them. Some portions and features of certain embodiments may be included in or replace portions and features of other embodiments. The scope of the embodiments herein includes the entire scope of the claims and all available equivalents thereof. The various embodiments described herein are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0074] The modules in the apparatus or system of this application can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0075] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0076] Figure 1 An embodiment of a power distribution line fault diagnosis method of the present invention is shown.

[0077] In this optional embodiment, the power distribution line fault diagnosis method includes:

[0078] Step S101: Collect the initial fault signal data corresponding to the power distribution line based on the simulation operation results of the power distribution line, preprocess the initial fault signal data, and output the fault signal data.

[0079] Step S102: Decompose the fault signal data using joint decomposition technology to obtain intrinsic mode components; introduce information mining technology to extract the distribution change feature vector of the fault signal, and fuse it with the intrinsic mode components to obtain a change image;

[0080] Step S103: Extract local features of the changed image through a convolutional neural network, and perform fault classification by combining the policy function optimized by Huffman tree, and output the fault diagnosis result.

[0081] In this optional embodiment, initial fault signal data corresponding to the power distribution line is collected based on the simulation operation results of the power distribution line, and the initial fault signal data is preprocessed to output fault signal data. This includes: simulating the actual operating conditions of the power distribution line under early fault risks and load changes using power simulation technology, and monitoring each node of the power distribution line in real time during the simulation to obtain voltage and current data; monitoring the environment of the power distribution line during the simulation to obtain temperature and humidity, and combining the voltage and current data to obtain initial fault signal data; describing the initial operating parameters using the minimum, first quartile, median, third quartile, and maximum value, and arranging the description results in ascending order, and using an outlier detection formula to remove abnormal data from the initial fault signal data; after outlier removal, using moving average filtering technology to perform noise reduction processing on the initial fault signal data, and outputting fault signal data.

[0082] In this optional embodiment, the fault signal data is decomposed using joint decomposition technology to obtain intrinsic mode components (IMCs). Information mining technology is introduced to extract the distribution variation feature vector of the fault signal, and this vector is fused with the IMCs to obtain a variation image. This includes: reconstructing the fault signal data into the current-order mode components and residual signals; establishing three joint constraint criteria based on the decomposition results to constrain the spectral discorrelation between the current-order mode components and historically determined modes, ensuring that the decomposed IMCs are mutually non-interfering in the frequency domain; and transforming the extracted mode components into a constrained minimization function using the balance parameters of the three joint constraint criteria, and introducing... An augmented Lagrange function is constructed using the Lagrange operator. The alternating multiplier algorithm is used to solve the minimization function and iteratively update the modal components, center frequency, and Lagrange multipliers. The iteration termination condition is set based on the variance and convergence parameters of the fault signal data. When the iteration update result meets the iteration termination condition, the extraction of the current order intrinsic modal components is completed. The updated residual signal is used as the new processing object to extract all valid intrinsic modal components in the fault signal data step by step. The fault signal data is coarsely processed, and the distribution change feature vector for extracting the fault signal data is formed based on the processing result. This vector is then fused with the intrinsic modal components to output an entropy change image.

[0083] In this optional embodiment, the extracted modal components are transformed into a constrained minimization function by combining the balance parameters of the three joint constraint criteria, and an augmented Lagrange function is constructed by introducing a Lagrange operator. The alternating multiplier algorithm is used to solve the minimization function and iteratively update the modal components, center frequency, and Lagrange multipliers. This includes: establishing a minimization objective function by introducing the balance factors of the three joint constraint criteria, and using the fault signal data reconstruction equation as the only constraint condition to transform the extracted modal components into a constrained minimization function; expanding the residual signal in the minimization function into the objective form to obtain the equation approximation. The residual terms are bundled, and an augmented Lagrangian function containing the dual terms of the introduced Lagrange multipliers and a quadratic penalty term is constructed based on the equality constraint residual terms. Partial derivatives of the modal components in the augmented Lagrangian function with respect to the Lagrange multipliers and the previous modal components are obtained to derive the frequency domain closed-form solution of the modal components for the target iteration. Partial derivatives of the center frequency in the augmented Lagrangian function are also obtained to derive the updated center frequency. The Lagrange multipliers are updated with a fixed step size along the gradient ascent direction of the dual function, and the frequency domain closed-form update solution of the Lagrange multipliers is obtained by combining the residuals, thus completing the iterative update of the modal components, center frequency, and Lagrange multipliers.

[0084] In this optional embodiment, coarse-graining processing of fault signal data, forming a distribution change feature vector for extracting fault signal data based on the processing result, and fusing it with intrinsic modal components to output an entropy change image includes: setting embedding dimension and similarity tolerance parameters based on the length of fault signal data, coarse-graining processing of fault signal data on a scale-by-scale, and constructing composite coarse-grained sequences at different scales by dividing the fault signal data at each scale according to length; generating composite coarse-grained sequence parameters with different offsets based on the composite coarse-grained sequences at the same scale, calculating the mean of the composite coarse-grained sequence parameters, defining the composite multi-scale entropy value based on the logarithm of the mean ratio, and arranging the composite multi-scale entropy values ​​corresponding to all scales in target order to form a distribution change feature vector for extracting fault signal data; fusing the distribution change features corresponding to different intrinsic modal components using direct connection or multiplication joint methods, and judging the loss of modal features during the fusion process by applying integrity verification technology, and outputting an entropy change image from the non-fault phase to the fault phase based on the judgment result.

