High-voltage cable intelligent fault diagnosis method and system based on machine learning

Abstract

The invention belongs to the technical field of cable fault diagnosis, and particularly relates to a high-voltage cable intelligent fault diagnosis method and system based on machine learning. The invention discloses a method for realizing fault diagnosis of a high-voltage cable by using a machine learning program. The method comprises the following specific steps: S1) decomposing a current waveform signal: collecting the current waveform signal of the high-voltage cable in a known fault state; extracting feature information of the current waveform signal by adopting a local mean value time-frequency analysis method; further decomposing the demodulated signal into a plurality of demodulated signal components, namely PF components; s2) constructing clustering prediction of a Gaussian mixture model: taking the plurality of pieces of demodulation signal component data after LMD processing as clustering factors to construct a feature matrix [X], and inputting the feature matrix [X] into the Gaussian mixture model for training; and further determining an optimal data classification result, namely a clustering label. The fault type of the high-voltage cable is identified; the reliability and stability of a power system are improved, and fault power failure time and economic loss are reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of cable fault diagnosis, and in particular relates to a high-voltage cable intelligent fault diagnosis method and system based on machine learning. Background Art

[0002] In modern power systems, the reliability and stability of high-voltage cables are of vital importance. However, since high-voltage cables may be affected by various factors during operation, various cable faults may occur due to overload operation or insulation aging, such as local insulation damage, insulation bubbles, flashover faults or fault arcs. These faults will not only affect the normal operation of the power system, but may also cause serious economic losses and safety hazards. However, traditional fault diagnosis methods have certain limitations:

[0003] It mainly relies on manual inspection and empirical judgment, which is inefficient and prone to misjudgment. In addition, the empirical judgment method requires complex detection methods, such as bridge method, pulse method, fault point burn-through method, etc. to perform fault testing and diagnosis. The diagnosis process is complex and changeable, resulting in untimely fault handling.

[0004] With the research and development of technologies such as machine learning, the existing machine learning methods have poor model stability and robustness when dealing with high-voltage cable fault diagnosis, and are prone to overfitting or underfitting. Summary of the invention

[0005] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a high-voltage cable intelligent fault diagnosis method and system based on machine learning.

[0006] The purpose of the present invention can be achieved through the following technical solutions: a high-voltage cable intelligent fault diagnosis system based on machine learning, including a data acquisition module, an industrial computer module and a host computer module; the data acquisition module includes a current transformer, a high-frequency current sensor and a temperature sensor; the current signal, high-frequency current signal and cable temperature signal in the cable are collected respectively; through these sensors, the system can obtain the operating status data of the cable and provide basic data support for fault diagnosis.

[0007] The industrial computer module includes an STC89C51 single-chip microcomputer, an AD converter and a communication circuit; the industrial computer module receives the signal data sent by the data acquisition module through the communication circuit and performs preliminary processing;

[0008] The STC89C51 single-chip microcomputer has solidified data acquisition program and serial communication program;

[0009] The AD converter is used to convert analog signals into digital signals for subsequent processing;

[0010] The communication circuit is divided into two paths, one is an RS232 communication circuit connected to the data acquisition module, and the other is a CAN bus circuit connected to the host computer module.

[0011] The host computer module includes a Xilinx chip and expandable memory; the Xilinx chip is used for data processing and data modeling of current signals; the expandable memory is used to store related data and machine learning applications.

[0012] Based on the above-mentioned intelligent fault diagnosis system for high-voltage cables, a method for high-voltage cable fault diagnosis is implemented using a machine learning program. The specific steps are as follows:

[0013] S1) Current waveform signal decomposition: collect the current waveform signal of the high-voltage cable when the high-voltage cable is in a known fault state; use the local mean value (LMD) time-frequency analysis method to extract the characteristic information of the current waveform signal, and then decompose it into multiple demodulated signal components, namely PF components;

[0014] S2) Construct a Gaussian mixture model for cluster prediction. The specific steps are as follows:

[0015] S21) using the multiple PF components after LMD processing as clustering factors to construct a feature matrix [X], and inputting it into a Gaussian mixture model for training; using an expectation maximization (EM) algorithm to maximize the log-likelihood function of the observed data, and then determining the best data classification result, i.e., the clustering label;

[0016] S22) initializing the Gaussian mixture model and setting the parameters of the Gaussian mixture model, including the Gaussian distribution number K and the covariance type;

[0017] S23) collecting the current waveform signal of the high-voltage cable in the unknown fault time period and inputting it into the trained Gaussian mixture model for cluster prediction to obtain the cluster label of each data point; and calculating the posterior probability that each data point belongs to each Gaussian component;

[0018] S24) According to the clustering labels and the posterior probability, the fault type is identified and determined to be local insulation damage, insulation bubbles, flashover fault or fault arc.

