CN109029699B - Transformer vibration online anomaly detection method - Google Patents

Transformer vibration online anomaly detection method Download PDF

Info

Publication number
CN109029699B
CN109029699B CN201810598845.3A CN201810598845A CN109029699B CN 109029699 B CN109029699 B CN 109029699B CN 201810598845 A CN201810598845 A CN 201810598845A CN 109029699 B CN109029699 B CN 109029699B
Authority
CN
China
Prior art keywords
sample
newly added
training
transformer
training sample
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810598845.3A
Other languages
Chinese (zh)
Other versions
CN109029699A (en
Inventor
徐卫
康驰
杜向京
钟斌
王元驰
王乃会
�金钟
罗剑
胡红
肖文章
宋加波
颜周锐
纪坤
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Leshan Power Supply Co Of State Grid Sichuan Electric Power Co
State Grid Corp of China SGCC
Original Assignee
Leshan Power Supply Co Of State Grid Sichuan Electric Power Co
State Grid Corp of China SGCC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Leshan Power Supply Co Of State Grid Sichuan Electric Power Co, State Grid Corp of China SGCC filed Critical Leshan Power Supply Co Of State Grid Sichuan Electric Power Co
Priority to CN201810598845.3A priority Critical patent/CN109029699B/en
Publication of CN109029699A publication Critical patent/CN109029699A/en
Application granted granted Critical
Publication of CN109029699B publication Critical patent/CN109029699B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The embodiment of the invention discloses a transformer vibration online anomaly detection method aiming at how to analyze the running state of a transformer and perform fault diagnosis based on a vibration method at present. The incremental learning algorithm is used for learning only the newly added data samples in the preset time period, so that the newly added data can be timely and effectively processed, the newly added training samples can be learned online in real time, the rapid upgrade of the detection model is realized, and the requirements of model updating on time and space can be reduced.

