A high-voltage cable partial discharge fault prediction method and system

Abstract

This invention discloses a method and system for predicting partial discharge (PD) faults in high-voltage cables. It collects multiple PD data, temperature data, and power data from online cable monitoring to locate PD faults in high-voltage cables. The method extracts morphological statistical features from the online PRPD (partial discharge probability) maps. Missing monitoring parameters are filled in. Using the remaining monitoring parameters and the extracted PD morphological statistical features, trend and periodic terms are extracted and combined to build a trend prediction model. The trend prediction results of the remaining monitoring parameters and PD morphological statistical features are used as input to the PD diagnosis model to obtain the PD fault diagnosis results for high-voltage cables. This invention enables the prediction, location, and diagnosis of PD faults in high-voltage cables. It also provides real-time location and decision support for high-voltage cable insulation operation risks, contributing to improved reliability early warning, global perception, and robust control capabilities of power system digital twins.

Description

Technical Field

[0001] This invention belongs to the field of power equipment condition prediction and fault diagnosis technology, specifically relating to a method and system for predicting partial discharge faults in high-voltage cables. Background Technology

[0002] To adapt to the development of clean energy and the trend of low-carbon transformation, a large number of new energy power plants will be connected to the power grid. This will present significant challenges to the safety, stability, and operational optimization of the power grid, especially for high-voltage and high-efficiency power systems. This poses unprecedented challenges to the routine prediction of grid operation and transmission control. Furthermore, with increasing demands for the safety and reliability of power system operation due to societal development, improving the multi-dimensional detection, condition prediction, and system management of the power system has become a major focus and requirement for current power grid digitalization.

[0003] Regarding data, with the construction and development of the power Internet of Things, massive amounts of historical data have been accumulated during the construction and operation of power grid equipment. A large amount of equipment data has accumulated and remained dormant, failing to effectively tap its value. There is a lack of a scientific analysis and efficient decision-making mechanism for historical tests, online monitoring, and other data, making it difficult to use big data to accurately guide current production operations. High-voltage cables, as important transmission equipment, involve numerous parameters characterizing their operating status, such as online monitoring data, electrical test data, power grid operation data, meteorological environmental data, insulation monitoring data, and equipment quality records. Because these characteristics are closely coupled and interrelated, and cable insulation issues have always been a core technical challenge for stable and reliable cable operation, a data-driven approach to comprehensively diagnose and locate the current state of high-voltage cables and predict their future trends can help to quickly, effectively, and specifically arrange equipment maintenance and formulate operation and maintenance strategies. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for predicting partial discharge faults in high-voltage cables, which addresses the shortcomings of the prior art and solves the technical problems of unpredictable partial discharge state, diagnosis and location of partial discharge fault types in high-voltage cables.

[0005] The present invention adopts the following technical solution:

[0006] A method for predicting partial discharge faults in high-voltage cables includes the following steps:

[0007] S1. Collect the status data of high-voltage cables, construct a cable partial discharge fault location dataset, a cable status prediction historical dataset, and the original cable fault diagnosis historical dataset.

[0008] S2. Use wavelet filtering to process the high-frequency signal in the cable partial discharge fault location dataset obtained in step S1, and then reconstruct the high-frequency signal in reverse.

[0009] S3. Extract modal components from the high-frequency signal obtained in step S2 to obtain the primary finite bandwidth intrinsic mode function of the high-frequency signal.

[0010] S4. Based on the two-end phase response method, the high-frequency signal in the cable partial discharge fault location dataset in step S1 is processed by the fast Fourier distribution transformation to obtain the corresponding phase response spectrum. The phase response where the peak value of the high-frequency signal in the primary finite bandwidth intrinsic mode function obtained in step S3 is located is combined with the phase response of the high-voltage cable partial discharge fault location using the cable partial discharge phase attenuation characteristic equation.

[0011] S5. Extract the morphological statistical features of the PRPD map from the historical data set of cable condition prediction obtained in step S1, and use the morphological statistical features of the PRPD map to replace the PRPD map to form a new historical data set of cable condition prediction.

[0012] S6. Use multiple linear regression to fill in the data of each cable feature parameter in the new cable condition prediction historical dataset obtained in step S5 to form a cable condition historical dataset.

[0013] S7. Using the time series trend decomposition algorithm, the historical data in the cable status historical dataset obtained in step S6 are decomposed into trend components, periodic components and residual terms. Then, the joint time series prediction of the trend components of each historical data is realized through the long short-term memory network to obtain the prediction results of various characteristic parameters of the high-voltage cable.

[0014] S8. Extract the morphological statistical features of the PRPD map from the historical data set for cable fault diagnosis obtained in step S1, and use the morphological statistical features of the PRPD map to replace the PRPD map to form a new historical data set for cable fault diagnosis.

[0015] S9. Using the new historical dataset for cable fault diagnosis obtained in step S8, establish a fault diagnosis model of deep convolutional belief network, save the model with the best training results as the DCBN cable fault diagnosis model, and use the prediction results of various characteristic parameters of high voltage cable obtained in step S7 as the input of the DCBN cable fault diagnosis model to realize the prediction of recent partial discharge faults of the cable.

[0016] Specifically, in step S1, the PRPD pattern, high-frequency signal, grounding current, temperature, and load current data of the high-voltage cable are collected; the high-frequency signals collected by two high-frequency sensors located close to each other on the high-voltage cable are used as the cable partial discharge fault location dataset; the PRPD pattern, grounding current, temperature, and load current are used as the cable condition prediction historical dataset; and the PRPD pattern, grounding current, temperature, and load current data of the high-voltage cable under normal conditions and when partial discharge occurs are collected to form the original cable fault diagnosis historical dataset.

[0017] Specifically, in step S3, the Lagrange multiplier method and penalty factor α are introduced to constrain the variational decomposition of the partial discharge high-frequency signal. The Lagrange multiplication factor and update factor σ are used to iteratively optimize the signal through an alternating direction algorithm until the convergence criterion tolerance ε is met, thereby obtaining a series of finite bandwidth intrinsic mode functions after decomposition. Finally, the primary finite bandwidth intrinsic mode functions of the two high-frequency signals are obtained.

[0018] Specifically, step S4 is as follows:

[0019] Acquire two adjacent high-frequency signals on a high-voltage cable; perform a Fast Fourier Transform on the two high-frequency signals to obtain their phase response spectra; and obtain the phase response at the peak of the primary finite-bandwidth intrinsic mode function of each high-frequency signal. and United and Determine the location of the partial discharge fault.

[0020] Specifically, in step S5, PRPD map feature extraction includes skew S k Steepness K u Discharge factor Q, cross-correlation coefficient CC, and phase asymmetry ψ.

[0021] Specifically, in step S6, the data imputation for multiple linear regression is as follows:

[0022] S601. Obtain the complete parameter set from the historical data, arrange and combine different parameters, fit the relationship of all combinations using multiple linear regression, obtain the data collected by online sensors, and identify the data type missing at a certain moment.