[0019] Preferably, a fault type threshold may be set, and when the posterior probability of a cluster label exceeds the fault type threshold, it is determined to be the fault type.

[0020] Preferably, in step S22), the Gaussian distribution quantity K is defined, and the sample entropy, permutation entropy or multi-scale entropy of one of the PF component data after the current waveform signal is decomposed is selected as the quantitative value.

[0021] Preferably, based on the preliminary selection of quantitative values, the sample entropy of the remaining PF component data is further verified by a cross-validation method, and then the value of the optimal Gaussian distribution number K is selected to improve the model performance.

[0022] Preferably, the fault type threshold of the local insulation damage is 0.8; the fault type threshold of the insulation bubble is 0.86; and the fault type threshold of the flashover fault is 0.9.

[0023] Preferably, the fault type threshold of the fault arc is 0.85, and the collected temperature of the high-voltage cable reaches and the collected temperature of the high-voltage cable reaches above 1200°C.

[0024] Preferably, the sample entropy of the current waveform signal is calculated as follows:

[0025] Define the sample sequence: the original current waveform signal is decomposed into multiple PF components, and one of the PF component data is selected as the sample sequence. The sample sequence can be expressed as: X(t) = {PF(t)}; where t = 1, 2, ..., N;

[0026] Construct an embedding vector: For any time series {PF(t)}, construct an embedding vector of dimension (m): X m (t)={x(t),x(t+1),...,x(t+m-1)};

[0027] Calculate the number of similar vector pairs: Calculate all embedding vector pairs X m (t) and X m (j) The distance between: And statistics satisfy d[X m (t), X m (j)]≤r

[0028] Calculate sample entropy: Among them, m represents the embedding dimension; r represents the similarity threshold; N represents the length of the time series;

[0029] represents the logarithm of similar vectors under the condition of embedding dimension (m+1) and similarity threshold (r);

[0030] It represents the sum of similarity vector logarithms of all embedded vectors under the condition that the embedding dimension is (m+1) and the similarity threshold is (r).

[0031] Beneficial effects of the present invention:

[0032] 1. Collect the current waveform signals of high-voltage cables under known fault conditions and perform LMD processing respectively; then, construct a Gaussian mixture model (GMM) for cluster analysis; then use the trained model to perform cluster prediction on new signal data, and obtain the cluster to which the new data belongs (i.e. the label corresponding to the Gaussian component), and then identify the fault type of the high-voltage cable; improve the reliability and stability of the power system, and reduce the fault outage time and economic losses.

[0033] 2. The present invention uses indicators such as sample entropy, permutation entropy or multi-scale entropy of the current waveform as the value of the Gaussian distribution quantity K to improve the stability and robustness of the model and reduce the risk of overfitting or underfitting.

[0034] 3. On the other hand, by real-time collection of the current waveform signal of the high-voltage cable and clustering prediction, real-time monitoring and early warning of high-voltage cable faults can be achieved, and faults can be discovered and handled in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0036] Figure 1 This is a schematic diagram of the layout structure of a high-voltage cable intelligent fault diagnosis system based on machine learning in the present invention;

[0037] Figure 2 A flowchart of a high-voltage cable intelligent fault diagnosis method based on machine learning in the present invention;

[0038] Figure 3 This is a flow chart of the method for calculating sample entropy in the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] See also Figure 1-Figure 2As shown, a high-voltage cable intelligent fault diagnosis system based on machine learning includes a data acquisition module, an industrial computer module and a host computer module; the data acquisition module includes a current transformer, a high-frequency current sensor and a temperature sensor; the current signal, high-frequency current signal and cable temperature signal in the cable are collected respectively; through these sensors, the system can obtain the operating status data of the cable and provide basic data support for fault diagnosis.