Description

Transformer vibration online anomaly detection method
Technical Field
The invention relates to the field of electric power, in particular to a transformer vibration online abnormity detection method.
Background
Power transformers are important devices in power systems, and their operating conditions have a significant impact on the safety and economic efficiency of the power system. The surface vibration of the transformer mainly originates from the vibration of a winding and an iron core under the excitation of current and voltage, and theoretical analysis and practical experience show that the working states of the winding and the iron core can be analyzed through the vibration signal of the transformer and fault diagnosis can be carried out. The vibration method is used for analyzing the running state and fault diagnosis of the transformer, and researchers at home and abroad do a great deal of research work and obtain a considerable result. The transformer vibration signal feature extraction is the premise and the basis for analyzing the transformer operation state and fault diagnosis by a vibration method. However, most of data of the existing research results are from laboratory or experimental transformer simulation, and because the surface vibration of the transformer is complexly influenced by multiple factors, the difference between the surface vibration signal of the transformer in actual operation and the surface vibration signal of the transformer obtained under theoretical analysis and laboratory test conditions is obvious.
In actual production, the transformer vibrates to generate stream data, and the stream data contains new sample knowledge, so how to timely and effectively process the stream data is one of the difficulties in detecting the abnormality of the transformer.
Disclosure of Invention
In order to solve the technical problems, the invention provides an online abnormality detection method for transformer vibration, which comprises the steps of obtaining a newly added data sample related to a vibration signal of a transformer operation state in a preset time period, extracting characteristic parameters of the newly added data sample based on wavelet packet analysis, finishing learning of the newly added data sample based on a rapid convex hull algorithm, training and updating a single-class abnormality detector model, calling the updated single-class abnormality detector model, and performing online abnormality detection on the current transformer vibration. The incremental learning algorithm is used for learning only the newly added data samples in the preset time period, so that the newly added data can be timely and effectively processed, the newly added training samples can be learned online in real time, the rapid upgrade of the detection model is realized, and the requirements of model updating on time and space can be reduced.
The embodiment of the invention provides the following technical scheme:
a transformer vibration online abnormity detection method comprises the following steps:
acquiring a newly added data sample of a vibration signal related to the running state of the transformer in a preset time period;
extracting characteristic parameters of the newly added data samples based on wavelet packet analysis;
learning of newly added data samples is completed based on a fast convex hull algorithm, and a single-class anomaly detector model is trained and updated;
and calling the updated single-type anomaly detector model to perform online anomaly detection on the current transformer vibration.
Wherein, accomplish the study of newly-increased data sample based on quick convex hull algorithm, train the single abnormal detector model of update, specifically include:
let initial training sample A0The newly added sample is B ═ B1,B2,…,BnAre multiplied by
Figure BDA0001692560630000021
New training sample Ai(i∈n)，
1) Initial training sample A0Training to obtain a support vector data description model omega0Support vector is SV0；
2) Adding a new sample Bi(i ∈ n), find BiSamples in violation of the KKT condition are scored as
Figure BDA0001692560630000031
If it is
Figure BDA0001692560630000032
Then returns to omegai-1(ii) a Otherwise calculate sample Ai-1Support vector removal SVi-1Latter sample Ai-1I.e. Ai'-1＝Ai-1-SVi-1And calculating A by using a fast convex hull algorithmi'-1Shell vector C ofi-1；
3) Will be provided with
Figure BDA0001692560630000033
Training as a new training sample to obtain a new support vector data description model omegaiI is i + 1; when i > n, the algorithm terminates, when i ≦ n goes to step 2).
Wherein the KKT condition is represented as follows:
Figure BDA0001692560630000034
wherein, αi(i ∈ n) is a Lagrange multiplier, v is used to balance the product of the hypersphere and the training error, called the balance parameter, R is the radius of the minimum hypersphere of the training sample as the sphere center, a is the radius of the minimum hypersphere of the training sample as the sphere center, z is the test sample, d is the distance between the minimum hypersphere of the training sample and the sphere center2＝||z-a||2。
Compared with the prior art, the technical scheme has the following advantages:
the method comprises the steps of obtaining a newly added data sample related to a vibration signal of a transformer in an operation state within a preset time period, extracting characteristic parameters of the newly added data sample based on wavelet packet analysis, finishing learning of the newly added data sample based on a rapid convex hull algorithm, training and updating a single-class anomaly detector model, calling the updated single-class anomaly detector model, and carrying out online anomaly detection on current transformer vibration. The incremental learning algorithm is used for learning only the newly added data samples in the preset time period, so that the newly added data can be timely and effectively processed, the newly added training samples can be learned online in real time, the rapid upgrade of the detection model is realized, and the requirements of model updating on time and space can be reduced.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a schematic diagram of a transformer vibration online anomaly detection method.
Detailed Description
