CN116664870A - Power quality disturbance identification method based on transfer learning - Google Patents

Power quality disturbance identification method based on transfer learning Download PDF

Info

Publication number
CN116664870A
CN116664870A CN202310658665.0A CN202310658665A CN116664870A CN 116664870 A CN116664870 A CN 116664870A CN 202310658665 A CN202310658665 A CN 202310658665A CN 116664870 A CN116664870 A CN 116664870A
Authority
CN
China
Prior art keywords
disturbance
layers
transfer learning
layer
power quality
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.)
Pending
Application number
CN202310658665.0A
Other languages
Chinese (zh)
Inventor
胡雪峰
陈兵
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 Institute of Technology
Original Assignee
Nanjing Institute of Technology
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 Institute of Technology filed Critical Nanjing Institute of Technology
Priority to CN202310658665.0A priority Critical patent/CN116664870A/en
Publication of CN116664870A publication Critical patent/CN116664870A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • 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
    • G06F18/2431—Multiple classes
    • 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/048—Activation functions
    • 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
    • G06N3/08—Learning methods
    • G06N3/096—Transfer learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06395—Quality analysis or management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Human Resources & Organizations (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Economics (AREA)
  • Databases & Information Systems (AREA)
  • Strategic Management (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Evolutionary Biology (AREA)
  • Marketing (AREA)
  • Educational Administration (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Primary Health Care (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)

Abstract

The application discloses a power quality disturbance identification method based on transfer learning, which comprises the following steps: firstly, sampling and constructing a power quality disturbance data set; and secondly, converting time sequence disturbance data of the electric energy quality disturbance data set into a two-dimensional track image by using a phase space reconstruction technology, finally constructing a convolutional neural network disturbance recognition model based on transfer learning, and recognizing the disturbance track image by using the transfer learning convolutional neural network. The application solves the problems that the disturbance signal feature set in the traditional disturbance recognition method is complicated in structure and too dependent on manual selection, the feature is selected without correct discrimination standard and general evaluation frame, so that the feature set contains a large number of redundant features, and the like, reduces the complexity of disturbance recognition, reduces calculation time, improves the accuracy of disturbance recognition, provides a certain reference basis for electric energy quality management, and further improves the reliability and economy of power supply.

