CN114595883A - Oil-immersed transformer residual life personalized dynamic prediction method based on meta-learning - Google Patents
Oil-immersed transformer residual life personalized dynamic prediction method based on meta-learning Download PDFInfo
- Publication number
- CN114595883A CN114595883A CN202210217435.6A CN202210217435A CN114595883A CN 114595883 A CN114595883 A CN 114595883A CN 202210217435 A CN202210217435 A CN 202210217435A CN 114595883 A CN114595883 A CN 114595883A
- Authority
- CN
- China
- Prior art keywords
- oil
- data
- prediction
- transformer
- network
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 239000007789 gas Substances 0.000 claims description 69
- 238000012360 testing method Methods 0.000 claims description 31
- 230000008569 process Effects 0.000 claims description 25
- 238000012549 training Methods 0.000 claims description 22
- 238000013528 artificial neural network Methods 0.000 claims description 16
- 238000007781 pre-processing Methods 0.000 claims description 15
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 claims description 10
- 238000011478 gradient descent method Methods 0.000 claims description 7
- OTMSDBZUPAUEDD-UHFFFAOYSA-N Ethane Chemical compound CC OTMSDBZUPAUEDD-UHFFFAOYSA-N 0.000 claims description 5
- VGGSQFUCUMXWEO-UHFFFAOYSA-N Ethene Chemical compound C=C VGGSQFUCUMXWEO-UHFFFAOYSA-N 0.000 claims description 5
- 239000005977 Ethylene Substances 0.000 claims description 5
- HSFWRNGVRCDJHI-UHFFFAOYSA-N alpha-acetylene Natural products C#C HSFWRNGVRCDJHI-UHFFFAOYSA-N 0.000 claims description 5
- 125000002534 ethynyl group Chemical group [H]C#C* 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 239000001257 hydrogen Substances 0.000 claims description 4
- 229910052739 hydrogen Inorganic materials 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 125000004122 cyclic group Chemical group 0.000 claims description 2
- 238000012795 verification Methods 0.000 description 8
- 230000036541 health Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000000052 comparative effect Effects 0.000 description 3
- 230000000306 recurrent effect Effects 0.000 description 3
- 239000004215 Carbon black (E152) Substances 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- HYBBIBNJHNGZAN-UHFFFAOYSA-N furfural Chemical compound O=CC1=CC=CO1 HYBBIBNJHNGZAN-UHFFFAOYSA-N 0.000 description 2
- 229930195733 hydrocarbon Natural products 0.000 description 2
- 150000002430 hydrocarbons Chemical class 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000007599 discharging Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 150000002431 hydrogen Chemical class 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013021 overheating Methods 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2462—Approximate or statistical queries
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
-
- 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
-
- 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)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Tourism & Hospitality (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Probability & Statistics with Applications (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Fuzzy Systems (AREA)
- Databases & Information Systems (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Housings And Mounting Of Transformers (AREA)
Abstract
The invention discloses a method for individually predicting the residual life of an oil-immersed transformer based on meta-learning, which comprises the steps of extracting common characteristics and rules in historical data of the transformer by constructing two shared generators, identifying the individual operation characteristics of the transformer when the dissolved gas data in oil of different transformers are obtained, generating an individual prediction model and a discrimination model aiming at each transformer, and respectively predicting the concentration data of the dissolved gas in the future oil of the transformer and the fault occurrence time so as to realize the individual prediction of the residual life of the transformer. The method can improve the accuracy of the prediction of the residual life of the oil-immersed transformer and meet the requirement of individualized prediction of the residual life of the transformer.
Description
Technical Field
The invention belongs to the field of transformer service life analysis, and particularly relates to a meta-learning-based personalized dynamic prediction method for residual service life of an oil-immersed transformer.
Background
The transformer is one of the main devices of the power system, and plays a role of the hub of grid interconnection and power exchange. With the continuous development of the complicated and intelligent power grid system in China, transformer equipment plays an important role in the aspects of extra-high voltage transmission and intelligent power grid regulation, and once a fault occurs, the important loss is caused to the power transmission and the social power utilization of the whole power grid system, so that the research on the fault prediction and the future residual life estimation of the transformer equipment has important significance for formulating equipment inspection and maintenance strategies and guaranteeing the stability and the safety of the power system.
