Multi-parameter fusion identification method, device, and electronic equipment for partial discharge of power equipment

Abstract

The present invention discloses a multi-parameter fusion identification method, device, and electronic device for partial discharge in power equipment. The method comprises: measuring partial discharge data from power equipment using multiple types of sensors, generating relative energy amplitudes of multiple sensor parameters based on the partial discharge data; extracting partial discharge features based on the relative energy amplitudes using a chi-square test and a maximum correlation minimum redundancy algorithm, and establishing a corresponding classification model using a support vector machine algorithm; and performing decision fusion based on the soft output results of the classification model using the Dempster-Shafer evidence theory. The present invention significantly improves the accuracy of discharge type identification and provides a high-confidence solution for partial discharge detection.

Description

Technical Field

[0001] The present invention belongs to the field of insulation of electric power equipment and relates to a multi-parameter fusion identification method, device and electronic equipment for partial discharge of electric power equipment. Background Art

[0002] Gas-insulated switchgear (GIS) typically uses gases such as SF6 as the insulating medium. Its advantages include high electrical strength, small footprint, and minimal maintenance, leading to its widespread application. However, due to unavoidable factors during the manufacturing, transportation, assembly, and long-term service of GIS equipment, insulation defects such as metal spikes, surface anomalies of insulation components, and floating potentials can sometimes develop within the equipment, posing varying degrees of safety hazards to the equipment. Partial discharge (PD), a physical phenomenon caused by insulation defects under the influence of strong external electric fields, is a primary manifestation of insulation degradation and a significant factor contributing to this degradation. The occurrence and evolution of PD is accompanied by various microphysical and energy release processes, such as charge transfer, optical radiation, electromagnetic radiation, and acoustic radiation, leading to the development of a variety of PD measurement methods. Currently, scholars at home and abroad have conducted extensive research in the field of PD detection. Using various measurement methods, significant progress has been made in signal-to-noise separation, feature extraction, and fault identification of discharge signals, and these methods have been applied to condition-based maintenance in production sites.

[0003] However, due to different measurement principles, various detection methods have significant differences in detection sensitivity, effectiveness, and scope of application for different operating conditions, partial discharge types, and discharge severity. Clear application specifications have not yet been established, and false fault alarms often occur. Various detection data suffer from obstacles such as measurement time asynchrony and data structure differences, making it difficult to effectively reuse them to obtain comprehensive diagnostic information and address the problem of data desensitization of certain fault conditions by a single measurement method. After each detection method independently analyzes and draws different conclusions, there is a lack of comparative correlation and comprehensive analysis methods between the conclusions, making it difficult to obtain highly confident guiding diagnostic conclusions. Furthermore, in the absence of a unified understanding of various partial discharge processes and microscopic mechanisms, the insulation condition and partial discharge severity of the equipment under test are usually assessed based on simple discharge amplitude or discharge frequency. Due to the changes in various physical processes and energy release mechanisms during the discharge development process, this method is prone to judgment differences between different measurement methods, which brings difficulties to condition-based maintenance work in actual projects. To overcome the shortcomings of single detection methods in measuring partial discharge and improve the accuracy of partial discharge diagnosis of gas-insulated electrical insulation, the present invention proposes a method for fusion identification of discharge defects using multiple physical parameters.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the invention and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention proposes a multi-parameter fusion identification method for partial discharge of power equipment. The method comprises the following steps:

[0006] Measuring partial discharge data of the power equipment using multiple types of sensors, and generating relative energy amplitudes of multiple sensing parameters based on the partial discharge data;

[0007] Based on the relative energy amplitude, a chi-square test and a maximum correlation minimum redundancy algorithm are used to extract partial discharge features, and a support vector machine algorithm is used to establish a corresponding classification model;

[0008] Based on the soft output results of the classification model, the Dempster-Shafer evidence theory is used to perform decision fusion.