[0085] In this optional embodiment, local features of the changing image are extracted using a convolutional neural network, and fault classification is performed using a policy function optimized by a Huffman tree. The output fault diagnosis result includes: sliding the convolution kernel of the convolutional neural network on the entropy change image, performing pixel-by-pixel weighted summation on the entropy change image through convolution operations and the sliding process, and extracting local features of the entropy change image based on the summation result; capturing the time-dependent features of fault signal data by introducing an input gate, a forget gate, and an output gate to form a gating mechanism; constructing a Huffman tree based on the occurrence probability of power line fault categories in a preset training set, and optimizing the output layer structure of the policy function using the Huffman tree; and performing fault classification based on the optimized policy function, combining the local features and time-dependent features of the entropy change image, and outputting the fault diagnosis result.

[0086] In this optional embodiment, constructing a Huffman tree based on the occurrence probability of power line fault categories in a preset training set, and optimizing the output layer structure of the policy function using the Huffman tree includes: creating leaf nodes for each fault category based on the occurrence frequency of each fault category in the preset training set, wherein the data structure of the leaf node contains the fault category tag and the occurrence frequency, and placing all leaf nodes into a minimum priority queue for iterative construction of the Huffman tree; starting from the root node of the Huffman tree, performing a depth-first traversal of the Huffman tree, assigning a binary code to each leaf node, and specifying the marker value for moving from the root node to the left and right leaf nodes to define the Huffman code corresponding to each leaf node, and determining the binary sequence on the path from the root node to the leaf node; optimizing the output layer structure of the policy function based on the binary sequence so that the fault type output by the policy function can be accessed through a unique path from the root node to the leaf node, thus completing the optimization of the policy function.

[0087] Figure 2 An embodiment of a power distribution line fault diagnosis system according to the present invention is shown.

[0088] In this optional embodiment, the power distribution line fault diagnosis system includes:

[0089] The data acquisition and processing unit 201 is used to acquire the initial fault signal data corresponding to the power distribution line based on the simulation operation results of the power distribution line, preprocess the initial fault signal data, and output the fault signal data.

[0090] The decomposition and fusion processing unit 202 is used to decompose the fault signal data using joint decomposition technology to obtain the intrinsic mode components; introduce information mining technology to extract the distribution change feature vector of the fault signal, and fuse it with the intrinsic mode components to obtain the change image;

[0091] The diagnostic result output unit 203 is used to extract local features of the changed image through a convolutional neural network, and perform fault classification by combining the policy function optimized by the Huffman tree, and output the fault diagnosis result.

[0092] In this optional embodiment, the data acquisition and processing unit 201 includes: simulating the actual operating conditions of early fault risks and load changes in power distribution lines using power simulation technology, and monitoring each node of the power distribution line in real time during the simulation to obtain voltage and current data; monitoring the environment of the power distribution line during the simulation to obtain temperature and humidity, and combining the voltage and current data to obtain initial fault signal data; describing the initial operating parameters using the minimum value, first quartile, median, third quartile, and maximum value, and arranging the description results in ascending order, and using an outlier detection formula to remove abnormal data from the initial fault signal data; and after the outlier removal is completed, using moving average filtering technology to perform noise reduction processing on the initial fault signal data and outputting fault signal data.

[0093] In this optional embodiment, the decomposition and fusion processing unit 202 includes: a criterion establishment module, used to reconstruct the fault signal data into the current-order modal components and residual signals, and establish three joint constraint criteria based on the decomposition results to constrain the spectral uncorrelation between the current-order modal components and historically determined modes, so as to ensure that the decomposed intrinsic modal components are mutually non-interfering in the frequency domain; and an iterative update module, used to combine the balance parameters of the three joint constraint criteria to transform the extracted modal components into a constrained minimization function, and introduce a Lagrange operator to construct an augmented Lagrange function, and use an alternating multiplier algorithm to solve the minimization function. The modal components, center frequency, and Lagrange multipliers are iteratively updated. The intrinsic modal component output module is used to set the iteration termination condition based on the variance and convergence parameters of the fault signal data. When the iteration update result meets the iteration termination condition, the extraction of the current order intrinsic modal components is completed. The updated residual signal is used as a new processing object to extract all valid intrinsic modal components in the fault signal data step by step. The change image output module is used to coarsely process the fault signal data, form a distribution change feature vector for extracting the fault signal data according to the processing result, and fuse it with the intrinsic modal components to output an entropy change image.