[0023] S602. Using missing data as the dependent variable ζ and non-missing data as the independent variable α, approximate αβ+ε=ζ using historical data, where β is the weight matrix and ε is the bias matrix.

[0024] S603. Use the fitting results to fill in the missing terms.

[0025] Specifically, in step S7, the joint time series prediction is performed as follows:

[0026] S701. Obtain historical time-series data of multiple parameters, and perform STL decomposition on the statistical characteristic quantity of the spectrum, grounding current, temperature and load current respectively to obtain the trend components of the historical data of each characteristic parameter.

[0027] S702. Specify the time span that needs to be predicted;

[0028] S703. Take the time span of the periodic quantity obtained from STL decomposition as the input window length of LSTM;

[0029] S704. Using the LSTM model, the trend components obtained from the decomposition of all parameters in step S701 are used as input for joint time series prediction to obtain the time series prediction results of the trend components of each characteristic parameter of the cable.

[0030] S705. Add the trend prediction results of different parameters obtained in step S704 to their respective periodic components to achieve multi-parameter joint time series prediction.

[0031] Specifically, in step S9, the fault diagnosis using a depthwise convolutional belief network (DCBN) is performed as follows:

[0032] The cable fault diagnosis historical dataset obtained in step S8 is normalized to [-1,1] and used as the input of the DCBN cable fault diagnosis model. The partial discharge type is one-hot encoded and used as the output.

[0033] Set the interface parameters of the DCBN cable fault diagnosis model, namely the number of neurons in the input layer and the output layer. The number of input layers is equal to the number of feature parameters, and the number of output layers is equal to the number of partial discharge types.

[0034] The internal network structure of the DCBN cable fault diagnosis model is set up. The effective feature values ​​of the output are extracted by the CNN at the front end of the internal network. Unsupervised optimization of the multi-layer RBM of the internal network structure is performed. The BP algorithm is used to perform global optimization of the internal network structure by utilizing the fully connected terminals at the end of the internal network structure.

[0035] The multi-parameter joint time series prediction results are obtained by normalizing the normalized scale of the historical dataset in step S7; the multi-parameter joint time series prediction results are input into the optimal DCBN cable fault diagnosis model to obtain the final diagnosis results, thereby realizing the prediction of cable partial discharge faults.

[0036] Furthermore, the optimal DCBN cable fault diagnosis model is as follows:

[0037] Obtain historical fault datasets, determine the number of neurons in the input and output layers, the number of layers in the network structure RBM, the number of neurons in each layer of the RBM, the parameters of the convolutional and pooling layers, the convolutional kernel, and padding; then, perform one-hot encoding on the partial discharge fault types in the dataset as the output of the DCBN cable fault diagnosis model, then normalize the feature parameters to [-1,1] and divide them into training and test sets, train using the training set and fine-tune using the test set, and save the model with the best performance on the test set as the DCBN cable fault diagnosis model.

[0038] Secondly, embodiments of the present invention provide a high-voltage cable partial discharge fault prediction system, comprising:

[0039] The data module collects the status data of high-voltage cables, constructs a cable partial discharge fault location dataset, a cable status prediction historical dataset, and a raw cable fault diagnosis historical dataset.

[0040] The reconstruction module uses wavelet filtering to process the high-frequency signals in the cable partial discharge fault location dataset obtained by the data module, and then reverse reconstructs the filtered high-frequency signals.

[0041] The extraction module extracts modal components from the high-frequency signal obtained by the reconstruction module to obtain the primary finite-bandwidth intrinsic mode function of the high-frequency signal.

[0042] The localization module, based on the two-end phase response method, uses the high-frequency signal in the cable partial discharge fault localization dataset in the fast Fourier distribution transformation processing data module to obtain the corresponding phase response spectrum. It then extracts the phase response where the peak value of the high-frequency signal in the primary finite bandwidth intrinsic mode function obtained by the extraction module is located, and uses the cable partial discharge phase attenuation characteristic equation to realize the fault localization of high-voltage cable partial discharge.

[0043] The first replacement module extracts the morphological statistical features of the PRPD map in the historical data set for cable condition prediction obtained by the data module, and uses the morphological statistical features of the PRPD map to replace the PRPD map, thus forming a new historical data set for cable condition prediction.

[0044] The data filling module uses multiple linear regression to fill in the various cable feature parameters in the new cable condition prediction historical dataset obtained by the first replacement module, thus forming the cable condition historical dataset.

[0045] The decomposition module uses a time series trend decomposition algorithm to decompose the historical data of the cable status obtained by the filling module into trend components, periodic components and remainders. Then, a long short-term memory network is used to achieve joint time series prediction of the trend components of each historical data to obtain the prediction results of various characteristic parameters of the high-voltage cable.

[0046] The second replacement module extracts the morphological statistical features of the PRPD map in the historical data set for cable fault diagnosis obtained by the data module, and uses the morphological statistical features of the PRPD map to replace the PRPD map, thus forming a new historical data set for cable fault diagnosis.

[0047] The prediction module uses the new historical dataset of cable fault diagnosis obtained by the second replacement module to establish a fault diagnosis model of deep convolutional belief network. The model with the best training results is saved as the DCBN cable fault diagnosis model. The prediction results of various feature parameters of high voltage cable obtained by the decomposition module are used as the input of the DCBN cable fault diagnosis model to realize the prediction of recent partial discharge faults of cable.

[0048] Compared with the prior art, the present invention has at least the following beneficial effects:

[0049] A method for predicting partial discharge (PD) faults in high-voltage cables utilizes historical and online data to locate and predict PD faults. Wavelet denoising and variational mode decomposition (VMD) are performed on the online high-frequency signals, selecting the phase response with the highest amplitude in the IMF1 (Integrated Mode Factor 1) and combining it with the cable's two-end signals to locate the PD. Based on the multiple linear regression (MLR) algorithm, historical data is used to approximate the correlation characteristics between grounding current, temperature, load current, and PRPD (Predictive Phase Discharge) pattern morphological statistical features, enabling real-time filling of missing items in the online monitoring data. A joint time-series prediction of multiple parameters is achieved based on a time-series decomposition model (STL) and a time-series prediction model (LSTM). A DCBN (Deep Convolutional Belief Network) is established to explore the potential correlations between various parameters and cable PD types. This method enables the prediction, location, and diagnosis of PD faults in high-voltage cables.

[0050] Furthermore, during the long-term operation and maintenance of cables, a large amount of historical and valid data is accumulated. Collecting this data into cable partial discharge fault location datasets, cable condition prediction historical datasets, and original cable fault diagnosis historical datasets can provide data support for cable fault location, condition prediction, and insulation fault diagnosis model training.

[0051] Furthermore, accurate partial discharge fault location can effectively reduce the manpower and material resources consumed in the operation and maintenance of high-voltage cables. Modal component extraction is performed on the high-frequency signal of the cable, and the Lagrange multiplier method and penalty factor α are introduced to constrain the variational decomposition of the high-frequency partial discharge signal, obtaining the primary finite-bandwidth intrinsic mode functions of the two high-frequency signals, providing a signal reference for dual-end phase response location.

