CN116304846B - CVT internal insulation abnormality online assessment method based on self-supervision learning - Google Patents
CVT internal insulation abnormality online assessment method based on self-supervision learning Download PDFInfo
- Publication number
- CN116304846B CN116304846B CN202310586721.4A CN202310586721A CN116304846B CN 116304846 B CN116304846 B CN 116304846B CN 202310586721 A CN202310586721 A CN 202310586721A CN 116304846 B CN116304846 B CN 116304846B
- Authority
- CN
- China
- Prior art keywords
- cvt
- self
- cvt voltage
- voltage measurement
- sequence
- 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
Links
- 230000005856 abnormality Effects 0.000 title claims abstract description 57
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000009421 internal insulation Methods 0.000 title claims abstract description 34
- 238000005259 measurement Methods 0.000 claims abstract description 75
- 238000012549 training Methods 0.000 claims abstract description 35
- 210000003608 fece Anatomy 0.000 claims abstract description 32
- 241000254173 Coleoptera Species 0.000 claims abstract description 29
- 230000002159 abnormal effect Effects 0.000 claims abstract description 23
- 238000001514 detection method Methods 0.000 claims abstract description 7
- 238000005457 optimization Methods 0.000 claims abstract description 5
- 206010008138 Cerebral venous thrombosis Diseases 0.000 claims description 176
- 239000013598 vector Substances 0.000 claims description 31
- 238000012360 testing method Methods 0.000 claims description 23
- 238000010606 normalization Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 9
- 241001229848 Aphodius granarius Species 0.000 claims description 6
- 238000009826 distribution Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000005096 rolling process Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 230000009194 climbing Effects 0.000 claims description 3
- 238000005553 drilling Methods 0.000 claims description 3
- 239000002245 particle Substances 0.000 claims description 3
- 230000003595 spectral effect Effects 0.000 claims description 3
- 238000009413 insulation Methods 0.000 abstract description 6
- 238000004590 computer program Methods 0.000 description 3
- 230000004224 protection Effects 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 230000032683 aging Effects 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 238000005520 cutting process Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000005329 antiferromagnetic resonance Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
-
- 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
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a CVT internal insulation abnormality online assessment method based on self-supervision learning, which comprises the steps of collecting CVT voltage measurement values, constructing a CVT voltage measurement value sequence, carrying out standardization, and converting the standardized CVT voltage measurement value sequence into a time sequence window; establishing a self-supervision learning model, optimizing super parameters of the self-supervision learning model through a dung beetle optimization algorithm, training the self-supervision learning model, and outputting a CVT voltage predicted value and a reconstruction probability through the trained self-supervision learning model; and establishing a CVT internal insulation state abnormality detection strategy according to the CVT voltage predicted value and the reconstruction probability, and detecting the abnormal state of the CVT voltage measured value. According to the CVT voltage prediction method, the CVT voltage prediction value is obtained through the self-supervision learning model, the CVT voltage abnormality score corresponding to the CVT voltage measurement value is detected, whether the CVT voltage abnormality score is abnormal or not is judged, and the insulation condition inside the CVT can be accurately evaluated in real time.
Description
Technical Field
The invention relates to the technical field of online monitoring of electric power metering, in particular to a CVT internal insulation abnormality online assessment method based on self-supervision learning.
Background
A capacitive voltage transformer (Capacitive Voltage Transformer, CVT) is one of the electrical parameter measuring devices that provides a reliable basis for electrical energy measurement, condition monitoring and relay protection. Compared with the traditional electromagnetic voltage transformer, the CVT has higher insulating strength and antiferromagnetic resonance characteristic, and is widely applied to high-voltage systems of 110kV and above. When the CVT operates, the primary high voltage is converted into the secondary low voltage, and the secondary low voltage is supplied to relay protection, an automatic device and a measuring instrument for use, so that the requirements of various secondary protections are met.
However, the capacitive voltage divider unit is affected by heat aging, electrical aging, and other factors during long-term operation, and is prone to the problem of reduced insulation performance, and even affects the normal operation of the CVT and the safe operation of the power system. Therefore, the accurate and real-time error evaluation of the CVT is of great importance to ensure safe, stable and economical operation of the electric power system. Analysis of a CVT by field tests is subject to a blackout schedule, resulting in failure to analyze many CVT insulation states in transit, and once insulation degradation to a certain extent results in insulation breakdown, equipment blackout accidents will result, and analysis of voltage data collected by the CVT is an effective real-time online analysis means.
Disclosure of Invention
Aiming at the fact that the traditional CVT internal insulation abnormality online assessment method mainly depends on experience rules and priori knowledge and lacks the capability of independently mining effective characteristics, the CVT internal insulation abnormality online assessment method based on self-supervision learning is provided, the influence of primary voltage fluctuation and voltage active change on online assessment of the CVT internal insulation state can be effectively eliminated, and the CVT secondary voltage abnormality characteristics are independently mined to achieve accurate assessment of the CVT internal insulation state.
