CN111428772A - Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting - Google Patents

Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting Download PDF

Info

Publication number
CN111428772A
CN111428772A CN202010195167.3A CN202010195167A CN111428772A CN 111428772 A CN111428772 A CN 111428772A CN 202010195167 A CN202010195167 A CN 202010195167A CN 111428772 A CN111428772 A CN 111428772A
Authority
CN
China
Prior art keywords
data
anomaly detection
voting
dimensional
adaptive
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.)
Granted
Application number
CN202010195167.3A
Other languages
Chinese (zh)
Other versions
CN111428772B (en
Inventor
张洁
张志昊
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.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
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 Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN202010195167.3A priority Critical patent/CN111428772B/en
Publication of CN111428772A publication Critical patent/CN111428772A/en
Application granted granted Critical
Publication of CN111428772B publication Critical patent/CN111428772B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Photovoltaic Devices (AREA)

Abstract

The invention provides a photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting, which comprises the steps of obtaining one-dimensional time sequence data inside a photovoltaic array accumulated in a photovoltaic power generation system, utilizing a sliding window to cut the original data at a fixed length, converting the original data into a two-dimensional frequency domain data image through fast Fourier transform, classifying external data by combining with the slicing time, and giving a label to construct a data set; and training a classification model on the data set by using a deep neural network, and extracting implicit characteristics as anomaly detection input. The method comprises the steps of carrying out data conversion on input data, extracting hidden layer coding features, firstly selecting voting points in a local hidden layer coding feature set by using double k-neighbors based on distance degrees, and obtaining voting weights for determining whether the voting points are abnormal or not through distance difference self-adaption between the voting points and a second k-neighbor point set, so that the method is robust and improves detection accuracy. The method does not need to compare the real data at the next moment, and improves the real-time detection efficiency.

