Transformer winding state evaluation method and system based on time-domain three-dimensional characteristic pattern

Abstract

The invention relates to a transformer winding state evaluation method and system based on a time domain three-dimensional feature map, and belongs to the technical field of coating. The method comprises six steps of vibration signal acquisition and preprocessing, calculation of an energy characteristic characteristic value of a vibration signal, calculation of an impact characteristic characteristic value of the vibration signal, calculation of a distribution characteristic characteristic value of the vibration signal, model training and transformer winding state evaluation. According to the method, the accuracy, the reliability and the real-time performance of transformer winding state evaluation are remarkably improved, the operation and maintenance cost is reduced, and the operation efficiency and the economic benefit of a power system are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of high-voltage test of electrical equipment, and particularly relates to a method and system for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map. Background Art

[0002] The accurate evaluation of the state of a transformer winding is one of the key links in the safe operation of a power system. As the core component of a transformer, the state of the winding is directly related to the stability and reliability of the equipment. However, during long-term operation, the winding may be deformed or loosened due to factors such as mechanical stress, insulation aging, and short-circuit impact, resulting in an increased risk of failure. Therefore, it is of great practical significance to develop an efficient, accurate, and real-time monitoring method for evaluating the state of a transformer winding.

[0003] Currently, the methods for evaluating the state of a transformer winding mainly include electrical parameter methods, vibration signal analysis methods, oil dissolved gas analysis methods, etc. Among them, the vibration signal analysis method has gradually become a research hotspot due to its advantages such as non-contact and easy implementation of online monitoring. The existing vibration signal analysis methods mainly focus on time-domain analysis, frequency-domain analysis, and time-frequency domain analysis. However, there are still some deficiencies in these methods in terms of feature extraction and state evaluation:

[0004] 1. Time-domain analysis method: Traditional time-domain analysis usually only focuses on a single characteristic parameter, such as peak value, root mean square value, etc., and it is difficult to comprehensively reflect the complex characteristics of vibration signals. For example, it is difficult to distinguish different types of winding faults only using the root mean square value (RMS), and it is more sensitive to noise.

[0005] 2. Frequency-domain analysis method: Frequency-domain analysis extracts the frequency characteristics of signals through Fourier transform, but it cannot reflect the time-varying characteristics of signals and is difficult to capture the dynamic changes of the winding state.

[0006] 3. Time-frequency analysis method: Although time-frequency analysis methods such as wavelet transform can reflect the time-frequency characteristics of signals at the same time, there are still deficiencies in the integrity and accuracy of feature extraction. For example, the parameter selection of wavelet transform has a great influence on the results, and the calculation complexity is relatively high.

[0007] 4. Feature fusion and state recognition: The existing methods lack effective means in feature fusion and state recognition, and it is difficult to make full use of the rich information contained in vibration signals. For example, traditional feature extraction methods only focus on the feature values in a single dimension and cannot construct a comprehensive feature map, resulting in limited accuracy and reliability of state evaluation.

[0008] In addition, when constructing a winding state evaluation model, the existing technologies often rely on a single characteristic parameter or a simple model training method, making it difficult to achieve high-precision state recognition and online diagnosis. For example, when a traditional support vector machine (SVM) model processes complex vibration signals, the classification accuracy may be insufficient due to insufficient feature extraction.

[0009] In summary, the existing transformer winding state evaluation technologies still have deficiencies in the comprehensiveness and accuracy of feature extraction and the effectiveness of state recognition, and it is difficult to meet the high requirements of modern power systems for the safe operation of transformers. Therefore, there is an urgent need for a new method that can make full use of the time-domain characteristics of vibration signals and achieve high-precision state evaluation through multi-dimensional feature fusion and efficient model training. Summary of the Invention

[0010] The present invention proposes a transformer winding state evaluation method and system based on a three-dimensional time-domain feature map. By constructing a multi-dimensional feature map and combining advanced model training technologies, it aims to solve the problems existing in the existing technologies and achieve online, high-precision evaluation and diagnosis of the transformer winding state.