[0009] In the multi-parameter fusion identification method for partial discharge of power equipment, when the sensor is a silicon photomultiplier, the relative energy amplitude P L for:

[0010]

[0011] Among them A L is the output current peak value of the silicon photomultiplier, t0 is the discharge start time, and t is the current time.

[0012] In the multi-parameter fusion identification method for partial discharge of power equipment, when the sensor is a UHF sensor, the relative energy amplitude P UHF for:

[0013]

[0014] Among them A U is the peak value of the UHF sensor output voltage, t0 is the discharge start time, and t is the current time.

[0015] In the multi-parameter fusion identification method for partial discharge of power equipment, when the sensor is an ultrasonic sensor, the relative energy amplitude P AE for:

[0016]

[0017] Where V i (t) is the voltage amplitude output by the ultrasonic sensor at time t, t0 is the discharge start time, and t is the current time.

[0018] In the multi-parameter fusion identification method for partial discharge of power equipment, the steps of extracting partial discharge features based on the relative energy amplitude using the chi-square test and maximum correlation minimum redundancy algorithm and establishing a corresponding classification model using the support vector machine algorithm include:

[0019] Extracting partial discharge features based on the relative energy amplitude using a chi-square test and a maximum correlation minimum redundancy algorithm;

[0020] The error-correcting output code is used to transform the multi-classification problem of partial discharge types into a binary classification problem that can be handled by support vector machines:

[0021] The ternary error correction output code strategy with a coding matrix of {1,0,1} is introduced to enhance the robustness of the model, and the probability estimation r of the binary classifier is j and the posterior probability p of the kth class k The relationship is:

[0022]

[0023] p k represents the posterior probability of the kth class, N c Indicates the number of all types;

[0024] I represents the indicator function, and its calculation rule is:

[0025]

[0026]

[0027] Optimize r and The Kullback-Leibler divergence between Is the probability estimate to be optimized:

[0028]

[0029] N is the number of binary classifiers, w j is the number of training samples for the j-th binary classifier;

[0030] The relative energy amplitudes are used as inputs, and classification is performed using a support vector machine. The hyperplane slope w and the hyperplane intercept b are solved by minimizing the following objective function L:

[0031]

[0032] α i is the Lagrange multiplier, which is a free variable; m is the number of data samples; y i is the binary classification label of the i-th data; x iis the classifier input vector for the i-th data, which is composed of the relative energy amplitude of each sensor;

[0033] Introducing slack variables and penalty factors, the objective function L is updated as follows:

[0034]

[0035] C represents the penalty parameter, which is a preset value; α i and μ i Both are Lagrange multipliers and are free variables; ξ i Denotes slack variables:

[0036]

[0037] Solve the updated objective function L to obtain the hyperplane slope w and hyperplane intercept b;

[0038] According to the trained support vector machine model, the sigmoid function is used for soft output transformation:

[0039]

[0040] P(y=1|x) represents the probability that the classifier classifies sample x as "1", and f(x) is the classification result of sample x;

[0041] For the values ​​of parameters A and B, we can solve the following regularized maximum likelihood function:

[0042]

[0043] In the above formula, m is the number of data samples;

[0044]

[0045] Among them, y i is the classification label of the i-th sample, N + and N - are the number of positive samples and the number of negative samples, respectively.

[0046] In the multi-parameter fusion identification method for partial discharge of power equipment, the steps of using the Dempster-Shafer evidence theory to perform decision fusion based on the soft output results of the classification model include:

[0047] Building an identification framework:

[0048] Θ={A1,A2,A3,…,A n ,θ}, the elements A1,A2,A3,…,A in the recognition frame nIndicates various discharge types, θ indicates the uncertain type;

[0049] Establish the basic probability distribution function:

[0050]

[0051]

[0052]

[0053] 2 Θ is the power set of Θ, which represents the set of all subsets of Θ; represents the empty set;

[0054] Each sensor classification result A i Corresponding to an m function, it is defined as:

[0055]

[0056] In the above formula, S is a subset of Θ, P(A i ) indicates classification as A i The posterior probability of

[0057] Introducing the discount factor γ ij As a weighted value of support:

[0058] m i (A j )=γ ij ·P ij ,

[0059] Among them, P ij is the observation source m i The SVM model for samples with classification label A j The posterior probability output value under ;

[0060] The observation support loss caused by the discount factor is converted into the uncertainty θ of the fusion decision.