[0052] Furthermore, current research on the phase attenuation characteristics of partial discharge in cables is relatively mature, and the mainstream wavelet analysis for denoising and variational mode analysis can effectively obtain the phase response from the original high-frequency signal. A fast Fourier transform is performed on the original high-frequency signal to obtain the sensor's phase response spectrum. The peak position of the intrinsic mode function (IMF1) obtained after filtering the original signal and performing variational mode decomposition (VMD) is used to obtain the sensor's phase response. Finally, the phase attenuation characteristic equation is solved to locate the partial discharge fault.

[0053] Furthermore, statistical physical quantities of the partial discharge (PRPD) spectrum in the cable partial discharge spectrum are extracted. Several numerical features are used to represent key information in the partial discharge spectrum, facilitating subsequent data supplementation and time series prediction. Specific statistical physical quantities of the spectrum include: skew S. k Steepness K u Discharge factor Q, cross-correlation coefficient CC, and phase asymmetry ψ. Tilt S k The degree of symmetry and tilt of the shape of the PRPD map represents the degree of tilt. If S k If S equals 0, it indicates that the spectrum is symmetrical. k A value greater than 0 indicates that the shape of the graph is tilted towards the value less than the arithmetic mean. If S k A value less than 0 indicates that the shape of the graph is sloping towards a slope greater than the arithmetic mean. Steepness K u The degree of bulge in the PRPD map relative to the normal distribution, if K u A value of 0 indicates that the shape of the spectrum and the normal distribution are generally smooth. If K u A value greater than 0 indicates a sharper distribution than a normal distribution; if K... u A value less than 0 indicates a flatter distribution than normal. The discharge factor Q represents the discharge difference in the positive and negative half-cycles of the PRPD pattern. The cross-correlation coefficient CC represents the similarity of the PRPD pattern profiles in the positive and negative half-cycles, ranging from 0 to 1, indicating decreasing profile differences. The phase asymmetry ψ represents the difference in the initial discharge phase of the positive and negative half-cycles of the PRPD pattern.

[0054] Furthermore, partial discharge in high-voltage cables is a process of breakdown discharge occurring in localized areas between conductors under applied voltage during the operation of electrical equipment. The type and extent of this activity inevitably lead to varying degrees of change in the cable's electrothermal characteristics. Load current, cable temperature, and grounding current are manifestations of these electrothermal characteristics, and the statistical physical quantities of the PRPD (Partial Discharge Perimeter) spectrum serve as statistical representations of the degree and phenomenon of partial discharge in cables. These factors are inherently strongly correlated. Using historical data and based on the multiple linear regression (MLR) algorithm, the potential relationships between these factors are explored. This allows for adaptive compensation of data gaps caused by different sensor sampling frequencies or sampling anomalies in the cable monitoring system, laying the foundation for subsequent prediction.

[0055] Furthermore, Seasonal Trend Decomposition (STL) is a widely used and robust time series decomposition method that can decompose a time series into a trend term, a periodic term, and a remainder term. Long Short-Term Memory (LSTM) networks, utilizing input, forget, and output gates, can effectively uncover the correlations between multiple parameters and the temporal characteristics of the data, thereby achieving multi-parameter time series prediction. To achieve higher prediction accuracy, the trend term obtained from STL decomposition is used as the input to the LSTM, and then the periodic term obtained from STL decomposition is added to the LSTM's prediction output.

[0056] Furthermore, Deep Convolution Belief Networks (DCBN) are a deep learning classification method based on a large training dataset. It utilizes convolutional layers to extract features from the input data, then employs several layers of Restricted Boltzmann Machines (RBMs) for feature refinement and transformation, and finally uses a Multi-layer Perceptron (MLP) and ReLU (Rectified Linear Units) as activation functions to achieve classification and diagnosis. Backpropagation (BP) is then used to fine-tune the overall network. DCBN achieves higher accuracy than general classifiers; therefore, it was chosen as the network template for high-voltage cable partial discharge fault diagnosis.

[0057] Furthermore, Deep Convolution Belief Networks (DCBN) are used as the terminal model for cable fault diagnosis. Due to its deep feature mining and ease of use, it can effectively improve the utilization rate of data and the timeliness of fault diagnosis, so as to achieve rapid and accurate diagnosis of cable partial discharge fault prediction.

[0058] It is understandable that the beneficial effects of the second aspect mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0059] In summary, this invention enables the location, prediction, diagnosis, and analysis of partial discharge faults in high-voltage power cables, effectively improving the insulation fault potential perception and adaptive monitoring capabilities of high-voltage cables.

[0060] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0061] Figure 1 This is a schematic diagram illustrating the entire process of high-voltage cable partial discharge fault prediction driven by state holographic perception data according to the present invention.

[0062] Figure 2 This is a flowchart illustrating the online location of partial discharge faults in high-voltage cables according to the present invention;

[0063] Figure 3 This is a flowchart illustrating the online imputation method based on multiple linear regression (MLR) of this invention.

[0064] Figure 4 This is a flowchart illustrating the multi-parameter joint time series prediction based on STL and LSTM of this invention.

[0065] Figure 5 This is a flowchart illustrating the construction and training of the DCBN network model for high-voltage cable insulation fault diagnosis in this invention.

[0066] Figure 6 This is a high-frequency filtered signal obtained from experimental monitoring on the A-phase circuit of the cable of the present invention.

[0067] Figure 7 This is another type of high-frequency filtered signal obtained from experimental monitoring on the A-phase circuit of the cable of the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0070] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0071] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" relationship.

[0072] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0073] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0074] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0075] This invention provides a method for predicting partial discharge (PD) faults in high-voltage cables. It collects multiple PD, temperature, and power data from online cable monitoring, and uses filtering and modal extraction to locate PD faults in the high-voltage cable during online monitoring. A historical fault dataset for the high-voltage cable is established, and a DCBN (Distributed Dynamic Block Design) model for PD diagnosis is trained. Morphological statistical features are extracted from the online PRPD (Pressure Partial Discharge) map. Missing monitoring parameters are filled in. Using the remaining monitoring parameters and the extracted PD morphological statistical features, trend and periodic terms are extracted and combined to build a trend prediction model. The trend prediction results of the remaining monitoring parameters and PD morphological statistical features are used as input to the PD diagnosis model to obtain the PD fault diagnosis result for the high-voltage cable. This method enables the prediction, location, and diagnosis of PD faults in high-voltage cables. Based on historical operating data and online monitoring data, this invention achieves real-time location and decision support for high-voltage cable insulation operation risks, contributing to improved reliability early warning, global perception, and robust control capabilities of power system digital twins.