[0094] In this optional embodiment, the extracted modal components are transformed into a constrained minimization function by combining the balance parameters of the three joint constraint criteria, and an augmented Lagrange function is constructed by introducing a Lagrange operator. The alternating multiplier algorithm is used to solve the minimization function and iteratively update the modal components, center frequency, and Lagrange multipliers. This includes: establishing a minimization objective function by introducing the balance factors of the three joint constraint criteria, and using the fault signal data reconstruction equation as the only constraint condition to transform the extracted modal components into a constrained minimization function; expanding the residual signal in the minimization function into the objective form to obtain the equation approximation. The residual terms are bundled, and an augmented Lagrangian function containing the dual terms of the introduced Lagrange multipliers and a quadratic penalty term is constructed based on the equality constraint residual terms. Partial derivatives of the modal components in the augmented Lagrangian function with respect to the Lagrange multipliers and the previous modal components are obtained to derive the frequency domain closed-form solution of the modal components for the target iteration. Partial derivatives of the center frequency in the augmented Lagrangian function are also obtained to derive the updated center frequency. The Lagrange multipliers are updated with a fixed step size along the gradient ascent direction of the dual function, and the frequency domain closed-form update solution of the Lagrange multipliers is obtained by combining the residuals, thus completing the iterative update of the modal components, center frequency, and Lagrange multipliers.

[0095] To facilitate understanding of the above technical solutions of the present invention, the following further describes the above technical solutions of the present invention from the perspectives of architecture and principle, as follows:

[0096] Step 1: Data Acquisition and Preprocessing;

[0097] The actual operating conditions of the power distribution line are simulated using power system simulation software such as Matlab / Simulink, and the operating data is comprehensively collected. The data collection content mainly includes environmental and operating parameters such as voltage (U), current (I), temperature (T), and humidity (H) (i.e., fault signal data). The specific data collection process is shown in the following example.

[0098] (1) Data acquisition of voltage, current, etc.: Using intelligent terminal equipment, real-time monitoring is performed on each node of the power distribution line to obtain voltage and current data. The data acquisition formula is as follows:

[0099] ;

[0100] ;

[0101] In the formula, and They represent the first Voltage and current at each monitoring point; and Indicates the base voltage and current; and Indicates the first Voltage and current changes caused by various influencing factors.

[0102] (2) Temperature and humidity data acquisition: A distributed sensor network is used to monitor the environment surrounding the power distribution lines in real time to acquire temperature and humidity data. The data acquisition formula is:

[0103] ;

[0104] ;

[0105] In the formula, and These represent the actual temperature and humidity, respectively. and These represent the temperature and humidity measured by the sensor, respectively. and This indicates the sensor calibration deviation.

[0106] The data preprocessing stage mainly includes the following steps:

[0107] Outlier handling: Outliers are identified and removed using methods such as box plots to ensure data consistency and reliability. Five statistical measures from the data collection process—minimum, first quartile, median, third quartile, and maximum—are used to describe the data. The first quartile, median, and third quartile, after being arranged in ascending order of the collected data, are the values ​​at the 25th, 50th, and 75th percentiles, respectively. Then, the outlier detection formula is applied.

[0108] ;

[0109] In the formula, and These represent the first and third quartiles of the data, respectively. Indicates the interquartile range, which is ; This indicates the data to be detected. Data outside the range of the outlier detection formula will be identified as outliers and removed.

[0110] Noise Reduction: Methods such as moving average filtering and wavelet transform are used to denoise the data and improve data quality. The moving average filtering formula is as follows:

[0111] ;

[0112] In the formula, This represents the filtered data; Represents the original data; Indicates the length of the filtering window.

[0113] Step 2: A joint feature extraction method based on SVMD-RCMSE-multimodal;

[0114] VMD is an adaptive, fully non-recursive modal variational and signal processing method. SVMD improves upon VMD by continuously extracting modes without pre-setting the K value. This continuous search mode not only improves convergence speed but also avoids extracting unwanted modes, reducing computation time. The steps of SVMD decomposition are as follows:

[0115] (1) For a time-domain signal (i.e., fault signal data) Suppose it is decomposed into two signals, as shown in the following equation:

[0116] ;

[0117] In the formula, Indicates the first First mode; Let L represent the residual signal, t represent the continuous-time independent variable, and L represent the modal order, including the th... Previous modes and unprocessed signals As shown in the following formula (i represents the traversal index of the extracted modal components):

[0118] ;

[0119] (2) To ensure the above assumptions hold and to achieve continuous mode extraction, new constraint criteria are established:

[0120] Each mode should closely revolve around its center frequency, achieved by minimizing the following constraint, the constraint criterion of which is as follows (modal bandwidth minimization constraint term):

[0121] ;

[0122] In the formula, Indicates the first The center frequency of the first mode, This represents the convolution operation. This represents the first-order partial differential operator with respect to time t. Represents the imaginary unit. This represents the complex exponential frequency shift operator.

[0123] In fault diagnosis, early fault signals are often weak and overlap with noise and interference spectra. This constraint allows the component to reconstruct the original signal well and concentrates the spectrum of the component as much as possible at the center frequency, effectively suppressing mode aliasing. This ensures that each decomposed mode corresponds to a relatively isolated frequency band, thereby enabling more accurate extraction of the time-frequency domain features of the fault.

[0124] residual signal With mode The spectral overlap should be minimized, meaning the energy of the residual signal is minimized within the frequency band of the desired mode. The established constraint criterion formula is (the constraint term for minimizing the spectral overlap between the mode component to be extracted and the residual signal):

[0125] ;

[0126] In the formula, The impulse response of the filter is its time-domain impulse response, designed for the center frequency of the Lth mode. Its function is to extract the components of the residual signal that fall within the Lth mode's frequency band. Its frequency response is:

[0127] ;

[0128] By minimizing and Constraints may not be able to effectively distinguish the first First mode and previous Since there are several modes, based on the idea of ​​establishing constraint J2, the established constraint criterion formula (the spectral isolation constraint term between the modal component to be extracted and the historically extracted modal components) is as follows:

[0129] ;

[0130] In the formula For filter The impulse response, representing the time-domain impulse response of the filter designed for the center frequency of the extracted i-th mode, is given by:

[0131] ;

[0132] (3) Based on the constraint criterion in (2), the problem of extracting modal components is transformed into a constrained minimization problem, and the constraint model is:

[0133] ;

[0134] This is the minimization process for the first three constraint criteria, where α is the balance factor for balancing the three constraints. To solve the problem of modal overlap, To ensure that frequency components already included in the residual signal are not repeatedly extracted when extracting the current mode, This is used to constrain the orthogonality or spectral uncorrelation between the current mode and all historically determined modes, ensuring that all decomposed modes are separated from each other in the frequency domain and do not interfere with each other. This represents the constraints that the optimization process must satisfy, i.e., the signal reconstruction equation.

[0135] To find the optimal solution to the variational problem, we introduce... and Lagrange operator Construct the augmented Lagrange function; the expanded Lagrange expression is:

[0136] + ;

[0137] (4) The minimization problem is solved by using the alternating multiplier algorithm, and the updates are iteratively alternated. , , To extend the above equation to include the saddle point, The expression is:

[0138] ;

[0139] In the formula, , , and They are , , and The result is obtained through Fourier transform, where n represents the iteration round, and n+1 represents the iteration update flag, indicating that the variable is the result after the (n+1)th iteration update. This represents the frequency domain expression (Fourier transform result) of the Lth modal component after the (n+1)th iteration update. Let represent the frequency domain prior value of the Lth modal component in the nth iteration, and represent the result of the previous iteration. This represents the frequency domain representation of the original input signal f(t). This represents the frequency domain representation of the Lagrange multiplier λ(t).

[0140] in, After the (n+1)th iteration update, the expression for the center angular frequency of the Lth mode is:

[0141] ;

[0142] In the formula, The dual ascent method is used to obtain it when , After the update, the gradient of the augmented Lagrange function with respect to λ is precisely the residual of the original equality constraint. The formula is as follows:

[0143] ;

[0144] In the formula, This represents the frequency domain representation of the i-th extracted mode.

[0145] The residual reflects the degree to which the current decomposition result violates the original equation. The rule of the dual ascent method is that it moves along the direction of the gradient of the dual function with a step size τ. This yields the frequency domain representation of the Lagrange multipliers corresponding to the Lth mode after n+1 iterations, where... The frequency domain prior value of the Lagrange multipliers in the nth iteration is represented by the following equation:

[0146] ;

[0147] Finally, we can obtain The expression is as follows:

[0148] ;

[0149] In the formula, To update the parameters, the iteration terminates under the following condition:

[0150] ;

[0151] In the formula, Let f(t) be the variance of the signal. These are the convergence parameters.

[0152] The CMSE algorithm delves deeper into useful information during the coarse-graining process. By considering the useful information from multiple coarse-grained time series at the same scale, it averages the entropy values ​​of all coarse-grained time series at that scale to obtain the CMSE value for that scale. Compared to the MSE algorithm, the CMSE algorithm can generate more coarse-grained sequences in data computation, thus obtaining more accurate entropy values. However, because the CMSE algorithm needs to average the entropy values ​​of all composite coarse-grained sequences, if even one composite coarse-grained sequence is too short, resulting in an undefined entropy value, the CMSE value at that scale will become an undefined entropy value. Therefore, compared to the MSE algorithm, the CMSE algorithm significantly increases the probability of generating undefined entropy when processing shorter data samples.

[0153] The RCMSE algorithm is an improvement on the CMSE algorithm. The specific process of the algorithm is as follows:

[0154] (1) For the original data Given a length of N, and setting the embedding dimension m and similarity tolerance r, construct a composite coarse-grained sequence:

[0155] ;

[0156] In the formula, In order to be in The sliding sequence number constructed below, at scale In contrast to MSE, which generates only one coarse-grained sequence, the CMSE algorithm generates sequences sequentially. A coarse-grained sequence.

[0157] (2) In terms of scale Next, calculate the parameters of any complex coarse-grained sequence. and As shown in the following formula:

[0158] ;

[0159] ;

[0160] In the formula, M is the length of the coarse-grained sequence, m is the embedding dimension, and r is the similarity tolerance (usually taken as 0.1 to 0.25 times the standard deviation of the original sequence).

[0161] (3) Calculate the mean value of the parameters of the composite coarse-grained sequence. and The RCMSE value is defined based on the logarithm of the mean ratio, as shown in the following formula:

[0162] ;

[0163] As can be seen from the above equation, compared with the CMSE algorithm, the RCMSE algorithm greatly reduces the probability of generating undefined entropy values, and the resulting entropy values ​​are more stable.

