Power transformer partial discharge source positioning method based on learning vector quantization network model

Abstract

The invention relates to a power transformer partial discharge source positioning method based on a learning vector quantization network model, and belongs to the technical field of power equipment detection, and the method comprises the following steps: S1, measuring voltage frequency responses of a transformer winding at different tapping positions; s2, obtaining an approximate fitting voltage transfer function through a vector fitting VF technology; s3, establishing a transformer segment winding transfer function SWTF library; s4, calculating a partial discharge PD reference signal; s5, constructing a learning vector quantization LVQ network model, and taking the marked partial discharge reference signals as a training set to train the LVQ network model; s6, injecting a partial discharge test pulse into the randomly selected winding tap connection, and taking the obtained response as a partial discharge test signal; and S7, inputting a partial discharge test signal into the trained LVQ network model, and testing the classification performance of the model.

Description

Technical Field

[0001] The invention belongs to the technical field of power equipment detection, and relates to a method for locating a local discharge source of a power transformer based on a learning vector quantization network model. Background Art

[0002] Power transformers are critical components of the power system, ensuring the performance and stability of transmission and distribution networks. Windings are the most common location for transformer failures, and dielectric failure modes such as partial discharge, tracking, or flashover are also common in transformers. Therefore, enhanced condition monitoring tools and early detection of power transformer faults are key priorities for preventing power supply interruptions, minimizing the risk of failures, and improving the economic and technical performance of power system grids.

[0003] The insulation health of power transformers significantly impacts their expected service life. Partial discharge (PD) is the root cause of the gradual breakdown of transformer insulation. If the problem is not addressed, complete breakdown can cause irreparable damage to the insulation material. Therefore, accurately identifying the scope and exact location of PD sources is essential to making informed decisions about whether to perform timely maintenance on power transformers or enhance monitoring during operation, thereby reducing maintenance costs. However, due to the non-uniformity and complexity of transformer windings and the random nature of PD, identifying the exact location of PD sources is a challenging and difficult task.

[0004] Various previous studies have investigated conventional methods for locating partial discharge (PD) sources in transformer windings, including digital filtering, capacitance component extraction, location of frequency response zeros, and time-coded signal processing and identification. However, these methods are susceptible to noise and may not accurately locate PDs, limiting their practical application in field power transformers. Therefore, solutions are needed to mitigate the impact of background noise on PD signal location. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method for locating partial discharge sources in power transformers based on a learning vector quantization network model. The method combines the learning vector quantization network with measured frequency response analysis to identify partial discharge sources. At the same time, the segmented winding transfer function is obtained through vector fitting technology to improve the accuracy of partial discharge localization. The method aims to achieve strong performance in noisy environments without the need for traditional noise reduction techniques.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for locating a local discharge source of a power transformer based on a learning vector quantization network model comprises the following steps:

[0008] S1: Measure the voltage frequency response of the transformer winding at different tap positions;

[0009] S2: Obtain an approximate fitting voltage transfer function by using vector fitting VF technology;

[0010] S3: Establish the transformer segment winding transfer function SWTF library;

[0011] S4: Calculate the partial discharge PD reference signal;

[0012] S5: constructing a learning vector quantization (LVQ) network model, and training the LVQ network model using the labeled partial discharge reference signal as a training set;

[0013] S6: injecting a partial discharge test pulse into a randomly selected winding tap connection and using the resulting response as the partial discharge test signal;

[0014] S7: Input the partial discharge test signal into the trained LVQ network model to test the classification performance of the model.

[0015] Furthermore, step S1 specifically includes: using a sweep frequency response method to continuously change the frequency of the external excitation source, measuring the ratio of the signal amplitude of the response terminal voltage to the excitation terminal voltage of the transformer winding at different frequencies, and obtaining the transfer function of the winding under the conditions of specified excitation terminal and response terminal;

[0016] The input voltage signal V Ti Inject all taps of the transformer winding and measure the output voltage V at the neutral point o , the winding transfer function is obtained as:

[0017] TF i =V o / V Ti

[0018] Transfer Function TF i is derived from the voltage frequency response corresponding to each tap.

[0019] Furthermore, step S2 specifically includes: using vector fitting technology, approximating the frequency domain function in the form of a sum of first-order rational fractions, and approximately fitting multiple transfer functions (TF1, TF2, ···, TF i ), these transfer functions are expressed in the form of rational functions:

[0020]

[0021] Where N(s) and D(s) are polynomials in the domain s=jw, with orders m and n respectively; i and p iare complex conjugate pairs, representing the residues and poles of the fitting expression respectively; d is a real number.