[0076] Please see Figure 1 This invention's prediction method consists of three parts: data acquisition, online location, and predictive diagnosis. Based on the phase attenuation characteristics of high-frequency partial discharge signals during propagation, online acquisition of high-frequency signals enables online location of partial discharge faults. Using multiple linear regression, based on complete time-node data sets, the potential relationships between parameters are arranged and combined to fill in missing items in the historical time series data, laying a solid data foundation for subsequent predictions. Then, based on STL and LSTM models, the development trends and periodic quantities in historical data are comprehensively mined to achieve multi-parameter joint time series prediction. Finally, based on the DCBN network, a partial discharge fault diagnosis model is built using PRPD morphological statistical features, grounding current, load current, and photometric temperature as input features, achieving the final prediction and diagnosis of high-voltage cable insulation faults.

[0077] Data acquisition is divided into two categories: one is partial discharge monitoring signals, and the other is other monitoring parameters.

[0078] Partial discharge monitoring signals include high-frequency signals and PRPD spectra; other monitoring parameters include grounding current, load current, and optical temperature.

[0079] The specific sampling cycle can be flexibly adjusted by the sensors in the on-site cable monitoring system.

[0080] Online positioning utilizes high-frequency signals in partial discharge monitoring signals, requiring a high-frequency signal group collected by two adjacent high-frequency sensors;

[0081] By using classical wavelet analysis, after selecting the wavelet basis and threshold, the original high-frequency signal is denoised and purified.

[0082] Then, a fast Fourier transform is performed on the acquired raw signals to obtain the phase response spectra of the two sensor signals.

[0083] Next, the finite-bandwidth intrinsic mode function (IMF1) of the denoised high-frequency signal is extracted using variational mode decomposition. Combined with the phase response spectrum and the peak position of IMF1, the phase response of the two high-frequency signals is obtained. and

[0084] Finally, utilizing the partial discharge phase attenuation characteristics of the cable, a combined... and The specific location of the partial discharge fault point between the two high-frequency sensors was obtained.

[0085] The data required for predictive diagnosis includes PRPD patterns, grounding current, load current, and photometric temperature. The operation process can be divided into four parts: PRPD pattern morphological statistical feature extraction, data imputation, time series prediction, and fault diagnosis. Among these, PRPD pattern morphological statistical feature extraction involves extracting physical quantities from the pattern that effectively represent statistical physical characteristics, including: skewness S... k Steepness K u Discharge factor Q, cross-correlation coefficient CC, and phase asymmetry ψ.

[0086] Data imputation utilizes multiple linear regression to complete missing feature parameter data at specific time points. This aims to prevent data asymmetry or omissions caused by different sensor sampling frequencies or equipment malfunctions. It uses complete datasets from historical time series data, with missing items as the dependent variable and non-missing items as independent variables, to approximate the relationship between the independent and dependent variables using historical data. To achieve excellent timeliness in field applications and enable rapid prediction and diagnosis, various feature parameters can be permuted and combined to obtain a comprehensive data imputation model that automatically completes missing items during sensor data acquisition and database upload.

[0087] Time series forecasting utilizes STL and LSTM to predict the future trends of various characteristic parameters of high-voltage cables based on complete historical time series data of multiple parameters. STL is used to decompose the historical time series data of each parameter, retaining the trend and periodic terms. LSTM is then used, with the period length of the optimal periodic term predicted by LSTM from the STL decomposition results set as the LSTM window length, enabling joint time series forecasting of multiple parameter periodic terms. Finally, each parameter is added with its respective periodic term to obtain the final forecast result.

[0088] Fault diagnosis utilizes a DCBN neural network model based on historical fault datasets, using multiple parameters as input features to obtain the final prediction and diagnosis results for high-voltage cable insulation faults. This includes model building, training, and usage. For DCBN model building and training, the number of neurons in the input and output layers needs to be determined based on the number of parameters and the number of possible cable insulation fault types. Parameter tuning and optimization determine the final network parameters and structure, including the number of layers in the RBM, the number of neurons per RBM layer, the parameters of convolutional and pooling layers, convolutional kernels, padding, etc. Finally, the DCBN model with the best performance is saved. Model usage involves using the prediction results of multiple parameters as input to the saved DCBN model, and obtaining the final prediction and diagnosis results based on the probabilities of various fault types output by the model.

[0089] Please see Figure 2 The present invention provides a method for predicting partial discharge faults in high-voltage cables, comprising the following steps:

[0090] S1. Collect the required cable status data using online monitoring devices, including PRPD patterns, high-frequency signals, grounding current, temperature, and load current; group the high-frequency signals collected by two high-frequency sensors located close to each other on the cable as a set, and use them as a cable partial discharge fault location dataset; use the PRPD patterns, grounding current, temperature, and load current as a cable status prediction historical dataset; collect the PRPD patterns, grounding current, temperature, and load current data of the cable under normal conditions and when partial discharge occurs to form the original cable fault diagnosis historical dataset.

[0091] S2. The two high-frequency signals in the cable partial discharge fault location dataset are obtained by using wavelet filtering in step S1. Wavelet decomposition is performed on the original signal to obtain the scale coefficients. Then, high-frequency noise is filtered out by thresholding and the filtered signal is obtained by reverse reconstruction.

[0092] S201. Selecting a wavelet basis: Considering that the partial discharge signal is an exponentially decaying oscillating type, selecting a similar wavelet basis is beneficial for noise reduction.

[0093] S202. Determine the decomposition scale. The decomposition scale should take into account both the purpose of signal-to-noise separation and the prevention of distortion.

[0094] S203. Define the threshold and threshold function;

[0095] S204. Reverse reconstruction yields the purified high-frequency signal.

[0096] S3. Modal component extraction is performed on the two high-frequency signals obtained in step S2 after data processing. By introducing the Lagrange multiplier method and the penalty factor α, the variational decomposition of the partial discharge high-frequency signal is constrained. The Lagrange multiplication factor and the update factor σ are used to iteratively optimize the signal through an alternating direction algorithm until the convergence criterion tolerance ε is met, thereby obtaining a series of finite bandwidth intrinsic mode functions (IMFs) after their decomposition. Finally, the primary finite bandwidth intrinsic mode functions (IMF1) of the two high-frequency signals are obtained.

[0097] S4. Based on the dual-end phase response method, the two high-frequency signals in the cable partial discharge fault location dataset in step S1 are processed by fast Fourier distribution transformation to obtain their respective phase response spectra. The phase responses of the two IMF1 peaks obtained in step S3 are combined with the distance between the two sensors, and the cable partial discharge phase attenuation characteristic equation is used to realize the fault location of cable partial discharge.

[0098] Fault location is achieved using the two-terminal phase response method as follows:

[0099] S401. Acquire signals from two high-frequency sensors located adjacent to each other on the cable;

[0100] S402. Perform a fast Fourier transform on the acquired signals to obtain the phase response spectra of the two signals respectively;

[0101] S403. After steps S2 and S3, the IMF1 of both is obtained;

[0102] S404. Obtain the phase response where the peak values ​​are located in the two IMF1 values ​​respectively. and

[0103] S405, Lianli and Use the following formula to determine the location of the partial discharge fault.

[0104]

[0105] in, Let d be the peak phase of the original partial discharge high-frequency signal, l be the location of the partial discharge, and β1 and β2 be the phase constants of the partial discharge signal at the cable locations of the two sensors, respectively. These phase constants are calculated using the following formula:

[0106]

[0107] Where R0, G0, L0, and C0 are the resistance, conductance, inductance, and capacitance values ​​per unit length of cable, respectively, and ω is the angular frequency of the peak position of the partial discharge IMF1.

[0108] S5. Extract the morphological statistical features of the PRPD map from the historical data set of cable condition prediction obtained in step S1, and use the morphological statistical features of the PRPD map to replace the PRPD map to form a new historical data set of cable condition prediction.

[0109] PRPD map feature extraction includes the following: skew S k Steepness K u Discharge factor Q, cross-correlation coefficient CC, phase asymmetry ψ;

[0110] Skew S k The calculation method is shown in equation (3):

[0111]

[0112] Where N is the number of phase windows within half a power frequency cycle of the spectrum. It is the phase of the i-th phase window. It is the phase width, μ, p i And σ are respectively based on When y is a variable, the mean, probability, and variance of partial discharge occurring in the i-th phase window of the spectrum are calculated using formulas (4) to (6), where y is the ordinate of the two-dimensional spectrum.

[0113]

[0114]

[0115]

[0116] Steepness K u The calculation formula is:

[0117]

[0118] The formula for calculating the discharge factor Q is:

[0119]

[0120] in, and These represent the total discharge amount during the positive and negative half-cycles of the phase, respectively. and These are the total number of discharges during the positive and negative half-cycles, respectively.

[0121] The formula for calculating the cross-correlation coefficient (CC) is:

[0122]

[0123] in, and These are the average discharge quantities during the positive and negative half-cycles of the i-th phase window of the spectrum, respectively; the formula for calculating the phase asymmetry ψ is:

[0124]

[0125] in, and These represent the initial discharge phases of the spectrum during the positive and negative half-cycles, respectively.

[0126] S6. For each cable feature parameter in the historical cable condition prediction dataset obtained in step S5, multiple linear regression is used to fill in the missing feature parameters, thereby forming a historical cable condition dataset, which provides reliable data support for subsequent time series prediction.

[0127] Please see Figure 3 The specific steps for imputation of missing data in multiple linear regression are as follows:

[0128] S601. Obtain the complete parameter set from the historical data, arrange and combine different parameters, fit the relationship of all combinations using multiple linear regression, obtain the data collected by online sensors, and identify the data type missing at a certain moment.

[0129] S602. Using missing terms as the dependent variable and non-missing terms as the independent variables, approximate the following formula using historical data:

[0130] αβ+ε=ζ (11)

[0131] Where α is the independent variable, ζ is the dependent variable, β is the weight matrix, and ε is the bias matrix.

[0132] S603. Use the fitting results to fill in the missing terms.

[0133] S7. Using the Time Series Trend Decomposition (STL) algorithm based on Loess, the historical data in the cable condition prediction historical dataset obtained in step S6 are decomposed into trend components, periodic components, and remainder components. Then, the joint time series prediction of the trend components of each historical data is achieved through the Long Short-Term Memory (LSTM) network to obtain the prediction results of various characteristic parameters of the cable.

[0134] Please see Figure 4 The joint time-series prediction of various characteristic parameters of the cable using the temporal decomposition system (STL) and the long short-term memory network (LSTM) is specifically as follows:

[0135] S701. Obtain historical time-series data of multiple parameters, and perform STL decomposition on the statistical characteristic quantity of the spectrum, grounding current, temperature and load current respectively to obtain the trend components of the historical data of each characteristic parameter.

[0136] S702. Specify the required forecast time span;

[0137] S703. Take the time span of the periodic quantity obtained from STL decomposition as the input window length of LSTM (if multiple different time spans are obtained from STL decomposition, select the one with the best prediction effect).

[0138] S704. Using the LSTM model, the trend components obtained from the decomposition of all parameters in step S701 are used as input for joint time series prediction to obtain the time series prediction results of the trend components of each characteristic parameter of the cable.

[0139] S705. The trend prediction results of different parameters are added to their respective periodic components to achieve multi-parameter joint time series prediction.

[0140] S8. Extract the morphological statistical features of the PRPD map from the historical cable fault diagnosis dataset obtained in step S1, including skewness S. k Steepness K u Discharge factor Q, cross-correlation coefficient CC, and phase asymmetry ψ are used to replace the PRPD spectrum with morphological statistical features of the PRPD spectrum, forming a new historical dataset for cable fault diagnosis.

[0141] S9. Using the historical cable fault diagnosis dataset obtained in step S8, establish a deep convolutional belief network (DCBN) fault diagnosis model, and save the model with the best training results as the DCBN cable fault diagnosis model. Use the prediction results of various cable feature parameters obtained in step S7 as the input of the DCBN cable fault diagnosis model to realize the prediction of recent partial discharge faults of the cable.

[0142] Please see Figure 5 The overall process for building and training a DCBN network model for high-voltage cable insulation fault diagnosis is as follows:

[0143] S901. Obtain historical fault dataset, then proceed to step S902;

[0144] S902. Determine the number of neurons in the input and output layers, the number of layers in the network structure RBM, the number of neurons in each layer of the RBM, the parameters of the convolutional and pooling layers, such as convolutional kernels and padding, and then proceed to step S903.

[0145] S903. Perform one-hot encoding on the partial discharge fault type of the dataset as the output of the DCBN network, and then proceed to step S904.

[0146] S904. Normalize the feature parameters to [-1,1] and divide the training set and test set, then proceed to step S905;

[0147] S905. Train using the training set and fine-tune using the test set, then proceed to step S906.

[0148] S906. Save the model with the best test set performance as the DCBN cable fault diagnosis model.

[0149] Please see Figure 6 The specific method for fault diagnosis using a depthwise convolutional belief network (DCBN) is as follows:

[0150] S907. Normalize the cable fault diagnosis historical dataset obtained in step S8 to [-1,1] and use it as the input of the DCBN network. Perform one-hot encoding on the partial discharge type and use it as the output.

[0151] S908. Set the interface parameters of DCBN, namely the number of neurons in the input layer and the output layer. The number of input layers is equal to the number of feature parameters, and the number of output layers is equal to the number of partial discharge types.

[0152] S909. Set the internal network structure of DCBN, namely the number of layers of RBM, the number of neurons in each layer of RBM, the parameters of convolutional and pooling layers, convolutional kernels, padding, etc.

[0153] S910. Using the CNN at the front end of the network, extract the effective feature values ​​of the output;

[0154] S911. Perform unsupervised optimization on the multi-layer RBM of DCBN;

[0155] S912. Utilize the full connectivity at the network endpoints and use the BP algorithm to optimize the entire network.

[0156] S913. Use the normalized dataset from step S907 to repeatedly train the network and save the model with the best diagnostic performance.

[0157] S914. Normalize the multi-parameter joint time series prediction results obtained in step S7 using the normalized scale of the historical dataset in step S907.

[0158] S915. Input the result of step S914 into DCBN to obtain the final diagnostic result and realize the prediction of cable partial discharge fault.