[0061] In the multi-parameter fusion identification method for partial discharge of electric power equipment, the discharge types include corona discharge, surface discharge and suspended potential discharge.

[0062] In addition, the present invention also discloses a multi-parameter fusion identification device for partial discharge of power equipment, comprising:

[0063] a generating unit, which uses multiple types of sensors to measure partial discharge data of the power equipment and generates relative energy amplitudes of multiple sensing parameters based on the partial discharge data;

[0064] a modeling unit, which extracts partial discharge features based on the relative energy amplitude using a chi-square test and a maximum correlation minimum redundancy algorithm and establishes a corresponding classification model using a support vector machine algorithm;

[0065] The fusion unit performs decision fusion based on the soft output results of the classification model using the Dempster-Shafer evidence theory.

[0066] In addition, the present invention also discloses an electronic device, comprising:

[0067] A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein:

[0068] When the processor executes the program, the method described above is implemented.

[0069] In addition, the present invention also discloses a computer storage medium, wherein the computer storage medium stores computer-executable instructions, and the computer-executable instructions are used to execute the method described above.

[0070] Beneficial effects

[0071] The method proposed in this paper utilizes multi-parameter data to identify discharge types at different scales, achieving normalization of different physical parameters. It also proposes a feature extraction method and classification model for pulse current data, optical signal data, ultra-high frequency data, and ultrasonic data. Furthermore, it utilizes the soft output of a support vector machine (SVM) to discriminate discharge types, combined with the Dempster-Shafer evidence theory, to achieve decision fusion for partial discharge diagnosis. This method significantly improves the accuracy of discharge type identification and provides a high-confidence solution for partial discharge detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Various other advantages and benefits of the present invention will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are intended only to illustrate preferred embodiments and are not to be construed as limiting the present invention. It should be understood that the drawings described below are merely examples of the present invention, and that those skilled in the art will be able to derive other drawings from these drawings without inventive effort. Throughout the drawings, identical reference numerals are used to denote identical components.

[0073] In the attached figure:

[0074] Figure 1 This is a flow chart of a multi-parameter fusion identification method for partial discharge of power equipment according to an embodiment of the present invention.

[0075] The present invention will be further explained below with reference to the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0076] The following will refer to the attached Figure 1 Specific embodiments of the present invention will now be described in greater detail. Although specific embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention may be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to facilitate a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0077] It should be noted that certain words are used in the specification and claims to refer to specific components. Those skilled in the art should understand that technicians may use different nouns to refer to the same component. This specification and claims do not use the difference in nouns as a way to distinguish components, but use the difference in the functions of the components as the criterion for distinction. As mentioned throughout the specification and claims, "including" or "comprising" is an open term, so it should be interpreted as "including but not limited to". The subsequent description of the specification is a preferred embodiment of the present invention, but the description is based on the general principles of the specification and is not intended to limit the scope of the invention. The scope of protection of the present invention shall be as defined in the attached claims.

[0078] To facilitate understanding of the embodiments of the present invention, further explanation will be given below using specific embodiments as examples in conjunction with the accompanying drawings, and the accompanying drawings do not constitute a limitation on the embodiments of the present invention.