[0022] Furthermore, step S3 specifically includes: calculating and saving all approximate fitting voltage transfer functions from the taps of the transformer winding to the neutral line, thereby creating a transformer segment winding transfer function SWTF library.

[0023] Furthermore, step S4 specifically includes: defining the actual partial discharge pulse by a Gaussian function, and determining a reference signal corresponding to each tap position in the transformer winding by capturing the PD pulse response of each calculated SWTF in the time domain; the mathematical expression of the PD pulse defined by the Gaussian function is as follows:

[0024]

[0025] Where t represents the time variable, A represents the amplitude of the PD pulse, β represents the retention time, and α represents the width of the control pulse; the calibration process uses the Gaussian function.

[0026] Furthermore, the LVQ network model is first trained using labeled partial discharge reference signals to establish hidden layers to develop a robust classification model. Then, arbitrary partial discharge responses measured at the neutral terminal are used as test signals to evaluate the accuracy of the model in detecting partial discharge locations.

[0027] Furthermore, the LVQ network model consists of an input layer, a hidden layer and an output layer; the input data of the LVQ network model consists of a partial discharge reference signal in the time domain, and the output is a category label, each category representing a specific winding tap position.

[0028] Furthermore, the LVQ network training process is as follows:

[0029] Step a: Transform the prototype vector {w j (0)|j=1,2,···,N} assigns N input vectors {x i |i=1,2,···,L} to initialize the prototype vector;

[0030] Step b: Randomly select input vector x from the sample set i ;

[0031] Step c: Calculate the input vector x i With the prototype vector w j Distance: d ij =||x i -w j || 2 ;

[0032] Step d: Find the value that matches the input vector xi The nearest prototype vector w c , c = argmin j∈{1,2,···,N} d ij ;

[0033] Step e: If the class of the input vector is consistent with the class of the winning prototype vector, the prototype vector is adjusted to be closer to the input vector; if the class of the input vector and the winning prototype do not match, the prototype vector is moved away from the input vector. The specific modifications are as follows:

[0034]

[0035] Among them, S wc Representation and prototype vector w c Related classes, S xi Represents the input vector x i Class label; 0<α n <1 is the learning constant;

[0036] Step f: Iterate step bf until the prototype vector is stable and converges.

[0037] Further, in step S7, the trained LVQ network model predicts the PD position based on matching the input vector category with the retained prototype vector obtained from the PD reference signal.

[0038] The present invention combines a learned vector quantization network with measured frequency response analysis to identify partial discharge (PD) sources. It also utilizes segmented winding transfer functions derived through vector fitting techniques to improve the accuracy of PD source location. Furthermore, the proposed classification algorithm maintains high accuracy even in noisy environments, eliminating the need for traditional noise reduction techniques.

[0039] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings, in which:

[0041] Figure 1 This is a flow chart of the method for locating the local discharge source of a power transformer based on a learning vector quantization network model according to the present invention;

[0042] Figure 2 Schematic diagram for setting up the measurement of PD reference signal;

[0043] Figure 3 Schematic diagram for setting up measurements of PD test signals;

[0044] Figure 4 Schematic diagram of the laboratory winding geometry and its tap connections. DETAILED DESCRIPTION

[0045] The following describes the embodiments of the present invention by means of specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0046] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0047] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0048] Example 1

[0049] like Figure 1 As shown, the present invention provides a method for locating the local discharge source of a power transformer based on a learning vector quantization network model, comprising the following steps:

[0050] Step 1: Using the swept frequency response method, the frequency response of the voltage corresponding to each tap of the transformer winding is measured using a frequency response analyzer (FRA) and the corresponding transfer function is derived;

[0051] Step 2: Use vector fitting technology to approximate the frequency domain function in the form of a sum of first-order rational fractions to obtain an approximate fitting transfer function;

[0052] Step 3: Calculate and save all approximate fitted voltage transfer functions from the transformer winding taps to the neutral line, thus creating a comprehensive library of segmented winding transfer functions (SWTFs);

[0053] Step 4: Inject a Gaussian function into the calculated SWTF at each winding tap and capture the time domain response as a partial discharge reference signal.

[0054] Step 5: The labeled partial discharge reference signals form training data, and the LVQ network model is trained and created.

[0055] Step 6: Generate partial discharge test signals by applying partial discharge pulses to any part of the winding using the partial discharge calibration device and measuring their response at the neutral terminal in the time domain.

[0056] Step 7: Use partial discharge test signals to test the performance of the proposed classification model.

[0057] Example 2

[0058] The test object in this embodiment is an air-core laboratory winding, and its LVQ model is established by SWTF obtained by a frequency response analyzer, wherein the FRA test circuit corresponding to each tap position is connected as follows: Figure 2 As shown. The configuration consists of a disc-type winding with 8 discs (coils), each of which contains 16 turns. A separate connection is taken from each disc, and a total of 8 separate connections can be used for partial discharge injection. The experimental equipment includes a PD calibrator for generating partial discharge pulses, a quadrupole and a coupling capacitor, which transmits the high-frequency PD signal from the winding to the measurement impedance with minimal attenuation. At the same time, in order to further improve the accuracy of the method, the influence of the coupling capacitor is integrated into the MATLAB calculation used to build the LVQ model. In order to simulate the transformer core, a slender cylindrical aluminum tube with a diameter of 520 mm is placed concentrically in the winding. The schematic diagram of the laboratory winding is shown in Figure 3 The geometry of the laboratory winding is shown in Figure 4 As shown, the first four discs are interleaved and arranged in series with the next four continuous discs. Each turn of conductor is rectangular with double-sided paper inserted between the turns. The discs are separated by key pads and air.

[0059] The implementation process of the present invention will be further described in detail below with reference to the accompanying drawings:

[0060] The first step is to use a frequency response analyzer to obtain the RLC model parameters to achieve accurate modeling of the transformer winding.

[0061] In this embodiment, a sweep frequency response method is used to inject a 20V low-amplitude input voltage signal into all taps of the transformer winding. The frequency of the external excitation source is continuously changed, and the ratio of the signal amplitude of the neutral point output voltage to the input voltage in the range of 20Hz to 2MHz is measured to obtain the transfer function of the winding under the specified excitation end and response end, that is, the transfer function TF of the segmented winding. i .

[0062] TF i =V o / V Ti

[0063] Where V Ti The transfer function TF is the measured voltage at each tap position along the air-core laboratory winding in the frequency domain, and V is the measured voltage at the neutral point in the frequency domain. i is derived from the voltage frequency response corresponding to each tap.

[0064] In the second step, the frequency domain function is approximated by the vector fitting technique in the form of a sum of first-order rational fractions to obtain the approximate fitting transfer function TF. The frequency response is then calculated based on the selected transfer function and compared with the measured frequency response.

[0065]

[0066] Where N(s) and D(s) are polynomials in the domain s=jw, with orders m and n respectively. i and p i are usually complex conjugate pairs, being the residues and poles of the fitted expression, respectively, while d is a real number.

[0067] In the third step, all the approximate fitted voltage transfer functions from the taps of the transformer winding to the neutral are calculated and saved, thus creating a comprehensive library of segmented winding transfer functions.

[0068] In the fourth step, the partial discharge pulse is simulated by a Gaussian function and injected into the transfer function of the segmented winding, and the captured time domain response is used as the partial discharge reference signal.

[0069] In this embodiment, the mathematical expression of the PD pulse defined by the Gaussian function is as follows:

[0070]

[0071] t represents the time variable, A represents the amplitude of the PD pulse, β represents the retention time (the time of the peak maximum), and α represents the width of the control pulse. In this example, the following parameters are used: α = 839.3 × 10-9, β = 2.155 μs, and A = 9.807 to construct the Gaussian function.

[0072] In the fifth step, the labeled partial discharge reference signals form the training data, and the LVQ network model is trained and created.

[0073] 1) Input and output of the LVQ model

[0074] In this embodiment, the LVQ model's input dataset is in matrix format, where each column corresponds to a sampled time-domain signal at a specific tap location along the transformer winding. This input matrix includes eight tap locations, organized into eight columns, each representing a tap. The rows of the matrix are populated with the time-domain sampling points of the PD signal. Preprocessing methods are used to normalize the input vectors to ensure signal amplitude uniformity and enhance the effectiveness of the training process.

[0075] In this embodiment, the output of the LVQ network is a class label, where each class represents a specific winding tap position. In this case, there are 8 classes of output, corresponding to the 8 available tap positions.

[0076] 2) LVQ model training process

[0077] In this embodiment, the LVQ network iteratively updates its prototype vector based on the input data to improve classification accuracy. This process includes the following steps:

[0078] Step a: Transform the prototype vector {w j (0)|j=1,2,···,N} assigns N input vectors {x i |i=1,2,···,L} to initialize the prototype vector. Generally speaking, L>N.

[0079] Step b: Randomly select input vector x from the sample set i ;

[0080] Step c: Calculate the input vector x i With the prototype vector w j Distance: d ij =||x i -w j || 2 ;

[0081] Step d: Find the value that matches the input vector x i The nearest prototype vector w c , c = argmin j∈{1,2,···,N} d ij ;

[0082] Step e: When the class of the input vector is consistent with the class of the winning prototype vector, the prototype vector will be adjusted to be closer to the input vector, thereby increasing its chance of winning in subsequent iterations; on the contrary, if the class of the input vector and the winning prototype does not match, the prototype vector will move away from the input vector, thereby reducing its chance of winning in the next round. The specific modifications are as follows:

[0083]

[0084] Among them, S wc Representation and prototype vector w c Related classes, S xi Represents the input vector x i The class label of . Learning constant α n (0<α n <1).

[0085] Step f: The training process iterates until the prototype vector stabilizes, ensuring convergence. If no further changes are observed, the process terminates; otherwise, it returns to step 2. Once converged, the final prototype vector is used to classify new partial discharge signals.

[0086] In this example, 90% of the collected PD signals were randomly allocated for training, while the remaining 10% were allocated to testing (5%) and validation. A "grid search" method was used to optimize key parameters, with a learning constant set to 0.001 and 700 neurons defined in the competitive layer. The sum of squared errors (SSE) was set to 0 as the target to guide the training process. After training, the network retained eight hidden nodes in the hidden layer, each representing a specific tap position. The final prototype vector was used to classify new PD signals.

[0087] In the sixth step, a 5000pC partial discharge pulse is randomly applied to any tap of the winding through the partial discharge calibration device, and the corresponding response is recorded on the four electrodes connected in series with the neutral end coupling capacitor to obtain the PD test signal.

[0088] In the seventh step, to evaluate the accuracy of the method under noisy conditions, Gaussian noise signals with varying signal-to-noise ratios were externally added to the PD test signals collected via a digitizing oscilloscope. These signals were then processed using MATLAB. The experimental results demonstrate that the method can successfully locate PD sources even in the presence of Gaussian noise with an SNR of -24 dB, without the need for noise suppression techniques. Table 1 lists the detailed results, demonstrating the robustness of the method in accurately identifying PD sources under challenging noise conditions.

[0089] Table 1

[0090]

[0091] In the above embodiments, references to "this embodiment" in the specification indicate that a particular feature, structure, or characteristic described in conjunction with the embodiment is included in at least some embodiments, but not necessarily all embodiments. Multiple occurrences of "this embodiment" do not necessarily refer to the same embodiment.

[0092] In the above embodiments, although the invention has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those skilled in the art based on the foregoing description. For example, other memory structures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed. The embodiments of the present invention are intended to encompass all such alternatives, modifications, and variations that fall within the broad scope of the appended claims.

[0093] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, any one of the methods in this embodiment is implemented.

[0094] This embodiment also provides an electronic terminal, including: a processor and a memory;

[0095] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes any one of the methods in this embodiment.

[0096] Regarding the computer-readable storage medium in this embodiment, those skilled in the art will appreciate that all or part of the steps in the aforementioned method embodiments can be implemented using hardware associated with the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps in the aforementioned method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0097] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication with each other. The memory is used to store computer programs, the communication interface is used for communication, and the processor and the transceiver are used to run computer programs so that the electronic terminal executes the various steps of the above method.

[0098] In this embodiment, the memory may include a random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage.

[0099] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0100] The present invention can be used in a wide variety of general-purpose or special-purpose computing system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments that include any of the above.

[0101] The present invention may be described in the general context of computer-executable instructions, such as program modules, executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media, including storage devices.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for locating partial discharge sources in power transformers based on a learning vector quantization network model, characterized by: The following steps are involved: S1: Measure the voltage frequency response of the transformer winding at different tap positions; S2: Obtain an approximate fitting voltage transfer function by using vector fitting VF technology; S3: Establish the transformer segment winding transfer function SWTF library; S4: Calculate the partial discharge PD reference signal; S5: constructing a learning vector quantization (LVQ) network model, and training the LVQ network model using the labeled partial discharge reference signal as a training set; S6: injecting a partial discharge test pulse into a randomly selected winding tap connection and using the resulting response as the partial discharge test signal; S7: Input the partial discharge test signal into the trained LVQ network model to test the classification performance of the model.