As described in the background art, how to analyze the operation state of the transformer and perform fault diagnosis based on the vibration method is a technical problem that needs to be solved urgently by those skilled in the art.
In order to solve the technical problems, the invention provides an online abnormality detection method for transformer vibration, which comprises the steps of obtaining a newly added data sample related to a vibration signal of a transformer operation state in a preset time period, extracting characteristic parameters of the newly added data sample based on wavelet packet analysis, finishing learning of the newly added data sample based on a rapid convex hull algorithm, training and updating a single-class abnormality detector model, calling the updated single-class abnormality detector model, and performing online abnormality detection on the current transformer vibration. The incremental learning algorithm is used for learning only the newly added data samples in the preset time period, so that the newly added data can be timely and effectively processed, the newly added training samples can be learned online in real time, the rapid upgrade of the detection model is realized, and the requirements of model updating on time and space can be reduced.
Fig. 1 is a schematic diagram of a transformer vibration online anomaly detection method, which includes:
step 101: and acquiring a newly added data sample of the vibration signal of the running state of the transformer in a preset time period.
In the actual operation process of the transformer, the acquisition of vibration signal data samples related to the operation state of the transformer is a continuous accumulation process, vibration stream data is formed, and the temporary data samples are continuously updated by taking newly added data samples in a preset time period as temporary data samples. While training of the anomaly detector model will also involve a large sample learning problem. If the anomaly detector model is relearned on the basis of the original training samples and the new training samples for each update upgrade, a large amount of computing resources and time are consumed. Therefore, on the basis of keeping the previous Learning result, the Incremental Learning (Incremental Learning) algorithm can learn the knowledge of the newly added sample, the newly added training sample can be learned online in real time, the anomaly detector model can be rapidly upgraded, and the requirements of model updating on time and space can be reduced. Therefore, the incremental learning algorithm is used for learning only the newly added data samples in the preset time period.
It should be noted that the preset time period may be set according to actual needs, for example, the preset time period may be set to be one week, or may be set to be one month, or may be set to be one quarter, which is not limited in this application.
Step 102: and extracting the characteristic parameters of the newly added data sample based on wavelet packet analysis.
The wavelet packet analysis can decompose the original vibration signals of the transformer in the running state on different frequency bands, the local energy of the signals on each frequency band can reflect the time-varying characteristic of the frequency characteristic of the signals, and the method has good time-frequency positioning characteristics and self-adaptive capacity to the signals. The method and the device establish a vibration characteristic parameter extraction model based on a wavelet packet analysis technology, and extract vibration state characteristic parameters representing the running state of the transformer.
Step 103: and finishing learning of the newly added data sample based on a fast convex hull algorithm, and training and updating the single-class anomaly detector model.
According to the method, a Support Vector Data Description (SVDD) model is selected as a single-class anomaly detector, and online anomaly detection is carried out on the actually measured vibration data of the transformer.
The main idea of SVDD is to find a minimum hypersphere boundary in the kernel feature space, which should surround all training samples as much as possible, classify and describe the samples with the boundary, and the described boundary can be used to reflect the distribution characteristics of the training samples.
Suppose that the training sample X contains n target class samples, i.e., X ═ X1,x2…,xnAnd in the mapped high-dimensional space, finding the sphere center a and the radius R of the smallest hypersphere that can contain the training sample, and then the optimization problem can be expressed as:
Figure BDA0001692560630000061
where v is used to balance the hypersphere sum and the training error, called the balance parameter, and is typically 0.1; ξiIs a relaxation factor.
Using Lagrange multiplier method and introducing kernel function K (x)i,xj) To solve the non-linearity problem, the
Figure BDA0001692560630000062
The formula translates into a dual problem:
Figure BDA0001692560630000063
α thereini(i ∈ n) is the Lagrangian multiplier.
In the present application, a gaussian kernel function is selected:
Figure BDA0001692560630000064
where s is the gaussian bandwidth factor.
By passing
Figure BDA0001692560630000065
The center a and radius of the sphere can be obtainedR：
Figure BDA0001692560630000071
αiSamples > 0 are called support vectors SV.
Figure BDA0001692560630000072
Called boundary support vector BSV, xk∈BSV。
For test sample z, if
Figure BDA0001692560630000073
The test sample z is considered to fall within the hypersphere, the accepted sample is of the target class, otherwise, the accepted sample is of the heterogeneous class.
When all αiThe condition of Karush-Kuhn-Tucker (KKT) satisfying the objective function can be regarded as the original equation
Figure BDA0001692560630000074
A solution of (2).
The KKT condition may be represented as follows:
Figure BDA0001692560630000075
according to the method, a transformer historical data sample set is used as a training sample to construct the SVDD model. According to the formula
Figure BDA0001692560630000076
And detecting the newly added data sample, and judging whether the newly added data sample is an abnormal sample, thereby realizing abnormal detection.
In the following examples, gaussian kernels are used. Is composed of
Figure BDA0001692560630000081