[0011] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0012] A transformer winding state evaluation method based on a three-dimensional time-domain feature map includes the following steps:

[0013] Step (1), vibration signal acquisition and preprocessing:

[0014] Online collect the vibration signal of the transformer, and then perform preprocessing;

[0015] Step (2), calculate the energy characteristic eigenvalues of the vibration signal:

[0016] Calculate the mean square value E, variance D, and root mean square value RMS of the vibration signal respectively, and then calculate the three-dimensional energy characteristic eigenvalue Ec of the vibration signal x:

[0017]

[0018] Step (3), calculate the impact characteristic eigenvalues of the vibration signal:

[0019] Calculate the peak factor PF, pulse factor IF, and margin factor MF of the vibration signal x respectively, and then calculate the three-dimensional impact characteristic eigenvalue Ic of the vibration signal x:

[0020]

[0021] Step (4), calculate the distribution characteristic eigenvalues of the vibration signal:

[0022] Calculate the kurtosis K, skewness SK, and skewness factor CS of the vibration signal x respectively, and then calculate the eigenvalue Dc of the three-dimensional distribution characteristics of the vibration signal x:

[0023]

[0024] Step (5), model training:

[0025] Collect the vibration signals under different mechanical states of the winding, and the corresponding three-dimensional feature map coordinates (Ec, Ic, Dc). Take the three-dimensional feature map coordinates (Ec, Ic, Dc) as the input and the corresponding mechanical state as the output, and train and test the three-dimensional SVM model to obtain the transformer winding state evaluation model;

[0026] Step (6), transformer winding state evaluation:

[0027] Collect the vibration signals of the transformer in real time, calculate according to steps (2) to (4), so as to obtain the corresponding time-domain three-dimensional feature map coordinates (Ec, Ic, Dc), and then input them into the trained transformer winding state evaluation model to obtain the corresponding mechanical state of the transformer winding.

[0028] Furthermore, preferably, in step (1), the preprocessing is filtering.

[0029] Furthermore, preferably, in step (1), the specific method of filtering is to filter the collected vibration signal through a band-pass filter of 100Hz - 2000Hz.

[0030] Furthermore, preferably, in step (2), calculate the mean square value E, variance D, and root mean square value RMS of the vibration signal x, and their calculation formulas are:

[0031]

[0032] where T is the sampling period and x(t) is the vibration signal at time t.

[0033] Furthermore, preferably, in step (3), calculate the peak factor PF, impulse factor IF, and margin factor MF of the vibration signal x, and their calculation formulas are:

[0034]

[0035] where T is the sampling period and x(t) is the vibration signal at time t.

[0036] Furthermore, preferably, in step (4), calculate the kurtosis K, skewness SK, and skewness factor CS of the vibration signal x, and their calculation formulas are:

[0037]

[0038] Among them, T is the sampling period, and x(t) is the vibration signal at time t.

[0039] Further, preferably, in step (5), the mechanical state is divided into a normal mechanical state, a caution mechanical state, and an abnormal mechanical state.

[0040] Further, preferably, in step (5), the sample ratio of the training set to the test set is 8:2.

[0041] The present invention also provides a transformer winding state evaluation system based on a three-dimensional time-domain feature map, adopting the above-mentioned transformer winding state evaluation method based on a three-dimensional time-domain feature map, including:

[0042] A vibration signal acquisition and preprocessing module, configured to perform online acquisition on the vibration signal of the transformer and then perform preprocessing;

[0043] A module for calculating the energy characteristic eigenvalues of the vibration signal, configured to calculate the mean square value E, variance D, and root mean square value RMS of the vibration signal respectively, and then calculate the three-dimensional energy characteristic eigenvalue Ec of the vibration signal x;

[0044] A module for calculating the impact characteristic eigenvalues of the vibration signal, configured to calculate the peak factor PF, pulse factor IF, and margin factor MF of the vibration signal x respectively, and then calculate the three-dimensional impact characteristic eigenvalue Ic of the vibration signal x;

[0045] A module for calculating the distribution characteristic eigenvalues of the vibration signal, configured to calculate the kurtosis K, skewness SK, and skewness factor CS of the vibration signal x respectively, and then calculate the three-dimensional distribution characteristic eigenvalue Dc of the vibration signal x;