2. The method for locating partial discharge sources of power transformers based on a learning vector quantization network model according to claim 1, characterized in that: Step S1 specifically includes: using a sweep frequency response method to continuously change the frequency of the external excitation source, measuring the ratio of the signal amplitude of the response terminal voltage to the excitation terminal voltage of the transformer winding at different frequencies, and obtaining the transfer function of the winding under the conditions of specified excitation terminal and response terminal; The input voltage signal V Ti Inject all taps of the transformer winding and measure the output voltage V at the neutral point o , the winding transfer function is obtained as: TF i =V o / V Ti Transfer Function TF i is derived from the voltage frequency response corresponding to each tap.

3. The method for locating partial discharge sources of power transformers based on a learning vector quantization network model according to claim 1, characterized in that: Step S2 specifically includes: using vector fitting technology to approximate the frequency domain function in the form of a sum of first-order rational fractions, and approximately fitting multiple transfer functions (TF1, TF2, ···, TF i ), these transfer functions are expressed in the form of rational functions: Where N(s) and D(s) are polynomials in the domain s=jw, with orders m and n respectively; i and p i are complex conjugate pairs, representing the residues and poles of the fitting expression respectively; d is a real number.

4. The method for locating partial discharge sources of power transformers based on a learning vector quantization network model according to claim 1, characterized in that: Step S3 specifically includes: calculating and saving all approximate fitting voltage transfer functions from the taps of the transformer winding to the neutral line, thereby creating a transformer segment winding transfer function SWTF library.

5. The method for locating partial discharge sources of power transformers based on a learning vector quantization network model according to claim 1, characterized in that: Step S4 specifically includes: defining the actual partial discharge pulse by a Gaussian function, and determining a reference signal corresponding to each tap position in the transformer winding by capturing the PD pulse response of each calculated SWTF in the time domain; the mathematical expression of the PD pulse defined by the Gaussian function is as follows: Where t represents the time variable, A represents the amplitude of the PD pulse, β represents the retention time, and α represents the width of the control pulse; the calibration process uses the Gaussian function.

6. The method for locating partial discharge sources of power transformers based on a learning vector quantization network model according to claim 1, characterized in that: The LVQ network model is first trained using labeled partial discharge reference signals to establish hidden layers to develop a robust classification model. Arbitrary partial discharge responses measured at the neutral terminal are then used as test signals to evaluate the accuracy of the model in detecting partial discharge locations.

7. The method for locating partial discharge sources of power transformers based on a learning vector quantization network model according to claim 1, characterized in that: The LVQ network model consists of an input layer, a hidden layer and an output layer; the input data of the LVQ network model consists of a partial discharge reference signal in the time domain, and the output is a category label, each category representing a specific winding tap position.

8. The method for locating partial discharge sources of power transformers based on a learning vector quantization network model according to claim 1, characterized in that: The LVQ network training process is as follows: Step a: Transform the prototype vector {w j (0)|j=1,2,···,N} assigns N input vectors {x i |i=1,2,···,L} to initialize the prototype vector; Step b: Randomly select input vector x from the sample set i ; Step c: Calculate the input vector x i With the prototype vector w j Distance: d ij =||x i -w j || 2 ; Step d: Find the value that matches the input vector x i The nearest prototype vector w c ,c=arg min j∈{1,2,···,N} d ij ; Step e: If the class of the input vector is consistent with the class of the winning prototype vector, the prototype vector is adjusted to be closer to the input vector; if the class of the input vector and the winning prototype do not match, the prototype vector is moved away from the input vector. The specific modifications are as follows: Among them, S wc Representation and prototype vector w c Related classes, S xi Represents the input vector x i Class label; 0<α n <1 is the learning constant; Step f: Iterate step bf until the prototype vector is stable and converges.

9. The method for locating partial discharge sources of power transformers based on a learning vector quantization network model according to claim 1, characterized in that: In step S7, the trained LVQ network model predicts the PD position based on matching the input vector category with the retained prototype vector obtained from the PD reference signal.

Images