Description

Power quality disturbance identification method based on transfer learning
Technical Field
The application relates to the technical field of power quality disturbance recognition, in particular to a power quality disturbance recognition method based on transfer learning.
Background
In recent years, with the integration of multiple energy sources, the power grid structure of China is greatly changed, and the power grid structure is mainly embodied on a power supply side and a user side. On the power supply side, the fans, the photovoltaic devices and the energy storage devices are connected into the power grid in a high proportion, so that serious electric energy quality problems are caused, the safe and economic operation of the power grid is seriously influenced, and the service life and the power conversion efficiency of the grid-connected renewable energy system are also reduced; on the user side, the distortion level of the voltage and the current of the power grid is aggravated by the application of a large number of nonlinear equipment such as asynchronous motors, electronic measurement and control devices and the like, and great interference is brought to the power grid; in addition, the increasingly widespread use of sensitive electronic devices such as computers, programmable logic controllers, and the like, also place higher demands on the quality of electrical energy, which can lead to failure or malfunction of electrical devices, and damage of automatic protection devices and devices. Therefore, the method has important significance for the identification and classification research of the electric energy quality.
The research aiming at the power quality disturbance recognition is mainly divided into two parts, namely feature extraction and classification. Feature extraction is the extraction of features from different types of power quality disturbances using digital signal processing techniques such as fourier transform (FFT), short Time Fourier Transform (STFT), wavelet Transform (WT), S-transform (ST), empirical Mode Decomposition (EMD), variational Mode Decomposition (VMD), and Independent Component Analysis (ICA). The classification is to perform different types of power quality disturbance recognition according to specific rules according to the characteristics. Typical recognition methods are Decision Trees (DT), support Vector Machines (SVM), artificial Neural Networks (ANN), probabilistic Neural Networks (PNN), and the like.
In the above two parts, the existing method is effective, but has certain limitations. The different types and numbers of features have a great influence on the recognition result, and in the power quality disturbance recognition process, it is difficult to know which features should be extracted. In the current research, the process of extracting the disturbance signal features is greatly dependent on various different types and numbers of manual features, such as statistical features based on the maximum value, the mean value, the extreme value, the standard deviation, the effective value and the like of waveform signals, and besides, skewness, singular value, spectral kurtosis, energy entropy, information entropy and the like are also proposed as the signal disturbance features. The use of unsuitable features may reduce the accuracy of the final disturbance identification.
Because the power quality monitoring needs more accurate measuring equipment, most monitoring equipment cannot be popularized to all power users in consideration of economy, meanwhile, the disturbance of a power system caused by loads or circuits has larger uncertainty, and a large number of disturbance signal data sets cannot be acquired under the common influence of various factors. The deep learning model usually requires a large amount of training data to realize accurate and efficient learning, so in order to solve the problem that the deep learning network cannot obtain enough training precision due to insufficient disturbance data, a power quality disturbance recognition method based on transfer learning is provided, and the power quality disturbance recognition method not only can automatically extract disturbance characteristics, but also is suitable for all nonlinear and non-stationary signals.
Disclosure of Invention
1. The technical problems to be solved are as follows:
aiming at the technical problems, the application provides a power quality disturbance identification method based on transfer learning,
1. the technical scheme is as follows:
a power quality disturbance identification method based on transfer learning is characterized by comprising the following steps of: the method comprises the following steps:
step one: sampling different types of power quality disturbance signals to construct a power quality disturbance data set;
step two: converting time sequence disturbance data of the electric energy quality disturbance data set into a two-dimensional track image by using a phase space reconstruction method;
step three: and identifying the converted two-dimensional track image by using a convolutional neural network of transfer learning.
Further, the different types of power quality disturbance signals in the first step include 7 types of single disturbance and 8 types of composite disturbance; the 7 types of single disturbance comprise dip disturbance, interrupt disturbance, flicker disturbance, harmonic disturbance, transient oscillation disturbance and transient pulse disturbance; the 8-class composite disturbance comprises dip + harmonic disturbance, voltage dip + harmonic disturbance, harmonic + voltage interruption disturbance, voltage dip + transient oscillation disturbance, harmonic + voltage flicker disturbance, voltage dip + harmonic + voltage interruption disturbance, voltage dip + harmonic + transient oscillation disturbance, voltage dip + harmonic + voltage flicker disturbance.
