Method and system for identifying mechanical defects of GIS equipment by variable frequency vibration characteristic screening

Abstract

The application discloses a GIS equipment mechanical defect identification method and system based on variable-frequency vibration feature screening, and relates to the technical field of electricity.The technical scheme is as follows: original vibration data under variable-frequency excitation is acquired, and the original vibration data is divided into a training sample set and a test sample set; composite feature extraction is performed on the training sample set and the test sample set, a feature data set is constructed, significant feature screening is performed on the feature data sets under different excitation current frequencies, and the significant feature data sets under the excitation current frequencies are obtained; a GIS equipment mechanical defect identification model is constructed through significant training feature data set and ABDT algorithm, accuracy verification is performed through significant test feature data set, the GIS equipment mechanical defect identification model that passes the accuracy verification is obtained, and the GIS equipment mechanical defect identification model is used to realize GIS equipment mechanical defect identification; and the variable-frequency feature screening and the GIS equipment mechanical defect diagnosis model are combined to realize GIS equipment mechanical defect type identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electrical technology, more particularly, it relates to a GIS device mechanical defect identification method and system based on variable frequency vibration feature screening. BACKGROUND

[0002] Gas insulated metal enclosed switchgear (GIS) devices play a role in control, protection and measurement in power systems, and internal insulation and mechanical defects are key factors threatening the safe and stable operation of the devices. GIS device mechanical defects are very harmful, which can cause structural parts to loosen, induce local overheating and gas leakage, and even lead to insulation damage and breakdown in the long run. According to statistical data, internal mechanical defects of GIS devices are mostly generated during manufacturing, transportation and installation. Once the devices are put into operation, undetected internal mechanical defects will be difficult to disassemble and maintain due to the fully enclosed structure of the devices. Therefore, it is urgent to effectively find internal mechanical defects of GIS devices at the factory or handover test stage to ensure their stable operation.

[0003] Based on the mechanical vibration signals generated by variable frequency large current excitation of power equipment, the signals have more obvious nonlinear dynamic behavior and more abundant mechanical state information than traditional single power frequency excitation signals, which have attracted the attention of researchers, and are particularly suitable for pre-detecting mechanical defects of GIS devices at the factory or handover test stage. CN101738567A discloses a system and method for detecting the state of transformer windings by using a constant current sweep power supply to excite and detect the vibration response signal after applying a constant current sweep excitation signal to the high voltage side of the transformer, and uses the peak value comparison method of the resonance frequency curve to analyze the internal mechanical state; CN114047410A discloses a GIS / GIL device mechanical vibration simulation system and method, which provides a simulation and detection system for applying voltage load and injecting variable frequency current to the primary main circuit, and can effectively detect internal mechanical defects of GIS / GIL devices. However, the existing variable frequency current excitation method focuses on internal defect detection, and lacks specific methods for identifying internal mechanical defects of GIS devices based on variable frequency current excitation.

[0004] To overcome the above problems, the present application provides a GIS device mechanical defect identification method based on adaptive boosting decision tree and variable frequency vibration feature screening. SUMMARY

[0005] The purpose of the present application is to provide a GIS device mechanical defect identification method and system based on variable frequency vibration feature screening, which combines variable frequency feature screening and GIS device mechanical defect diagnosis model to realize mechanical defect type identification of GIS devices.

[0006] The technical purposes of the present application are achieved by the following technical solutions: the method comprises

[0007] S1, obtaining original vibration data under variable frequency excitation, the original vibration data comprising: vibration signals, defect types, excitation current frequencies and operating currents, and dividing a plurality of sets of the original vibration data into: a training sample set and a test sample set;

[0008] S2, performing composite feature extraction on the training sample set and the test sample set, constructing a feature data set, and performing significant feature screening on the feature data sets of different excitation current frequencies to obtain significant feature data sets under each excitation current frequency, the significant feature data sets comprising: a significant training feature data set and a significant test feature data set;

[0009] S3, constructing a GIS device mechanical defect recognition model through the significant training feature data set and an ABDT algorithm, verifying the accuracy of the GIS device mechanical defect recognition model through the significant test feature data set, and obtaining a GIS device mechanical defect recognition model that passes the accuracy verification, for realizing GIS device mechanical defect recognition.

[0010] By the above technical solutions, the original vibration data under variable frequency excitation is first obtained, and the training sample set and the test sample set are divided according to a certain proportion; the training sample set and the test sample set are subjected to composite feature extraction to construct the training feature data set and the test feature data set, and then the significant features of different excitation current frequencies are screened to construct the significant training feature data set and the significant test feature data set; then, the significant training feature data set is combined with the ABDT adaptive boosting decision tree algorithm to train the GIS device mechanical defect recognition model, and the model accuracy is verified based on the significant test feature data set to obtain an output GIS device mechanical defect recognition model for GIS device mechanical defect recognition.