The invention is realized by the following technical scheme: a CVT internal insulation abnormality online assessment method based on self-supervision learning comprises the following steps:
s1: collecting CVT voltage measurement values, constructing a CVT voltage measurement value sequence, carrying out standardization, and dividing the standardized CVT voltage measurement value sequence into training data and test data; converting the training data into a time sequence window to serve as a training set, and converting the test data into a time sequence window to serve as a test set;
s2: establishing a self-supervision learning model, optimizing super parameters of the self-supervision learning model through a dung beetle optimization algorithm, training the self-supervision learning model by using a training set, and outputting a CVT voltage predicted value and a reconstruction probability through the trained self-supervision learning model;
s3: and establishing a CVT internal insulation state abnormality detection strategy according to the CVT voltage predicted value and the reconstruction probability, and detecting the abnormal state of the CVT voltage measured value.
Further preferably, in step S1, CVT voltage measurement values of the same phase of k branches on the same bus are collected, CVT voltage measurement values of the same phase of each CVT branch are taken as a feature, a CVT voltage measurement value sequence is constructed, and the CVT voltage measurement value sequence is obtainedAnd (3) carrying out standardization, wherein the standardization formula is as follows:
wherein ,for timestamp->CVT voltage measurement, ±>For timestamp->CVT voltage measurement normalized data of +.>Is the CVT voltage measurement sequence minimum, and +.>Is the CVT voltage measurement sequence maximum; r is a non-zero constant vector, < >>T is the number of time stamps.
Further preferably, the process of converting training data or test data into time series windows is: detecting abnormal time stamps in each normalized CVT voltage measurement value sequence in the training data by using a spectral residual algorithm, and replacing CVT voltage measurement value normalization data with adjacent time stamps by using the CVT voltage measurement value normalization data with the abnormal time stamps; and converting the normalized CVT voltage measurement value sequence into a time sequence windowTime series window->Is +.>Step size e, thereby obtaining w local context windows of the normalized CVT voltage measurement value sequence, i.e，/>Is->A local context window->The method comprises the steps of carrying out a first treatment on the surface of the For the purpose of->CVT Voltage measurement normalization data at +.>Modeling the dependency of (2) with length +.>Is represented as:
for a pair ofConverts the input normalized CVT voltage measurement sequence into a time sequence windowOne length is +.>Is added to the normalized CVT voltage measurement sequence +.>To keep each time series window length +.>。
Further preferably, the self-supervised learning model includes a self-encoder, a parallel graph attention network, a fully connected layer, a gated loop network, a predictive model, and a reconstruction model; the training set is input into a self-encoder, the residual sequence output by the self-encoder is used as input of a parallel graph attention network, then the dependency relationship of the residual sequence on time and characteristic dimension is captured through the parallel graph attention network, after the characteristics and time relationship of the parallel graph attention network and the residual sequence are input into a full-connection layer, the characteristic of the normalized CVT voltage measured value sequence is captured by feeding a gating loop network, and the output of the gating loop network is fed into a prediction model and a reconstruction model to obtain CVT voltage predicted values and reconstruction probabilities.
Further preferably, taking the super-parameters of the self-supervision learning model as optimizing variables of a dung beetle optimizing algorithm, and iteratively obtaining the optimal super-parameters of the self-supervision learning model by updating individual positions and fitness values in the population:
s21: initializing a population, randomly generating N individuals, and enabling each individual to express a potential solution;
s22: each individual is subjected to fitness evaluation, and the population is divided into four sub-populations according to fitness values: rolling ball dung beetles, egg ball dung beetles, small dung beetles and stealing dung beetles;
s23: performing the corresponding operations on each sub-population:
ball dung beetle: rolling the faecal ball along a given direction, navigating by using celestial bodies, climbing onto the faecal ball to dance when encountering an obstacle, and re-determining the direction;
egg ball dung beetle: spawning near the locally optimal solution to form egg balls, and providing safe environment and food for offspring;
small dung beetles: finding optimal food source positions, namely globally optimal solutions, from underground drilling to find food;
theft of dung beetles: stealing dung balls from other dung beetles, namely a local optimal solution or a global optimal solution;
s24: updating individual positions and fitness values in the population, and recording the current optimal solution;
s25: judging whether a termination condition is reached, wherein the termination condition is as follows: maximum iteration number or a preset error threshold; if the termination condition is reached, outputting an optimal solution and ending the algorithm; if the termination condition is not reached, returning to step S23, and if the termination condition is reached, outputting the optimal solution and ending the algorithm.
Further preferably, the method for obtaining the residual sequence is as follows:
the self-encoder learns the effective encoding of a set of data in an unsupervised manner forThe activity value of the intermediate hidden layer of the self-encoder is +.>Is encoded by (a), namely:
the output data from the encoder are:
by passing throughRepresenting a residual sequence of error change information; wherein->Weight matrix for middle hidden layer, +.>Bias term for middle hidden layer, +.>Weight matrix for output layer of self-encoder, < ->Is a bias term from the output layer of the encoder, < ->To activate the function +.>For->Encoded data.