The parameter s is known to be related to the relative position between the training samples, so the initial value of s can be set between the minimum and maximum values of the training sample spacing, followed by an intersectionThe fork verification method determines an optimal value.
As known from the theoretical introduction of SVDD, the SVDD model is searched for a sample of support vectors that are determinative of classification during the training process. In the process of training the samples, samples near class boundaries are likely to become support vectors, so non-boundary samples can be eliminated, and the scale of the training sample set is further reduced.
For the change of the support vector set after the new data sample is added, the following theorem is provided:
theorem 1, if a sample meeting the KKT condition exists in the newly added samples, the part of samples certainly does not have a new support vector; if there is a sample that violates the KKT condition, then there must be a new support vector in the sample that violates the KKT condition.
Theorem 2 if there is a sample violating the KKT condition in the newly added sample, the non-support vector in the original sample may be converted into the support vector.
From theorem 1, the new support vector information is provided in the sample which violates the KKT condition in the newly added sample, so that whether the newly added sample violates the KKT condition is screened, and the newly added samples participating in training are reduced. From theorem 2, the non-support vector in the original sample can be converted into a new support vector after the incremental learning.
These possible transitions to new support vector samples are typically located at sample boundaries. The shell vector is all samples located at class boundaries in the training sample, i.e., the shell vector is the training sample convex vertex. The fast convex hull algorithm (Quickhull algorithm) deletes points inside a convex hull step by screening points in a training sample, so that a hull vector is calculated quickly. Therefore, in order to reduce the size of the training sample set while not losing the support vector information, the shell vector of the training sample except the support vector is calculated by using the Quickhull algorithm to ensure the integrity of the support vector information.
This application is when studying newly-increased data sample, when the increment study promptly, and the constitution of new training sample is former support vector, the shell vector of former non-support vector sample and violate the sample of KKT condition in the newly-increased sample.
The application learns newly-added data samples based on a fast convex hull algorithm, and the specific process of updating the anomaly detector model is as follows:
initial training sample A0The newly added sample is B ═ B1,B2,…,BnAre multiplied by
Figure BDA0001692560630000091
New training sample Ai(i∈n)。
1) Initial training sample A0Training to obtain a support vector data description model omega0Support vector is SV0；
2) Adding a new sample Bi(i ∈ n), find BiSamples in violation of the KKT condition are marked as Bif, if
Figure BDA0001692560630000092
Then returns to omegai-1(ii) a Otherwise calculate sample Ai-1Support vector removal SVi-1Latter sample Ai-1I.e. Ai'-1＝Ai-1-SVi-1And calculating A by using Quickhull algorithmi'-1Shell vector C ofi-1；
3) Will be provided with
Figure BDA0001692560630000093
Training as a new training sample to obtain a new support vector data description model omegaiAnd obtaining a new sphere center and a new radius:
Figure BDA0001692560630000094
i is i + 1; when i > n, the algorithm terminates, when i ≦ n goes to step 2).
In the process of establishing the SVDD model, a standard quadratic programming problem needs to be solved, and the complexity of the algorithm is O ((n + m)3) Wherein n is the number of original training samples, and m is the newly added sample. Increment by application
And (3) calculating the time complexity of the training sample m of which the newly added sample m violates the KKT condition to be O (m multiplied by l multiplied by k), wherein l is the number of the original support vectors, and k is the dimensionality of the training set. From the theory of computational geometry, the shell direction
The quantity n is far less than the quantity of the original training samples, and the time complexity of the application is O (m × l × k) + O ((m '+ n')3). Because m ', n', l and k are far smaller than n, and m is smaller than (n + m), the training efficiency can be effectively improved by the algorithm.
Step 104: and calling the updated single-type anomaly detector model to perform online anomaly detection on the current transformer vibration.
That is, the updated single-class anomaly detector model yields new sphere centers and sum radii:
Figure BDA0001692560630000101
then, for the data z to be tested, if the formula is satisfied
Figure BDA0001692560630000102
The test data z is considered to fall within the hypersphere and the sample is accepted as the target class, otherwise it is heterogeneous.
In summary, the method for detecting the online abnormality of the transformer vibration includes the steps of obtaining a newly added data sample related to a vibration signal of the transformer in an operation state within a preset time period, extracting characteristic parameters of the newly added data sample based on wavelet packet analysis, completing learning of the newly added data sample based on a fast convex hull algorithm, training and updating a single-class abnormality detector model, calling the updated abnormality detector model, and performing online abnormality detection on the current transformer vibration. The incremental learning algorithm is used for learning only the newly added data samples in the preset time period, so that the newly added data can be timely and effectively processed, the newly added training samples can be learned online in real time, the rapid upgrade of the detection model is realized, and the requirements of model updating on time and space can be reduced.
In the description, each part is described in a progressive manner, each part is emphasized to be different from other parts, and the same and similar parts among the parts are referred to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (1)