[0011] Further, the step S2 comprises:

[0012] S21, performing composite feature extraction on the original vibration data in the training sample set and the test sample set, the features comprising: total harmonic distortion, total harmonic square sum, signal amplitude, frequency domain kurtosis value and barycenter frequency value, and constructing a feature data set of the original vibration data;

[0013] S22, dividing the feature data set of the original vibration data according to the excitation current frequencies to obtain the feature data sets under each excitation current frequency;

[0014] S23, performing significant feature screening on the feature data sets under each excitation current frequency to obtain the significant feature data sets under each excitation current frequency, the significant feature data sets comprising: a significant training feature data set and a significant test feature data set.

[0015] Further, the step S23 comprises: taking the total harmonic distortion, the total harmonic square sum, the barycenter frequency value, the maximum value matrix of the signal amplitude, and the minimum value matrix of the frequency domain kurtosis value in the feature data set at each excitation current frequency, to construct the significant feature data set at each excitation current frequency.

[0016] Further, the step S3 comprises:

[0017] S31, dividing the operating current into multiple intervals, taking the significant training feature data set of the operating current in the interval as input to the ADT algorithm, and training the GIS equipment mechanical defect recognition model of the interval;

[0018] S32, inputting the significant test feature data set of the operating current in the interval to the GIS equipment mechanical defect recognition model of the interval, and verifying the accuracy;

[0019] S33, taking the GIS equipment mechanical defect recognition model that passes the accuracy verification as output, to realize GIS equipment mechanical defect recognition.

[0020] Further, the step S31 comprises:

[0021] S311, dividing the operating current into intervals, and extracting the significant training feature data set group in the interval to form a training set;

[0022] S312, inputting any one group of significant training feature data set in the training set to the ADT algorithm, calculating the weight distribution of the weak learner, and repeating the step until all significant training feature data sets in the operating current interval are inputted;

[0023] S313, linearly weighting the multiple weak learners to obtain the GIS equipment mechanical defect recognition model of the interval, and repeating the above steps until the GIS equipment mechanical defect recognition models of all intervals are obtained.

[0024] In another aspect, the application provides a GIS equipment mechanical defect recognition system based on variable-frequency vibration feature screening, comprising:

[0025] An original vibration data acquisition module is configured to acquire original vibration data under variable-frequency excitation, wherein the original vibration data comprises vibration signals, defect types, excitation current frequencies, and operating currents, and a plurality of groups of the original vibration data are divided into a training sample set and a test sample set;

[0026] The significant feature dataset construction module is configured to perform composite feature extraction on the training sample set and the test sample set, construct a feature dataset, perform significant feature screening on the feature datasets of different excitation current frequencies, and obtain significant feature datasets under different excitation current frequencies, wherein the significant feature datasets include a significant training feature dataset and a significant test feature dataset.

[0027] The GIS device mechanical defect identification model construction module is configured to construct a GIS device mechanical defect identification model by using the significant training feature dataset and the ABDT algorithm, verify the accuracy of the GIS device mechanical defect identification model by using the significant test feature dataset, and obtain a GIS device mechanical defect identification model that passes the accuracy verification, so as to realize GIS device mechanical defect identification.

[0028] Further, the significant feature dataset construction module further includes:

[0029] The composite feature extraction module is configured to perform composite feature extraction on the original vibration data in the training sample set and the test sample set, and the features include total harmonic distortion, total harmonic square sum, signal amplitude, frequency domain kurtosis value, and barycenter frequency value, so as to construct a feature dataset of the original vibration data.

[0030] The feature dataset division module is configured to divide the feature dataset of the original vibration data according to the excitation current frequencies, and obtain feature datasets under different excitation current frequencies.

[0031] The significant feature screening module is configured to perform significant feature screening on the feature datasets under different excitation current frequencies, and obtain significant feature datasets under different excitation current frequencies, wherein the significant feature datasets include a significant training feature dataset and a significant test feature dataset.

[0032] Further, the significant feature screening module is further configured to: in the feature datasets under different excitation current frequencies, take the maximum value matrix of the total harmonic distortion, the total harmonic square sum, the barycenter frequency value, the signal amplitude, and the minimum value matrix of the frequency domain kurtosis value, and construct the significant feature datasets under different excitation current frequencies.

[0033] Further, the GIS device mechanical defect identification model construction module further includes:

[0034] The model construction module divides the operating current into multiple intervals, inputs the significant training feature dataset of the operating current in the interval into the ABDT algorithm, and trains a GIS device mechanical defect identification model of the interval.