Further preferred, the dependency of the residual sequence in time and feature dimensions is captured by a parallel graph attention network and />The input to the parallel graph attention network is a set of node feature vectors +.>Expressed as，/>Representing node->Feature vector of>,/>Is the number of nodes, equal to the number of branches, wherein +.>Representing the dimension of the feature vector for each node; the output of the parallel graph attention network is a new set of node feature vectors, i.e. +.>，/>Representing nodes after parallel graph attention network processing +.>Wherein>,/>Representing nodes after parallel graph attention network processing +.>Is a dimension of the feature vector of (a); the parallel graph attention network processing procedure is as follows:
where sigmoid represents the activation function,representation and node->Feature vector set of neighboring nodes, +.>Representation and node->Adjacent node->Feature vector of>For node->And node->Attention score between->Is a learnable column vector.
Further preferably, the method for obtaining the CVT voltage prediction value and the reconstruction probability is as follows: the CVT voltage predicted value is obtained through a predicted model, and the probability is reconstructed through a reconstruction model; the prediction model predicts the CVT voltage value of the next time stamp through the CVT voltage value of the previous time sequence window, and the reconstruction model is used for analyzing the data distribution of the whole time sequence window; the loss function is, wherein ,/>Indicating total loss->Representing the loss of the predictive model, +.>Representing the loss of the reconstructed model;
wherein ,representation->Is>CVT voltage measurement normalization data for each feature; />When expressedTimestamp->Is the first of (2)CVT voltage predictions for each feature; />For vector representation in potential space, +.>For input, & lt + & gt>In order to obtain the posterior density of the particles,for edge density +.>Representing parallel operation; />A reconstruction model for approximating the posterior distribution;representing the expected negative log likelihood of a given input, +.>Representing encoder distribution-> and />Kullback-Leibler divergence.
Further preferably, the procedure of step S3 is as follows:
calculating a time stamp by a predictive modelCVT voltage prediction value +.>Obtaining a time stamp by reconstructing the model +.>Reconstruction probability +.>，/>Representing a time stamp->Is>Reconstruction probability of individual features; time stampCVT voltage abnormality score of +.>：
wherein ,the self-supervision learning model super-parameters are used for balancing errors of the prediction model and reconstruction probability of the reconstruction model; according to the characteristics of primary CVT voltage fluctuation, combining a prediction model and a reconstruction model to obtain CVT voltage abnormality score +.>Determining an optimal threshold value by using an adaptive threshold value method; and outputting an abnormality label when the CVT voltage abnormality score corresponding to the detected CVT voltage measured value exceeds an optimal threshold.
Further preferably, according to the characteristics of the CVT in the real running system when the internal insulation state is abnormal, a single abnormality cutting strategy is adopted to optimize the output abnormal label, that is, when the abnormality of the single timestamp occurs, the corresponding CVT voltage predicted value is changed to the normal label.
The invention also provides a CVT internal insulation abnormality online assessment system based on self-supervision learning, which comprises a data acquisition module, a data preprocessing module, a voltage prediction module and an abnormality judgment module; the data acquisition module is used for acquiring CVT voltage measurement values; the data preprocessing module is used for constructing and normalizing a CVT voltage measurement value sequence, dividing the normalized CVT voltage measurement value sequence into training data and test data, converting the training data into a time sequence window as a training set, and converting the test data into a time sequence window as a test set; the voltage prediction module is internally provided with a self-supervision learning model and is used for outputting a CVT voltage predicted value and a reconstruction probability; and the abnormality judgment module establishes an abnormality detection strategy of the insulating state in the CVT according to the CVT voltage predicted value and the reconstruction probability, and detects the abnormal state of the CVT voltage measured value.
The invention has the beneficial effects that: according to the invention, the CVT voltage measured value is converted into a time sequence window, the time sequence window is input into the self-supervision learning model, the CVT voltage predicted value is obtained through the self-supervision learning model, then the CVT voltage abnormality score of each timestamp is obtained by combining the predicting model and the reconstructing model, and the CVT voltage abnormality score is compared with a set threshold value to judge whether the CVT voltage abnormality is abnormal or not, so that the insulation condition in the CVT can be accurately evaluated in real time.
Aiming at the fact that the traditional CVT internal insulation abnormality online assessment method mainly depends on experience rules and priori knowledge and lacks the capability of independently excavating effective characteristics, the method can effectively eliminate the influence of primary voltage fluctuation and voltage active change on the CVT internal insulation state online assessment, and the CVT secondary voltage abnormality characteristic is automatically excavated to achieve accurate assessment of the CVT internal insulation state.
Drawings
Fig. 1 is a schematic diagram of a self-supervision learning model structure.
Detailed Description
The invention is illustrated in further detail below with reference to examples.