Description

Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting
Technical Field
The invention relates to a photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting, and belongs to the technical field of photovoltaic systems.
Background
The photovoltaic power generation system is a complex system influenced by multiple natural factors such as temperature, irradiance and haze, and mainly comprises a grid-connected inverter, a junction box, a photovoltaic array and the like. As a dc power generation unit, a photovoltaic array plays an essential role in the normal operation of the entire photovoltaic power generation system. When the abnormity detection and fault location of related equipment in a photovoltaic system are researched, on one hand, various natural factors are considered, and on the other hand, various types of fault factors inside the photovoltaic power generation system are also considered.
The existing photovoltaic array anomaly detection and early warning method is characterized in that data generated by a photovoltaic array is used as input, various models are built, an anomaly rule evaluation method is set, abnormal parts in the data are analyzed, and anomaly detection of the photovoltaic array is realized.
Disclosure of Invention
The invention aims to solve the technical problem of providing a photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting, which overcomes the defects of the prior art.
The invention provides a photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting, which comprises the following steps of:
s1, preprocessing data, converting the one-dimensional time sequence data of the photovoltaic array into two-dimensional frequency domain data, generating a label by using external data, and constructing a labeled data set;
s2, taking the two-dimensional frequency domain data as the input of a deep neural network, and implicitly extracting features;
and S3, deep anomaly detection, namely, realizing real-time anomaly detection by improving an original k-nearest neighbor algorithm and utilizing a self-adaptive weight voting method.
The method of the invention designs a complete process from data preprocessing to real-time anomaly detection by combining and improving the deep learning and the k-nearest neighbor algorithm. The method comprises the steps of obtaining one-dimensional time sequence data inside a photovoltaic array accumulated in a photovoltaic power generation system, cutting original data at a fixed length by using a sliding window, converting the original data into a two-dimensional frequency domain data image through fast Fourier transform, classifying recorded external data such as temperature and the like by combining with cutting time, and giving a label construction data set; and training a classification model on the constructed data set by using a deep neural network, and extracting implicit characteristics as anomaly detection input. The method comprises the steps of carrying out data conversion on input data to be detected, extracting hidden layer coding features, firstly selecting voting points in a local hidden layer coding feature set by using double k-neighbors based on distance degrees, and adaptively determining whether the voting weights are abnormal or not through the distance difference between the voting points and a second k-neighbor point set, so that the deep anomaly detection method is more robust, and the detection accuracy is improved. In addition, the method does not need to compare the real data at the next moment, and improves the real-time detection efficiency.
The further technical scheme of the invention is as follows:
preferably, in step S1, the data preprocessing includes data transformation and data set construction; the process of the deep anomaly detection comprises feature extraction, voting detection and adaptive weight calculation.
Preferably, the specific method of the data conversion is as follows:
the data to be converted is internal one-dimensional time sequence data in the photovoltaic power generation system, the data conversion is carried out by slicing the one-dimensional time sequence data by using a sliding window, partial overlapping is allowed among the slices, and then the multi-dimensional characteristic vector is extracted through fast Fourier transform to obtain a two-dimensional frequency domain characteristic diagram.
Preferably, the constructed data set comprises two-dimensional frequency domain data and a label, wherein the two-dimensional frequency domain data is obtained by preprocessing the internal one-dimensional time sequence data; the label category is constructed by different numerical combinations of external data, namely, the two-dimensional data is obtained by processing the slices by data conversion, the external data corresponding to the time is found, and the external data is classified into one of the label categories.
Preferably, in the step S2, the feature extraction is to use a deep neural network in computer vision to train on the data set constructed above to extract implicit features of the data inside the photovoltaic array as input data for anomaly detection.
Preferably, the specific method of voting detection is as follows:
after one-dimensional time sequence data are converted into two-dimensional frequency domain images through data preprocessing, m points closest to test data in a local data set are selected as voting points and corresponding distances d are calculatediThen calculating the nearest n points in each voting point and calculating the average distance D, and finally comparing DiAnd D size to determine if an anomaly has occurred.
Preferably, the method of adaptive weight calculation is as follows:
by using diThe distance difference of the sum D calculates an interpretable dynamic coefficient as a weight when the final anomaly detection sum votes.
Compared with the prior art, the invention adopting the technical scheme has the following technical effects:
according to the invention, two-dimensional frequency domain data conversion based on sliding window segmentation is realized for accumulated one-dimensional time sequence historical data in a photovoltaic power generation system, characteristics are implicitly extracted from two-dimensional input data by using a deep neural network, and the abnormal condition of a photovoltaic array is detected in real time based on a k-nearest neighbor self-adaptive voting method, so that the abnormal condition or the fault is timely and accurately responded, and the abnormal or fault post-processing is accelerated.
Detailed Description
The technical scheme of the invention is further explained in detail as follows: the present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the protection authority of the present invention is not limited to the following embodiments.
The embodiment provides a photovoltaic system depth anomaly detection method based on k-neighbor adaptive voting, which comprises two parts, namely a data preprocessing part and a depth anomaly detection part.
The first section introduces a photovoltaic data pre-processing procedure. The process comprises the steps of converting one-dimensional time sequence data of the photovoltaic array into two-dimensional frequency domain data, generating a label by utilizing external data, and constructing a labeled data set.
The flow includes data transformation and data set construction. The data conversion is carried out by slicing internal one-dimensional time sequence data in the photovoltaic power generation system by using a sliding window, wherein partial overlapping is allowed between slices, and then a two-dimensional frequency domain characteristic diagram is obtained by extracting multi-dimensional characteristic vectors through fast Fourier transform. The constructed data set comprises two-dimensional frequency domain data and a label, wherein the two-dimensional frequency domain data is obtained by preprocessing internal one-dimensional time sequence data; the category of the label is constructed by different numerical combinations of external data; and processing the slices by using data conversion to obtain two-dimensional data, finding external data corresponding to time of the two-dimensional data, and classifying the external data into a certain category in the label categories.
The internal data of the photovoltaic array comprises time domain one-dimensional continuous data such as current, voltage, power and the like, and the external data comprises quantitative data such as temperature, irradiance and the like. The internal and external data together form the state condition of the photovoltaic array expressed by each parameter value under a certain natural condition at a certain moment. Since there is a linear relationship between the internal data, the method uses the current in the internal data as the data in the dataset to be processed, and the temperature, irradiance, etc. as the label in the dataset to be processed. Converting the one-dimensional current in the time domain into two-dimensional data in the frequency domain, and realizing the data preprocessing and constructing a data set by the following processes:
s101, filtering internal and external data in normal operation from the original data, and combining the external data such as temperature, irradiance and the like into C categories through different numerical values;
step S102, segmenting continuous current into sub-current data with equal time by using a sliding window, and calling the sub-current data into frames;
step S103, performing left-right end continuity processing on each frame f, and setting N as the frame size and a as the continuity processing coefficient, then
Figure RE-GDA0002528393540000051
F (n, a) is a function, the function F (n, a) is to take a part of continuous data in the whole one-dimensional linear data, similar to the function of the window function, n is a coefficient, and the value range of n is (0, n);
step S104, by using fast Fourier transform DFT (DFT is an algorithm for converting one-dimensional time domain data into two-dimensional frequency spectrum by using a computer to realize Fast Fourier Transform (FFT)), one-dimensional current data f extracts multi-dimensional vector features to obtain required two-dimensional data m, wherein f is input data, m is output data, the conversion process is realized by the following formula,
Figure RE-GDA0002528393540000052
in the above formula, Xa(k) Representing the final output data (i.e. the above-mentioned m-value), x (n) representing the sampled input data (i.e. the above-mentioned f-value), e-j2πk/NIs a standard formula of DFT for data conversion;
step S105, classifying the external data corresponding to the one-dimensional current data f into a certain class C in the C classes of the structureiCombining them into a tagged data tuple (m, c)i) The data set X required for the second part is constructed.
The second part introduces the deep anomaly detection of the photovoltaic power generation system based on hidden layer coding, and the anomaly detection application principle realized by the method is described in detail by combining pseudo codes.
And (4) taking the two-dimensional frequency domain data as the input of the deep neural network, and implicitly extracting the features. Feature extraction is the use ofAnd the deep neural network in computer vision is trained on the constructed data set to extract implicit characteristics of data in the photovoltaic array as input data for anomaly detection. After feature extraction, real-time deep anomaly detection is realized by improving an original k-nearest neighbor algorithm and utilizing a self-adaptive weight voting method. Thus, the anomaly detection process includes feature extraction, voting detection processes, and adaptive weight calculation. When voting detection is carried out, m points closest to test data in a local data set are selected as voting points, and corresponding distances d are calculatediThen calculating the nearest n points in each voting point and calculating the average distance D, and finally comparing DiAnd D size to determine if an anomaly has occurred. The adaptive weight calculation is based on diThe distance difference of the sum D calculates an interpretable dynamic coefficient as a weight when the final anomaly detection sum votes.
In addition, the relevant pseudo code is as follows:
parameters are as follows: coefficients m, n of the twofold kNN;
inputting: testing a sample X, a constructed labeled data set X and a model f;
and (3) outputting: x is abnormal.
The method comprises the following specific steps:
1. training a model f with high classification accuracy based on the tagged data set X;
2. initialization
Figure RE-GDA0002528393540000061
3. Firstly, f (x) is calculated to obtain a hidden layer code z;
4. find m nearest z point sets A (a) in the local sample set X1,a2,...,am)；
5.FOR aiIN A：
6. Calculating z to aiDistance d ofi；
7. Finding n distances a in a local sample set XiNearest set of points B (B)1,b2,...,bn)；
8. Calculating aiAnd a set of points B (B)1,b2,...,bn) The average distance D of;
9. calculating the weight w of the normal samplei＝-exp(-|D-di|)+1；
10.IF di<D，THEN：
11. Receiving the sample, pi＝wi，P←<ai:pi>；
12.ELSE：
13. Reject the sample, ni＝wi，N←<ai:ni>；
Sum of weights in IF N set
Figure RE-GDA0002528393540000071
Greater than the sum of weights in the P set
Figure RE-GDA0002528393540000072
THEN：
15. The test sample x is abnormal;
16.ELSE：
17. no abnormality occurred in the test sample x.
In the above steps, P, N is a collection for voting, which indicates normal or abnormal; (x) is an output value of the training model in the first step, x is two-dimensional input data similar to a picture, and a vector is calculated and output through the model f, namely z is f (x); piThe ith ticket is judged to be normal and is assigned with weight, and the ith ticket is put into a P (Positive) set; n isiThe ith ticket is judged to be abnormal and the weight is assigned, and the ith ticket is put into a set of N (negative). For example, there are (0.1, 0.5, 1) in P and (0.2, 0.5, 0.8) in N, where P isi，niIs the voting of the set with weight, the voting means that the sum of P is 1.6, the sum of N is 1.5, and the voting result is 1.6>1.5, so the sample is judged to be P, namely a normal sample.
The photovoltaic power generation system anomaly detection method provided by the embodiment is specifically explained by combining the pseudo code as follows:
step S201, training a classification model f based on a constructed data set X in a deep neural network image training mode to serve as a method for implicitly extracting hidden layer coding features, and extracting all hidden layer codes from data in the data set X to obtain a feature set Z;
step S301, in an anomaly detection stage, firstly, input one-dimensional current data x are subjected to data preprocessing of a first part to complete one-dimensional to two-dimensional data conversion, and a hidden layer code z is obtained through a classification model f;
step S302, m point sets with the shortest distance to the test data hidden layer code Z are searched in the hidden layer code set Z extracted from the data set and are used as m points A (a) to be voted1,a2,...,am)；
Step S303, traversing the m points in the step S302, and finding out n point sets B (B) with the nearest distance from each point1,b2,...,bn) Then calculate z to aiDistance d ofiAnd aiAverage distance D to B;
step S304, using diDistance difference calculation weight w of sum DiI.e. wi＝-exp(-|D-di|)) + 1. Then compare diAnd the size of D, if Di<D, the test data is considered to be normal, the sample is accepted, pi＝wi，P←<ai:pi>(ii) a Otherwise, the test data is considered abnormal, the sample is rejected, ni＝wi，N←<ai:ni>；
Step S305 obtains m weighted inference results through steps S303 and S304, votes whether the point is an abnormal point or not, and determines whether the point is abnormal or not. Namely, judging the weight sum in the N sets
Figure RE-GDA0002528393540000081
Whether is greater than the sum of weights in the P set
Figure RE-GDA0002528393540000082
If yes, the test sample x is abnormal, otherwise, the test sample x is abnormalNo abnormality occurred in this x.
In summary, the weight calculation utilizes z to aiDistance d ofi、aiThe distance difference between the average distances D to B is larger, and the confidence that the sample point is determined to be normal/abnormal is larger when the distance difference is larger, and the weight of the sample point in voting is also larger. The weight calculation formula is specifically as follows:
wi＝-exp(-|D-di|)+1
through the content, the adaptive voting depth anomaly detection method based on the k-nearest neighbor is realized. Compared with the method for judging the abnormality by comparing whether the classification result output by the model inference is the same as the real result at the time t +1 or not, the method provided by the invention does not need to compare the data at the time t +1 on one hand, and improves the response speed of the model abnormality detection; on the other hand, hidden layer codes with higher original data information retention degree are applied to anomaly detection of the photovoltaic power generation system, and the accuracy of the anomaly detection of the model is improved.
The above description is only an embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can understand that the modifications or substitutions within the technical scope of the present invention are included in the scope of the present invention, and therefore, the scope of the present invention should be subject to the protection scope of the claims.