[0035] The model verification module is configured to input the significant test feature data set of the operating current in the interval into a GIS equipment mechanical defect identification model of the interval to perform accuracy verification.

[0036] The model output module is configured to output the GIS equipment mechanical defect identification model that passes the accuracy verification to implement GIS equipment mechanical defect identification.

[0037] Further, the model construction module comprises:

[0038] The interval division module is configured to divide the operating current into intervals, extract a significant training feature data set group of the operating current in the interval to form a training set.

[0039] The model training module is configured to input any one significant training feature data set in the training set into an ABDT algorithm to calculate the weight distribution of a weak learner, and repeat the step until all significant training feature data sets in the operating current interval are input.

[0040] The model combination module is configured to linearly weight a plurality of weak learners to obtain a GIS equipment mechanical defect identification model of the interval, and repeat the above steps until all GIS equipment mechanical defect identification models of the intervals are obtained.

[0041] Compared with the prior art, the present application has the following beneficial effects: the present application provides a GIS equipment mechanical defect identification method and system based on variable frequency vibration feature screening, and has three advantages compared with the existing invention patent:

[0042] 1. The GIS equipment mechanical defect diagnosis is carried out around the variable frequency excitation vibration signal, and the basic data information is more comprehensive compared with the single power frequency detection and diagnosis method;

[0043] 2. The composite feature extraction and significant feature screening algorithm is used to extract and fuse the vibration signals of different frequency current excitation, and the feature vector is more representative;

[0044] 3. The GIS equipment mechanical defect diagnosis model of different current intervals constructed by using the adaptive boosting decision tree (ABDT) algorithm is proposed, the algorithm combines a plurality of weak classifiers into a strong classifier for GIS equipment mechanical defects, and the diagnosis effect is more accurate. The model is suitable for mechanical defect diagnosis and analysis of GIS equipment in the factory, handover test and operation maintenance stages, can realize effective identification and exclusion of defects, and prevent further deterioration of defects and occurrence of equipment accidents. BRIEF DESCRIPTION OF DRAWINGS

[0045] The accompanying drawings, which are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:

[0046] Figure 1 A flowchart of a method provided by an embodiment of the application;

[0047] Figure 2 A schematic diagram of a data set relationship provided by an embodiment of the application;

[0048] Figure 3 A schematic diagram of GIS equipment mechanical defect identification model construction provided by an embodiment of the application;

[0049] Figure 4 A schematic diagram of the structure of a system provided by an embodiment of the application;

[0050] Figure 5 A method verification flowchart provided by an embodiment of the application;

[0051] Figure 6 A raw vibration data graph provided by an embodiment of the application;

[0052] Figure 7 A vibration signal graph of different defect types in different running current information intervals provided by an embodiment of the application;

[0053] Figure 8 Diagnostic data of a GIS equipment defect identification model provided by an embodiment of the application. DETAILED DESCRIPTION

[0054] Hereinafter, the term "include" or "may include" used in various embodiments of the present application indicates existence of the applied function, operation, or element, and does not limit one or more functions, operations, or elements to be added. Also, as used in various embodiments of the present application, the term "include", "have", and their conjugates merely indicate that specific features, numbers, steps, operations, elements, components, or combinations thereof are present, and do not exclude the existence or addition of one or more other features, numbers, steps, operations, elements, components, or combinations thereof.

[0055] In various embodiments of the present application, the expression "or" or "at least one of B or / and C" includes any combination of the listed terms or all combinations thereof. For example, the expression "B or C" or "at least one of B or / and C" can include B, can include C, or can include both B and C.

[0056] The terminology used in the various embodiments of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the various embodiments of the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of the present application belong. The terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0057] For the purposes of the present application, the technical solutions and advantages will be more clearly apparent from the following further detailed description of embodiments of the application, given as a mere indication and without limitation to the same, in conjunction with the accompanying drawings.

[0058] Embodiment 1

[0059] The embodiment provides a GIS equipment mechanical defect identification method based on variable frequency vibration feature screening, combines variable frequency feature screening and a GIS equipment mechanical defect diagnosis model, realizes defect type identification, improves identification accuracy, is suitable for carrying out mechanical defect diagnosis analysis on GIS equipment at the stages of factory delivery, handover test and operation maintenance, and can realize effective identification and exclusion of defects, prevent further degradation of defects and occurrence of equipment accidents.