A CVT internal insulation abnormality online assessment method based on self-supervision learning comprises the following steps:
s1: collecting CVT voltage measurement values, constructing a CVT voltage measurement value sequence, carrying out standardization, and dividing the standardized CVT voltage measurement value sequence into training data and test data; the training data is converted into a time sequence window to be used as a training set, and the test data is converted into a time sequence window to be used as a test set.
Collecting CVT voltage measurement values of which the k branch lines are in phase on the same bus, taking the CVT voltage measurement value of which the k branch lines are in phase on each bus as a characteristic, constructing a CVT voltage measurement value sequence, and taking the CVT voltage measurement value sequenceAnd (3) carrying out standardization, wherein the standardization formula is as follows:
wherein ,for timestamp->CVT voltage measurement, ±>For timestamp->Is a normalized data of CVT voltage measurements of (a),is the CVT voltage measurement sequence minimum, and +.>Is the CVT voltage measurement sequence maximum; r is a non-zero constant vector to prevent zero as a divisor,/o>T is the number of time stamps.
Because the self-supervision learning model is sensitive to irregularities and abnormal conditions in the training data, detecting abnormal timestamps in each normalized CVT voltage measurement value sequence in the training data by using a spectral residual algorithm, and replacing CVT voltage measurement value normalization data with abnormal timestamps by using CVT voltage measurement value normalization data with adjacent timestamps; and converting the normalized CVT voltage measurement value sequence into a time sequence windowTime series window->Is +.>Step size e, thereby obtaining w local context windows of the normalized CVT voltage measurement value sequence, i.e，/>Is->A local context window->The method comprises the steps of carrying out a first treatment on the surface of the For the purpose of->CVT Voltage measurement normalization data at +.>Modeling the dependency of (2) with length +.>Is represented as:
for a pair ofConverts the input normalized CVT voltage measurement sequence into a time sequence windowOne length is +.>Is added to the normalized CVT voltage measurement sequence +.>To keep each time series window length +.>。
The time series window through the transformation allows the self-supervised learning model to use its local context instead of the independent vector to give the data points.
S2: establishing a self-supervision learning model, optimizing super parameters of the self-supervision learning model through a dung beetle optimization algorithm, training the self-supervision learning model by using a training set, and outputting a CVT voltage predicted value and a reconstruction probability through the trained self-supervision learning model;
referring to fig. 1, the self-supervised learning model includes a self-encoder, a parallel graph attention network (GATv 2), a full connection layer, a gated loop network (GRU), a prediction model, and a reconstruction model; the training set is input into a self-encoder, the residual sequence output by the self-encoder is used as input of a parallel graph attention network, then the dependency relationship of the residual sequence on time and characteristic dimension is captured through the parallel graph attention network, after the characteristics and time relationship of the parallel graph attention network and the residual sequence are input into a full-connection layer, the characteristic of the normalized CVT voltage measured value sequence is captured by feeding a gating loop network, and the output of the gating loop network is fed into a prediction model and a reconstruction model to obtain CVT voltage predicted values and reconstruction probabilities.
In this embodiment, the method for obtaining the residual sequence is: the self-encoder learns the effective encoding of a set of data in an unsupervised manner forThe activity value of the intermediate hidden layer of the self-encoder is +.>Is encoded by (a), namely:
the output data from the encoder are:
by passing throughRepresenting a residual sequence of error change information; wherein->Weight matrix for middle hidden layer, +.>Bias term for middle hidden layer, +.>Weight matrix for output layer of self-encoder, < ->Is a bias term from the output layer of the encoder, < ->To activate the function +.>For->Encoded data.
In this embodiment, the dependency of the residual sequence on time and feature dimensions is captured by a parallel graph attention network and />The input to the parallel graph attention network is a set of node feature vectors +.>Expressed as，/>Representing node->Feature vector of>,/>Is the number of nodes, equal to the number of branches, wherein +.>Representing the dimension of the feature vector for each node; the output of the parallel graph attention network is a new set of node feature vectors, i.e. +.>，/>Representing nodes after parallel graph attention network processing +.>Wherein,/>Representing nodes after parallel graph attention network processing +.>Is a dimension of the feature vector of (a); the parallel graph attention network processing procedure is as follows:
where sigmoid represents the activation function,representation and node->Feature vector set of neighboring nodes, +.>Representation and node->Adjacent node->Feature vector of>For node->And node->Attention score between->Is a learnable column vector.
In this embodiment, the method for obtaining the CVT voltage prediction value and the reconstruction probability includes: the CVT voltage predicted value is obtained through a predicted model, and the probability is reconstructed through a reconstruction model; the prediction model predicts the CVT voltage value of the next time stamp through the CVT voltage value of the previous time sequence window, and the reconstruction model is used for analyzing the data distribution of the whole time sequence window; the loss function is, wherein ,/>Indicating total loss->Representing the loss of the predictive model, +.>Representing the loss of the reconstructed model;
wherein ,representation->Is>CVT voltage measurement normalization data for each feature; />Representing a time stamp->Is the first of (2)CVT voltage predictions for each feature; />For vector representation in potential space, +.>For input, & lt + & gt>In order to obtain the posterior density of the particles,for edge density +.>Representing parallel operation; />A reconstruction model for approximating the posterior distribution;representing the expected negative log likelihood of a given input, +.>Representing encoder distribution-> and />Kullback-Leibler divergence.