1. A transformer vibration online abnormity detection method is characterized by comprising the following steps:
acquiring a newly added data sample of a vibration signal related to the running state of the transformer in a preset time period;
extracting characteristic parameters of the newly added data samples based on wavelet packet analysis;
learning of newly added data samples is completed based on a fast convex hull algorithm, and a single-class anomaly detector model is trained and updated;
calling the updated single-type anomaly detector model to perform online anomaly detection on the current transformer vibration;
wherein, accomplish the study of newly-increased data sample based on quick convex hull algorithm, train the single abnormal detector model of update, specifically include:
let initial training sample A0The newly added sample is B ═ B1,B2,…,BnAre multiplied by
Figure FDA0002581959670000011
New training sample Ai,i∈n，
1) Initial training sample A0Training to obtain a support vector data description model omega0Support vector is SV0；
2) Adding a new sample BiI ∈ n, find BiSamples in violation of the KKT condition are scored as
Figure FDA0002581959670000012
If it is
Figure FDA0002581959670000013
Then returns to omegai-1(ii) a Otherwise calculate sample Ai-1Support vector removal SVi-1Latter sample Ai′-1I.e. A'i-1＝Ai-1-SVi-1And calculating A 'by using a fast convex hull algorithm'i-1Shell vector C ofi-1；
3) Will be provided with
Figure FDA0002581959670000014
Training as a new training sample to obtain a new support vector data description model omegaiI is i + 1; when i is larger than n, the algorithm is terminated, and when i is smaller than or equal to n, the step 2) is carried out;
wherein the KKT condition is represented as follows:
Figure FDA0002581959670000021
Figure FDA0002581959670000022
Figure FDA0002581959670000023
wherein, αiI ∈ n is a Lagrange multiplier, v is used to balance the hypersphere sum and the training error, called the balance parameter, R is the radius of the minimum hypersphere of the training sample as the sphere center, a is the radius of the minimum hypersphere of the training sample as the sphere center, z is the test sample, d is the lagrange multiplier, v is the minimum hypersphere of the training sample as the sphere center, b is the minimum hypersphere of the training sample as the sphere center, d is2＝||z-a||2。
CN201810598845.3A 2018-06-12 2018-06-12 Transformer vibration online anomaly detection method Active CN109029699B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810598845.3A CN109029699B (en) 2018-06-12 2018-06-12 Transformer vibration online anomaly detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810598845.3A CN109029699B (en) 2018-06-12 2018-06-12 Transformer vibration online anomaly detection method