[0041] Among them, the current transformer is mainly used to collect the current signal in the cable.

[0042] High-frequency current sensors are mainly used to collect high-frequency signals generated by partial discharge. Partial discharge is an early manifestation of fault arc and has high-frequency characteristics.

[0043] The temperature sensor is used to collect the temperature signal of the cable. The temperature change can reflect the operating status of the cable and assist in determining the type of cable fault.

[0044] The industrial computer module includes an STC89C51 single-chip microcomputer, an AD converter and a communication circuit; the industrial computer module receives the signal data sent by the data acquisition module through the communication circuit and performs preliminary processing;

[0045] The STC89C51 single-chip microcomputer has solidified data acquisition program and serial communication program;

[0046] The AD converter is used to convert analog signals into digital signals for subsequent processing;

[0047] The communication circuit is divided into two paths, one is an RS232 communication circuit connected to the data acquisition module, and the other is a CAN bus circuit connected to the host computer module.

[0048] The host computer module includes a Xilinx chip and expandable memory; the Xilinx chip is used for data processing and data modeling of current signals; the expandable memory is used to store related data and machine learning applications.

[0049] Based on the above-mentioned intelligent fault diagnosis system for high-voltage cables, a method for high-voltage cable fault diagnosis is implemented using a machine learning program. The specific steps are as follows:

[0050] The characteristic information of fault arc detection is extracted by using the time-frequency analysis method based on the local mean (LMD). First, all the local maximum and local minimum points of the signal are found, and then the local mean function and envelope estimation function of the signal are obtained by sliding average. The local mean function is removed from the original signal and demodulated with the envelope estimation function until the standard pure frequency modulation function is obtained to terminate the loop iteration. The product of all envelope estimation functions generated during the iteration process is used as the envelope terminal function and multiplied with the last pure frequency modulation function to obtain the first-order PF component; then the PF components of each order are obtained repeatedly.

[0051] S1) Current waveform signal decomposition: collect the current waveform signal of the high-voltage cable under a known fault state; use the local mean value (LMD) time-frequency analysis method to extract the characteristic information of the current waveform signal, and then decompose it into multiple demodulated signal components, namely PF components; specifically, the specific process of the LMD time-frequency analysis method in step S1) is as follows:

[0052] S11) Assume that the current time-frequency signal during arc fault detection is x(t), find all local extreme points n i , and calculate n i and n i+1 The average value of

[0053]

[0054] All the average values ​​m i At the corresponding extreme point time t ni and t ni+1 The linear extension is performed between the two values, and the extended straight line is smoothed by the sliding average method to obtain the local mean function m i (t);

[0055] S12) Calculate the local amplitude a i ;

[0056]

[0057] All local amplitudes a i At the corresponding extreme point time t n and t ni+1 A straight line is extended between the two values, and the extended straight line is smoothed by the sliding average method to obtain the local mean function a 11 (t);

[0058] S13) Separate the local mean function m from the current signal x(t) 11 (t):

[0059] h 11 (t) = x(t) - m 11 (t);

[0060] S14) Use a 11 (t) for h 11 (t) is demodulated to obtain:

[0061] S 11 (t) = h 11 (t) / a 11 (t);

[0062] At this time, judge s 11 (t) Is it a pure frequency modulation function (the amplitude of a pure frequency modulation function is always 1, and -1 11 (t)<1; if a 12 (t) is S 11 (t) function is the envelope estimation function, then a 12 (t) = 1,) if it is not a pure frequency modulation function, return to step S14) for S 11 (t) Repeat the above iterative process until a pure FM signal S is obtained. 11 (t), then:

[0063] h 11 (t) = x(t) - m 11 (t);

[0064] h 12 (t) = s 11 (t)-m 12 (t);

[0065] …

[0066] H 1n (t) = s 1 (n-1)(t)-m 1n (t);

[0067] S15) Multiply all local envelope functions generated in the entire iteration process to obtain envelope signal a 1 (t):

[0068]

[0069] S16) The first PF component of the current signal is the envelope signal a 1 (t) and pure FM signal s 1n The product of (t):

[0070] PF 1 (t) = a 1 (t)s 1n (t);

[0071] S17) Subtract PF from the original signal x(t) 1 ​The (t) component is separated and the obtained new signal u(t) is used as a new original signal to repeat steps (1)-(6) and perform multiple cycles until u n (t) is a monotonic function;

[0072] u 1 (t) = x(t) - PF 1 (t);

[0073] u 2 (t) = u 1 (t)-PF 2 (t);

[0074] …

[0075] u n (t) = u n-1 (t)-PF n (t);

[0076] After multiple cycles of iterative decomposition, the original current waveform signal is finally decomposed into the sum of k PF components and a residual u(t).