S3: and establishing a CVT internal insulation state abnormality detection strategy according to the CVT voltage predicted value and the reconstruction probability, and detecting the abnormal state of the CVT voltage measured value.
Calculating a time stamp by a predictive modelCVT voltage prediction value +.>Obtaining a time stamp by reconstructing the model +.>Reconstruction probability +.>，/>Representing a time stamp->Is>Reconstruction probability of individual features; time stampCVT voltage abnormality score of +.>：
wherein ,the self-supervision learning model super-parameters are used for balancing errors of the prediction model and reconstruction probability of the reconstruction model; according to the characteristics of primary CVT voltage fluctuation, combining a prediction model and a reconstruction model to obtain CVT voltage abnormality score +.>Determining an optimal threshold value by using an adaptive threshold value method; and outputting an abnormality label when the CVT voltage abnormality score corresponding to the detected CVT voltage measured value exceeds an optimal threshold.
Inputting the test set into a self-supervision learning model which is trained by the training set and optimized by the dung beetle optimization algorithm, and outputting an abnormal label of the test set. When the actual online evaluation is carried out, according to the characteristics of the CVT in the actual operation system when the internal insulation state is abnormal, a single abnormal cutting strategy is adopted to optimize the output abnormal label, namely when the abnormality of a single time stamp occurs, the corresponding CVT voltage predicted value is changed into a normal label, and the false positive rate can be reduced.
In this embodiment, the super parameter of the self-supervision learning model is used as the optimizing variable of the dung beetle optimizing algorithm, and the optimal super parameter of the self-supervision learning model is obtained through iteration by updating the individual position and the fitness value in the population, and the process is as follows:
s21: initializing a population, randomly generating N individuals, and enabling each individual to express a potential solution;
s22: each individual is subjected to fitness evaluation, and the population is divided into four sub-populations according to fitness values: rolling ball dung beetles, egg ball dung beetles, small dung beetles and stealing dung beetles;
s23: performing the corresponding operations on each sub-population:
ball dung beetle: rolling the faecal ball along a given direction, navigating by using celestial bodies, climbing onto the faecal ball to dance when encountering an obstacle, and re-determining the direction;
egg ball dung beetle: spawning near the locally optimal solution to form egg balls, and providing safe environment and food for offspring;
small dung beetles: finding optimal food source positions, namely globally optimal solutions, from underground drilling to find food;
theft of dung beetles: stealing dung balls from other dung beetles, namely a local optimal solution or a global optimal solution;
s24: updating individual positions and fitness values in the population, and recording the current optimal solution;
s25: judging whether a termination condition is reached, wherein the termination condition is as follows: maximum iteration number or a preset error threshold; if the termination condition is reached, outputting an optimal solution and ending the algorithm; if the termination condition is not reached, returning to step S23, and if the termination condition is reached, outputting the optimal solution and ending the algorithm.
The embodiment also provides a CVT internal insulation abnormality online evaluation system based on self-supervision learning, which comprises a data acquisition module, a data preprocessing module, a voltage prediction module and an abnormality judgment module; the data acquisition module is used for acquiring CVT voltage measurement values; the data preprocessing module is used for constructing and normalizing a CVT voltage measurement value sequence, dividing the normalized CVT voltage measurement value sequence into training data and test data, converting the training data into a time sequence window as a training set, and converting the test data into a time sequence window as a test set; the voltage prediction module is internally provided with a self-supervision learning model and is used for outputting a CVT voltage predicted value and a reconstruction probability; and the abnormality judgment module establishes an abnormality detection strategy of the insulating state in the CVT according to the CVT voltage predicted value and the reconstruction probability, and detects the abnormal state of the CVT voltage measured value.
In another embodiment, a non-volatile computer storage medium is provided, the computer storage medium storing computer executable instructions capable of executing the CVT internal insulation anomaly online assessment method based on self-supervised learning in any of the above embodiments.
The present embodiment also provides a computer program product comprising a computer program stored on a non-volatile computer storage medium, the computer program comprising program instructions which, when executed by a computer, cause the computer to perform the CVT internal insulation abnormality online assessment method based on self-supervised learning of the above embodiments.
The present embodiment provides an electronic device including: the system comprises at least one processor and a memory communicatively connected with the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a CVT internal insulation anomaly online assessment method based on self-supervised learning.
The above-described specific embodiments further illustrate the objects, technical solutions and technical effects of the present invention in detail. It should be understood that the foregoing is only illustrative of the present invention and is not intended to limit the scope of the invention, and that all equivalent changes and modifications that may be made by those skilled in the art without departing from the spirit and principles of the invention shall fall within the scope of the invention.