Further, the phase space reconstruction method in the second step is to reconstruct a disturbance signal containing a one-dimensional time sequence into Gao Weixiang space, and then map the disturbance signal to a two-dimensional plane to obtain a corresponding track image; the track image not only comprises the time sequence characteristics of the initial signal, but also comprises disturbance characteristics hidden in the original signal; wherein, reconstructing a disturbance signal containing a one-dimensional time sequence signal into Gao Weixiang space and mapping the disturbance signal to a two-dimensional plane to obtain a corresponding track image, wherein the disturbance signal is expressed by the following formula:
Y(t i )＝[x(t i ),x(t i +2τ),…,x(t i +(m-1)τ)] (1)；
in the formula (1), x (t) i ) Representing univariate time series, x (t) i ) E R, i=1, 2,..n; τ is a time delay parameter; m is the embedding dimension; y (t) i ) The track matrix is a track matrix corresponding to the reconstructed two-dimensional plane; wherein the embedded spatial dimension m satisfies the following relationship:
m≥2·d+1 (2)；
(2) Wherein d is the dimension of the dynamic system;
the track matrix form of the one-dimensional time sequence signal after phase space reconstruction is as follows:
k in the formula (3) is the number of phase points in the phase space X.
Further, the convolutional neural network in the third step is an AlexNet network and consists of 5 modularized structures, wherein the first modularized structure is a convolutional layer 1 and a convolutional layer 2, and the convolutional neural network mainly comprises 2 convolutional layers, 2 activating layers, 2 batch normalization layers and 2 maximum pooling layers; the second modular structure is convolutional layers 3, 4, which mainly comprise 2 convolutional layers and 2 active layers; the third modular structure is convolution layer 5, which mainly contains 1 convolution layer, 1 activation layer and 1 max pooling layer; the fourth modular structure is the full connection layers 6, 7, which mainly comprise 2 full connection layers, 2 activation layers and 2 Dropout layers; the fifth modular structure is the full tie layer 8, which contains mainly 1 full tie layer and 1 Softmax layer; a total of 5 convolutional layers and 3 fully-connected layers.
Further, the convolutional neural network for transfer learning is: performing secondary training on the convolutional neural network model trained by the data set in a new task target data set, namely, utilizing the trained network model, and solving a new target task by adjusting fine parameters of the model;
in the process of recognizing the disturbance track image by using the transfer learning convolutional neural network, two network layers in a fifth module in the AlexNet network are replaced by a transfer module, and the output size of an output layer is adjusted to be the total number of disturbance signal types; other relevant parameters are kept unchanged, and then training is carried out;
the convolutional neural network is a probability value for calculating which disturbance type the disturbance signal reconstruction track image belongs to in the AlexNet network after transfer learning, the disturbance type with the largest probability is selected as the predicted output, the calculation error of the output layer is obtained through a cross entropy loss function, and the calculation error is used for evaluating the similarity degree between the actual output and the expected output, wherein the cross entropy loss function expression is as follows:
in the formula (4), W is a weight matrix; b is a bias vector; n is the number of samples; k is the number of disturbance types; t is t ij The probability that the ith sample belongs to class j; y is ij The probability that the j-th sample belongs to class i.
Further, the migration module adopts two full-connection layers to further increase the automatic extraction capability of the network to the disturbance track characteristics; the active layer adopts a ReLU function, so that the problem that the model gradually degenerates along with the increase of the network layer number is solved; the added Dropout layer improves the generalization capability and robustness of the model; the Softmax layer is used to determine the type of disturbance.
3. The beneficial effects are that:
(1) The application replaces the whole process of disturbance signal feature extraction, feature optimization and disturbance recognition in the traditional method by the transfer-learning convolutional nerves, which is a closed-loop feedback idea, namely, the signal feature extraction process and the disturbance classification process are put into the same closed loop for training, thereby realizing the end-to-end recognition of disturbance signals to disturbance types.
(2) The convolutional neural network for transfer learning in the scheme is a method for performing secondary training on a convolutional neural network model trained by a large number of data sets in a new task target data set, namely, utilizing the trained network model and solving a new target task by adjusting fine parameters of the model. Compared with the original model, the transfer learning has the advantages that the existing optimal network model is utilized, and the model required by the research is quickly built by fine adjustment of parameters of the optimal network model; meanwhile, the original optimal network model is already trained by a large amount of data, so that training time can be greatly reduced.
Drawings
FIG. 1 is a schematic flow chart of a power quality disturbance recognition method based on transfer learning;
FIG. 2 is a schematic diagram of waveforms of some of the different types of power quality disturbance signals involved in the present application;
FIG. 3 is a schematic diagram of a phase space reconstruction trace of a portion of the different types of power quality disturbance signals in FIG. 2; the method comprises the steps of carrying out a first treatment on the surface of the
Fig. 4 is a schematic diagram of an AlexNet network structure in the present application;
FIG. 5 is a block diagram of a migration module according to the present application;