Publications (2)

Publication Number Publication Date
CN109029699A CN109029699A (en) 2018-12-18
CN109029699B true CN109029699B (en) 2020-09-11

Family

ID=64612631

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810598845.3A Active CN109029699B (en) 2018-06-12 2018-06-12 Transformer vibration online anomaly detection method

Country Status (1)

Country Link
CN (1) CN109029699B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110749793A (en) * 2019-10-31 2020-02-04 杭州中恒云能源互联网技术有限公司 Dry-type transformer health management method and system based on width learning and storage medium
CN112257737A (en) * 2020-06-29 2021-01-22 北京沃东天骏信息技术有限公司 Abnormal commodity object identification method, equipment and storage medium
CN113670428B (en) * 2021-07-19 2023-12-22 国网四川省电力公司电力科学研究院 Transformer vibration online abnormality detection method
CN115905974B (en) * 2022-11-04 2024-02-27 北京科技大学 Method for detecting abnormal furnace condition of blast furnace

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103294716A (en) * 2012-02-29 2013-09-11 佳能株式会社 On-line semi-supervised learning method and device for classifier, and processing equipment
CN103645249A (en) * 2013-11-27 2014-03-19 国网黑龙江省电力有限公司 Online fault detection method for reduced set-based downsampling unbalance SVM (Support Vector Machine) transformer
CN107144430A (en) * 2017-06-27 2017-09-08 电子科技大学 A kind of Method for Bearing Fault Diagnosis based on incremental learning
WO2018010028A1 (en) * 2016-07-14 2018-01-18 HYDRO-QUéBEC Detection of anomalies in an electrical network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103294716A (en) * 2012-02-29 2013-09-11 佳能株式会社 On-line semi-supervised learning method and device for classifier, and processing equipment
CN103645249A (en) * 2013-11-27 2014-03-19 国网黑龙江省电力有限公司 Online fault detection method for reduced set-based downsampling unbalance SVM (Support Vector Machine) transformer
WO2018010028A1 (en) * 2016-07-14 2018-01-18 HYDRO-QUéBEC Detection of anomalies in an electrical network
CN107144430A (en) * 2017-06-27 2017-09-08 电子科技大学 A kind of Method for Bearing Fault Diagnosis based on incremental learning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
基于支持向量数据描述的变压器故障检测方法;刘福荣等;《黑龙江电力》;20131215(第06期);第474-477页 *
支持向量数据描述法在变压器故障预警中的应用;陈利等;《西南民族大学学报(自然科学版)》;20080615(第03期);第568-571页 *

Also Published As

Publication number Publication date
CN109029699A (en) 2018-12-18

Similar Documents

Publication Publication Date Title
CN109029699B (en) Transformer vibration online anomaly detection method
Li et al. A dual weighting label assignment scheme for object detection
CN108073158A (en) Based on PCA and KNN density algorithm Wind turbines Method for Bearing Fault Diagnosis
CN112508442B (en) Transient stability assessment method and system based on automatic and interpretable machine learning
WO2018009733A1 (en) Computer system and method for monitoring key performance indicators (kpis) online using time series pattern model
CN109165160A (en) Software defect prediction model design method based on core principle component analysis algorithm
WO2021004198A1 (en) Plate performance prediction method and apparatus
JP2007140694A (en) Device and method for analyzing relevance of operation and quality in manufacturing process, computer program and computer-readable recording medium
CN114239404A (en) Intelligent material optimization design method based on multi-objective optimization
Badrinarayanan et al. Symmetry-invariant optimization in deep networks
Sun et al. Application of time series based SVM model on next-day electricity price forecasting under deregulated power market
Ye et al. A multi-model approach for soft sensor development based on feature extraction using weighted kernel fisher criterion
He et al. Research on hybrid ARIMA and support vector machine model in short term load forecasting
CN111444963A (en) Blast furnace molten iron silicon content prediction method based on SSA-SVR model
CN115775231A (en) Cascade R-CNN-based hardware defect detection method and system
WO2022174436A1 (en) Incremental learning implementation method and apparatus for classification model, and electronic device and medium
CN115617666A (en) GPT2 model-based Chinese test case completion method
CN115577290A (en) Distribution network fault classification and source positioning method based on deep learning
Liu et al. Facial expression recognition approach based on least squares support vector machine with improved particle swarm optimization algorithm
CN107506461A (en) A kind of method of intellectual analysis and risk assessment on video display drama
El Azami et al. Robust outlier detection with L0-SVDD.
CN113239562A (en) Method for predicting mechanical property of whole plate surface of strip steel
CN105825205A (en) Cooperative sparse representation self-adaptive rapid face recognition method
CN111950652A (en) Semi-supervised learning data classification algorithm based on similarity
CN115687899B (en) Hybrid feature selection method based on high-dimensional spinning data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant