Dry-type transformer winding defect identification method

Abstract

The invention discloses a dry-type transformer winding defect identification method. The method comprises the following steps: step 1, collecting vibration quantity, electrical quantity and temperature data of a winding and an iron core; step 2, extracting time domain features and frequency domain features of each type of data of the vibration quantity and the electrical quantity, and forming a data set after feature expansion; step 3, clustering data in the data set obtained in the step 2 to obtain a plurality of clusters; judging the vibration quantity data in each cluster to identify an abnormal cluster; electrical quantity data and temperature data in the abnormal clusters are judged based on national standards; and step 4, clustering each abnormal cluster to obtain a plurality of clustering centers, and performing qualitative analysis in combination with an electrical quantity and temperature data judgment result, thereby obtaining a defect type of each abnormal cluster. According to the embodiment of the invention, the turn-to-turn short circuit and deformation defect of the dry-type transformer winding can be effectively identified.

Description

Technical Field

[0001] The present invention relates to the field of dry-type transformer defect identification, and in particular to a dry-type transformer winding defect identification method. Background Art

[0002] With the increasing development of power systems, electrical equipment failures are attracting increasing attention. Analysis of the causes of power transformer failures, particularly dry-type transformers, shows that winding failures account for over 60% of all failures. Common faults include winding deformation and depression, interturn short circuits, and partial discharge. In recent years, with the increase in power loads and the increasing complexity of transformer operating environments, early detection of winding defects has become particularly important.

[0003] Since winding deformation and inter-turn short circuits are common defects in dry-type transformers, different defects produce different acoustic signal characteristics. Therefore, existing methods for detecting dry-type transformer winding defects based on soundprints can also identify and distinguish winding deformation and inter-turn short circuits. Soundprint detection technology uses the characteristics of sound signals to analyze the physical state of a target. In dry-type transformer winding defect detection, soundprint detection technology uses acoustic sensors such as high-frequency microphones to collect the sound signals emitted by the dry-type transformer windings and core during operation. It then extracts multi-dimensional features from the sound signals as a basis for judgment.

[0004] Voiceprint-based dry-type transformer winding defect identification methods typically require the integration of machine learning models, such as support vector machines (SVM), random forests (RF), and convolutional neural networks (CNN), to classify acoustic signals. By training large amounts of data from both normal and abnormal operating conditions, the accuracy of winding defect identification can be improved. However, because acoustic signals are susceptible to interference from external noise, the generalization capabilities of machine learning models are difficult to optimize, resulting in low accuracy for existing voiceprint-based dry-type transformer winding defect identification methods. Furthermore, over time, the aging of dry-type transformer windings, changes in the characteristics of the internal voiceprint, and changes in external noise can cause the false alarm and missed alarm rates of identification and classification to skyrocket, failing to meet practical standards.

[0005] Therefore, although the existing technology of dry-type transformer winding defect detection based on voiceprint can serve as a supplementary means of vibration detection to a certain extent, it is currently difficult to use as an independent online monitoring method because it is easily affected by environmental noise and has poor reliability. Summary of the Invention

[0006] The present invention provides a dry-type transformer winding defect identification method to solve the problem of poor reliability of the dry-type transformer winding defect detection and identification method in the prior art.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] The dry-type transformer winding defect identification method includes the following steps:

[0009] Step 1: Collect three types of data from the dry-type transformer, namely vibration data, electrical quantity data, and temperature data of the three-phase winding and the iron core, thereby forming a data set;

[0010] Step 2: extracting the time domain features and frequency domain features of the electrical quantity data and vibration quantity data collected in step 1 respectively, and adding the time domain features and frequency domain features to the data set obtained in step 1, thereby forming a data set with expanded features;

[0011] Step 3: Cluster the data in the dataset after the expanded features obtained in step 2 to obtain multiple clusters; then, based on the Lyapunov stability principle, judge the vibration data in each cluster to identify clusters with abnormal winding vibration and normal core vibration as abnormal clusters; and, in combination with national standards for power quality and temperature of dry-type transformers, judge the electrical quantity data and temperature data in the abnormal clusters to determine whether the electrical quantity data and temperature data in each abnormal cluster are normal;

[0012] Step 4: Cluster the abnormal clusters obtained in step 3 to obtain multiple cluster centers, and perform a qualitative analysis on each cluster center based on the judgment results of the electrical quantity data and temperature data in the abnormal cluster obtained in step 3, thereby judging the defect type to which each cluster center belongs, and then identifying the defect type of each abnormal cluster according to the defect type to which each cluster center belongs.