Claims (7)

1. The photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting is characterized by comprising the following steps of:
s1, preprocessing data, converting the one-dimensional time sequence data of the photovoltaic array into two-dimensional frequency domain data, generating a label by using external data, and constructing a labeled data set;
s2, taking the two-dimensional frequency domain data as the input of a deep neural network, and implicitly extracting features;
and S3, deep anomaly detection, namely, realizing real-time anomaly detection by improving an original k-nearest neighbor algorithm and utilizing a self-adaptive weight voting method.
2. The photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting according to claim 1, wherein in step S1, the data is preprocessed, and the process includes data conversion and data set construction; the process of the deep anomaly detection comprises feature extraction, voting detection and adaptive weight calculation.
3. The photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting according to claim 2, wherein the specific method of data conversion is as follows:
the data to be converted is internal one-dimensional time sequence data in the photovoltaic power generation system, the data conversion is carried out by slicing the one-dimensional time sequence data through a sliding window, and then a two-dimensional frequency domain characteristic diagram is obtained by extracting multi-dimensional characteristic vectors through fast Fourier transform.
4. The photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting is characterized in that a constructed data set comprises two-dimensional frequency domain data and a label, wherein the two-dimensional frequency domain data are obtained by preprocessing one-dimensional time sequence data; the category of the tag is constructed by different combinations of values of the external data.
5. The photovoltaic system deep anomaly detection method based on k-nearest neighbor adaptive voting is characterized in that in step S2, a deep neural network in computer vision is utilized to train on a constructed data set so as to extract implicit characteristics of data inside a photovoltaic array as input data for anomaly detection.
6. The photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting according to claim 5, wherein the voting detection method is as follows:
after one-dimensional time sequence data are converted into two-dimensional frequency domain images through data preprocessing, m points closest to test data in a local data set are selected as projectionTicket points and calculating the corresponding distance diThen calculating the nearest n points in each voting point and calculating the average distance D, and finally comparing DiAnd D size to determine if an anomaly has occurred.
7. The photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting according to claim 6, wherein the adaptive weight calculation method comprises the following steps:
by using diThe distance difference of the sum D calculates an interpretable dynamic coefficient as a weight when the final anomaly detection sum votes.
CN202010195167.3A 2020-03-19 2020-03-19 Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting Active CN111428772B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010195167.3A CN111428772B (en) 2020-03-19 2020-03-19 Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010195167.3A CN111428772B (en) 2020-03-19 2020-03-19 Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting

Publications (2)

Publication Number Publication Date
CN111428772A true CN111428772A (en) 2020-07-17
CN111428772B CN111428772B (en) 2022-08-23

Family

ID=71548328

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010195167.3A Active CN111428772B (en) 2020-03-19 2020-03-19 Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting

Country Status (1)

Country Link
CN (1) CN111428772B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113780417A (en) * 2021-09-10 2021-12-10 泰康保险集团股份有限公司 Medical insurance data detection method and device
CN117272211A (en) * 2023-11-20 2023-12-22 北京邮电大学 Lightweight spacecraft fault detection classification method based on long-short-term memory model
CN118040910A (en) * 2024-04-12 2024-05-14 国网山东省电力公司菏泽供电公司 Online monitoring method for energy storage abnormal state of micro-grid in remote area

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109858509A (en) * 2018-11-05 2019-06-07 杭州电子科技大学 Based on multilayer stochastic neural net single classifier method for detecting abnormality
CN110119755A (en) * 2019-03-22 2019-08-13 国网浙江省电力有限公司信息通信分公司 Electricity method for detecting abnormality based on Ensemble learning model

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109858509A (en) * 2018-11-05 2019-06-07 杭州电子科技大学 Based on multilayer stochastic neural net single classifier method for detecting abnormality
CN110119755A (en) * 2019-03-22 2019-08-13 国网浙江省电力有限公司信息通信分公司 Electricity method for detecting abnormality based on Ensemble learning model

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113780417A (en) * 2021-09-10 2021-12-10 泰康保险集团股份有限公司 Medical insurance data detection method and device
CN117272211A (en) * 2023-11-20 2023-12-22 北京邮电大学 Lightweight spacecraft fault detection classification method based on long-short-term memory model
CN117272211B (en) * 2023-11-20 2024-02-13 北京邮电大学 Lightweight spacecraft fault detection classification method based on long-short-term memory model
CN118040910A (en) * 2024-04-12 2024-05-14 国网山东省电力公司菏泽供电公司 Online monitoring method for energy storage abnormal state of micro-grid in remote area

Also Published As

Publication number Publication date
CN111428772B (en) 2022-08-23

Similar Documents

Publication Publication Date Title
CN110823574B (en) Fault diagnosis method based on semi-supervised learning deep countermeasure network
CN111428772B (en) Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting
CN112036301B (en) Driving motor fault diagnosis model construction method based on intra-class feature transfer learning and multi-source information fusion
CN107563433B (en) Infrared small target detection method based on convolutional neural network
WO2023216583A1 (en) Hail weather identification and classification method based on multi-channel deep residual shrinkage network
Xiao et al. Maximal overlap discrete wavelet transform and deep learning for robust denoising and detection of power quality disturbance
CN114282571B (en) Method, system, equipment and medium for constructing multidimensional health index of bearing
CN113158722A (en) Rotary machine fault diagnosis method based on multi-scale deep neural network
CN110472671A (en) Based on multistage oil-immersed transformer fault data preprocess method
CN116503399B (en) Insulator pollution flashover detection method based on YOLO-AFPS
Afrasiabi et al. Two-stage deep learning-based wind turbine condition monitoring using SCADA data
CN112419243B (en) Power distribution room equipment fault identification method based on infrared image analysis
CN113449618A (en) Method for carrying out deep learning rolling bearing fault diagnosis based on feature fusion and mixed enhancement
CN112802011A (en) Fan blade defect detection method based on VGG-BLS
CN116952554A (en) Multi-sensor mechanical equipment fault diagnosis method and device based on graph rolling network
CN116484258A (en) Elevator traction machine bearing fault diagnosis method
Sinha et al. Power system fault detection using image processing and pattern recognition
CN115661543A (en) Multi-scale industrial part defect detection method based on generation countermeasure network
CN115564709A (en) Evaluation method and system for robustness of power algorithm model in confrontation scene
CN115130516A (en) GIS equipment state identification method and device based on spectrum power spectrum entropy
CN114760128A (en) Network abnormal flow detection method based on resampling
CN110223295B (en) Significance prediction method and device based on deep neural network color perception
CN110689071B (en) Target detection system and method based on structured high-order features
CN114764867A (en) Fan fault diagnosis system and method based on image main feature extraction and application
CN117152422B (en) Ultraviolet image anchor-free frame target detection method, storage medium and electronic equipment

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