[0159] In another embodiment of the present invention, a high-voltage cable partial discharge fault prediction system is provided. This system can be used to implement the above-mentioned high-voltage cable partial discharge fault prediction method. Specifically, the high-voltage cable partial discharge fault prediction system includes a data module, a reconstruction module, an extraction module, a positioning module, a first replacement module, a filling module, a decomposition module, a second replacement module, and a prediction module.

[0160] The data module collects the status data of high-voltage cables, constructs a partial discharge fault location dataset, a historical dataset for cable status prediction, and a historical dataset for original cable fault diagnosis.

[0161] The reconstruction module uses wavelet filtering to process the high-frequency signals in the cable partial discharge fault location dataset obtained by the data module, and then reverse reconstructs the filtered high-frequency signals.

[0162] The extraction module extracts modal components from the high-frequency signal obtained by the reconstruction module to obtain the primary finite-bandwidth intrinsic mode function of the high-frequency signal.

[0163] The localization module, based on the two-end phase response method, uses the high-frequency signal in the cable partial discharge fault localization dataset in the fast Fourier distribution transformation processing data module to obtain the corresponding phase response spectrum. It then extracts the phase response where the peak value of the high-frequency signal in the primary finite bandwidth intrinsic mode function obtained by the extraction module is located, and uses the cable partial discharge phase attenuation characteristic equation to realize the fault localization of high-voltage cable partial discharge.

[0164] The first replacement module extracts the morphological statistical features of the PRPD map in the historical data set for cable condition prediction obtained by the data module, and uses the morphological statistical features of the PRPD map to replace the PRPD map, thus forming a new historical data set for cable condition prediction.

[0165] The data filling module uses multiple linear regression to fill in the various cable feature parameters in the new cable condition prediction historical dataset obtained by the first replacement module, thus forming the cable condition historical dataset.

[0166] The decomposition module uses a time series trend decomposition algorithm to decompose the historical data of the cable status obtained by the filling module into trend components, periodic components and remainders. Then, a long short-term memory network is used to achieve joint time series prediction of the trend components of each historical data to obtain the prediction results of various characteristic parameters of the high-voltage cable.

[0167] The second replacement module extracts the morphological statistical features of the PRPD map in the historical data set for cable fault diagnosis obtained by the data module, and uses the morphological statistical features of the PRPD map to replace the PRPD map, thus forming a new historical data set for cable fault diagnosis.

[0168] The prediction module uses the new historical dataset of cable fault diagnosis obtained by the second replacement module to establish a fault diagnosis model of deep convolutional belief network. The model with the best training results is saved as the DCBN cable fault diagnosis model. The prediction results of various feature parameters of high voltage cable obtained by the decomposition module are used as the input of the DCBN cable fault diagnosis model to realize the prediction of recent partial discharge faults of cable.

[0169] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in the operation of a high-voltage cable partial discharge fault prediction method, including:

[0170] High-voltage cable condition data is collected to construct a cable partial discharge fault location dataset, a cable condition prediction historical dataset, and the original cable fault diagnosis historical dataset. High-frequency signals in the cable partial discharge fault location dataset are processed using wavelet filtering, and the filtered high-frequency signals are reconstructed in reverse. Modal component extraction is performed on the high-frequency signals to obtain the primary finite bandwidth intrinsic mode function (PRPD). Based on the two-terminal phase response method, the high-frequency signals in the cable partial discharge fault location dataset are processed using Fast Fourier Transform (FFT) to obtain the corresponding phase response spectra. The phase response at the peak value in the primary finite bandwidth intrinsic mode function of the high-frequency signal is combined with the phase attenuation characteristic equation of cable partial discharge to achieve fault location of high-voltage cable partial discharge. Morphological statistical features of the PRPD spectra in the cable condition prediction historical dataset are extracted and used to replace the PRPD spectra, forming a new cable condition prediction historical dataset. Multiple linear regression analysis is then performed. Regression is used to fill in the data for various cable feature parameters in the new historical cable condition prediction dataset, forming a historical cable condition dataset. A time series trend decomposition algorithm is used to decompose the historical data in the historical cable condition dataset into trend components, periodic components, and remainders. Then, a long short-term memory network is used to perform joint time-series prediction of the trend components of each historical data item, obtaining the prediction results for various feature parameters of the high-voltage cable. Morphological statistical features of the PRPD (Partial Discharge Peripheral) map are extracted from the historical cable fault diagnosis dataset and used to replace the PRPD map, forming a new historical cable fault diagnosis dataset. A fault diagnosis model using a deep convolutional belief network is established using the new historical cable fault diagnosis dataset. The model with the best training results is saved as the DCBN (Distributed Partial Discharge Network) cable fault diagnosis model. The prediction results of various feature parameters of the high-voltage cable are used as the input of the DCBN cable fault diagnosis model to achieve recent partial discharge fault prediction of the cable.

[0171] In another embodiment of the present invention, a storage medium is also provided, specifically a computer-readable storage medium (memory). This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0172] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the high-voltage cable partial discharge fault prediction method in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor to perform the following steps:

[0173] High-voltage cable condition data is collected to construct a cable partial discharge fault location dataset, a cable condition prediction historical dataset, and the original cable fault diagnosis historical dataset. High-frequency signals in the cable partial discharge fault location dataset are processed using wavelet filtering, and the filtered high-frequency signals are reconstructed in reverse. Modal component extraction is performed on the high-frequency signals to obtain the primary finite bandwidth intrinsic mode function (PRPD). Based on the two-terminal phase response method, the high-frequency signals in the cable partial discharge fault location dataset are processed using Fast Fourier Transform (FFT) to obtain the corresponding phase response spectra. The phase response at the peak value in the primary finite bandwidth intrinsic mode function of the high-frequency signal is combined with the phase attenuation characteristic equation of cable partial discharge to achieve fault location of high-voltage cable partial discharge. Morphological statistical features of the PRPD spectra in the cable condition prediction historical dataset are extracted and used to replace the PRPD spectra, forming a new cable condition prediction historical dataset. Multiple linear regression analysis is then performed. Regression is used to fill in the data for various cable feature parameters in the new historical cable condition prediction dataset, forming a historical cable condition dataset. A time series trend decomposition algorithm is used to decompose the historical data in the historical cable condition dataset into trend components, periodic components, and remainders. Then, a long short-term memory network is used to perform joint time-series prediction of the trend components of each historical data item, obtaining the prediction results for various feature parameters of the high-voltage cable. Morphological statistical features of the PRPD (Partial Discharge Peripheral) map are extracted from the historical cable fault diagnosis dataset and used to replace the PRPD map, forming a new historical cable fault diagnosis dataset. A fault diagnosis model using a deep convolutional belief network is established using the new historical cable fault diagnosis dataset. The model with the best training results is saved as the DCBN (Distributed Partial Discharge Network) cable fault diagnosis model. The prediction results of various feature parameters of the high-voltage cable are used as the input of the DCBN cable fault diagnosis model to achieve recent partial discharge fault prediction of the cable.