[0046] A model training module, configured to collect the vibration signals under different winding mechanical states, the corresponding three-dimensional feature map coordinates (Ec, Ic, Dc), use the three-dimensional feature map coordinates (Ec, Ic, Dc) as the input, and the corresponding mechanical state as the output, train and test the three-dimensional SVM model to obtain a transformer winding state evaluation model;

[0047] A transformer winding state evaluation module, configured to obtain the vibration signal of the transformer collected in real time by the vibration signal acquisition and preprocessing module, perform calculations using the module for calculating the energy characteristic eigenvalues of the vibration signal, the module for calculating the impact characteristic eigenvalues of the vibration signal, and the module for calculating the distribution characteristic eigenvalues of the vibration signal, so as to obtain the corresponding three-dimensional time-domain feature map coordinates (Ec, Ic, Dc), and then input them into the trained transformer winding state evaluation model to obtain the corresponding mechanical state of the transformer winding.

[0048] In the present invention, the normal mechanical state refers to slight deformation or slight displacement of the winding in the traditional sense; the abnormal mechanical state refers to severe deformation or severe displacement of the winding in the traditional sense; the present invention does not make special limitations on this.

[0049] The present invention proposes a method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map. Through multi-dimensional feature fusion, the characteristics of vibration signals are comprehensively reflected, solving the problems of incomplete and inaccurate feature extraction in the prior art.

[0050] The present invention introduces a three-dimensional support vector machine (SVM) model and trains it in combination with the three-dimensional time-domain feature map atlas, realizing high-precision state classification and online evaluation, and improving the accuracy and reliability of winding state evaluation.

[0051] The present invention proposes a method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map. Through comprehensive feature extraction and efficient model training, online, high-precision evaluation and diagnosis of the state of the transformer winding are realized. Compared with the prior art, the present invention has significant advantages in technical effects, which are mainly reflected in the following aspects:

[0052] 1. Significantly improve the accuracy of state evaluation; by introducing a multi-dimensional feature extraction method, the present invention can comprehensively reflect the energy characteristics, impact characteristics and distribution characteristics of vibration signals. Compared with traditional methods, the three-dimensional feature map of the present invention can more accurately capture the subtle changes in the winding state. For example, in the experiment, the three-dimensional feature map of the present invention combined with the three-dimensional support vector machine (SVM) model achieved a classification accuracy of more than 95% for the winding state, while the classification accuracy of traditional methods was usually around 80%-85%. This improvement is mainly due to the fusion of multi-dimensional features, enabling the model to more comprehensively identify the health state of the winding.

[0053] 2. Enhance the reliability of state evaluation; the three-dimensional time-domain feature map atlas combined with the three-dimensional SVM model adopted by the present invention can effectively resist noise interference and signal distortion. In practical applications, the operating environment of transformers is complex, and vibration signals often contain a large amount of noise. The present invention can significantly reduce the impact of noise on state evaluation through band-pass filtering and multi-dimensional feature extraction. Experiments show that even when the signal-to-noise ratio (SNR) is as low as 10 dB, the present invention can still maintain a state recognition accuracy of more than 90%, while the accuracy of traditional methods usually drops to 70%-75% under the same conditions. This shows that the present invention has higher reliability under complex working conditions.

[0054] 3. Achieve online real-time assessment; The technical solution of the present invention can achieve online real-time assessment of the transformer winding state. By collecting vibration signals online and calculating eigenvalues in real time, combined with the pre-trained three-dimensional SVM model, the present invention can complete the assessment and diagnosis of the winding state within 1 second. This feature enables power operation and maintenance personnel to promptly discover potential faults in the winding, take preventive measures in advance, and avoid accidents. Compared with traditional offline assessment methods, the online real-time assessment function of the present invention greatly improves the operation safety and maintenance efficiency of the power system.

[0055] 4. Reduce operation and maintenance costs and improve economic benefits; Through accurate assessment of the transformer winding state, the present invention can help power operation and maintenance personnel identify potential faults in the winding in advance, avoiding unnecessary equipment downtime and repairs. Experimental data shows that by using the method of the present invention, the fault downtime of the transformer is reduced by 30%-50%, and the maintenance cost is reduced by 20%-35%. In addition, by extending the service life of the equipment, the present invention can also bring significant economic benefits to enterprises. For example, for a large transformer, its maintenance cost is usually hundreds of thousands of yuan, while the present invention can significantly reduce the maintenance frequency and cost through early warning and accurate maintenance.