[0079] For better understanding, Figure 1 It is a multi-parameter fusion identification method for partial discharge of power equipment, such as Figure 1 As shown, the multi-parameter fusion identification method for partial discharge of power equipment includes the following steps:

[0080] Step 1: using multiple types of sensors to measure partial discharge data of power equipment, and generating relative energy amplitudes of multiple sensing parameters based on the partial discharge data;

[0081] Step 2: extracting partial discharge features based on the relative energy amplitude and establishing a corresponding classification model;

[0082] Step 3: Perform decision fusion based on the soft output results of the classification model.

[0083] In a preferred embodiment of the method, in step 1, the sensor is a silicon photomultiplier, and the relationship between its output current and incident light flux is:

[0084] I=q×PDE×Ф,

[0085] Where q is the output charge of the silicon photomultiplier, PDE is the light quantum efficiency, Φ is the incident light flux, and the light radiation power of the partial discharge is:

[0086]

[0087] Where k is the absorption coefficient of the medium to photons, ρ is the gas density, μ is the attenuation coefficient at the light source position, L is the distance between the sensor and the light source, δV is the integral value of the standard Gaussian pulse, and A L is the voltage amplitude output by the optical sensor. When the sensor position is fixed, except for A L The influence of other parameters on the measured amplitude is reflected by the fixed coefficient k L Therefore, when the sensor is a silicon photomultiplier, the relative energy amplitude is:

[0088]

[0089] In a preferred embodiment of the method, in step 1, when the sensor is a UHF sensor, the electromagnetic radiation energy flux density is:

[0090]

[0091] j represents the imaginary unit, I represents the instantaneous current generated by the discharge, l represents the distance between the positive and negative poles of the discharge gap, r represents the distance between the UHF sensor and the discharge position, ω represents the electromagnetic wave frequency, ε represents the dielectric constant of the space, and θ represents the angle formed by the line connecting the discharge position and the sensor position and the direction of the discharge current. is the unit vector in the spherical coordinate system. The relationship between the induced voltage U and the electric field strength E on the UHF sensor is:

[0092] U=k·E,

[0093] Where k is a coefficient, which is determined by the specific structure of the sensor. When the sensor is determined and the layout position is fixed, the influence of parameters such as l, r, ε, ω, θ on the measurement results is reflected as the proportional coefficient k EM , thus the relationship between the energy coupled to the UHF sensor and the output voltage of the UHF sensor is:

[0094]

[0095] Therefore, when the sensor is a UHF sensor, the relative energy amplitude is:

[0096]

[0097] In a preferred embodiment of the method, in step 1, when the sensor is an ultrasonic sensor, the relative energy amplitude is:

[0098]

[0099] In a preferred embodiment of the method, in step 2,

[0100] Extracting partial discharge features based on the relative energy amplitude using a chi-square test and a maximum correlation minimum redundancy algorithm;

[0101] The error-correcting output code is used to transform the multi-classification problem of partial discharge types into a binary classification problem that can be handled by support vector machines:

[0102] The ternary error correction output code strategy with a coding matrix of {1, 0, 1} is introduced to enhance the robustness of the model. The probability estimate r of the binary classifier is related to the posterior probability p of the kth class. k The relationship is:

[0103]

[0104] p k represents the posterior probability of the kth class, N c Indicates the number of all types;

[0105] I represents the indicator function, and its calculation rule is:

[0106]

[0107]

[0108] Optimize r and The Kullback-Leibler divergence between Is the probability estimate to be optimized:

[0109]

[0110] The relative energy amplitudes are used as inputs, and classification is performed using a support vector machine. The hyperplane parameters, the hyperplane slope w and the hyperplane intercept b, are solved by minimizing the following objective function L:

[0111]

[0112] Introducing slack variables and penalty factors, the objective function L is updated as follows:

[0113]

[0114] Solve the above objective function to obtain w and b;

[0115] According to the trained support vector machine model, the sigmoid function is used for soft output transformation:

[0116]

[0117] For the values ​​of parameters A and B, we can solve the following regularized maximum likelihood function:

[0118]

[0119] In a preferred embodiment of the method, in step 3,

[0120] Building an identification framework:

[0121] Θ={A1,A2,A3,…,A n ,θ}, the elements A1,A2,A3,…,A in the recognition frame n Indicates various discharge types, θ indicates the uncertain type;

[0122] Establish the basic probability distribution function:

[0123]

[0124]

[0125]

[0126] Each sensor classification result A i Corresponding to an m function, it is defined as:

[0127]

[0128] In the above formula, S is a subset of Θ, P(A i ) indicates classification as A i The posterior probability of

[0129] Introducing the discount factor γ ij As a weighted value of support:

[0130] m i (A j )=γ ij ·P ij

[0131] Among them, P ij is the observation source m i The SVM model for samples with classification label A j The posterior probability output value under ;

[0132] The observation support loss caused by the discount factor is converted into the uncertainty θ of the fusion decision.

[0133] In a preferred embodiment of the method, the discharge types include corona discharge, creeping discharge and floating potential discharge.

[0134] In one embodiment, a high frequency Rogowski coil (HFCT), a silicon photomultiplier (SiPM), an ultra-high frequency sensor (UHF), and an ultrasonic sensor (AE) are used to synchronously collect data on partial discharge signals.

[0135] The flow chart of the identification method is as follows Figure 1 Shown, including:

[0136] Step 1: Calculate the relative energy amplitude:

[0137] 1.1 A silicon photomultiplier (SiPM) device contains a large number of avalanche diodes (Avalanche Diodes) made from doped silicon. When a sufficient bias voltage is applied, the Avalanche Diodes trigger a Geiger avalanche upon receiving an incident photon, generating a photocurrent. When the SiPM is operating normally, each Avalanche Diode has two states: a "1" state (when a photon is detected) and a "0" state (when no photon is detected). The SiPM's response is the sum of the responses of all Avalanche Diodes:

[0138]

[0139] Where Q is the output charge of SiPM / C, q is the charge generated by the avalanche diode APD unit / C, N ph is the number of photons incident on the device surface, M is the number of avalanche diodes of the SiPM, and PDE is the photon detection efficiency of the SiPM.

[0140] For partial discharge optical signals, before the discharge forms a breakdown arc, the radiation intensity is not enough to saturate the SiPM response:

[0141] N ph ·PDE<<M

[0142] therefore:

[0143]

[0144] Q≈q·N ph PDE

[0145] Taking the derivative of both sides with respect to t, we can get:

[0146]

[0147] The luminous flux Φ decays as the distance from the light source increases, and decays to the initial μe at the distance L. -kρL times, so the optical radiation power of partial discharge is:

[0148]

[0149] Where k is the absorption coefficient of the medium to photons, ρ is the gas density, μ is the attenuation coefficient at the light source position, L is the distance between the sensor and the light source, δV is the integral value of the standard Gaussian pulse, and A L is the voltage amplitude output by the optical sensor. Under the same measurement conditions, k L is a constant, so the relative light radiation power is defined as the relative energy amplitude:

[0150]

[0151] 1.2 For ultra-high frequency sensors (UHF), the energy received by the sensor is proportional to the integral of the electromagnetic energy on its receiving surface. According to electromagnetic theory, the electromagnetic radiation energy flux density at the sensor is:

[0152]

[0153] j represents the imaginary unit, I represents the instantaneous current generated by the discharge, l represents the distance between the positive and negative poles of the discharge gap, r represents the distance between the UHF sensor and the discharge position, ω represents the electromagnetic wave frequency, ε represents the dielectric constant of the space, and θ represents the angle formed by the line connecting the discharge position and the sensor position and the direction of the discharge current. is a unit vector in the spherical coordinate system. The relationship between the induced voltage U on the sensor and the electric field strength E at that location is:

[0154] U=k·E

[0155] Where k is a parameter determined by the antenna characteristics and position. Therefore, the energy coupled to the sensor is proportional to the square of the voltage. At this time, its relative energy is:

[0156]

[0157] Among them A U is the output signal amplitude of the UHF sensor. Under the same measurement conditions, k EM is a constant, so the relative electromagnetic radiation power is defined as the relative energy amplitude:

[0158]

[0159] 1.3 For ultrasonic sensors (AE), when the discharge source and sensor are relatively fixed, the partial discharge ultrasonic signal has a fixed attenuation factor. The relative sound power is defined as the relative energy amplitude:

[0160]

[0161] In step 2, the relative energy amplitude of each sensor is used, and the commonly recognized feature statistics are selected: average discharge amplitude A, average discharge frequency N, skewness Sk, steepness Ku, phase asymmetry Φ, discharge asymmetry Q, cross-correlation factor Cc, and ultrasonic signal dispersion entropy DE. A total of 43 feature quantities are obtained and feature extraction is performed.

[0162] The chi-square test measures the deviation between the actual observed values ​​of the sample and the theoretically predicted values. A larger chi-square value indicates a stronger orthogonality between the variables, while a lower chi-square value indicates a stronger dependence between the variables. Using a chi-square value of 0.6 as the scoring standard for feature extraction, the results are shown in Table 1.

[0163] Table 1 Optimized set of chi-square test for partial discharge characteristics

[0164]

[0165] In Table 1 above, the meanings of the subscripts of the characteristic quantities are: H corresponds to HFCT (i.e., high-frequency current sensor), L corresponds to optical sensor, and U corresponds to UHF (i.e., ultra-high frequency sensor).

[0166] The maximum relevance and minimum redundancy (mRMR) algorithm defines the "mutual information" between random variables. It quantitatively analyzes the redundancy and correlation between different features and finds an optimal set of features that maximizes the difference between them, effectively representing the response variable. In this implementation example, the condition with the largest difference between adjacent scores was selected as the feature screening boundary. This resulted in the optimal maximum relevance and minimum redundancy set after feature selection. The results are shown in Table 2:

[0167] Table 2 mRMR optimization set of partial discharge characteristics

[0168]

[0169] In Table 2 above, compared with Table 1, the additional subscripts of each characteristic quantity have the following meanings: - represents negative pulse discharge, and + represents positive pulse discharge.

[0170] The classification results of the support vector machine are shown in Appendix 3, with the original data set, chi-square test optimization set, and mRMR optimization set as input respectively:

[0171] Appendix 3: Classification accuracy of partial discharge diagnosis model samples

[0172]

[0173] The chi-square test optimization set is used as the feature parameter set for partial discharge multi-physics detection. Classification is performed for each single physical quantity. The obtained sample classification accuracy TPR is shown in Appendix 4, and the prediction accuracy PPV is shown in Appendix 5:

[0174] Appendix 4: TPR of single physical quantity partial discharge diagnosis model

[0175]

[0176] Appendix 5: PPV of single physical quantity partial discharge diagnosis model

[0177]

[0178] Step 3: In the partial discharge diagnosis problem, the propositions included in the identification framework Θ are: A1 corona discharge, A2 surface discharge, A3 suspension discharge, and θ uncertainty. There are 4 m functions, and the observation sources are m1 pulse current method, m2 optical measurement method, m3 ultra-high frequency method,

[0179] m4 ultrasound method. After obtaining the BPA of each piece of evidence within the framework, the Dempster synthesis formula was used to obtain the final decision result. The diagnostic accuracy rates obtained by the DS decision fusion method are shown in Appendix 6.

[0180] Appendix 6 Accuracy of fusion decision diagnosis results

[0181]

[0182] The diagnostic effect after multi-physics fusion is significantly improved compared with the single physical quantity model.