[0174] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0175] The invention will be explained using experimental data from a 220kV three-phase cable as an example. Figure 6 and Figure 7 The image shows the high-frequency signals obtained from filtering at both ends of a test monitoring of phase A circuit of this cable. According to the high-voltage cable partial discharge fault prediction method and system proposed in this invention, a Fast Fourier Transform (FFT) is performed on the original signals at both ends, followed by mode decomposition, yielding two IMF1 peak phase responses of 1.37 rad and 1.25 rad, respectively. Formulas 1 and 2 are combined to determine the partial discharge location, with a final location error of 0.023%.

[0176] The historical data of the cable's load current, photometric temperature, and A-phase grounding current are shown in the table.

[0177] Table 1 Historical Cable Data

[0178]

[0179]

[0180] Using load current, photometric temperature, and phase A grounding current as the prediction targets, the PRPD morphological statistical characteristics were first supplemented using MLR, and then a prediction model was established based on the STL-LSTM model. After multiple adjustments, the prediction model with the best fit was selected, achieving fitting accuracies of 93.235%, 95.856%, and 96.519% for load current, photometric temperature, and phase A grounding current, respectively. Therefore, the STL-LSTM model can be used as a basic model for predicting cable conditions. The prediction results for this cable over the next two days are shown in the table.

[0181] Table 2 Cable Forecast Data

[0182] Predicted time Load current / A Photometric temperature / °C Grounding current / A 2022-08-19 272 35.6 6.95 2022-08-20 265 35.8 6.86

[0183] Based on historical fault datasets, a high-voltage cable partial discharge fault diagnosis model based on DCBN was established. The historical fault dataset was divided into training and test sets in an 8:2 ratio. After model parameter optimization using DCBN, the DCBN model parameters were: 8 hidden layers, rbm learning rate of 0.1, ReLU activation function, and padding of 2. Results showed that the diagnostic accuracy on the training set was 96.568%, and the diagnostic accuracy on the test set was 94.056%. The established DCBN fault diagnosis model can effectively detect cable faults using multiple monitoring signals. Therefore, the prediction results in Table 2 were used for diagnosis, and the results are shown in Table 3. The probability of surface discharge faults in phase A of this high-voltage cable within the next two days is the highest, followed by air gap discharge faults.

[0184] Table 3 Cable Fault Prediction and Diagnosis Results

[0185]

[0186] This invention has the following characteristics:

[0187] (1) By combining online location and fault prediction methods for high-voltage cables and following the sequence of location first and then diagnosis, online monitoring, rapid location, trend prediction, and fault pre-diagnosis of high-voltage cables can be achieved. This not only provides guidance and reference for the reliable operation and maintenance of high-voltage cables, but also effectively saves the required time and resources.

[0188] (2) By extracting the statistical characteristics of physical quantities from the partial discharge PRPD spectrum of high-voltage cables, we can not only retain data with less storage resources, but also facilitate data supplementation and prediction. This reduces the computational load of the program, saves time, and improves accuracy.

[0189] (3) Using the physical quantities of PRPD spectrum, load current, cable temperature and grounding current as feature data, the MLR algorithm is used to explore the potential relationships between them and establish a data gap model, so that the adaptability of the overall solution is improved and it is not limited by the asynchronous sampling frequency and instability of the high-voltage cable monitoring system.

[0190] (4) By combining STL and LSTM algorithms, reliable time-series predictions of PRPD spectrum statistical physical quantities, load current, cable temperature, and grounding current are achieved, providing a source of features for high-voltage cable fault prediction and diagnosis, making it possible. This is of great significance for guiding the operation and maintenance of high-voltage cables for insulation faults.

[0191] In summary, this invention provides a method and system for predicting partial discharge faults in high-voltage cables. It utilizes the phase attenuation characteristics of high-frequency partial discharge signals during propagation to locate the partial discharge position. To lay a solid data foundation for time-series prediction, multiple linear regression is used to fill in missing terms in the required time series data. By combining online available signals from high-voltage cables with a time-series decomposition (STL) method and a long short-term memory (LSTM) network, the temporal trends of the data and the correlation between data structure components and fault types are jointly mined. This enables predictive diagnosis of insulation faults in high-voltage cables, effectively utilizing historical cable data and mining its potential value. It achieves partial discharge fault location and predictive diagnosis in high-voltage cables and provides guidance for the operation and maintenance of high-voltage cables.

[0192] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0193] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0194] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0195] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0196] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0197] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0198] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0199] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0200] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0201] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0202] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for partial discharge fault prediction of a high voltage cable, characterized by, Includes the following steps: S1. Collect the status data of high-voltage cables, construct a cable partial discharge fault location dataset, a cable status prediction historical dataset, and the original cable fault diagnosis historical dataset. S2. Use wavelet filtering to process the high-frequency signal in the cable partial discharge fault location dataset obtained in step S1, and then reconstruct the high-frequency signal in reverse. S3. Extract modal components from the high-frequency signal obtained in step S2 to obtain the primary finite bandwidth intrinsic mode function of the high-frequency signal. S4. Based on the two-end phase response method, the high-frequency signal in the cable partial discharge fault location dataset in step S1 is processed by the fast Fourier distribution transformation to obtain the corresponding phase response spectrum. The phase response where the peak value of the high-frequency signal in the primary finite bandwidth intrinsic mode function obtained in step S3 is located is combined with the phase response of the high-voltage cable partial discharge fault location using the cable partial discharge phase attenuation characteristic equation. S5. Extract the morphological statistical features of the PRPD map from the historical data set of cable condition prediction obtained in step S1, and use the morphological statistical features of the PRPD map to replace the PRPD map to form a new historical data set of cable condition prediction. S6. Use multiple linear regression to fill in the data of each cable feature parameter in the new cable condition prediction historical dataset obtained in step S5 to form a cable condition historical dataset. S7. Using the time series trend decomposition algorithm, the historical data in the cable status historical dataset obtained in step S6 are decomposed into trend components, periodic components and remainders. Then, the joint time series prediction of the trend components of each historical data is achieved through the long short-term memory network to obtain the prediction results of various characteristic parameters of the high-voltage cable. S8. Extract the morphological statistical features of the PRPD map from the historical data set for cable fault diagnosis obtained in step S1, and use the morphological statistical features of the PRPD map to replace the PRPD map to form a new historical data set for cable fault diagnosis. S9. Using the new historical dataset for cable fault diagnosis obtained in step S8, establish a fault diagnosis model of deep convolutional belief network, save the model with the best training results as the DCBN cable fault diagnosis model, and use the prediction results of various characteristic parameters of high voltage cable obtained in step S7 as the input of the DCBN cable fault diagnosis model to realize the prediction of recent partial discharge faults of the cable.