[0013] In the further step 1, the collected electrical quantities of the three-phase winding and the core include the effective value of voltage, the effective value of current, the total current distortion rate, the total voltage distortion rate, the load rate, and the active power.

[0014] In the further step 2, a sequence of a set time window length is selected from each type of electrical quantity data and vibration quantity data based on the sliding window method as a basic feature, and then the time domain features of the corresponding type of data are constructed based on the basic features.

[0015] In the further step 2, the time domain features constructed for the vibration amount include the average value, skewness value, peak-to-peak value, and kurtosis value.

[0016] In the further step 2, the fast Fourier transform method is used to extract frequency domain features from each type of electrical quantity data and vibration quantity data.

[0017] In a further step 3, a self-organizing map algorithm is used to cluster the data in the dataset after the expanded features obtained in step 2.

[0018] In the further step 4, the fuzzy C-means clustering method is used to cluster the abnormal clusters obtained in step 3.

[0019] Compared with the prior art, the present invention has the following advantages:

[0020] 1. Multi-category data collection improves the real-time and accuracy of defect identification.

[0021] 2. Combining vibration, electrical and temperature data, it can fully reflect the winding operating status and enhance diagnostic capabilities.

[0022] 3. The self-organizing map algorithm adopted can adaptively form abnormal clusters, improve the sensitivity of abnormality detection, and reduce the false alarm rate.

[0023] 4. The fuzzy C-means clustering adopted can effectively distinguish between winding deformation and inter-turn short circuit, making the recognition and classification accuracy reach more than 97%. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flow chart of a method according to an embodiment of the present invention.

[0025] Figure 2 It is a flowchart of the self-organizing map training according to an embodiment of the present invention.

[0026] Figure 3 It is a schematic diagram of the process of fuzzy C-means clustering according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The present invention will be further described below with reference to the accompanying drawings and examples.

[0028] like Figure 1 As shown, this embodiment discloses a method for identifying defects in a dry-type transformer winding, comprising the following steps:

[0029] Step 1: Collect four types of data from dry-type transformers.

[0030] Specifically, in this embodiment, vibration sensors and electrical sensors are installed at multiple different locations on the three-phase windings and iron core of the dry-type transformer to collect vibration and electrical data of the three-phase windings and iron core, and an infrared imager is used to collect temperature data of the three-phase windings and iron core. Three types of data are obtained to construct a data set.

[0031] Electrical quantities can reflect the electrical performance of dry-type transformers, temperature values can reflect the energy loss status of dry-type transformers, and vibration values can indicate mechanical problems with the dry-type transformer winding components. These multiple types of data can comprehensively characterize the operating status of dry-type transformers. The electrical quantities collected for the three-phase windings and core in this embodiment include the three-phase RMS voltage, RMS current, total current distortion rate, total voltage distortion rate, load factor, and active power.

[0032] After data collection, the collected data is preprocessed to improve the accuracy of subsequent defect identification. The preprocessing adopts conventional preprocessing methods including correction and filtering.

[0033] Step 2: extract the time domain features and frequency domain features of the electrical quantity data and vibration quantity data collected in step 1 respectively, and add the time domain features and frequency domain features to the data set obtained in step 1, thereby forming a data set with expanded features.

[0034] In this embodiment, a sliding window method is used to select sequences of a set length from each type of electrical quantity data and vibration quantity data collected in step 1 as basic features. Then, based on the basic features, time domain features of the corresponding data types are constructed. The constructed time domain features can more comprehensively and accurately capture the fluctuation degree of dry-type transformers, thereby improving the accuracy and sensitivity of defect detection. The time domain feature acquisition process is as follows:

[0035] (A1) Step 1: Collect electrical quantity data and vibration quantity data. Each type of data is time series data. Each type of data can be expressed as S = {s1, s2, ..., s t ,…}, where s i is the data value corresponding to time index i, i∈{1,2,…}.

[0036] Based on the sliding window method, a sequence S(t) with a set time window length of l is selected from each type of data. t-(l-1) ,s t-l ,…,s t}, as the basic features of the corresponding class data at time t. The same sliding window operation is performed on each type of data to obtain the basic features of each type of data.

[0037] (A2) Based on statistical knowledge, the time domain features of the corresponding data are constructed according to the basic features. Among them, the time domain features constructed according to the basic features for the vibration data are the average value, skewness value, peak-to-peak value, and kurtosis value.