FIG. 6 is a schematic diagram of a disturbance recognition model based on transfer learning in the present application.
Detailed Description
The present application will be described in detail with reference to the accompanying drawings.
As shown in fig. 1 and 6, a power quality disturbance recognition method based on transfer learning is characterized in that: the method comprises the following steps:
step one: sampling different types of power quality disturbance signals to construct a power quality disturbance data set;
according to the power quality disturbance model in the standards of IEEE Std.1159-2019 and the like, 7 single power quality disturbances and 15 disturbances combined by 8 single power quality disturbances are generated as shown in table 1. The amplitude of the disturbance signal is normalized to 1p.u, the fundamental frequency is set to 50Hz, the sampling frequency is 3.2KHz, 10 cycles are sampled, and one cycle contains 64 sampling points. 500 samples are randomly generated for each type of disturbance signal, 400 disturbance signals are selected as training sets, and 100 disturbance signals are selected as test sets. A diagram of a portion of different types of power quality disturbance waveforms is shown in fig. 2.
Meter 1 electric energy quality disturbance model
Step two: converting time sequence disturbance data of the electric energy quality disturbance data set into a two-dimensional track image by using a phase space reconstruction method;
the track image not only comprises the time sequence characteristics of the initial signal, but also comprises disturbance characteristics hidden in the original signal; wherein, reconstructing a disturbance signal containing a one-dimensional time sequence signal into Gao Weixiang space and mapping the disturbance signal to a two-dimensional plane to obtain a corresponding track image, wherein the disturbance signal is expressed by the following formula:
Y(t i )＝[x(t i ),x(t i +2τ),…,x(t i +(m-1)τ)] (1)；
in the formula (1), x (t) i ) Representing univariate time series, x (t) i ) E R, i=1, 2,..n; τ is a time delay parameter; m is the embedding dimension; y (t) i ) The track matrix is a track matrix corresponding to the reconstructed two-dimensional plane; wherein the embedded spatial dimension m satisfies the following relationship:
m≥2·d+1 (2)；
d in the formula (2) is the dimension of a dynamic system;
the track matrix form of the one-dimensional time sequence signal after phase space reconstruction is as follows:
k in the formula (3) is the number of phase points in the phase space X.
According to the principles of the Takens (takes) and the G-P algorithm, the phase space reconstruction trajectories of the different types of power quality disturbance signals obtained after determining the embedding dimension m=3 and the delay time τ=20 are shown in fig. 3, respectively.
Step three: and identifying the converted two-dimensional track image by using a convolutional neural network of transfer learning.
The convolutional neural network in the process of recognizing the disturbance track image by utilizing the convolutional neural network of transfer learning is an AlexNet network, is a classical convolutional neural network, and mainly comprises 5 modularized structures as shown in fig. 4, wherein the first modularized structure is a convolutional layer 1 and a convolutional layer 2, and mainly comprises 2 convolutional layers, 2 activating layers, 2 batch normalization layers and 2 maximum pooling layers; the second modular structure is convolutional layers 3, 4, which mainly comprise 2 convolutional layers and 2 active layers; the third modular structure is convolution layer 5, which mainly contains 1 convolution layer, 1 activation layer and 1 max pooling layer; the fourth modular structure is the full connection layers 6, 7, which mainly comprise 2 full connection layers, 2 activation layers and 2 Dropout layers; the fifth modular structure is the full tie layer 8, which contains mainly 1 full tie layer and 1 Softmax layer; a total of 5 convolutional layers and 3 fully-connected layers.
The method mainly comprises the steps of replacing two network layers in a fifth module of an AlexNet network model by a migration module, adjusting the output size of an output layer to the total number of disturbance signal types, keeping other relevant parameters unchanged, and training. A specific structural diagram for constructing the disturbance recognition model based on the transfer learning is shown in fig. 6.
The convolutional neural network is used for calculating probability values of disturbance types to which disturbance signal reconstruction track images belong in the AlexNet network after transfer learning, the disturbance type with the largest probability is selected as predicted output, calculation errors of the output layer are obtained through a cross entropy loss function, and meanwhile the calculation errors are used for evaluating the approximation degree of actual output and expected output, wherein the cross entropy loss function expression is as follows:
in the formula (4), W is a weight matrix; b is a bias vector; n is the number of samples; k is the number of disturbance types; t is t ij The probability that the ith sample belongs to class j; y is ij The probability that the j-th sample belongs to class i.
As shown in fig. 5, the migration module includes two fully-connected layers for further increasing the automatic extraction capability of the network to the disturbance track features; the active layer adopts a ReLU function, so that the problem that the model gradually degenerates along with the increase of the network layer number is solved; the added Dropout layer improves the generalization capability and robustness of the model; the Softmax layer is used to determine the type of disturbance.
While the application has been described with reference to the preferred embodiments, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the application, and it is intended that the scope of the application shall be defined by the appended claims.