[0183] In addition, in one embodiment, the present invention also discloses a multi-parameter fusion identification device for partial discharge of power equipment, comprising:

[0184] a generating unit, which uses multiple types of sensors to measure partial discharge data of the power equipment and generates relative energy amplitudes of multiple sensing parameters based on the partial discharge data;

[0185] a modeling unit, which extracts partial discharge features based on the relative energy amplitude using a chi-square test and a maximum correlation minimum redundancy algorithm and establishes a corresponding classification model using a support vector machine algorithm;

[0186] The fusion unit performs decision fusion based on the soft output results of the classification model using the Dempster-Shafer evidence theory.

[0187] In addition, in one embodiment, the present invention further discloses an electronic device, comprising:

[0188] A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein:

[0189] When the processor executes the program, the method described above is implemented.

[0190] Furthermore, in one embodiment, the present invention further discloses a computer storage medium storing computer-executable instructions for executing the method described above.

[0191] Although the embodiments of the present invention have been described above with reference to the accompanying drawings, the present invention is not limited to the above-mentioned specific embodiments and application fields. The above-mentioned specific embodiments are merely illustrative and instructive, and are not restrictive. A person skilled in the art, guided by this specification and without departing from the scope of protection of the claims of the present invention, may also devise various forms, all of which fall within the scope of protection of the present invention.

Claims

1. A multi-parameter fusion identification method for partial discharge of power equipment, characterized by: The method comprises the following steps: Measuring partial discharge data of the power equipment using multiple types of sensors, and generating relative energy amplitudes of multiple sensing parameters based on the partial discharge data; Based on the relative energy amplitude, a chi-square test and a maximum correlation minimum redundancy algorithm are used to extract partial discharge features, and a support vector machine algorithm is used to establish a corresponding classification model; Based on the soft output results of the classification model, the Dempster-Shafer evidence theory is used to perform decision fusion; The steps of extracting partial discharge features based on the relative energy amplitude using a chi-square test and a maximum correlation minimum redundancy algorithm and establishing a corresponding classification model using a support vector machine algorithm include: Extracting partial discharge features based on the relative energy amplitude using a chi-square test and a maximum correlation minimum redundancy algorithm; The error-correcting output code is used to transform the multi-classification problem of partial discharge types into a binary classification problem that can be handled by support vector machines: The encoding matrix is ​​introduced as The ternary error correction output code strategy is used to enhance the robustness of the model and the probability estimation of the binary classifier and the posterior probability of the kth class The relationship is: ， represents the posterior probability of the kth class, Indicates the number of all types; Represents the indicator function, and its calculation rule is: ， ， optimization and The Kullback-Leibler divergence between Is the probability estimate to be optimized: ， N is the number of binary classifiers, is the number of training samples for the j-th binary classifier; The relative energy amplitudes are input and the support vector machine is used for classification by minimizing the following objective function To solve the hyperplane slope w and hyperplane intercept b: ， is the Lagrange multiplier, which is a free variable; is the number of data samples; is the binary classification label of the i-th data; is the classifier input vector for the i-th data, which is composed of the relative energy amplitude of each sensor; Introducing slack variables and penalty factors, the objective function Updated to: ， Represents the penalty parameter, which is a preset value; and All are Lagrange multipliers and are free variables; Denotes slack variables: ， For the updated objective function Solve to obtain the hyperplane slope w and hyperplane intercept b; According to the trained support vector machine model, the sigmoid function is used for soft output transformation: ， Represents the probability that the classifier classifies sample x as "1", is the classification result of sample x; For the values ​​of parameters A and B, we can solve the following regularized maximum likelihood function: ， In the above formula, , is the number of data samples; ， in, is the classification label of the i-th sample, and are the number of positive samples and the number of negative samples, respectively.

2. The multi-parameter fusion identification method for partial discharge of power equipment according to claim 1 is characterized in that: When the sensor is a silicon photomultiplier, the relative energy amplitude for: ， in A L is the peak output current of the silicon photomultiplier, t 0 is the discharge start time, and t is the current time.