[0038] average value Characterizes the symmetry of the vibration quantity, and its calculation formula is:

[0039]

[0040] Where: represents the average value of the signal in a time window of length l; s i represents the data value corresponding to the time index i; l represents the total amount of data of the vibration sample.

[0041] Skew (S) is a statistic that reflects the direction and degree of skew in the distribution of vibration data. It characterizes the degree of asymmetry of the waveform relative to the average value. Its calculation formula is:

[0042]

[0043] Where: Skew(S) represents the data skewness; s i Represents the data value corresponding to time index i; represents the signal mean; σ represents the signal variance.

[0044] Peak-to-peak value S pp It refers to the difference between the maximum and minimum vibration values within a cycle, reflecting the magnitude of the vibration intensity of the dry-type transformer. The calculation formula is:

[0045] S pp =S max -S min

[0046] Where: S max Indicates the maximum value of vibration within a period; S min Indicates the minimum value of the signal within the period.

[0047] Kurtosis K is a numerical statistic that describes the peak degree of the signal waveform and can reflect the distribution characteristics of the data. Its calculation formula is:

[0048]

[0049] Where, Represents the expected value of data S(t); represents the mean value of the data S(t); σ represents the standard deviation of the data S(t).

[0050] In this embodiment, the fast Fourier transform method is used to extract frequency domain features from each type of electrical quantity and vibration quantity data. When the state of the dry-type transformer changes, it is usually accompanied by changes in nonlinear characteristics to varying degrees. The change in nonlinear characteristics can be characterized by the mutual relationship between the frequency components of the signal. Extracting the fundamental frequency and harmonic information of each type of electrical quantity data and vibration quantity data can help determine the state of the dry-type transformer. Fast Fourier transform (FFT) is an algorithm for efficiently calculating discrete Fourier transform (DFT). The process of extracting frequency domain features using the fast Fourier transform method in this embodiment is as follows:

[0051] (B1) Select the appropriate transformation length;

[0052] (B2) The input data sequence is divided into two subsequences according to the odd-even order, and then the two subsequences are respectively extracted by time. Finally, the DFT results of the two subsequences are combined to obtain the DFT of the entire sequence.

[0053] (B3) For each subsequence, continue the decomposition by decimation in time until the length of the subsequence is 2, and calculate its DFT.

[0054] (B4) Combining the results of the sub-DFTs according to the time extraction method to obtain the DFT of the entire sequence.

[0055] (B5) The amplitude of the FFT result is obtained by complex modulus, and the effective value of the amplitude is calculated to obtain the amplitude of the original data such as vibration quantity and electrical quantity at different frequencies.

[0056] Step 3: Use the self-organizing map algorithm to perform unsupervised clustering on the data in the data set after the expanded features obtained in step 2 to obtain multiple clusters; then judge the vibration quantity data in each cluster based on the Lyapunov stability principle to identify the cluster with abnormal winding vibration and normal core vibration as the abnormal cluster, thereby obtaining all abnormal clusters; and judge the electrical quantity data and temperature data in the abnormal cluster in combination with the national power quality standard to determine whether the electrical quantity data and temperature data in each abnormal cluster meet the national standard.

[0057] like Figure 2 As shown, the process of unsupervised clustering using the self-organizing map algorithm in this embodiment is as follows:

[0058] (C1) Calculate the Z-score of each feature to normalize the data in the input dataset.

[0059] (C2) The number of neurons is calculated based on the number of data in the dataset, thereby determining the size of the map.

[0060] (C3) Map the neurons into a two-dimensional map and initialize the weight vector of each neuron.

[0061] (C4) For each input data sample, calculate the Euclidean distance between all neurons.

[0062] (C5) Select the neuron closest to the current input data sample as the best matching unit.

[0063] (C6) Update the formula W according to the weight i (t+1)=W i (t)+h ci (t)[S(t)-W i (t)] Update the neuron weights.

[0064] Where: W i (t) is the weight vector of neuron i at the tth training iteration; S(t) is the input sample value of the best matching unit; h ci (t) is a neighborhood function, and has:

[0065]

[0066] Where: h ci (t) is the neighborhood function between the best matching unit c and neuron i; t is the index of training iteration; a(t) is the learning rate; r c (t) is the vector of the best matching unit c; r i (t) is the vector of neuron i; σ(t) is the radius around c.

[0067] (C7) Iterate the weight and Euclidean distance calculation process until the weights of all neurons no longer change significantly, and obtain the optimal weight vector of each neuron in the output layer.