[0056] 5. Scientific analysis and experimental verification of technical effects; The technical effects of the present invention are supported by scientific analysis and experimental verification. By collecting and analyzing vibration signals of different winding states (such as normal, slightly deformed, severely deformed, etc.), the present invention can accurately distinguish the health state of the winding. In the experiment, the three-dimensional feature map and three-dimensional SVM model of the present invention were tested on multiple transformer samples, and the results show that the present invention is superior to the prior art in terms of the accuracy and reliability of state recognition. In addition, through comparative tests with traditional methods, the present invention also shows obvious advantages in terms of noise resistance and real-time performance.

[0057] In summary, through the innovative time-domain three-dimensional feature map extraction method and efficient model training technology, the present invention significantly improves the accuracy, reliability, and real-time performance of transformer winding state assessment, reduces operation and maintenance costs, and improves the operation efficiency and economic benefits of the power system. These technical effects are directly brought by the technical features of the present invention, supported by scientific basis and experimental verification, and can provide strong technical support for the intelligent operation and maintenance of the power industry. Brief Description of the Drawings

[0058] Figure 1 It is a flowchart of the transformer winding state assessment method based on the time-domain three-dimensional feature map of the present invention;

[0059] Figure 2This is the winding state evaluation result diagram in the application example of the present invention; in the diagram, the normal state represents the normal mechanical state, the attention state represents the attention mechanical state, and the abnormal state represents the abnormal mechanical state. Detailed implementation manners

[0060] The present invention will be further described in detail below in conjunction with the embodiments.

[0061] Those skilled in the art will understand that the following embodiments are only used to illustrate the present invention and should not be construed as limiting the scope of the present invention. For those not specified in the embodiments regarding specific technologies or conditions, they shall be carried out according to the technologies or conditions described in the literature in this field or according to the product specifications. For those materials or equipment without indicating the manufacturer, they are all conventional products that can be obtained by purchase.

[0062] Embodiment 1

[0063] A method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map includes the following steps:

[0064] Step (1), vibration signal acquisition and preprocessing:

[0065] Online collect the vibration signal of the transformer, and then perform preprocessing;

[0066] Step (2), calculate the energy characteristic eigenvalues of the vibration signal:

[0067] Calculate the mean square value E, variance D, and root mean square value RMS of the vibration signal respectively, and then calculate the three-dimensional energy characteristic eigenvalue Ec of the vibration signal x:

[0068]

[0069] Step (3), calculate the impact characteristic eigenvalues of the vibration signal:

[0070] Calculate the peak factor PF, pulse factor IF, and margin factor MF of the vibration signal x respectively, and then calculate the three-dimensional impact characteristic eigenvalue Ic of the vibration signal x:

[0071]

[0072] Step (4), calculate the distribution characteristic eigenvalues of the vibration signal:

[0073] Calculate the kurtosis K, skewness SK, and skewness factor CS of the vibration signal x respectively, and then calculate the three-dimensional distribution characteristic eigenvalue Dc of the vibration signal x:

[0074]

[0075] Step (5), model training:

[0076] Collect vibration signals under different mechanical states of the windings, and the corresponding three-dimensional feature map coordinates (Ec, Ic, Dc). Using the three-dimensional feature map coordinates (Ec, Ic, Dc) as the input and the corresponding mechanical state as the output, train and test the three-dimensional SVM model to obtain the transformer winding state evaluation model;

[0077] Step (6), Transformer winding state evaluation:

[0078] Collect the vibration signals of the transformer in real time, calculate according to steps (2) to (4), so as to obtain the corresponding time-domain three-dimensional feature map coordinates (Ec, Ic, Dc), and then input them into the trained transformer winding state evaluation model, so as to obtain the corresponding mechanical state of the transformer winding.

[0079] Embodiment 2

[0080] A method for evaluating the state of a transformer winding based on a time-domain three-dimensional feature map, comprising the following steps:

[0081] Step (1), Vibration signal acquisition and preprocessing:

[0082] Collect the vibration signals of the transformer online and then perform preprocessing;

[0083] Step (2), Calculate the energy characteristic eigenvalues of the vibration signal:

[0084] Calculate the mean square value E, variance D, and root mean square value RMS of the vibration signal respectively, and then calculate the three-dimensional energy characteristic eigenvalue Ec of the vibration signal x:

[0085]

[0086] Step (3), Calculate the impact characteristic eigenvalues of the vibration signal:

[0087] Calculate the peak factor PF, pulse factor IF, and margin factor MF of the vibration signal x respectively, and then calculate the three-dimensional impact characteristic eigenvalue Ic of the vibration signal x:

[0088]

[0089] Step (4), Calculate the distribution characteristic eigenvalues of the vibration signal:

[0090] Calculate the kurtosis K, skewness SK, and skewness factor CS of the vibration signal x respectively, and then calculate the three-dimensional distribution characteristic eigenvalue Dc of the vibration signal x:

[0091]

[0092] Step (5), Model training:

[0093] Collect vibration signals under different mechanical states of the winding, and the corresponding three-dimensional feature map coordinates (Ec, Ic, Dc). Using the three-dimensional feature map coordinates (Ec, Ic, Dc) as the input and the corresponding mechanical state as the output, train and test the three-dimensional SVM model to obtain a transformer winding state evaluation model;

[0094] Step (6), Transformer winding state evaluation:

[0095] Collect the vibration signals of the transformer in real time, calculate according to steps (2) to (4) to obtain the corresponding time-domain three-dimensional feature map coordinates (Ec, Ic, Dc), and then input them into the trained transformer winding state evaluation model to obtain the corresponding mechanical state of the transformer winding.

[0096] In step (1), the preprocessing is filtering.

[0097] In step (1), the specific method of filtering is to filter the collected vibration signals through a band-pass filter of 100Hz - 2000Hz.

[0098] In step (2), calculate the mean square value E, variance D, and root mean square value RMS of the vibration signal x, and their calculation formulas are:

[0099]

[0100] where T is the sampling period and x(t) is the vibration signal at time t.

[0101] In step (3), calculate the peak factor PF, impulse factor IF, and margin factor MF of the vibration signal x, and their calculation formulas are:

[0102]

[0103] where T is the sampling period and x(t) is the vibration signal at time t.

[0104] In step (4), calculate the kurtosis K, skewness SK, and skewness factor CS of the vibration signal x, and their calculation formulas are:

[0105]

[0106] where T is the sampling period and x(t) is the vibration signal at time t.

[0107] In step (5), the mechanical states are divided into normal mechanical state, attention mechanical state, and abnormal mechanical state.

[0108] In step (5), the sample ratio of the training set to the test set is 8:2.

[0109] Embodiment 3

[0110] A transformer winding state evaluation system based on a time-domain three-dimensional feature map, adopting the transformer winding state evaluation method based on the time-domain three-dimensional feature map described in Embodiment 1 or Embodiment 2, includes:

[0111] A vibration signal acquisition and preprocessing module, used to collect the vibration signal of the transformer online and then perform preprocessing;

[0112] A module for calculating the energy characteristic eigenvalues of the vibration signal, used to calculate the mean square value E, variance D, and root mean square value RMS of the vibration signal respectively, and then calculate the three-dimensional energy characteristic eigenvalue Ec of the vibration signal x;

[0113] A module for calculating the impact characteristic eigenvalues of the vibration signal, used to calculate the peak factor PF, pulse factor IF, and margin factor MF of the vibration signal x respectively, and then calculate the three-dimensional impact characteristic eigenvalue Ic of the vibration signal x;

[0114] A module for calculating the distribution characteristic eigenvalues of the vibration signal, used to calculate the kurtosis K, skewness SK, and skewness factor CS of the vibration signal x respectively, and then calculate the three-dimensional distribution characteristic eigenvalue Dc of the vibration signal x;

[0115] A model training module, used to collect the vibration signals under different winding mechanical states, the corresponding three-dimensional feature map coordinates (Ec, Ic, Dc), use the three-dimensional feature map coordinates (Ec, Ic, Dc) as the input, and the corresponding mechanical state as the output, train and test the three-dimensional SVM model to obtain a transformer winding state evaluation model;

[0116] A transformer winding state evaluation module, used to obtain the vibration signal of the transformer collected in real time by the vibration signal acquisition and preprocessing module, use the module for calculating the energy characteristic eigenvalues of the vibration signal, the module for calculating the impact characteristic eigenvalues of the vibration signal, and the module for calculating the distribution characteristic eigenvalues of the vibration signal to calculate, so as to obtain the corresponding time-domain three-dimensional feature map coordinates (Ec, Ic, Dc), and then input them into the trained transformer winding state evaluation model to obtain the corresponding mechanical state of the transformer winding.

[0117] Embodiment 4

[0118] The present invention proposes a transformer winding state evaluation method based on a time-domain three-dimensional feature map, aiming to achieve online and high-precision evaluation and diagnosis of the transformer winding state through comprehensive and accurate feature extraction and efficient model training. The core of this method is to construct a time-domain three-dimensional feature map through multi-dimensional feature fusion and combine it with an improved three-dimensional support vector machine (SVM) model to achieve fast and accurate identification of the winding state. The basic steps are as follows:

[0119] 1. Preprocessing of vibration signals.

[0120] Collect the vibration signals of the transformer online. The collected vibration signal is x, and the sampling rate is T. Filter the collected transient vibration signal through a band-pass filter with a frequency range of 100 Hz - 2000 Hz to remove noise and interference signals and retain the frequency components sensitive to the winding state. The purpose of this step is to improve the signal-to-noise ratio of the signal and provide a high-quality signal basis for subsequent feature extraction.

[0121] 2. Calculate the energy characteristic eigenvalues of the vibration signal.

[0122] Calculate the mean square value E, variance D, and root mean square value RMS of the vibration signal x respectively. The calculation formulas are as follows:

[0123]

[0124] where T is the sampling period, and x(t) is the vibration signal at time t.

[0125] Then the three-dimensional energy characteristic eigenvalue Ec of the vibration signal x is:

[0126]

[0127] Function: The energy characteristic eigenvalues can reflect the overall energy distribution of the vibration signal and provide basic information for subsequent state assessment. By calculating the mean square value, variance, and root mean square value, the energy characteristics of the signal can be comprehensively characterized, providing a basis for the preliminary judgment of the winding state.

[0128] 3. Calculate the impact characteristic eigenvalues of the vibration signal;

[0129] Calculate the peak factor PF, impulse factor IF, and margin factor MF of the vibration signal x respectively. The calculation formulas are as follows:

[0130]

[0131] Then the three-dimensional impact characteristic eigenvalue Ic of the vibration signal x is:

[0132]

[0133] Function: The impact characteristic eigenvalues can reflect the impact components in the vibration signal, and these components are usually closely related to abnormal states such as mechanical shock and looseness of the winding. By calculating the peak factor, impulse factor, and margin factor, the impact characteristics in the signal can be effectively identified, providing an important basis for the further assessment of the winding state.

[0134] 4. Calculate the distribution characteristic eigenvalues of the vibration signal.

[0135] Calculate the kurtosis K, skewness SK, and skewness factor CS of the vibration signal x respectively. The calculation formulas are as follows:

[0136]

[0137] Then the eigenvalue Dc of the three-dimensional distribution characteristics of the vibration signal x is:

[0138]

[0139] Function: The eigenvalue of the distribution characteristics can reflect the probability distribution characteristics of the vibration signal. Kurtosis and skewness are important parameters to describe the distribution form of the signal. By calculating these eigenvalues, the change of the signal distribution form can be effectively identified, providing an important supplement for the comprehensive evaluation of the winding state.

[0140] 5. Let the coordinates of the three-dimensional time-domain feature map of this section of the vibration signal be (Ec, Ic, Dc). Collect the vibration signals under different mechanical states of the winding (divided into normal mechanical state, attention mechanical state, and abnormal mechanical state), calculate the coordinates of the three-dimensional time-domain feature map of each section of the signal respectively, and construct the three-dimensional time-domain feature map atlas through the three-dimensional coordinate system based on the coordinates of the three-dimensional time-domain feature map of each section of the signal;

[0141] Taking the three-dimensional feature map coordinates (Ec, Ic, Dc) as the input and the corresponding mechanical state as the output, train and test the three-dimensional SVM model to obtain the transformer winding state evaluation model; this model can realize the online evaluation and diagnosis of the mechanical state of the transformer winding.

[0142] Function: By constructing the three-dimensional time-domain feature map, the energy characteristics, impact characteristics, and distribution characteristics are integrated into a comprehensive feature atlas, which can effectively reflect the multi-dimensional characteristics of the vibration signal. Combined with the three-dimensional SVM model, high-precision classification and recognition of the winding state can be realized, improving the accuracy and reliability of the state evaluation.

[0143] When collecting and preprocessing the vibration signal of the present invention, preferably, the vibration sensor is installed on the surface of the transformer box and connected to the signal acquisition card. The signal acquisition card is connected to the processing unit for data transmission. The vibration sensor collects the vibration signal of the transformer winding in real time and converts it into an electrical signal. The signal acquisition card samples the signal at a sampling rate of T and converts the analog signal into a digital signal.

[0144] Filter the collected vibration signal through a band-pass filter (100Hz - 2000Hz) to remove high-frequency noise and low-frequency interference and retain the frequency components sensitive to the winding state.

[0145] In the present invention, the processing unit calculates the eigenvalue of the energy characteristic, the eigenvalue of the impact characteristic, and the eigenvalue of the distribution characteristic of the vibration signal, obtains the three-dimensional characteristic map coordinates (Ec, Ic, Dc), constructs a three-dimensional characteristic map atlas, and then stores it in the database of the processing unit.

[0146] Vibration signals under different mechanical states of the winding are collected, and the three-dimensional characteristic map coordinates (Ec, Ic, Dc) of each segment of the signal are calculated respectively; these coordinates are used to construct a three-dimensional characteristic map atlas in the time domain and used as training samples to train a three-dimensional support vector machine (SVM) model. The trained model is used to evaluate the mechanical state of the transformer winding in real time.

[0147] In the present invention, by constructing a three-dimensional characteristic map atlas in the time domain, the energy characteristic, the impact characteristic, and the distribution characteristic are integrated into a comprehensive characteristic atlas, which can effectively reflect the multi-dimensional characteristics of the vibration signal. Combined with the three-dimensional SVM model, it realizes high-precision classification and identification of the winding state, and improves the accuracy and reliability of the state evaluation.

[0148] Application Example

[0149] A certain 220 kV transformer has been in operation for more than 15 years and has suffered 2 near-area short-circuit impacts in the past 3 years. To detect the mechanical state of the transformer winding, vibration signals of the transformer are collected. High-precision vibration sensors are installed in the middle and bottom of the transformer shell facing the winding. The sampling frequency is set to 1000 times per second, and the vibration signals during the operation of the transformer are collected in real time. The collected vibration signals are filtered to remove high-frequency noise and low-frequency drift.

[0150] Then, the method of steps (2) to (6) of the present invention is used for evaluation. The vibration signals sampled every 5 minutes are used for evaluation once, and the evaluation results within one day are statistically analyzed, as Figure 2 shown. According to the evaluation results, it can be seen that there is an abnormal mechanical state in this winding, that is, there may be a serious deformation situation, and further inspection and maintenance are required.

[0151] Afterwards, the transformer winding is inspected and maintained, and it is found that there is indeed a serious deformation in a small part of the winding. That is, through the present invention, it can help power operation and maintenance personnel identify potential faults of the winding in advance and avoid unnecessary equipment shutdowns and repairs.