[0060] Referring to Figure 1 as shown, Figure 1 The method is a flowchart, and the method comprises the following steps:

[0061] S1, acquiring original vibration data under variable frequency excitation, wherein the original vibration data comprises a vibration signal, a defect type, an excitation current frequency and an operating current; and a plurality of groups of the original vibration data are divided into a training sample set and a test sample set;

[0062] S2, performing complex feature extraction on the training sample set and the test sample set, constructing a feature data set, and performing significant feature screening on the feature data sets of different excitation current frequencies to obtain significant feature data sets under each excitation current frequency, wherein the significant feature data sets comprise a significant training feature data set and a significant test feature data set;

[0063] S3, constructing a GIS equipment mechanical defect identification model through the significant training feature data set and an ABDT algorithm, verifying the accuracy of the GIS equipment mechanical defect identification model through the significant test feature data set, obtaining a GIS equipment mechanical defect identification model that passes accuracy verification, and using the GIS equipment mechanical defect identification model to realize GIS equipment mechanical defect identification.

[0064] Specifically, see Figure 2 As shown, this method first acquires the original vibration data S of the variable frequency current excitation. The original vibration data S is a vibration signal containing excitation current frequency information, defect type information, and operating current information. The original vibration data S is then divided into a training sample set S according to a certain ratio. TR and test sample set S TE After partitioning, the training sample set S TR and test sample set S TE Perform composite feature extraction and construct a training feature dataset C. TR and test feature dataset C TE The features include: total harmonic distortion rate, total harmonic sum of squares, centroid frequency, signal amplitude, and frequency domain kurtosis; then, from the training feature dataset C... TR and test feature dataset C TE Significant features of different excitation current frequencies were selected from the data to construct a significant training feature dataset S. RM and significant test feature dataset S RE Then, through the significant training feature dataset S RM A combined ABDT adaptive boosting decision tree algorithm is used to train a GIS equipment mechanical defect identification model, and the model is then based on the salient test feature dataset S. RE The accuracy of the model was verified to obtain the final mechanical defect identification model for GIS equipment, which was then used to identify mechanical defects in GIS equipment.

[0065] In step S1, based on the original vibration data, training sample sets S are constructed, for example, at a ratio of 2:1. TR ={D TRi ,Y TRi ,F TRi ,I TRi i = 1, 2, 3, N TR} and test sample set S TE ={D TEi ,Y TEi ,F TEi ,I TEi i = 1, 2, 3, N TE}. Among them, {D TRi ,Y TRi ,F TRi ,I TRi} and {D TEi ,Y TEi ,F TEi ,I TEi} represents the vibration signal, defect type, excitation current frequency, and operating current of the training and test sample sets, respectively, N. TR and N TEThe number of training and test samples, respectively.

[0066] Since the GIS device has different inherent frequencies and vibration frequencies, the response amplitudes of the GIS device vibration signals under different excitation current frequencies and the defect types are also different, and at the same time, the distortion characteristics of different vibration signals have great differences, therefore, the step S2 is used to extract the composite features, and the feature extraction and quantization from different angles are beneficial to further grasp the defect types of the GIS device. The step S2 comprises:

[0067] S21, composite feature extraction is performed on the original vibration data in the training sample set and the test sample set, the features include: total harmonic distortion rate, total harmonic square sum, signal amplitude, frequency domain kurtosis value and barycenter frequency value, and a feature data set of the original vibration data is constructed;

[0068] S22, the feature data set of the original vibration data is divided according to the excitation current frequency, and a feature data set under each excitation current frequency is obtained;

[0069] S23, significant feature screening is performed on the feature data set under each excitation current frequency, and a significant feature data set under each excitation current frequency is obtained, the significant feature data set includes: a significant training feature data set and a significant test feature data set.

[0070] Specifically, in the step S21, composite feature extraction is performed on the original vibration data in the training sample set and the test sample set, the total harmonic distortion rate THD, the total harmonic square sum E f , the signal amplitude A m , the frequency domain kurtosis value K U and the barycenter frequency value F G of the original vibration data are extracted, and a feature data set of the original vibration data is constructed, including a training feature data set C TR and a test feature data set C TE ; the calculation formula of each feature is as follows:

[0071] A, the total harmonic distortion rate THD: the ratio of the square root value of the harmonic content in the periodic flow of the vibration signal to the square root value of the fundamental component.

[0072]

[0073] wherein G T0 is the energy sum of the 50Hz and 100Hz spectral amplitudes, G TH is the energy of the multiple frequency (i.e. A i×100 ) of 100Hz and the odd multiple frequency (i.e. A j×50 ) of 50Hz, and the observation stationary range is recommended to be greater than 3500Hz, therefore H is greater than or equal to 35.

[0074] B, signal amplitude A m : the half of the average value of the difference between the maximum instantaneous amplitude and the minimum instantaneous amplitude of the time domain vibration signal in a certain time period.