[0068] (C8) According to the weights of the output layer neurons, similar input samples are mapped to the same or adjacent neurons, and clustering is performed to obtain multiple different clusters.

[0069] In this embodiment, the dry-type transformer system can be considered a dynamic system based on the vibration data in each cluster, and the stability of the dry-type transformer is analyzed using Lyapunov's transient stability theory. Based on the Lyapunov stability principle, if the input varies within a certain range, the system's output state will also fluctuate within a certain range. When the vibration response or other output indicators of the dry-type transformer system exceed the historical output range, the system is considered to be in an abnormal state. Based on this principle, based on the vibration data in each cluster, abnormal clusters with abnormal winding vibration and normal core vibration are identified.

[0070] In this embodiment, for the electrical quantity data in the abnormal cluster, the normal range of the electrical quantity characteristics is evaluated based on the relevant national standards for power quality and combined with historical operating experience. If the electrical quantity characteristics in the abnormal cluster exceed the range of the national standards for power quality, the electrical quantity of the abnormal cluster is considered abnormal.

[0071] In this embodiment, the maximum allowable temperature of a dry-type transformer winding under maximum load conditions is equal to the ambient temperature plus the temperature rise. The temperature rise of a dry-type transformer is typically determined by the heat resistance rating of its insulation material, and there are corresponding national standards. For example, according to the national standard GB / T 10228-2008, "Technical Parameters and Requirements for Dry-Type Transformers," for dry-type transformers with Class B insulation, the maximum operating temperature limit is 130°C, and the maximum allowable temperature rise is 80°C. If the temperature data in an abnormal cluster exceeds the range of the corresponding national standard, the temperature value of the abnormal cluster is considered abnormal.

[0072] Abnormal analysis of electrical quantity and temperature data can more comprehensively evaluate the causal characteristics of abnormal winding vibration, which can be used to verify the effectiveness of vibration anomaly identification and provide a basis for subsequent further defect classification.

[0073] Step 4: Use the fuzzy C-means clustering method to cluster the abnormal clusters obtained in step 3 to obtain multiple cluster centers, and perform a qualitative analysis on each cluster center based on the judgment results of the electrical quantity data and temperature data in the abnormal cluster obtained in step 3, thereby judging the defect type to which each cluster center belongs, and then identifying the defect type of each abnormal cluster according to the defect type to which each cluster center belongs.

[0074] like Figure 3 As shown, the process of clustering using the fuzzy C-means clustering method in this embodiment is as follows:

[0075] For fuzzy C-means clustering, its objective function J FCM The (U,V) calculation formula is as follows:

[0076]

[0077] Where: m>1 is called fuzzy index; fuzzy membership matrix U={u ki}, where u ki Represents the i-th data point s i The membership value of the kth class; V = {v1, v2, ..., v c} is a matrix consisting of c cluster center vectors; u ki =‖s i -v k ‖ represents the number of points from the outlier point s i To the center v k The distance measure of To represent the data point s i To cluster center v k The square of the distance between them; X is the total number of sample points in the data set, that is, the number of all data points that need to be clustered.

[0078] This embodiment aims to achieve intelligent identification of inter-turn short circuits and deformations in windings. However, actual abnormal data sets on site contain at least one defect type. The number of data clustering categories c is determined through empirical methods and limited trial and error. Cross-validation is used to determine the fuzzy index m.

[0079] Initialize the cluster center to V (0) , initialize the membership matrix to U (0) , set the value of convergence accuracy ε (ε>0) and set the number of iterations l = 0.

[0080] Using the formula Calculate and update the membership matrix, and get the updated membership matrix U (l+1) . Among them ‖s i -v l ‖ represents the data point s i To the lth cluster center v l The Euclidean distance between .

[0081] Using the formula Calculate and update the cluster center to get the updated cluster center V (l+1) If the convergence condition max{J (l+1) -J (l)}<ε, then stop the operation, otherwise let l=l+1, repeatedly calculate and update the cluster center and membership matrix until the convergence condition is met. (l) Represents the objective function value of the 1st iteration of the fuzzy C-means clustering algorithm.

[0082] Finally, according to the highest degree of membership of the data, the data is assigned to the corresponding cluster, and the cluster center of each cluster is obtained.

[0083] In this embodiment, a qualitative analysis is performed on each cluster center in combination with the electrical quantity data and temperature data judgment results in the abnormal cluster obtained in step 3. The analysis process is as follows:

[0084] When a short circuit occurs between turns of a dry-type transformer winding, the winding current increases sharply. The nonlinear load characteristics cause a significant increase in harmonic components, and the main frequency is concentrated at the multiple of the fundamental frequency. The overheating effect caused by the short-circuit current causes the winding temperature to rise rapidly. The imbalance of electromagnetic force caused by the short circuit and the instantaneous high current cause the vibration amplitude to change significantly, which generally manifests as a sharp increase.

[0085] When the transformer winding is deformed, the current and voltage do not change significantly, but the current harmonic components increase, especially at the odd harmonics. Deformation causes uneven heat dissipation inside the winding and a gentle temperature rise. Due to changes in the mechanical structure, the amplitude of the vibration harmonics increases. As the degree of deformation deepens, the main frequency of the vibration changes from 100Hz to an integer multiple of it, and the amplitude increases.

[0086] Based on the above analysis process, the defect type of each cluster center can be determined, and then according to the defect type of each cluster center, the defect type of each abnormal cluster in the cluster corresponding to the cluster center can be identified.

[0087] The dry-type transformer winding defect identification method described in this embodiment can effectively identify inter-turn short circuits and deformation defects in dry-type transformer windings, issue early alarms, and reduce the damage caused by winding defects to dry-type transformers.

[0088] The preferred embodiments of the present invention are described in detail above with reference to the accompanying drawings. The embodiments described in the present invention are merely descriptions of the preferred embodiments of the present invention and do not limit the concept and scope of the present invention. The various specific technical features described in the above specific embodiments can be combined in any suitable manner unless there is any contradiction. Such combinations should also be regarded as the contents disclosed in this disclosure as long as they do not violate the concept of the present invention. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0089] The present invention is not limited to the specific details of the above-mentioned embodiments. Within the scope of the technical concept of the present invention and without departing from the design concept of the present invention, various modifications and improvements made to the technical solution of the present invention by those skilled in the art should fall within the scope of protection of the present invention. The technical contents for which protection is sought in the present invention have been fully recorded in the claims.

Claims

1. A dry-type transformer winding defect identification method, characterized in that: The following steps are involved: Step 1: Collect three types of data from the dry-type transformer, namely vibration data, electrical quantity data, and temperature data of the three-phase winding and the iron core, thereby forming a data set; Step 2: extracting the time domain features and frequency domain features of the electrical quantity data and vibration quantity data collected in step 1 respectively, and adding the time domain features and frequency domain features to the data set obtained in step 1, thereby forming a data set with expanded features; Step 3: Cluster the data in the dataset after the expanded features obtained in step 2 to obtain multiple clusters; then, based on the Lyapunov stability principle, judge the vibration data in each cluster to identify clusters with abnormal winding vibration and normal core vibration as abnormal clusters; and, in combination with national standards for power quality and temperature of dry-type transformers, judge the electrical quantity data and temperature data in the abnormal clusters to determine whether the electrical quantity data and temperature data in each abnormal cluster are normal; Step 4: Cluster the abnormal clusters obtained in step 3 to obtain multiple cluster centers, and perform a qualitative analysis on each cluster center based on the judgment results of the electrical quantity data and temperature data in the abnormal cluster obtained in step 3, thereby judging the defect type to which each cluster center belongs, and then identifying the defect type of each abnormal cluster according to the defect type to which each cluster center belongs.

2. The dry-type transformer winding defect identification method according to claim 1, characterized in that: In step 1, the electrical quantities collected of the three-phase winding and the core include the effective value of voltage, effective value of current, total current distortion rate, total voltage distortion rate, load factor, and active power.

3. The dry-type transformer winding defect identification method according to claim 1, characterized in that: In step 2, based on the sliding window method, a sequence of set time window length is selected from each type of electrical quantity data and vibration quantity data as the basic feature, and then the time domain feature of the corresponding type of data is constructed based on the basic feature.

4. The dry-type transformer winding defect identification method according to claim 1, characterized in that: In step 2, the time domain features constructed for the vibration amount include the average value, skewness value, peak-to-peak value, and kurtosis value.

5. The dry-type transformer winding defect identification method according to claim 1, characterized in that: In step 2, the fast Fourier transform method is used to extract frequency domain features from each type of electrical quantity data and vibration quantity data.

6. The dry-type transformer winding defect identification method according to claim 1, characterized in that: In step 3, the self-organizing map algorithm is used to cluster the data in the dataset after the expanded features obtained in step 2.

7. The dry-type transformer winding defect identification method according to claim 1, characterized in that: In step 4, the fuzzy C-means clustering method is used to cluster the abnormal clusters obtained in step 3.

Images