[0164] In summary, the SVMD algorithm decomposes signals such as voltage, current, and temperature to obtain their intrinsic mode components. The RCMSE algorithm is introduced to characterize the fault signal, extracting feature vectors showing the distribution changes from low-frequency to high-frequency components at different scales. Finally, the extracted feature vectors are concatenated to form a set of data samples. In the multimodal data fusion process, after feature extraction, the features of different modes are fused to form a comprehensive feature vector. Multiple strategies are employed for fusion, one of which is a direct connection method, i.e., a joint method, which combines the semantics of feature vectors from different modes to achieve multimodal fusion. The specific formula is as follows:

[0165] ;

[0166] In the formula, For the output results in the shared semantic subspace; For each single mode input; The weights are represented by subscripts, which indicate different modes, and are mapped... Transform all submodal semantics into a shared subspace.

[0167] Another approach is the multiplication-joint method, which constructs a tensor by element-wise multiplying the outputs of feature vectors from different modalities (or using other specific multiplication methods) to achieve multimodal fusion. The specific formula is as follows:

[0168] ;

[0169] In the formula, This is the output result after fusing the tensors; For different modes; It is the outer product operator.

[0170] After multimodal fusion, the entropy change image from the non-faulty phase to the faulty phase is output. The x-axis represents the scale and the y-axis represents the RCMSE value (based on the essence of RCMSE multi-scale entropy, each scale has an entropy value. The stitching method does not lose the scale dimension information, so it should normally be a multi-scale curve. Here, the result can be directly plotted as an image).

[0171] The combined feature extraction based on SVMD-RCMSE-multimodal features can be visualized using the following steps to obtain the changed image:

[0172] (1) Perform SVMD decomposition and RCMSE feature extraction on each signal separately. For each monitoring signal such as voltage, current and temperature, process them independently first. Among them, the voltage signal is divided into three phase signals A, B and C.

[0173] (2) The RCMSE feature vectors calculated from multiple signals such as phase A voltage, phase B voltage, phase C voltage, current, and temperature are fused to form a comprehensive feature vector.

[0174] (3) Using two sets of data samples: samples in non-fault state and samples in fault state, perform steps (1) and (2), and the comprehensive feature vector formed by splicing still retains the scale dimension information. The vector curve that changes according to the scale can be drawn directly based on the results.

[0175] For the verification of multimodal fusion features, an integrity verification based on reconstruction error is adopted to avoid the loss of modal features or severe distortion during the fusion process, as shown in the following formula:

[0176] ;

[0177] In the formula, This represents the number of modes participating in the fusion process. If the reconstruction error exceeds a preset threshold, it is determined that modal features have been lost or severely distorted during the fusion process, and the modal data will be recalculated or corrected.

[0178] Step 3: Fault classification and diagnosis based on CNN-LSTM-Softmox;

[0179] To address the complex problem of power equipment fault diagnosis, this embodiment utilizes the powerful pattern recognition capabilities of deep learning. Through unsupervised learning, it automatically extracts high-order, abstract fault features to effectively identify the differences between the healthy and abnormal states of power equipment. Deep learning methods construct multi-layer neural networks to automatically extract high-level features from the data, thereby achieving fault diagnosis. The deep learning methods used in this paper include Convolutional Neural Networks (CNNs) and their variant, Long Short-Term Memory Networks (LSTMs).

[0180] CNNs are suitable for processing image data. Their core operation is the convolutional layer, which slides a convolutional kernel (filter) across the input image, performing a weighted summation operation pixel-by-pixel to extract local features. The mathematical formula for the convolution operation is as follows:

[0181] ;

[0182] In the formula, Input image; For convolution kernel; The output feature map (the result after convolution); For the input image at position Pixel values; For the convolution kernel at position The weights; To output feature map at location The value; The size of the convolution kernel is an odd number (such as 3, 5, etc.), representing the width and height of the convolution kernel.

[0183] LSTM is suitable for processing time-series data, capturing the temporal dependencies within the data. By introducing gating mechanisms, including input gates, forget gates, and output gates, it controls the inflow and outflow of information. The update formula is:

[0184] ;

[0185] In the formula, The output of the input gate; , These are the weight matrices for the input and the hidden state at the previous time step, respectively. For bias terms; It is the sigmoid activation function.

[0186] ;

[0187] In the formula, The output of the forget gate; , These are the weight matrices for the input and the hidden state at the previous time step, respectively. This is a bias term.

[0188] ;

[0189] In the formula, This represents the current state of the cell. represents the cell state at the previous time step; ⊙ represents element-wise multiplication; tanh is the hyperbolic tangent activation function.

[0190] ;

[0191] In the formula, This is the output of the output gate; , These are the weight matrices for the input and the hidden state at the previous time step, respectively. This is a bias term.

[0192] ;

[0193] In the formula, This represents the current hidden state.

[0194] Furthermore, for CNNs, the recognition process is optimized by adjusting the size and number of convolutional kernels and increasing the number of pooling layers. For LSTMs, optimization is achieved by adjusting the size of hidden layers and adding gating mechanisms, combined with transfer learning and online learning methods to improve the model's adaptability and generalization ability.