2. The high voltage cable partial discharge fault prediction method of claim 1, wherein, In step S1, the PRPD pattern, high-frequency signal, grounding current, temperature and load current data of the high-voltage cable are collected; the high-frequency signals collected by two high-frequency sensors located close to each other on the high-voltage cable are used as the cable partial discharge fault location dataset; the PRPD pattern, grounding current, temperature and load current are used as the cable condition prediction historical dataset; the PRPD pattern, grounding current, temperature and load current data of the high-voltage cable under normal condition and when partial discharge occurs are collected to form the original cable fault diagnosis historical dataset.

3. The high voltage cable partial discharge fault prediction method of claim 1, wherein, In step S3, the Lagrange multiplier method and penalty factor α are introduced to constrain the variational decomposition of the partial discharge high-frequency signal. The Lagrange multiplication factor and update factor σ are used to iteratively optimize the signal through an alternating direction algorithm until the convergence criterion tolerance ε is met, thereby obtaining a series of finite bandwidth intrinsic mode functions after decomposition. Finally, the primary finite bandwidth intrinsic mode functions of the two high-frequency signals are obtained.

4. The high voltage cable partial discharge fault prediction method of claim 1, wherein, Step S4 is as follows: Acquire two adjacent high-frequency signals on a high-voltage cable; perform a Fast Fourier Transform on the two high-frequency signals to obtain their phase response spectra; and obtain the phase response at the peak of the primary finite-bandwidth intrinsic mode function of each high-frequency signal. and United and Determine the location of the partial discharge fault.

5. The high voltage cable partial discharge fault prediction method of claim 1, wherein, In step S5, the PRPD pattern feature extraction includes skew S k , kurtosis K u , discharge factor Q, cross-correlation number CC and phase asymmetry ψ.

6. The high voltage cable partial discharge fault prediction method of claim 1, wherein, In step S6, the data imputation for multiple linear regression is specifically as follows: S601. Obtain the complete parameter set from the historical data, arrange and combine different parameters, fit the relationship of all combinations using multiple linear regression, obtain the data collected by online sensors, and identify the data type missing at a certain moment. S602. Using missing data as the dependent variable ζ and non-missing data as the independent variable α, approximate αβ+ε=ζ using historical data, where β is the weight matrix and ε is the bias matrix. S603. Use the fitting results to fill in the missing terms.

7. The high voltage cable partial discharge fault prediction method of claim 1, wherein, In step S7, the joint time series prediction is performed as follows: S701. Obtain historical time-series data of multiple parameters, and perform STL decomposition on the statistical characteristic quantity of the spectrum, grounding current, temperature and load current respectively to obtain the trend components of the historical data of each characteristic parameter. S702. Specify the time span that needs to be predicted; S703. Take the time span of the periodic quantity obtained from STL decomposition as the input window length of LSTM; S704. Using the LSTM model, the trend components obtained from the decomposition of all parameters in step S701 are used as input for joint time series prediction to obtain the time series prediction results of the trend components of each characteristic parameter of the cable. S705. Add the trend prediction results of different parameters obtained in step S704 to their respective periodic components to achieve multi-parameter joint time series prediction.

8. The high voltage cable partial discharge fault prediction method of claim 1, wherein, In step S9, the fault diagnosis using a depthwise convolutional belief network (DCBN) is specifically as follows: The cable fault diagnosis historical dataset obtained in step S8 is normalized to [-1,1] and used as the input of the DCBN cable fault diagnosis model. The partial discharge type is one-hot encoded and used as the output. Set the interface parameters of the DCBN cable fault diagnosis model, namely the number of neurons in the input layer and the output layer. The number of input layers is equal to the number of feature parameters, and the number of output layers is equal to the number of partial discharge types. The internal network structure of the DCBN cable fault diagnosis model is set up. The effective feature values ​​of the output are extracted by the CNN at the front end of the internal network. Unsupervised optimization of the multi-layer RBM of the internal network structure is performed. The BP algorithm is used to perform global optimization of the internal network structure by utilizing the fully connected terminals at the end of the internal network structure. The multi-parameter joint time series prediction results are obtained by normalizing the normalized scale of the historical dataset in step S7; the multi-parameter joint time series prediction results are input into the optimal DCBN cable fault diagnosis model to obtain the final diagnosis results, thereby realizing the prediction of cable partial discharge faults.

9. The high voltage cable partial discharge fault prediction method of claim 8, wherein, The optimal DCBN cable fault diagnosis model is as follows: Obtain historical fault datasets, determine the number of neurons in the input and output layers, the number of layers in the network structure RBM, the number of neurons in each layer of the RBM, the parameters of the convolutional and pooling layers, the convolutional kernel, and padding; then, perform one-hot encoding on the partial discharge fault types in the dataset as the output of the DCBN cable fault diagnosis model, then normalize the feature parameters to [-1,1] and divide them into training and test sets, train using the training set and fine-tune using the test set, and save the model with the best performance on the test set as the DCBN cable fault diagnosis model.

10. A high voltage cable partial discharge fault prediction system characterized by, include: The data module collects the status data of high-voltage cables, constructs a cable partial discharge fault location dataset, a cable status prediction historical dataset, and a raw cable fault diagnosis historical dataset. The reconstruction module uses wavelet filtering to process the high-frequency signals in the cable partial discharge fault location dataset obtained by the data module, and then reverse reconstructs the filtered high-frequency signals. The extraction module extracts modal components from the high-frequency signal obtained by the reconstruction module to obtain the primary finite-bandwidth intrinsic mode function of the high-frequency signal. The localization module, based on the two-end phase response method, uses the high-frequency signal in the cable partial discharge fault localization dataset in the fast Fourier distribution transformation processing data module to obtain the corresponding phase response spectrum. It then extracts the phase response where the peak value of the high-frequency signal in the primary finite bandwidth intrinsic mode function obtained by the extraction module is located, and uses the cable partial discharge phase attenuation characteristic equation to realize the fault localization of high-voltage cable partial discharge. The first replacement module extracts the morphological statistical features of the PRPD map in the historical data set for cable condition prediction obtained by the data module, and uses the morphological statistical features of the PRPD map to replace the PRPD map, thus forming a new historical data set for cable condition prediction. The data filling module uses multiple linear regression to fill in the various cable feature parameters in the new cable condition prediction historical dataset obtained by the first replacement module, thus forming the cable condition historical dataset. The decomposition module uses a time series trend decomposition algorithm to decompose the historical data of the cable status obtained by the filling module into trend components, periodic components and remainders. Then, a long short-term memory network is used to achieve joint time series prediction of the trend components of each historical data to obtain the prediction results of various characteristic parameters of the high-voltage cable. The second replacement module extracts the morphological statistical features of the PRPD map in the historical cable fault diagnosis dataset obtained by the data module, and uses the morphological statistical features of the PRPD map to replace the PRPD map, thus forming a new historical cable fault diagnosis dataset. The prediction module uses the new historical dataset of cable fault diagnosis obtained by the second replacement module to establish a fault diagnosis model of deep convolutional belief network. The model with the best training results is saved as the DCBN cable fault diagnosis model. The prediction results of various feature parameters of high voltage cable obtained by the decomposition module are used as the input of the DCBN cable fault diagnosis model to realize the prediction of recent partial discharge faults of cable.

Images