Claims (9)
1. The CVT internal insulation abnormality online assessment method based on self-supervision learning is characterized by comprising the following steps of:
s1: collecting CVT voltage measurement values, constructing a CVT voltage measurement value sequence, carrying out standardization, and dividing the standardized CVT voltage measurement value sequence into training data and test data; converting the training data into a time sequence window to serve as a training set, and converting the test data into a time sequence window to serve as a test set;
s2: establishing a self-supervision learning model, optimizing super parameters of the self-supervision learning model through a dung beetle optimization algorithm, training the self-supervision learning model by using a training set, and outputting a CVT voltage predicted value and a reconstruction probability through the trained self-supervision learning model; the self-supervision learning model comprises a self-encoder, a parallel graph attention network, a full-connection layer, a gating circulation network, a prediction model and a reconstruction model; inputting a training set into a self-encoder, taking a residual sequence output by the self-encoder as input of a parallel graph attention network, capturing the dependency relationship of the residual sequence on time and feature dimensions through the parallel graph attention network, inputting the features and time relationship from the parallel graph attention network and the residual sequence into a full-connection layer, feeding a gating loop network to capture the features of the standardized CVT voltage measurement value sequence, and feeding the output of the gating loop network into a prediction model and a reconstruction model to obtain a CVT voltage predicted value and reconstruction probability;
taking the super-parameters of the self-supervision learning model as optimizing variables of a dung beetle optimizing algorithm, and iteratively obtaining the optimal super-parameters of the self-supervision learning model by updating individual positions and fitness values in the population:
s21: initializing a population, randomly generating N individuals, and enabling each individual to express a potential solution;
s22: each individual is subjected to fitness evaluation, and the population is divided into four sub-populations according to fitness values: rolling ball dung beetles, egg ball dung beetles, small dung beetles and stealing dung beetles;
s23: performing the corresponding operations on each sub-population:
ball dung beetle: rolling the faecal ball along a given direction, navigating by using celestial bodies, climbing onto the faecal ball to dance when encountering an obstacle, and re-determining the direction;
egg ball dung beetle: spawning near the locally optimal solution to form egg balls, and providing safe environment and food for offspring;
small dung beetles: finding optimal food source positions, namely globally optimal solutions, from underground drilling to find food;
theft of dung beetles: stealing dung balls from other dung beetles, namely a local optimal solution or a global optimal solution;
s24: updating individual positions and fitness values in the population, and recording the current optimal solution;
s25: judging whether a termination condition is reached, wherein the termination condition is as follows: maximum iteration number or a preset error threshold; if the termination condition is reached, outputting an optimal solution and ending the algorithm; if the termination condition is not reached, returning to the step S23, and outputting the optimal solution and ending the algorithm if the termination condition is reached;
s3: and establishing a CVT internal insulation state abnormality detection strategy according to the CVT voltage predicted value and the reconstruction probability, and detecting the abnormal state of the CVT voltage measured value.
2. The online assessment method of CVT internal insulation abnormality based on self-supervision learning according to claim 1, wherein in step S1, CVT voltage measurement values of k branch lines in phase on the same bus are collected, CVT voltage measurement values of each CVT branch line in phase are taken as a feature, a CVT voltage measurement value sequence is constructed, and the CVT voltage measurement value sequence is obtainedAnd (3) carrying out standardization, wherein the standardization formula is as follows:
（1）；
wherein ,for timestamp->CVT voltage measurement, ±>For timestamp->Is a normalized data of CVT voltage measurements of (a),is the CVT voltage measurement sequence minimum, and +.>Is the CVT voltage measurement sequence maximum; r is a non-zero constant vector, < >>T is the number of time stamps.
3. The CVT internal insulation abnormality online assessment method based on self-supervised learning as claimed in claim 2, wherein the process of converting training data or test data into time series windows is: detecting abnormal time stamps in each normalized CVT voltage measurement value sequence in the training data by using a spectral residual algorithm, and replacing CVT voltage measurement value normalization data with adjacent time stamps by using the CVT voltage measurement value normalization data with the abnormal time stamps; and converting the normalized CVT voltage measurement value sequence into a time sequence windowTime series window->Length of (2)Is->Step size e, thus obtaining w local context windows of the normalized CVT voltage measurement sequence, i.e.>，/>Is->A local context window->The method comprises the steps of carrying out a first treatment on the surface of the For the purpose of->CVT Voltage measurement normalization data at +.>Modeling the dependency of (2) with length +.>Is represented as:
（2）；
for a pair ofConverting the input normalized CVT voltage measurement value sequence into a time sequence window +.>One length is +.>Is added to the normalized CVT voltage measurement sequence +.>To keep each time series window length +.>。
4. The CVT internal insulation abnormality online assessment method based on self-supervised learning as claimed in claim 3, wherein the method for obtaining the residual sequence is as follows:
the self-encoder learns the effective encoding of a set of data in an unsupervised manner forThe activity value of the intermediate hidden layer of the self-encoder is +.>Is encoded by (a), namely:
（3）；
the output data from the encoder are:
（4）；
by passing throughRepresenting a residual sequence of error change information; wherein->Weight matrix for middle hidden layer, +.>Bias term for middle hidden layer, +.>Weight matrix for output layer of self-encoder, < ->Is a bias term from the output layer of the encoder, < ->To activate the function +.>For->Encoded data.