[0075]

[0076] wherein A m is the average value of the peak-to-peak value A ppi of the time domain vibration sequence in N observed fundamental period T0, the calculation formula is as formula (2). Wherein A ppi represents the value range of i is 1~N, the value of T0 is 0.01s, and the value of N is recommended to be greater than 100.

[0077] C, total harmonic square sum E f : the square sum of the amplitude of the harmonic component in the frequency spectrum of the vibration signal.

[0078]

[0079] wherein A(k) is the amplitude of x(k) at the kth harmonic frequency point, generally k>200.

[0080] D, frequency domain kurtosis value K U : the parameter value describing the pulse and attenuation characteristics of the frequency spectrum of the vibration signal.

[0081]

[0082] wherein E(x) and σ(x) represent the expectation and standard deviation calculated for x(k) respectively.

[0083] E, center of gravity frequency value F G : the frequency value corresponding to the center of gravity of the area under the frequency domain spectrum curve of the vibration signal.

[0084]

[0085] wherein f(k) is the frequency value corresponding to the kth frequency domain signal point, N is the length of the frequency domain signal sequence, and A(k) is the amplitude of x(k) at the kth frequency point.

[0086] In step S22, the characteristic data set of the original vibration data is divided according to different excitation current frequencies (such as 20Hz, 30Hz, 40Hz, 50Hz, 60Hz, 70Hz, 80Hz, 90Hz, 100Hz), and the characteristic data set under each excitation current frequency is obtained, for example, the characteristic data set under 20Hz in the training sample set and the characteristic data set under 20Hz in the test sample set.

[0087] In step S23, the significant feature set of each excitation current frequency is screened, and the total harmonic distortion rate, the total harmonic square sum, the barycenter frequency value, the maximum value matrix of signal amplitude, and the minimum value matrix of frequency domain kurtosis value at the excitation current frequency are taken as the significant feature data set at the excitation current frequency, including: the significant training feature data set S RM RMi RMi RMi and the significant test feature data set S RE EMi EMi EMi , wherein {x RMi , y RMi , I RMi} and {x EMi , y EMi , I EMi} represent the vibration signal, the defect type, and the operating current in the significant training feature data set and the significant test feature data set.

[0088] After the significant training feature data set and the significant test feature data set are constructed, step S3 is entered to construct the GIS device mechanical defect recognition model through the significant training feature data set and the ABDT algorithm. Step S3 includes:

[0089] S31, the operating current is divided into multiple intervals, the significant training feature data set of the operating current in the interval is input into the ABDT algorithm, and the GIS device mechanical defect recognition model of the interval is trained;

[0090] S32, the significant test feature data set of the operating current in the interval is input into the GIS device mechanical defect recognition model of the interval, and the accuracy is verified;

[0091] S33, the output of the GIS device mechanical defect recognition model that passes the accuracy verification is taken to realize the GIS device mechanical defect recognition.

[0092] Among them, step S31 includes:

[0093] S311, the operating current is divided into intervals, and the significant training feature data set of the operating current in the interval is extracted to form a training set;

[0094] S312, any one group of significant training feature data set in the training set is input into the ABDT algorithm, the weight distribution of the weak learner is calculated, and the step is repeated until all the significant training feature data sets in the operating current interval are input;

[0095] ​​​​​​S313, linearly weighting the plurality of weak learners to obtain a GIS device mechanical defect identification model of the interval, and repeating the above steps until GIS device mechanical defect identification models of all intervals are obtained.

[0096] Specifically, the ABDT algorithm is a strong classifier formed by reasonably combining a plurality of weak classifiers, and the weak classifier is a single-layer classification and regression tree (CART). Figure 3 As shown in FIG. 6, the main steps are as follows:

[0097] 1) Take the significant training feature data set of a certain operating current interval to form a training set S = {(x1, y1), (x2, y2), …, (x n ,y n )}, wherein x n represents a vibration signal, y n represents a defect type, (x n ,y n ) represents a sample in the training set, and the weight distribution of the samples in the training set is initialized, and each sample has the same weight W0 = {ω (1) ,ω (2) ,…,ω (b)};

[0098] 2) In the process of iteratively training the weak classifier using the training set S = {(x1, y1), (x2, y2), …, (x n ,y n )}, if the sample is classified correctly, its weight W i = {ω i(1) ,ω i(2) ,…,ω i(b)} will be reduced in the training set for constructing the next weak learner, wherein b is the number of training samples; otherwise, the weight W j = {ω j(1) ,ω j(2) ,…,ω j(b)} of the misclassified sample will be constantly increased, so that the classifier is gradually improved. The result of the previous weak learner is used to update the training set weight of the next weak learner each time, until all the classifiers are trained. The strong learner Φ k (x) of the kth round is:

[0099]