[0077] S2) Construct a Gaussian mixture model for cluster prediction: GMM is a probabilistic model that assumes that the data consists of multiple Gaussian distributions. The EM algorithm estimates the parameters of each Gaussian distribution through iterative optimization, so that the model can best fit the data. The number of Gaussian distributions (K) determines the number of Gaussian distributions in the model, and the covariance type determines the shape and direction of each Gaussian distribution. By reasonably selecting these parameters, the model's fitting ability and clustering effect can be improved. The technical principle is that different fault types show different characteristics in the current waveform signal, and the fault type can be accurately identified and determined through clustering labels and posterior probabilities.

[0078] On the other hand, by collecting current waveform signals in real time and performing cluster prediction, real-time monitoring and early warning of high-voltage cable faults can be achieved, and faults can be discovered and handled in a timely manner.

[0079] S21) using the multiple PF components after LMD processing as clustering factors to construct a feature matrix X, In the formula, PF(t) represents the time series data of the demodulated signal component, t1 and tm represent the data at each time point, and k represents the data of the kth demodulated signal component; the feature matrix X is input into the Gaussian mixture model for training; the expectation maximization (EM) algorithm is used to maximize the log-likelihood function of the observed data, and then the best data classification result, that is, the cluster label, is determined;

[0080] S22) initializing the Gaussian mixture model and setting the parameters of the Gaussian mixture model, including the Gaussian distribution number K and the covariance type;

[0081] S23) collecting the current waveform signal of the high-voltage cable in the unknown fault time period and inputting it into the trained Gaussian mixture model for cluster prediction to obtain the cluster label of each data point; and calculating the posterior probability that each data point belongs to each Gaussian component;

[0082] S24) According to the clustering labels and the posterior probability, the fault type is identified and determined to be local insulation damage, insulation bubbles, flashover fault or fault arc.

[0083] Specifically, the function formula of the Gaussian mixture model is:

[0084]

[0085] In the above formula, x is the multidimensional data point to be modeled;

[0086] K is the number of Gaussian distributions, that is, the number of potential Gaussian clusters;

[0087] α k represents the mixing weight of the k-th Gaussian distribution;

[0088] represents the kth Gaussian distribution, whose mean is μ k , the covariance matrix is ​​∑ jk ;

[0089] The parameter set (θ) of the Gaussian mixture model includes all the mixture weights, mean vector, and covariance matrix;

[0090] The clustering process of Gaussian mixture model is:

[0091] 1: Define the number of Gaussian distributions K and set the initial parameter set for each Gaussian distribution: θ (0) =α (0) , μ (0) ,∑ (0) ;

[0092] 2: According to the current parameter θ (t) , calculate the posterior probability distribution γ of each data point belonging to each Gaussian distribution t (z (i) );

[0093]

[0094] 3: According to the posterior probability distribution calculated in step 2, further calculate the new θ (t+1) ;

[0095]

[0096] 4: Repeat steps 2 and 3 until convergence, and get the optimal parameter set θ (t+1) .

[0097] As a preferred embodiment, Figure 3 As shown, in step S22), the Gaussian distribution quantity K is defined, and the sample entropy, permutation entropy or multi-scale entropy of one of the PF component data after the current waveform signal is decomposed is selected as the quantitative value.

[0098] It is worth noting that the selection of the Gaussian distribution number K has a significant impact on the model performance; the present invention utilizes the distortion point of the current waveform signal as the value of the Gaussian distribution number (K).

[0099] Specifically, the information entropy of the current waveform signal: such as sample entropy, permutation entropy or multi-scale entropy, determines the value of the Gaussian distribution number (K). By using complexity indicators such as sample entropy, permutation entropy or multi-scale entropy, among which sample entropy can more accurately reflect the different complexity characteristics and distortion points in the signal, the sample entropy indicator can capture these characteristics, thereby more reasonably determining the Gaussian distribution number (K); improving the stability and robustness of the model, and reducing the risk of overfitting or underfitting.