[0152] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map, characterized in that, It includes the following steps: Step (1), vibration signal acquisition and preprocessing: Online acquire the vibration signal of the transformer and then perform preprocessing; Step (2), calculate the energy characteristic eigenvalues of the vibration signal: Respectively calculate the mean square value E, variance D and root mean square value RMS of the vibration signal, and then calculate the three-dimensional energy characteristic eigenvalue Ec of the vibration signal x; Step (3), calculate the impact characteristic eigenvalues of the vibration signal: Respectively calculate the peak factor PF, pulse factor IF and margin factor MF of the vibration signal x, and then calculate the three-dimensional impact characteristic eigenvalue Ic of the vibration signal x; Step (4), calculate the distribution characteristic eigenvalues of the vibration signal: Respectively calculate the kurtosis K, skewness SK and skewness factor CS of the vibration signal x, and then calculate the three-dimensional distribution characteristic eigenvalue Dc of the vibration signal x; Step (5), model training: Acquire the vibration signals under different mechanical states of the windings, and the corresponding three-dimensional feature map coordinates (Ec, Ic, Dc). Use the three-dimensional feature map coordinates (Ec, Ic, Dc) as the input and the corresponding mechanical state as the output to train and test the three-dimensional SVM model to obtain the transformer winding state evaluation model; Step (6), transformer winding state evaluation: Real-time acquire the vibration signal of the transformer, calculate according to steps (2) to (4) to obtain the corresponding time-domain three-dimensional feature map coordinates (Ec, Ic, Dc), and then input them into the trained transformer winding state evaluation model to obtain the corresponding mechanical state of the transformer winding.

2. The method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map according to claim 1, characterized in that In step (1), the preprocessing is filtering.

3. The method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map according to claim 2, wherein In step (1), the specific method of filtering is to filter the acquired vibration signal through a band-pass filter of 100Hz - 2000Hz.

4. The method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map according to claim 1, wherein In step (2), calculate the mean square value E, variance D and root mean square value RMS of the vibration signal x, and their calculation formulas are: where T is the sampling period and x(t) is the vibration signal at time t.

5. The method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map according to claim 1, wherein In step (3), calculate the peak factor PF, pulse factor IF and margin factor MF of the vibration signal x, and their calculation formulas are: where T is the sampling period and x(t) is the vibration signal at time t.

6. The method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map according to claim 1, wherein In step (4), calculate the kurtosis K, skewness SK and skewness factor CS of the vibration signal x, and their calculation formulas are: where T is the sampling period and x(t) is the vibration signal at time t.

7. The method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map according to claim 1, characterized in that, In step (5), the mechanical states are divided into normal mechanical state, attention mechanical state and abnormal mechanical state.

8. The method for evaluating the state of a transformer winding based on a three-dimensional time-domain feature map according to claim 1, wherein In step (5), the sample ratio of the training set to the test set is 8:

2.

9. A transformer winding state evaluation system based on a three-dimensional time-domain feature map, which adopts the transformer winding state evaluation method based on a three-dimensional time-domain feature map described in any one of claims 1 to 8, characterized in that, It includes: A vibration signal acquisition and preprocessing module, which is used to online acquire the vibration signal of the transformer and then perform preprocessing; A module for calculating the energy characteristic eigenvalues of the vibration signal, which is used to respectively calculate the mean square value E, variance D and root mean square value RMS of the vibration signal, and then calculate the three-dimensional energy characteristic eigenvalue Ec of the vibration signal x; A module for calculating the impact characteristic eigenvalues of the vibration signal, which is used to respectively calculate the peak factor PF, pulse factor IF and margin factor MF of the vibration signal x, and then calculate the three-dimensional impact characteristic eigenvalue Ic of the vibration signal x; The module for calculating the distribution characteristic eigenvalues of the vibration signal is used to calculate the kurtosis K, skewness SK, and skewness factor CS of the vibration signal x respectively, and then calculate the three-dimensional distribution characteristic eigenvalue Dc of the vibration signal x; The model training module is used to collect the vibration signals under different mechanical states of the winding, the corresponding three-dimensional feature map coordinates (Ec, Ic, Dc), use the three-dimensional feature map coordinates (Ec, Ic, Dc) as the input, and the corresponding mechanical state as the output, train and test the three-dimensional SVM model to obtain the transformer winding state evaluation model; The transformer winding state evaluation module is used to obtain the vibration signal collected by the vibration signal acquisition and preprocessing module in real time for the transformer, and use the module for calculating the energy characteristic eigenvalues of the vibration signal, the module for calculating the impact characteristic eigenvalues of the vibration signal, and the module for calculating the distribution characteristic eigenvalues of the vibration signal to calculate, so as to obtain the corresponding time-domain three-dimensional feature map coordinates (Ec, Ic, Dc), and then input them into the trained transformer winding state evaluation model to obtain the corresponding mechanical state of the transformer winding.

Images