[0195] For time-series data or similar linear data, standardization is performed to ensure a mean of 0 and a standard deviation of 1. Fault classification is then performed using the Softmax policy function, ultimately outputting the fault diagnosis results.

[0196] ;

[0197] In the formula, The output value of the i-th node, where C is the number of output nodes, i.e. the number of categories.

[0198] To improve the computational efficiency of the Softmax strategy, a Huffman tree is introduced to optimize the classification inference path, thereby enhancing the model's computational efficiency. The Softmax strategy essentially employs a probability tree structure, leveraging the characteristics of Huffman trees to optimize the output layer's structure. First, a Huffman tree is constructed based on the class probabilities in the training data. Here, the classes correspond to different fault types of the equipment, and the frequency reflects the number of times the fault occurs in the dataset. This tree is constructed using the Huffman coding algorithm, ensuring that high-frequency classes are assigned shorter codes, while low-frequency classes receive longer codes. In this way, the Huffman tree's coding can efficiently manage the model's parameter count, making the model structure more lightweight and thus improving the model's decoding and judgment speed. The specific steps of Huffman coding are as follows:

[0199] (1) Statistically analyze the frequency of occurrence of each fault category in the training set;

[0200] (2) Initialize the leaf node set: Create a leaf node for each fault category. The data structure of each node includes a category label and frequency. Place all leaf nodes into a minimum priority queue, sorted by node frequency;

[0201] (3) Iteratively merge to construct the Huffman tree;

[0202] Repeat the following steps until only one node remains in the priority queue; this node is the root node:

[0203] Extract the two nodes with the lowest frequencies; create a new internal node whose frequency is the sum of the frequencies of the two nodes with the lowest frequencies; set the two nodes with the lowest frequencies as the left and right children of the new internal node; insert the new internal node into the priority queue.

[0204] (4) Assign binary codes: Starting from the root node, perform a depth-first traversal of the tree and assign a unique binary code to each leaf node (i.e., the fault category). It is stipulated that moving from the root node to the left child node is denoted as 0 (or 1), and moving to the right child node is denoted as 1 (or 0), and the two must be consistent. It is stipulated that the Huffman code corresponding to each leaf node is the binary sequence on the path from the root node to that leaf node.

[0205] Since high-frequency categories are closer to the root node, their paths are shorter and their encoding lengths are shorter; while low-frequency categories have longer paths and their encoding lengths are longer. This unequal-length encoding method is transformed into a series of conditional probabilities multiplied by a small number of internal nodes on the path in the Softmax hierarchical computation.

[0206] In summary, by applying the Softmax strategy of Huffman trees, each fault type can be accessed through a unique path from the root node to a leaf node, expressed as shown in the following formula:

[0207] ;

[0208] In the formula, For category labels or identifiers, The first in the tree The fault types represented by each leaf node Represents the first node on the path from the root node to the leaf node. 1 node This represents a branch that leads to a leaf node.

[0209] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores static and dynamic information data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0210] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0211] In addition, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0212] In addition, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0213] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0214] This invention is not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this invention is limited only by the appended claims.

Claims

1. A power distribution line fault diagnostic method characterized by, include: Based on the simulation operation results of the power distribution line, the initial fault signal data corresponding to the power distribution line is collected, and the initial fault signal data is preprocessed and the fault signal data is output. The fault signal data is decomposed using joint decomposition technology to obtain intrinsic mode components; information mining technology is introduced to extract the distribution change feature vector of the fault signal, and then fused with the intrinsic mode components to obtain the change image. The system extracts local features from the changing image using a convolutional neural network, and combines this with a policy function optimized by a Huffman tree to classify faults and output fault diagnosis results.

2. The power distribution line fault diagnostic method according to claim 1, characterized by, The method involves collecting initial fault signal data corresponding to the power distribution line based on the simulation operation results, preprocessing the initial fault signal data, and outputting fault signal data including: The power simulation technology is used to simulate the actual operating conditions of power distribution lines, such as early faults and sudden load changes. The voltage and current data of each node of the power distribution line are monitored in real time during the simulation. During the simulation, the environment of the power distribution line is monitored to obtain temperature and humidity, and the initial fault signal data is obtained by combining voltage and current data. The initial operating parameters are described using the minimum, first quartile, median, third quartile, and maximum value. The description results are arranged from smallest to largest, and abnormal data in the initial fault signal data are removed by combining the outlier detection formula. After outlier removal, the initial fault signal data is processed by moving average filtering to eliminate noise and output fault signal data.

3. The power distribution line fault diagnostic method of claim 1, wherein The process involves decomposing the fault signal data using joint decomposition techniques to obtain intrinsic mode components; then, information mining techniques are introduced to extract the distribution variation feature vector of the fault signal, which is then fused with the intrinsic mode components to obtain a variation image. The fault signal data is reconstructed into the modal components of the current order and the residual signal. Based on the decomposition results, three joint constraint criteria are established to constrain the spectral uncorrelation between the modal components of the current order and the historically determined modes, so as to ensure that the decomposed intrinsic modal components are mutually non-interfering in the frequency domain. The extracted modal components are transformed into a constrained minimization function by combining the equilibrium parameters of the three joint constraint criteria. An augmented Lagrange function is constructed by introducing a Lagrange operator. The alternating multiplier algorithm is used to solve the minimization function and iteratively update the modal components, center frequency and Lagrange multipliers. Based on the variance and convergence parameters of the fault signal data, the iteration termination condition is set, and when the iteration update result meets the iteration termination condition, the extraction of the current order intrinsic mode components is completed. The updated residual signal is used as a new processing object to extract all valid intrinsic mode components in the fault signal data step by step. The fault signal data is coarsely processed, and a feature vector for extracting the distribution change of the fault signal data is formed based on the processing results. This vector is then fused with the intrinsic mode components to output an entropy change image.

