CN-122017486-A - Inter-turn arc severity light-weight judging system and method based on multiple physical quantities

CN122017486ACN 122017486 ACN122017486 ACN 122017486ACN-122017486-A

Abstract

A turn-to-turn arc severity light-weight judging system and method based on multiple physical quantities. The system comprises a multi-source signal acquisition and windowing module, a full feature vector construction module, a feature normalization and dimension reduction module, a model training module and a lightweight discrimination module, wherein the multi-source signal acquisition and windowing module is used for acquiring multi-source physical signals of multi-turn windings and dividing the multi-source physical signals into a plurality of sliding time windows, the full feature vector construction module is used for extracting multi-dimensional feature parameters from each physical signal aiming at each sliding time window and splicing the multi-dimensional feature parameters into full feature vectors, the feature normalization and dimension reduction module is used for processing the full feature vectors to obtain multi-physical-quantity low-dimensional feature subsets, the model training module is used for constructing and training a support vector machine discrimination model based on the multi-physical-quantity low-dimensional feature subsets, and the lightweight discrimination module is used for achieving lightweight discrimination of the severity of arc discharge of the multi-source physical signals to be determined. The system has higher identification accuracy and stable discrimination results.

Inventors

DONG MING
Chang Haoxin
ZHANG CHONGXING
CHEN JI
HU YIZHUO
LI SHUHUA
ZENG QIANG
REN MING

Assignees

西安交通大学

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (10)

1. A multi-physical quantity-based inter-turn arc severity lightweight discrimination system, the system comprising: The multi-source signal acquisition and windowing module is used for acquiring multi-source physical signals of the multi-turn winding when arc discharge occurs in oil and dividing the multi-source physical signals into a plurality of sliding time windows; the full feature vector construction module is used for extracting multidimensional feature parameters for representing the arc discharge intensity and the evolution state from each physical signal aiming at each sliding time window, and splicing the multidimensional feature parameters into a full feature vector; the feature standardization and dimension reduction module is used for carrying out dimension-by-dimension standardization processing on the full feature vector, and carrying out feature screening and dimension reduction on the standardized feature vector to obtain a multi-physical-quantity low-dimension feature subset; The model training module is used for constructing and training a support vector machine discrimination model based on the multi-physical-quantity low-dimensional feature subsets; The light weight discriminating module is used for inputting the multisource physical signals to be judged, the multisource physical quantity low-dimensional feature subsets obtained after processing into a support vector machine discriminating model after training is completed, and light weight discriminating of the inter-turn arc severity is completed.
2. The system of claim 1, wherein preferably in the multi-source signal acquisition and windowing module, the number of turns of the multi-turn winding is at least 4 turns and the last turn is grounded.
3. The system of claim 1, wherein in the multi-source signal acquisition and windowing module, the multi-source physical signal is transmitted to the computer via an acquisition card having a sampling rate of 60-65 MHz.
4. The system of claim 1, wherein in the multi-source signal acquisition and windowing module, the multi-source physical signal comprises at least a voltage signal of a non-grounded winding, an arc loop current signal, an acoustic emission signal, and an uhf electromagnetic radiation signal.
5. The system of claim 1, wherein in the full feature vector construction module, the multi-dimensional feature parameters comprise maximum discharge amplitude, discharge amplitude standard deviation, average discharge amplitude, skewness, kurtosis, pulse repetition rate, spectral center frequency, spectral bandwidth, band energy duty cycle rate of change.
6. The system of claim 1, wherein in the feature normalization and dimension reduction module, the dimension-by-dimension normalization process comprises performing zero-mean and unit variance normalization process on each dimension feature parameter in the full feature vector, so that the mean value of each dimension feature on a sample set is 0 and the standard deviation is 1.
7. The system of claim 1, wherein in the model training module, the input of the support vector machine discriminant model is a data set formed by gathering feature vectors in a multi-physical-quantity low-dimensional feature subset corresponding to each sliding time window.
8. The utility model provides a turn-to-turn arc severity lightweight discrimination method based on multiple physical quantities, which is characterized in that the method comprises the following steps: Step 1, acquiring multi-source physical signals of a multi-turn winding during arc discharge in oil, and dividing the multi-source physical signals into a plurality of sliding time windows; Step 2, extracting multidimensional characteristic parameters for representing the intensity and evolution state of arc discharge from each physical signal aiming at each sliding time window, and splicing the multidimensional characteristic parameters into a full characteristic vector; step 3, carrying out dimension-by-dimension standardization processing on the full feature vector, and carrying out feature screening and dimension reduction on the standardized feature vector to obtain a multi-physical-quantity low-dimension feature subset; step 4, constructing and training a support vector machine discrimination model based on the multi-physical-quantity low-dimensional feature subset; And 5, inputting the multisource physical signals to be judged, which are processed in the steps 1-3, into a trained support vector machine judgment model to finish the light-weight judgment of the severity of the inter-turn arc.
9. The method according to claim 8, wherein the step 1 comprises: Step 1-1, acquiring multi-source physical signals of a multi-turn winding when arc discharge occurs in oil; Step 1-2, denoising the multi-source physical signals, and dividing each physical signal into a plurality of sliding time windows based on a uniform time reference.
10. A computer readable storage medium comprising a stored complete computer program, characterized in that the method according to claim 7 or 8 is implemented when the computer program is run.