Claims (6)

1. A power quality disturbance identification method based on transfer learning is characterized by comprising the following steps of: the method comprises the following steps:
step one: sampling different types of power quality disturbance signals to construct a power quality disturbance data set;
step two: converting time sequence disturbance data of the electric energy quality disturbance data set into a two-dimensional track image by using a phase space reconstruction method;
step three: and identifying the converted two-dimensional track image by using a convolutional neural network of transfer learning.
2. The power quality disturbance recognition method based on transfer learning according to claim 1, wherein: the different types of power quality disturbance signals in the first step comprise 7 types of single disturbance and 8 types of compound disturbance; the 7 types of single disturbance comprise dip disturbance, interrupt disturbance, flicker disturbance, harmonic disturbance, transient oscillation disturbance and transient pulse disturbance; the 8-class composite disturbance comprises dip + harmonic disturbance, voltage dip + harmonic disturbance, harmonic + voltage interruption disturbance, voltage dip + transient oscillation disturbance, harmonic + voltage flicker disturbance, voltage dip + harmonic + voltage interruption disturbance, voltage dip + harmonic + transient oscillation disturbance, voltage dip + harmonic + voltage flicker disturbance.
3. The power quality disturbance recognition method based on transfer learning according to claim 1, wherein: reconstructing a disturbance signal containing a one-dimensional time sequence into Gao Weixiang space, and mapping the disturbance signal to a two-dimensional plane to obtain a corresponding track image; the track image not only comprises the time sequence characteristics of the initial signal, but also comprises disturbance characteristics hidden in the original signal; wherein, reconstructing a disturbance signal containing a one-dimensional time sequence signal into Gao Weixiang space and mapping the disturbance signal to a two-dimensional plane to obtain a corresponding track image, wherein the disturbance signal is expressed by the following formula:
Y(t i )＝[x(t i ),x(t i +2τ),…,x(t i +(m-1)τ)] (1)；
in the formula (1), x (t) i ) Representing univariate time series, x (t) i ) E R, i=1, 2,..n; τ is a time delay parameter; m is the embedding dimension; y (t) i ) The track matrix is a track matrix corresponding to the reconstructed two-dimensional plane; wherein the embedded spatial dimension m satisfies the following relationship:
m≥2·d+1 (2)；
d in the formula (2) is the dimension of a dynamic system;
the track matrix form of the one-dimensional time sequence signal after phase space reconstruction is as follows:
k in the formula (3) is the number of phase points in the phase space X.
4. The power quality disturbance recognition method based on transfer learning according to claim 1, wherein: the convolutional neural network in the third step is an AlexNet network and consists of 5 modularized structures, wherein the first modularized structure is a convolutional layer 1 and a convolutional layer 2, and the convolutional neural network mainly comprises 2 convolutional layers, 2 activating layers, 2 batch normalization layers and 2 maximum pooling layers; the second modular structure is convolutional layers 3, 4, which mainly comprise 2 convolutional layers and 2 active layers; the third modular structure is convolution layer 5, which mainly contains 1 convolution layer, 1 activation layer and 1 max pooling layer; the fourth modular structure is the full connection layers 6, 7, which mainly comprise 2 full connection layers, 2 activation layers and 2 Dropout layers; the fifth modular structure is the full tie layer 8, which contains mainly 1 full tie layer and 1 Softmax layer; a total of 5 convolutional layers and 3 fully-connected layers.
5. The method for recognizing power quality disturbance based on transfer learning according to claim 4, wherein:
the convolutional neural network for transfer learning is as follows: performing secondary training on the convolutional neural network model trained by the data set in a new task target data set, namely, utilizing the trained network model, and solving a new target task by adjusting fine parameters of the model;
in the process of recognizing the disturbance track image by using the transfer learning convolutional neural network, two network layers in a fifth module in the AlexNet network are replaced by the transfer learning module, and the output size of the output layer is adjusted to the total number of disturbance signal types; other relevant parameters are kept unchanged, and then training is carried out;
the convolutional neural network is a probability value for calculating which disturbance type the disturbance signal reconstruction track image belongs to in the AlexNet network after transfer learning, the disturbance type with the largest probability is selected as the predicted output, the calculation error of the output layer is obtained through a cross entropy loss function, and the calculation error is used for evaluating the similarity degree between the actual output and the expected output, wherein the cross entropy loss function expression is as follows:
in the formula (4), W is a weight matrix; b is a bias vector; n is the number of samples; k is the number of disturbance types; t is t ij The probability that the ith sample belongs to class j; y is ij The probability that the j-th sample belongs to class i.
6. The method for recognizing the power quality disturbance based on the transfer learning according to claim 5, wherein the method comprises the following steps: the migration module comprises two full-connection layers for further increasing the automatic extraction capability of the network to disturbance track characteristics; the active layer adopts a ReLU function, so that the problem that the model gradually degenerates along with the increase of the network layer number is solved; the added Dropout layer improves the generalization capability and robustness of the model; the Softmax layer is used to determine the type of disturbance.
CN202310658665.0A 2023-06-05 2023-06-05 Power quality disturbance identification method based on transfer learning Pending CN116664870A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310658665.0A CN116664870A (en) 2023-06-05 2023-06-05 Power quality disturbance identification method based on transfer learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310658665.0A CN116664870A (en) 2023-06-05 2023-06-05 Power quality disturbance identification method based on transfer learning