[0100] The calculation method of the sample entropy of the current waveform signal is as follows:

[0101] Define the sample sequence: The original current waveform signal is decomposed into multiple PF components, and the sample entropy of one of the PF component data is selected as the quantitative value, then the sample sequence can be expressed as: X(t) = {PF(t)};

[0102] Where t=1,2,…,N;

[0103] Construct an embedding vector: For any time series {PF(t)}, construct an embedding vector of dimension (m): X m (t)={x(t),x(t+1),...,x(t+m-1)};

[0104] The embedding dimension (m) is usually 2 or 3.

[0105] Calculate similarity vector pairs Calculate all embedding vector pairs X m (t) and X m (j) The distance between: And statistics satisfy d[X m (t), X m (j)]≤r, and normalized;

[0106] Among them, the similarity threshold (r) is usually 0.2-0.3 times the standard deviation of the time series;.

[0107] Calculate sample entropy: Among them, m represents the embedding dimension; r represents the similarity threshold; N represents the length of the time series;

[0108] represents the logarithm of similar vectors under the condition of embedding dimension (m+1) and similarity threshold (r);

[0109] It represents the sum of similarity vector logarithms of all embedded vectors under the condition that the embedding dimension is (m+1) and the similarity threshold is (r).

[0110] As a preferred implementation scheme, based on the preliminary selection of quantitative values, the sample entropy of the remaining PF component data is further verified through a cross-validation method, and then the optimal value of the Gaussian distribution number K is selected to improve the model performance.

[0111] As a preferred implementation, the covariance type is 'full', 'tied', 'diag' or 'spherical'.

[0112] As a preferred implementation, a fault type threshold is set, and when the posterior probability of a cluster label exceeds the fault type threshold, it is determined to be the fault type.

[0113] Specifically, the fault type threshold of the local insulation damage is 0.8; the fault type threshold of the insulation bubble is 0.86; and the fault type threshold of the flashover fault is 0.9.

[0114] Specifically, the fault type threshold of the fault arc is 0.85, and the collected high-voltage cable temperature reaches above 1200°C; since the occurrence of fault arcs is often accompanied by high temperature and ionization discharge, and the occurrence of fault arcs is relatively random; therefore, a dual condition judgment is adopted. According to public data, the center temperature of the fault arc can be as high as 3000°C-4000°C. It should be noted that this temperature range refers to the center temperature of the arc itself, not the temperature collected by the sensor. The temperature collected by the sensor may be reduced due to the influence of the cable material, insulation layer thickness, heat dissipation conditions and surrounding environmental factors. But it may also be enough to ignite surrounding flammable materials, causing a fire.

[0115] The present invention adopts an adaptive LMD method and a flexible Gaussian mixture model, which can adapt to current waveform signals of different types and complexities and has strong adaptability and robustness. By accurately identifying and promptly handling high-voltage cable faults, the reliability and stability of the power system can be improved, and the fault outage time and economic losses can be reduced.

[0116] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0117] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0118] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0119] If the integrated unit is implemented in the form of 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, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.

[0120] In the description of the specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0121] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A method for intelligent fault diagnosis of high-voltage cables based on machine learning, It is characterized in that The following steps are involved: S1) obtaining a current waveform signal of a high-voltage cable under a known fault state, extracting characteristic information of the current waveform signal by a time-frequency analysis method of a local mean, and decomposing the current waveform signal into a plurality of fundamental frequency components according to the characteristic information, wherein the fundamental frequency component is a PF component; S2) constructing a Gaussian mixture model based on a clustering prediction method, taking the PF component processed by the local mean method as a clustering factor, and constructing a feature matrix according to the clustering factor. The feature matrix expression is: Where X represents the feature matrix, k is the total number of PF components, m is the total number of time nodes, and PF k (t m ) is the kth PF component at time node t m The time series data of the demodulated signal components; The parameters of the Gaussian mixture model include the number of Gaussian distributions and the covariance type; The feature matrix is ​​used as model observation data, and the Gaussian mixture model maximizes the log-likelihood function of the model observation data according to the expectation maximization algorithm, and the corresponding data classification result is used as the cluster label of the model observation data; supervised learning is performed through back propagation according to the known fault states corresponding to the cluster labels and the model observation data obtained by calculation, and the parameters of the Gaussian mixture model are updated; Obtaining a current waveform signal of a high-voltage cable during an unknown fault time period, taking the current waveform signal during the unknown fault time period as a test set, calculating a cluster label corresponding to the test set through the trained Gaussian mixture model, and calculating the posterior probability that each data point in the test set belongs to each Gaussian component; According to the cluster labels and posterior probabilities of the test set, the fault types are identified and determined, and the fault types include local insulation damage, insulation bubbles, flashover faults and fault arcs.