4. The power distribution line fault diagnostic method according to claim 3, characterized by, The equilibrium parameters combining the three joint constraint criteria transform the extracted modal components into a constrained minimization function. An augmented Lagrange function is constructed by introducing a Lagrange operator. The alternating multiplier algorithm is used to solve the minimization function, iteratively updating the modal components, center frequency, and Lagrange multipliers. A balance factor based on three joint constraint criteria is introduced to establish a minimization objective function, and the fault signal data reconstruction equation is used as the only constraint condition to transform the extracted modal components into a constrained minimization function. Expand the residual signal in the minimization function into the objective form to obtain the equality-constrained residual term, and construct the augmented Lagrange function based on the equality-constrained residual term, which includes the dual term of the introduced Lagrange multipliers and the quadratic penalty term. Based on the partial derivatives of the modal components in the augmented Lagrange function with respect to the Lagrange multipliers and the previous modal components, the frequency domain closed-form solution of the modal components of the target wheel iteration is derived, and the center frequency is derived by taking the partial derivative with respect to the center frequency in the augmented Lagrange function. The Lagrange multipliers are updated with a fixed step size along the gradient ascent direction of the dual function. The frequency domain closed-form update solution of the Lagrange multipliers is obtained by combining the residuals, thus completing the iterative update of the modal components, center frequency and Lagrange multipliers.

5. The power distribution line fault diagnostic method according to claim 4, characterized by, The expression for the minimization function is: ； wherein denotes the balance factor, denotes the modal bandwidth minimization constraint term, denotes the spectral overlap minimization constraint term of the modal component to be extracted with the residual signal, denotes the spectral isolation constraint term of the modal component to be extracted with the historical extracted modal components, denotes the modal component, denotes the residual signal, denotes the fault signal data; The expression for the constraint term minimizing the spectral overlap between the modal component to be extracted and the residual signal is as follows: ； The expression for the spectral isolation constraint term between the modal component to be extracted and the historically extracted modal components is as follows: ； wherein represents the impulse response of the filter corresponding to the Lth modal center frequency, represents the impulse response of the filter corresponding to the ith extracted modal center frequency.

6. The power distribution line fault diagnostic method according to claim 5, characterized by, The process of coarse-graining the fault signal data, forming a feature vector for extracting the distribution variation of the fault signal data based on the processing result, and fusing it with the intrinsic mode components to output an entropy value variation image includes: Based on the length of the fault signal data, the embedding dimension and similarity tolerance are set as parameters. The fault signal data is coarsened on a scale-by-scale, and the fault signal data at each scale is divided according to the length to construct composite coarsened sequences at different scales. Based on the composite coarse-grained sequence at the same scale, composite coarse-grained sequence parameters with different offsets are generated. The mean of the composite coarse-grained sequence parameters is calculated. The composite multi-scale entropy value is defined according to the logarithm of the mean ratio. The composite multi-scale entropy values ​​corresponding to all scales are arranged in the target order to form a distribution change feature vector for extracting fault signal data. The distribution variation characteristics of different intrinsic modal components are fused by direct connection or multiplication combination. The integrity verification technology is applied to judge the loss of modal features during the fusion process. Based on the judgment result, the entropy change image from non-faulty phase to faulty phase is output.

7. The power distribution line fault diagnostic method of claim 1, wherein The process involves extracting local features from the changed image using a convolutional neural network and combining this with a policy function optimized by a Huffman tree for fault classification, outputting fault diagnosis results including: The convolution kernel of a convolutional neural network is slid across an image with entropy changes. Through the convolution operation and the sliding process, a pixel-by-pixel weighted summation is performed on the image with entropy changes, and local features of the image with entropy changes are extracted based on the summation result. By introducing an input gate, a forget gate, and an output gate to form a gating mechanism to capture the time-dependent characteristics of fault signal data, a Huffman tree is constructed based on the occurrence probability of fault categories in the power distribution line in the preset training set, and the output layer structure of the policy function is optimized using the Huffman tree. Based on the optimized policy function, fault classification is performed by combining local features and temporal dependency features of the entropy change image, and fault diagnosis results are output.

8. The power distribution line fault diagnostic method according to claim 7, characterized by, The expression for the convolution operation is: ； In the formula, Indicates the output feature map at position I represents the entropy change graph. Represents the convolution kernel. Indicates the position of the convolution kernel The weight, This indicates the kernel size.