5. The online assessment method of CVT internal insulation abnormality based on self-supervised learning as claimed in claim 4, wherein the dependency of the residual sequence on time and feature dimension is captured through a parallel graph attention networkAndthe input to the parallel graph attention network is a set of node feature vectors +.>Expressed as->，Representing node->Feature vector of>,/>Is the number of nodes, equal to the number of branches, wherein +.>Representing the dimension of the feature vector for each node; the output of the parallel graph attention network is a new set of node feature vectors, i.e，/>Representing nodes after parallel graph attention network processing +.>Wherein>,/>Representing nodes after parallel graph attention network processing +.>Is a dimension of the feature vector of (a); the parallel graph attention network processing procedure is as follows:
（5）；
wherein sigmoidThe activation function is represented as a function of the activation,representation and node->Feature vector set of neighboring nodes, +.>Representation and node->Adjacent node->Feature vector of>For node->And node->Attention score between->Is a learnable column vector.
6. The online CVT internal insulation abnormality assessment method based on self-supervised learning as claimed in claim 5, wherein the method for obtaining CVT voltage prediction values and reconstruction probabilities is as follows: the CVT voltage predicted value is obtained through a predicted model, and the probability is reconstructed through a reconstruction model; the prediction model predicts the CVT voltage value of the next time stamp through the CVT voltage value of the previous time sequence window, and the reconstruction model is used for analyzing the data distribution of the whole time sequence window; the loss function is, wherein ,/>Indicating total loss->Representing the loss of the predictive model, +.>Representing the loss of the reconstructed model;
（6）；
（7）；
wherein ,representation->Is>CVT voltage measurement normalization data for each feature; />Representing a time stamp->Is>CVT voltage predictions of individual characteristics; ->For vector representation in potential space, +.>For input, & lt + & gt>In order to obtain the posterior density of the particles,for edge density +.>Representing parallel operation; />A reconstruction model for approximating the posterior distribution;representing the expected negative log likelihood of a given input, +.>Representing encoder distribution-> and />Kullback-Leibler divergence.
7. The CVT internal insulation abnormality online assessment method based on self-supervised learning as claimed in claim 6, wherein the process of step S3 is as follows:
calculating a time stamp by a predictive modelCVT voltage prediction value +.>Obtaining a time stamp by reconstructing the model +.>Reconstruction probability +.>，/>Representing a time stamp->Is>Reconstruction probability of individual features; timestamp->CVT voltage abnormality score of +.>：
（8）；
wherein ,the self-supervision learning model super-parameters are used for balancing errors of the prediction model and reconstruction probability of the reconstruction model; according to the characteristics of primary CVT voltage fluctuation, combining a prediction model and a reconstruction model to obtain CVT voltage abnormality score +.>Determining an optimal threshold value by using an adaptive threshold value method; and outputting an abnormality label when the CVT voltage abnormality score corresponding to the detected CVT voltage measured value exceeds an optimal threshold.
8. The online CVT internal insulation abnormality assessment method based on self-supervision learning according to claim 1, wherein a single abnormality clipping strategy is adopted according to characteristics of a real operation system when internal insulation state abnormality occurs in the CVT, and an output abnormal label is optimized, namely when abnormality of a single time stamp occurs, a corresponding CVT voltage predicted value is changed into a normal label.