[0100] wherein α iThe linear weighting coefficients of the i-th weak classifier can be obtained using the forward distribution learning algorithm. The exponential loss function L(,) for n samples S is:

[0101]

[0102] The weight ω of the i-th sample in the k-th weak classifier k,i =exp[-y i Φ k-1 (x i The solution formula for the weak classifier is as follows: It can be represented as:

[0103]

[0104] Then classifier Classification error rate e k The solution formula and weight ω k,i The update formulas are as follows:

[0105]

[0106] ω k+1,i =ω k,i exp[-y i α k φ k (x i (10)

[0107] 3) After the training process of each weak classifier is completed, all weak classifiers are... The final strong classifier Φ(x) is formed by linear weighted combination, and the weights of weak classifiers with small classification error rates are increased, while the weights of weak classifiers with large classification error rates are decreased, as shown in Equation (10), where θ i Let be the weight of the i-th weak classifier.

[0108]

[0109] Finally, the accuracy of the GIS equipment mechanical defect identification model was verified using a significant test feature dataset. The significant test feature dataset for the operating current range was input into the corresponding GIS equipment mechanical defect identification model to perform defect type diagnosis and verify the model's accuracy.

[0110] The method collects vibration signals of different defect types, different excitation current frequencies and different operating currents in step S1 to carry out GIS equipment mechanical defect diagnosis, and the basic data information is more comprehensive compared with the single power frequency detection and diagnosis method; in step S2, the composite features are extracted, and the composite features are subjected to significant feature screening, the vibration signals of different excitation current frequencies are subjected to feature extraction and fusion, and the feature vector is more representative; in step S3, the ABDT adaptive boosting decision tree algorithm is combined to construct a GIS equipment mechanical defect diagnosis model in different current intervals, the algorithm combines a plurality of weak classifiers into a strong classifier for GIS equipment mechanical defects, and the diagnosis effect is more accurate.

[0111] Embodiment 2

[0112] The embodiment provides a GIS equipment mechanical defect identification system for variable frequency vibration feature screening, which is used to realize the above-mentioned GIS equipment mechanical defect identification method for variable frequency vibration feature screening.

[0113] Referring to Figure 4 As shown in the figure, the system comprises:

[0114] An original vibration data acquisition module is configured to acquire original vibration data under variable frequency excitation, wherein the original vibration data comprises vibration signals, defect types, excitation current frequencies and operating currents, and a plurality of groups of the original vibration data are divided into a training sample set and a test sample set;

[0115] A significant feature dataset construction module is configured to extract composite features from the training sample set and the test sample set, construct a feature dataset, and screen significant features from the feature dataset under different excitation current frequencies to obtain significant feature datasets under different excitation current frequencies, wherein the significant feature datasets comprise a significant training feature dataset and a significant test feature dataset;

[0116] A GIS equipment mechanical defect identification model construction module is configured to construct a GIS equipment mechanical defect identification model by using the significant training feature dataset and an ABDT algorithm, verify the accuracy of the GIS equipment mechanical defect identification model by using the significant test feature dataset, obtain a GIS equipment mechanical defect identification model that passes the accuracy verification, and implement GIS equipment mechanical defect identification.

[0117] Further, the significant feature dataset construction module further comprises:

[0118] A composite feature extraction module is configured to extract composite features from the original vibration data in the training sample set and the test sample set, wherein the features include total harmonic distortion, total harmonic square sum, signal amplitude, frequency domain kurtosis value and barycenter frequency value, and a feature dataset of the original vibration data is constructed;

[0119] a feature dataset division module configured to divide feature datasets of the original vibration data according to the excitation current frequency, to obtain feature datasets under each excitation current frequency;

[0120] a significant feature screening module configured to screen significant feature datasets under each excitation current frequency from the feature datasets under each excitation current frequency, to obtain significant feature datasets under each excitation current frequency, wherein the significant feature datasets include significant training feature datasets and significant test feature datasets.

[0121] Further, the significant feature screening module is further configured to: in the feature datasets under each excitation current frequency, take total harmonic distortion, total harmonic square sum, barycenter frequency value, maximum value matrix of signal amplitude, and matrix of minimum value of kurtosis value in the frequency domain, to construct the significant feature datasets under each excitation current frequency.

[0122] Further, the GIS device mechanical defect identification model construction module further includes:

[0123] a model construction module configured to divide the operating current into multiple intervals, and input the significant training feature datasets of the operating current in the interval into the ADT algorithm to train the GIS device mechanical defect identification model of the interval;

[0124] a model verification module configured to input the significant test feature datasets of the operating current in the interval into the GIS device mechanical defect identification model of the interval to perform accuracy verification;

[0125] a model output module configured to output the GIS device mechanical defect identification model that passes the accuracy verification, to implement GIS device mechanical defect identification.