9. The method for diagnosing power distribution line faults according to claim 8, characterized in that, The method of constructing a Huffman tree based on the probability of fault categories in a preset training set of power distribution lines, and optimizing the output layer structure of the policy function using the Huffman tree, includes: Based on the occurrence frequency of each fault category in the preset training set, a leaf node is created for each fault category. The data structure of the leaf node contains the fault category tag and the occurrence frequency. All leaf nodes are put into the minimum priority queue for iterative construction of a Huffman tree. Starting from the root node of the Huffman tree, perform a depth-first traversal of the Huffman tree to assign a binary code to each leaf node, and specify the marker value for moving from the root node to the left and right leaf nodes to define the Huffman code corresponding to each leaf node, thus determining the binary sequence on the path from the root node to the leaf node. The output layer structure of the policy function is optimized based on binary sequence optimization, so that the fault type output by the policy function can be accessed through a unique path from the root node to the leaf node, thus completing the optimization of the policy function.

10. A power distribution line fault diagnosis system, characterized in that, include: The data acquisition and processing unit is used to acquire the initial fault signal data corresponding to the power distribution line based on the simulation operation results of the power distribution line, preprocess the initial fault signal data, and output the fault signal data. The decomposition and fusion processing unit is used to decompose the fault signal data using joint decomposition technology to obtain the intrinsic mode components; information mining technology is introduced to extract the distribution change feature vector of the fault signal, and then fused with the intrinsic mode components to obtain the change image. The diagnostic result output unit is used to extract local features of the changed image through a convolutional neural network, and combine them with a policy function optimized by a Huffman tree to perform fault classification and output fault diagnosis results.

11. The power distribution line fault diagnosis system according to claim 10, characterized in that, The data acquisition and processing unit includes: The power simulation technology is used to simulate the actual operating conditions of power distribution lines, such as early faults and sudden load changes. The voltage and current data of each node of the power distribution line are monitored in real time during the simulation. During the simulation, the environment of the power distribution line is monitored to obtain temperature and humidity, and the initial fault signal data is obtained by combining voltage and current data. The initial operating parameters are described using the minimum, first quartile, median, third quartile, and maximum value. The description results are arranged from smallest to largest, and abnormal data in the initial fault signal data are removed by combining the outlier detection formula. After outlier removal, the initial fault signal data is processed by moving average filtering to eliminate noise and output fault signal data.

12. The power distribution line fault diagnosis system according to claim 11, characterized in that, The decomposition and fusion processing unit includes: The criteria establishment module is used to reconstruct fault signal data into the modal components of the current order and the residual signal, and establish three joint constraint criteria based on the decomposition results to constrain the spectral uncorrelation between the modal components of the current order and the historically determined modes, so as to ensure that the decomposed intrinsic modal components are mutually non-interfering in the frequency domain. The iterative update module is used to combine the equilibrium parameters of the three joint constraint criteria to transform the extracted modal components into a constrained minimization function, and to introduce the Lagrange operator to construct the augmented Lagrange function. The alternating multiplier algorithm is used to solve the minimization function to iteratively update the modal components, center frequency and Lagrange multipliers. The intrinsic mode component output module is used to set the iteration termination condition based on the variance and convergence parameters of the fault signal data, and to complete the extraction of the current order intrinsic mode components when the iteration update result meets the iteration termination condition. The updated residual signal is used as a new processing object to extract all valid intrinsic mode components in the fault signal data step by step. The change image output module is used to coarsely process the fault signal data, form a distribution change feature vector for extracting the fault signal data based on the processing results, and fuse it with the intrinsic mode components to output an entropy value change image.

13. The power distribution line fault diagnosis system according to claim 12, characterized in that, The equilibrium parameters combining the three joint constraint criteria transform the extracted modal components into a constrained minimization function. An augmented Lagrange function is constructed by introducing a Lagrange operator. The alternating multiplier algorithm is used to solve the minimization function, iteratively updating the modal components, center frequency, and Lagrange multipliers. A balance factor based on three joint constraint criteria is introduced to establish a minimization objective function, and the fault signal data reconstruction equation is used as the only constraint condition to transform the extracted modal components into a constrained minimization function. Expand the residual signal in the minimization function into the objective form to obtain the equality-constrained residual term, and construct the augmented Lagrange function based on the equality-constrained residual term, which includes the dual term of the introduced Lagrange multipliers and the quadratic penalty term. Based on the partial derivatives of the modal components in the augmented Lagrange function with respect to the Lagrange multipliers and the previous modal components, the frequency domain closed-form solution of the modal components of the target wheel iteration is derived, and the center frequency is derived by taking the partial derivative with respect to the center frequency in the augmented Lagrange function. The Lagrange multipliers are updated with a fixed step size along the gradient ascent direction of the dual function. The frequency domain closed-form update solution of the Lagrange multipliers is obtained by combining the residuals, thus completing the iterative update of the modal components, center frequency and Lagrange multipliers.

14. The power distribution line fault diagnosis system according to claim 13, characterized in that, The expression for the minimization function is: ； In the formula, Represents the balance factor. This represents the modal bandwidth minimization constraint. This represents the constraint term that minimizes the spectral overlap between the modal component to be extracted and the residual signal. This represents the spectral isolation constraint term between the modal components to be extracted and those previously extracted. Represents modal components, Represents the residual signal. This indicates fault signal data.

15. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 9.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9.

Images