2. According to claim 1, a high-voltage cable intelligent fault diagnosis method based on machine learning, Features: The posterior probability sets a fault type threshold. When the posterior probability of a cluster label exceeds the fault type threshold, it is determined to be the fault type.

3. According to claim 1, a high-voltage cable intelligent fault diagnosis method based on machine learning, Features: In the step S2), the Gaussian distribution quantity K is defined, and the sample entropy of one of the PF component data after the current waveform signal is decomposed is selected as the quantitative value.

4. According to claim 3, a high-voltage cable intelligent fault diagnosis method based on machine learning, Features: On the basis of the preliminary selection of quantitative values, the sample entropy of the remaining PF component data is further verified through the cross-validation method, and then the optimal value of the Gaussian distribution number K is selected.

5. According to claim 3, a high-voltage cable intelligent fault diagnosis method based on machine learning, Features: The calculation method of the sample entropy of the current waveform signal is as follows: Define the sample sequence: The original current waveform signal is decomposed into multiple PF components. One of the PF component data is selected as the sample sequence. The sample sequence is expressed as: X(t) = {PF(t)}; Where t=1,2,…,N; Construct an embedding vector: For any time series {PF(t)}, construct an embedding vector of dimension (m): X m (t)={x(t),x(t+1),…,x(t+m-1)}; Calculate similarity vector pairs Calculate all embedding vector pairs X m (t) and X m (j) The distance between: And statistics satisfy d[X m (t), X m (j)]≤r, and normalize them; Calculate sample entropy: Among them, m represents the embedding dimension; r represents the similarity threshold; N represents the length of the time series; represents the logarithm of similar vectors under the condition of embedding dimension (m+1) and similarity threshold (r); It represents the sum of similarity vector logarithms of all embedded vectors under the condition that the embedding dimension is (m+1) and the similarity threshold is (r).

6. A method and system for intelligent fault diagnosis of high-voltage cables based on machine learning according to claim 1, Features: In step S2, the covariance type is selected from 'full', 'tied', 'diag' or 'spherical'.

7. A high-voltage cable intelligent fault diagnosis method and system based on machine learning according to claim 2, Features: The fault type threshold of the local insulation damage is 0.8; the fault type threshold of the insulation bubble is 0.86; and the fault type threshold of the flashover fault is 0.

9.

8. A method and system for intelligent fault diagnosis of high-voltage cables based on machine learning according to claim 2, Features: The fault type threshold of the fault arc is 0.85, and the collected high-voltage cable temperature reaches above 1200°C.

9. The system of a high-voltage cable intelligent fault diagnosis method based on machine learning according to claim 1, Features: It includes a data acquisition module, an industrial computer module and a host computer module; the data acquisition module includes a current transformer, a high-frequency current sensor and a temperature sensor; the current signal, high-frequency current signal and cable temperature signal in the cable are collected respectively; The industrial computer module includes an STC89C51 single-chip microcomputer, an AD converter and a communication circuit; the industrial computer module receives the signal data sent by the data acquisition module through the communication circuit and performs preliminary processing; the STC89C51 single-chip microcomputer solidifies the data acquisition program and the serial port communication program; the AD converter is used to convert the analog signal into a digital signal; the communication circuit is divided into two paths, one is an RS232 communication circuit connected to the data acquisition module, and the other is a CAN bus circuit connected to the host computer module.

10. A system for intelligent fault diagnosis of high-voltage cables based on machine learning according to claim 9, Features: The host computer module includes a Xilinx chip and an expandable memory; the Xilinx chip is used for data processing and data modeling of current signals; the expandable memory is used for storing related data and machine learning applications.

Images