9. A CVT internal insulation anomaly online assessment system based on self-supervised learning for implementing the method of any one of claims 1-8, comprising a data acquisition module, a data preprocessing module, a voltage prediction module, and an anomaly determination module; the data acquisition module is used for acquiring CVT voltage measurement values; the data preprocessing module is used for constructing and normalizing a CVT voltage measurement value sequence, dividing the normalized CVT voltage measurement value sequence into training data and test data, converting the training data into a time sequence window as a training set, and converting the test data into a time sequence window as a test set; the voltage prediction module is internally provided with a self-supervision learning model and is used for outputting a CVT voltage predicted value and a reconstruction probability; and the abnormality judgment module establishes an abnormality detection strategy of the insulating state in the CVT according to the CVT voltage predicted value and the reconstruction probability, and detects the abnormal state of the CVT voltage measured value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310586721.4A CN116304846B (en) | 2023-05-24 | 2023-05-24 | CVT internal insulation abnormality online assessment method based on self-supervision learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310586721.4A CN116304846B (en) | 2023-05-24 | 2023-05-24 | CVT internal insulation abnormality online assessment method based on self-supervision learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116304846A CN116304846A (en) | 2023-06-23 |
| CN116304846B true CN116304846B (en) | 2023-09-12 |
Family
ID=86789167
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310586721.4A Active CN116304846B (en) | 2023-05-24 | 2023-05-24 | CVT internal insulation abnormality online assessment method based on self-supervision learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116304846B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113447783A (en) * | 2021-08-30 | 2021-09-28 | 武汉格蓝若智能技术有限公司 | Voltage transformer insulation fault identification model construction method and device |
| CN114757305A (en) * | 2022-06-13 | 2022-07-15 | 华中科技大学 | Voltage transformer insulation fault identification method and system based on ensemble learning |
| WO2022160902A1 (en) * | 2021-01-28 | 2022-08-04 | 广西大学 | Anomaly detection method for large-scale multivariate time series data in cloud environment |
| CN115275991A (en) * | 2022-07-29 | 2022-11-01 | 国网上海市电力公司 | Active power distribution network operation situation prediction method based on IEMD-TA-LSTM model |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10606353B2 (en) * | 2012-09-14 | 2020-03-31 | Interaxon Inc. | Systems and methods for collecting, analyzing, and sharing bio-signal and non-bio-signal data |
| US11448671B2 (en) * | 2019-09-24 | 2022-09-20 | General Electric Company | Signature identification for power system events |
| CN112653142B (en) * | 2020-12-18 | 2022-09-02 | 武汉大学 | Wind power prediction method and system for optimizing depth transform network |
-
2023
- 2023-05-24 CN CN202310586721.4A patent/CN116304846B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022160902A1 (en) * | 2021-01-28 | 2022-08-04 | 广西大学 | Anomaly detection method for large-scale multivariate time series data in cloud environment |
| CN113447783A (en) * | 2021-08-30 | 2021-09-28 | 武汉格蓝若智能技术有限公司 | Voltage transformer insulation fault identification model construction method and device |
| CN114757305A (en) * | 2022-06-13 | 2022-07-15 | 华中科技大学 | Voltage transformer insulation fault identification method and system based on ensemble learning |
| CN115275991A (en) * | 2022-07-29 | 2022-11-01 | 国网上海市电力公司 | Active power distribution network operation situation prediction method based on IEMD-TA-LSTM model |
Non-Patent Citations (1)
| Title |
|---|
| 魏金萧 ; 周步祥 ; 张冰 ; .综合数据清洗及无监督学习技术的电力设备状态评估.水电能源科学.2016,(09),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116304846A (en) | 2023-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Belaout et al. | Multiclass adaptive neuro-fuzzy classifier and feature selection techniques for photovoltaic array fault detection and classification | |
| US10955456B2 (en) | Method and apparatus for automatic localization of a fault | |
| CN110221225B (en) | Spacecraft lithium ion battery cycle life prediction method | |
| CN111914873A (en) | Two-stage cloud server unsupervised anomaly prediction method | |
| CN109034191B (en) | ELM-based one-dimensional telemetry data abnormal interpretation method | |
| Yu | Health degradation detection and monitoring of lithium-ion battery based on adaptive learning method | |
| Singh et al. | Role of artificial intelligence in the reliability evaluation of electric power systems | |
| Jiang et al. | A fault diagnosis method for electric vehicle power lithium battery based on wavelet packet decomposition | |
| Mustafa et al. | Fault identification for photovoltaic systems using a multi-output deep learning approach | |
| CN116244617A (en) | Intelligent high-voltage switch cabinet fault diagnosis method and system based on heterogeneous graph structure learning | |
| CN113409166A (en) | XGboost model-based method and device for detecting abnormal electricity consumption behavior of user | |
| CN115859058A (en) | UPS (uninterrupted Power supply) fault prediction method and system based on width learning network | |
| CN107862375A (en) | A kind of two stage equipment fault diagnosis method | |
| CN116777088B (en) | Power supply emergency repair environment monitoring method and system for guaranteeing life safety | |
| CN115878992A (en) | Monitoring method and monitoring system for comprehensive pipe rack power supply system | |
| Liu et al. | State of health estimation of lithium-ion batteries based on multi-feature extraction and temporal convolutional network | |
| CN116840765B (en) | Voltage transformer error state evaluation method based on multivariate time sequence analysis | |
| CN116304846B (en) | CVT internal insulation abnormality online assessment method based on self-supervision learning | |
| Edelen et al. | Anomaly detection in particle accelerators using autoencoders | |
| Zhou et al. | Real-time anomaly detection in distribution grids using long short term memory network | |
| ul Hassan et al. | Online static security assessment for cascading failure using stacked De-noising Auto-encoder | |
| Pan et al. | Assessment of MV XLPE cable aging state based on PSO-XGBoost algorithm | |
| CN115632887A (en) | Block chain network abnormal data detection method, device and equipment | |
| Liu et al. | Adaptive robustness evaluation for complex system prognostics and health management software platform | |
| Du et al. | Fault diagnosis method of automation equipment in independent and controllable substation based on deep reinforcement learning |
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 |