3. The multi-parameter fusion identification method for partial discharge of power equipment according to claim 1, characterized in that: When the sensor is a UHF sensor, the relative energy amplitude for: ， in A U is the peak value of the UHF sensor output voltage, t 0 is the discharge start time, and t is the current time.

4. The multi-parameter fusion identification method for partial discharge of power equipment according to claim 1, characterized in that: When the sensor is an ultrasonic sensor, the relative energy amplitude for: ， in V i (t) is the voltage amplitude output by the ultrasonic sensor at time t, t 0 is the discharge start time, and t is the current time.

5. The multi-parameter fusion identification method for partial discharge of power equipment according to claim 1, characterized in that: Based on the soft output results of the classification model, the steps of using Dempster-Shafer evidence theory to perform decision fusion include: Building an identification framework: , the elements in the identification frame Indicates various types of discharge, Indicates uncertain type; Establish the basic probability distribution function: ， ， ， for The power set of The set of all subsets of ; represents the empty set; Each sensor classification result Corresponding to an m function, it is defined as: ， In the above formula, S is A subset of Indicates classification as The posterior probability of Introducing a discount factor As a weighted value of support: ， in, Observation source The SVM model is for samples in the classification label The posterior probability output value under ; The observation support loss caused by the discount factor is converted into the uncertainty of the fusion decision .

6. The multi-parameter fusion identification method for partial discharge of power equipment according to claim 5, characterized in that: Discharge types include corona discharge, creeping discharge and suspended potential discharge.

7. A multi-parameter fusion identification device for partial discharge of power equipment, characterized in that: include: a generating unit, which uses multiple types of sensors to measure partial discharge data of the power equipment and generates relative energy amplitudes of multiple sensing parameters based on the partial discharge data; a modeling unit, which extracts partial discharge features based on the relative energy amplitude using a chi-square test and a maximum correlation minimum redundancy algorithm and establishes a corresponding classification model using a support vector machine algorithm; a fusion unit, which performs decision fusion based on the soft output results of the classification model using the Dempster-Shafer evidence theory; The modeling units are: Extracting partial discharge features based on the relative energy amplitude using a chi-square test and a maximum correlation minimum redundancy algorithm; The error-correcting output code is used to transform the multi-classification problem of partial discharge types into a binary classification problem that can be handled by support vector machines: The encoding matrix is ​​introduced as The ternary error correction output code strategy is used to enhance the robustness of the model and the probability estimation of the binary classifier and the posterior probability of the kth class The relationship is: ， represents the posterior probability of the kth class, Indicates the number of all types; Represents the indicator function, and its calculation rule is: ， ， optimization and The Kullback-Leibler divergence between Is the probability estimate to be optimized: ， N is the number of binary classifiers, is the number of training samples for the j-th binary classifier; The relative energy amplitudes are input and the support vector machine is used for classification by minimizing the following objective function To solve the hyperplane slope w and hyperplane intercept b: ， is the Lagrange multiplier, which is a free variable; is the number of data samples; is the binary classification label of the i-th data; is the classifier input vector for the i-th data, which is composed of the relative energy amplitude of each sensor; Introducing slack variables and penalty factors, the objective function Updated to: ， Represents the penalty parameter, which is a preset value; and All are Lagrange multipliers and are free variables; Denotes slack variables: ， For the updated objective function Solve to obtain the hyperplane slope w and hyperplane intercept b; According to the trained support vector machine model, the sigmoid function is used for soft output transformation: ， Represents the probability that the classifier classifies sample x as "1", is the classification result of sample x; For the values ​​of parameters A and B, we can solve the following regularized maximum likelihood function: ， In the above formula, , is the number of data samples; ， in, is the classification label of the i-th sample, and are the number of positive samples and the number of negative samples, respectively.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

9. A computer storage medium, characterized in that The computer storage medium stores computer-executable instructions, and the computer-executable instructions are used to execute the method according to any one of claims 1 to 6.

Images