Publications (1)

Publication Number Publication Date
CN116664870A true CN116664870A (en) 2023-08-29

Family

ID=87718707

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310658665.0A Pending CN116664870A (en) 2023-06-05 2023-06-05 Power quality disturbance identification method based on transfer learning

Country Status (1)

Country Link
CN (1) CN116664870A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118395244A (en) * 2024-06-28 2024-07-26 北京汇思慧能科技有限公司 Petrochemical industry electric energy quality disturbance identification method based on multi-scale TCN and multi-head self-attention mechanism

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118395244A (en) * 2024-06-28 2024-07-26 北京汇思慧能科技有限公司 Petrochemical industry electric energy quality disturbance identification method based on multi-scale TCN and multi-head self-attention mechanism

Similar Documents

Publication Publication Date Title
CN109842373B (en) Photovoltaic array fault diagnosis method and device based on space-time distribution characteristics
Li et al. An intelligent transient stability assessment framework with continual learning ability
CN111880044B (en) Online fault positioning method for distribution network containing distributed power supply
CN109635928A (en) A kind of voltage sag reason recognition methods based on deep learning Model Fusion
Dou et al. Hybrid model for renewable energy and loads prediction based on data mining and variational mode decomposition
WO2019056753A1 (en) Dynamic equivalent modeling method for distributed photovoltaic power station cluster
CN114792156B (en) Photovoltaic output power prediction method and system based on curve characteristic index clustering
CN113011481B (en) Electric energy meter function abnormality assessment method and system based on decision tree algorithm
Li et al. High-precision dynamic modeling of two-staged photovoltaic power station clusters
CN110726898A (en) Power distribution network fault type identification method
CN116664870A (en) Power quality disturbance identification method based on transfer learning
CN112633658A (en) Low-voltage distribution area topological relation identification method based on CNN-LSTM
CN106980044A (en) A kind of Harmonious Waves in Power Systems current estimation method for adapting to wind power integration
CN110941909A (en) Method and system for identifying and calculating dominance of parameters of photovoltaic power station and wind power plant
CN115469184A (en) New energy transmission line fault identification method based on convolutional network
CN115758188A (en) Non-invasive load identification method, device, equipment and medium
Ye et al. Single pole‐to‐ground fault location method for mmc‐hvdc system using wavelet decomposition and dbn
CN117151770A (en) Attention mechanism-based LSTM carbon price prediction method and system
CN114818827A (en) Non-invasive load decomposition method based on seq2point network
CN113702767B (en) Island direct-current microgrid fault diagnosis method based on wavelet sliding window energy
CN112036718B (en) Electric power system safety risk assessment method considering new energy uncertainty
CN115983347A (en) Non-invasive load decomposition method, device and storage medium
CN116340753A (en) Energy monitoring method for distributed energy area considering non-electrical characteristic quantity
He et al. Partial discharge pattern recognition algorithm based on sparse self-coding and extreme learning machine
CN107069710B (en) Power system state estimation method considering new energy space-time correlation

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