Description

Inter-turn arc severity light-weight judging system and method based on multiple physical quantities Technical Field The invention belongs to the technical field of on-line monitoring of inter-turn arc faults of power equipment windings and intelligent judgment of fault degrees in a power system, and relates to a multi-physical-quantity-based inter-turn arc severity light judgment system and method. Background In the long-term operation process of oil paper insulating power equipment (such as an oil immersed transformer, a reactor and the like), winding turn-to-turn insulation is subjected to multi-factor coupling effects such as electric field stress, thermal stress, mechanical vibration, oil paper medium aging and the like, so that an insulating weak point is easy to appear and turn-to-turn discharge is induced. The discharge tends to exhibit weak-to-strong, partial-to-extended evolution characteristics, which may manifest as intermittent discharge or short-time arc at an early stage, then gradually transition to sustainable arc discharge, and may extend along inter-turn paths in multi-turn directions, eventually inducing a greater range of inter-turn breakdown or short-circuiting. Because the multi-turn arc discharge has the characteristics of concentrated energy release, high development speed, uncertain expansion path and the like, once entering a moderately severe stage, the insulation degradation of the oilpaper can be obviously accelerated and irreversible damage can be caused. Therefore, the severity of turn-to-turn arc discharge is classified and judged, and the method has important significance for operation and maintenance to take maintenance measures and early warning. In addition, in the prior art, aiming at monitoring and identifying the winding discharge faults, a discrimination method is constructed by relying on single physical quantity signals (such as voltage/current only, acoustic signals only and electromagnetic radiation signals only). However, arc discharge in oil is a typical multi-physical process, and sensitivity of different sensing links to discharge energy release, channel form change and medium response is different, namely, a single signal is easily influenced by factors such as noise, installation position, propagation attenuation, working condition fluctuation and the like, so that feature stability is insufficient, stage separability is not strong, and reliable discrimination performance is difficult to maintain under complex scenes such as multi-turn continuous breakdown and the like. Meanwhile, although the multi-physical-quantity combined monitoring is beneficial to improving the discrimination accuracy, the problems of high feature dimension, multiple redundant information, large model parameter quantity and the like are also brought, so that the algorithm is difficult to deploy in real time at a field terminal or an edge side. Particularly, in an online early warning scene, if the calculation cost of the discrimination model is too large, the warning timeliness and the engineering usability are affected. Therefore, how to construct a lightweight severity degree judging method with small model scale and still high judging accuracy by removing redundant information through feature screening and dimension reduction on the basis of multi-physical-quantity joint detection is a technical problem to be solved. Disclosure of Invention In order to overcome the problems, the invention provides a turn-to-turn arc severity light-weight judging system and method based on multiple physical quantities. In particular, it is an object of the present invention to provide the following aspects: in a first aspect, there is provided a multi-physical-quantity-based inter-turn arc severity lightweight discrimination system, the system comprising: The multi-source signal acquisition and windowing module is used for acquiring multi-source physical signals of the multi-turn winding when arc discharge occurs in oil and dividing the multi-source physical signals into a plurality of sliding time windows; the full feature vector construction module is used for extracting multidimensional feature parameters for representing the arc discharge intensity and the evolution state from each physical signal aiming at each sliding time window, and splicing the multidimensional feature parameters into a full feature vector; the feature standardization and dimension reduction module is used for carrying out dimension-by-dimension standardization processing on the full feature vector, and carrying out feature screening and dimension reduction on the standardized feature vector to obtain a multi-physical-quantity low-dimension feature subset; The model training module is used for constructing and training a support vector machine discrimination model based on the multi-physical-quantity low-dimensional feature subsets; the light weight judging module is used for inputting the multisource physical signals to be