[0126] Further, the model construction module includes:

[0127] an interval division module configured to divide the operating current into intervals, and extract the significant training feature dataset of the operating current in the interval to form a training set;

[0128] a model training module configured to input any one group of significant training feature datasets in the training set into the ADT algorithm to calculate the weight distribution of the weak learner, and repeat the step until all the significant training feature datasets in the operating current interval are inputted;

[0129] a model combination module configured to linearly weight multiple weak learners to obtain the GIS device mechanical defect identification model of the interval, and repeat the above steps until the GIS device mechanical defect identification models of all intervals are obtained.

[0130] Embodiment 3

[0131] This embodiment takes 550 kV real GIS equipment as an example, adopts variable frequency large current generating device, 550 kV real GIS equipment, vibration detection device to build a primary main loop, and simulates three kinds of defect types of normal, busbar guide rod base loosening (defect 1), disconnector guide rod stroke not in place (defect 2). The verification embodiment 1 proposes a variable frequency vibration characteristic screening GIS equipment mechanical defect identification method, and the specific verification process is as shown in Figure 5 .

[0132] Firstly, the original vibration data of variable frequency excitation is obtained: the excitation current in the range of 500A-5000A with 500A as the gradient is applied, and the excitation current frequency is 20Hz, 40Hz, 50Hz, 60Hz, 80Hz and 100Hz respectively. Further, the mechanical vibration detection of vibration signals of different operating loads and different defect types is carried out, and the vibration signal graphs under different excitation current frequencies, different defect types and different operating currents are obtained. Referring to Figure 6 , Figure 6 (a)-(f) are time domain and frequency domain vibration signal graphs of three defect types under 3000A excitation current.

[0133] Further, the training sample set and the test sample set are constructed according to the ratio of 2:1, the original vibration data in the training sample set is extracted for features, including: total harmonic distortion THD, total harmonic square sum E f , signal amplitude A m and frequency domain kurtosis value K U , the feature data set of the training sample is constructed, and the feature data set is further divided into feature data sets under each excitation current frequency according to the excitation current frequency, and significant feature screening is carried out to obtain the significant feature data set under each excitation current frequency. Figure 7 is the three-dimensional visualization distribution graph of the vibration signals of the GIS equipment of three defect types in different operating current intervals obtained by significant feature screening. It can be seen that the discrimination degree of different defect types is more obvious when the current is larger, and the boundary of the feature data is also more obvious.

[0134] Then, the significant training feature data set is combined with the ABDT algorithm to train the GIS equipment mechanical defect identification model in different operating current intervals, and the significant test feature data set in the operating current interval is used to verify the trained diagnosis model.

[0135] Figure 7The diagnostic data of the GIS device mechanical defect identification model provided by the method for variable frequency current excitation and the GIS device defect identification model for single power frequency excitation are respectively provided. Among them, the identification accuracy of the GIS device mechanical defect identification model provided by the method for the three defect types of normal state, bus conductor base loosening and disconnector guide rod stroke not in place in the small current interval is 0.96, 0.981 and 0.995 respectively, the overall identification accuracy is 0.975, and the identification accuracy in the large current interval is 0.995, 0.986 and 1 respectively, and the overall identification accuracy is 0.994, which realizes accurate diagnosis of different defect types, and the diagnosis accuracy is higher than that of the GIS device defect identification model based on single power frequency excitation, verifying the effectiveness of the method.

[0136] The above specific embodiments further illustrate the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. GIS equipment mechanical defect identification method of variable frequency vibration characteristic screening, characterized in that: Comprising ​ S1, obtaining original vibration data of variable frequency excitation, the original vibration data comprising: vibration signal, defect type, excitation current frequency and operating current, and dividing a plurality of sets of the original vibration data into: a training sample set and a test sample set; S2, performing composite feature extraction on the training sample set and the test sample set, constructing a feature data set, and performing significant feature screening on the feature data set of different excitation current frequencies to obtain a significant feature data set under each excitation current frequency, the significant feature data set comprising: a significant training feature data set and a significant test feature data set; wherein the step S2 comprises: S21, performing composite feature extraction on the original vibration data in the training sample set and the test sample set, the features comprising: total harmonic distortion rate, total harmonic square sum, signal amplitude, frequency domain peak value and barycenter frequency value, and constructing a feature data set of the original vibration data; S22, dividing the feature data set of the original vibration data according to the excitation current frequency to obtain a feature data set under each excitation current frequency; S23, performing significant feature screening on the feature data set under each excitation current frequency to obtain a significant feature data set under each excitation current frequency, the significant feature data set comprising: a significant training feature data set and a significant test feature data set; S3, constructing a GIS device mechanical defect recognition model through the significant training feature data set and the ABDT algorithm, verifying the accuracy of the GIS device mechanical defect recognition model through the significant test feature data set, obtaining a GIS device mechanical defect recognition model that passes the accuracy verification, and using the model to realize GIS device mechanical defect recognition; wherein the step S3 comprises: S31, dividing the operating current into multiple intervals, inputting the significant training feature data set of the operating current in the interval into the ABDT algorithm, and training a GIS device mechanical defect recognition model for the interval; S32, inputting the significant test feature data set of the operating current in the interval into the GIS device mechanical defect recognition model for the interval, and performing accuracy verification; S33, outputting the GIS device mechanical defect recognition model that passes the accuracy verification, and using the model to realize GIS device mechanical defect recognition; wherein the step S31 comprises: S311, dividing the operating current into intervals, and extracting the significant training feature data set of the operating current in the interval to form a training set; S312, inputting any one group of significant training feature data set in the training set into the ABDT algorithm to calculate the weight distribution of the weak learner, and repeating the step until all significant training feature data sets in the operating current interval are inputted; S313, linearly weighting a plurality of weak learners to obtain a GIS device mechanical defect recognition model for the interval, and repeating the above steps until all interval GIS device mechanical defect recognition models are obtained.

2. The method of claim 1, wherein the method further comprises: The step S23 comprises: in the feature data set under each excitation current frequency, taking the maximum matrix of the total harmonic distortion rate, the total harmonic square sum, the barycenter frequency value, the signal amplitude, and the minimum matrix of the frequency domain peak value to construct a significant feature data set under each excitation current frequency.

3. GIS equipment mechanical defect identification system of variable frequency vibration characteristic screening, characterized in that: Comprising An original vibration data acquisition module for acquiring original vibration data of variable frequency excitation, the original vibration data including vibration signals, defect types, excitation current frequencies and operating currents, and dividing a plurality of sets of the original vibration data into a training sample set and a test sample set; A significant feature data set construction module for performing composite feature extraction on the training sample set and the test sample set, constructing a feature data set, and performing significant feature screening on the feature data sets of different excitation current frequencies to obtain significant feature data sets under each excitation current frequency, the significant feature data sets including a significant training feature data set and a significant test feature data set; wherein the significant feature data set construction module further includes a composite feature extraction module for performing composite feature extraction on the original vibration data in the training sample set and the test sample set, the features including total harmonic distortion, total harmonic square sum, signal amplitude, frequency domain peak value and barycenter frequency value, and constructing a feature data set of the original vibration data; a feature data set division module for dividing the feature data set of the original vibration data according to the excitation current frequencies to obtain the feature data sets under each excitation current frequency; and a significant feature screening module for performing significant feature screening on the feature data sets under each excitation current frequency to obtain the significant feature data sets under each excitation current frequency, the significant feature data sets including a significant training feature data set and a significant test feature data set; The GIS device mechanical defect recognition model construction module is configured to construct a GIS device mechanical defect recognition model by using the significant training feature data set and the ABDT algorithm, to verify the accuracy of the GIS device mechanical defect recognition model by using the significant test feature data set, and to obtain a GIS device mechanical defect recognition model that passes the accuracy verification, thereby realizing GIS device mechanical defect recognition.

4. The variable frequency vibration signature screening GIS equipment mechanical defect identification system of claim 3, wherein: The model construction module is configured to divide the operating current into multiple intervals, to input the significant training feature data set of the operating current in the interval into the ABDT algorithm, and to train a GIS device mechanical defect recognition model for the interval; the model verification module is configured to input the significant test feature data set of the operating current in the interval into the GIS device mechanical defect recognition model for the interval, and to perform accuracy verification; and the model output module is configured to output the GIS device mechanical defect recognition model that passes the accuracy verification, thereby realizing GIS device mechanical defect recognition. The model construction module includes an interval division module configured to divide the operating current into intervals, to extract the significant training feature data set of the operating current in the interval to form a training set, a model training module configured to input any one group of significant training feature data set in the training set into the ABDT algorithm, to calculate the weight distribution of the weak learner, and to repeat the step until all significant training feature data sets in the operating current interval are inputted, and a model combination module configured to linearly weight multiple weak learners to obtain the GIS device mechanical defect recognition model for the interval, and to repeat the above steps until the GIS device mechanical defect recognition models for all intervals are obtained. The significant feature screening module is further configured to: in the feature data set at each excitation current frequency, take the total harmonic distortion rate, the total harmonic square sum, the barycenter frequency value, the maximum value matrix of the signal amplitude, and the minimum value matrix of the frequency domain peak value to construct the significant feature data set at each excitation current frequency.

Images