The method can judge and predict whether potential faults (such as overheating faults, discharging faults and the like) exist in transformer equipment under the condition of not disassembling and shutting down the transformer equipment by detecting and analyzing the components and concentration values of hydrocarbon gases such as hydrogen, methane, ethane, ethylene, acetylene and the like dissolved in the transformer insulating oil. In recent years, some researchers have proposed a fault prediction and remaining life estimation method based on dissolved gas in transformer oil. For example, patent application publication No. CN201710193347.6 discloses a hidden markov model-based method for estimating remaining life of a transformer, which uses the hidden markov model to extract typical dynamic characteristics of the transformer from normal state to fault state, and predicts the remaining life by estimating the transition probability of the device state; the patent application with the publication number of CN201610455561.X discloses a health index-based method for detecting the residual life of a transformer, wherein the health index is constructed according to indexes such as the aging coefficient, the carbon and oxygen content and the furfural gas content of the transformer, and the residual life of the transformer is estimated by analyzing the statistical relationship between the health index and the fault time.
The main idea of the existing method in the field is to extract typical fault characteristics or modes of equipment from dissolved gas data in transformer oil through a statistical model or a machine learning model, give fault early warning and estimate the remaining life when the operation characteristics of the equipment to be tested are consistent with the typical characteristics. However, due to the influence of the personalized operation environment of the equipment, the transformer fault is not only dependent on typical characteristics and rules in historical data, but also related to personalized characteristics shown in the process from normal to fault of the equipment. The existing model and method only carry out fault prediction through extracted typical characteristics and statistical rules, and are difficult to describe the influence of the difference between equipment and personalized operation characteristics on fault prediction, so that the phenomena of low prediction accuracy, fault report failure and report omission exist in actual application, and the requirement of personalized prediction of the residual life of the transformer cannot be met.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a meta-learning-based personalized prediction method for the residual life of an oil-immersed transformer, so as to improve the accuracy and the personalization of the prediction of the residual life of the oil-immersed transformer.
In order to achieve the above object, an embodiment of the present invention provides a meta-learning based method for individually predicting a remaining life of an oil-immersed transformer, including the following steps:
(1) collecting concentration data of dissolved gas in oil generated in the process that a fault transformer is changed from a normal state to a fault state, and carrying out data preprocessing to obtain time sequence data serving as a training sample;
(2) constructing two sharing generators with different structures, and training the sharing generators by adopting a meta-learning mode according to training samples so as to optimize parameters of the two sharing generators;
(3) collecting concentration data of dissolved gas in oil at the current time generated in the normal operation process of each normal transformer, and performing data preprocessing to obtain time sequence data as a test sample;
(4) respectively inputting the test samples into two sharing generators for parameter optimization, and respectively taking the two groups of output parameters as network parameters of a prediction network and a judgment network to construct a prediction model and a judgment model for each normal transformer;
(5) predicting the test sample by using a prediction model to obtain concentration data of dissolved gas in oil at a future time, and distinguishing the concentration data of the dissolved gas in the oil at the future time by using a distinguishing model to obtain an operation state distinguishing result;
(6) and when the operation state judgment result of the judgment model is a fault state, subtracting the current time from the time of the fault state to obtain the prediction result of the residual life of the oil-immersed transformer.
Wherein, the gas dissolved in the oil comprises methane, ethane, ethylene, acetylene and hydrogen; the dissolved gas concentration data in the oil comprises continuous measurement data of mass concentration or volume concentration of the dissolved gas in the oil within a given time range; the data preprocessing comprises linear interpolation processing, namely linear interpolation is adopted to replace missing values in the data.
In one embodiment, two structurally distinct shared generators are designated as shared generator Gen (θ)1) And share Generator Gen (θ)2) Wherein generator Gen (θ) is shared1) A cyclic neural network is adopted, the data of the dissolved gas in the oil is used as input to learn the time sequence characteristics and the law of the data of the dissolved gas in the oil in the process that different transformers are changed from a normal state to a fault state, and a group of parameters are output and recorded as alphanRepresentative and shared Generator Gen (θ)1) Prediction network Net (alpha) with same structuren) The network parameter of (2); sharing Generator Gen (θ)2) Adopting a feedforward neural network, taking the dissolved gas data in the oil as input to learn the personalized fault characteristics of each transformer, and outputting another group of parameters to be recorded as betanRepresentative and shared Generator Gen (θ)2) Discrimination network Net (beta) with same structuren) In which θ is1And theta2For two shared generators Gen (theta)1) And share Generator Gen (θ)2) The network parameter of (2).
In step (2) of an embodiment, training the shared generator according to the training samples and by adopting a meta-learning manner, includes:
(a) random initialization shared generator Gen (theta)1) And share generator Gen (θ)2) Parameter theta of1And theta2；
(b) Extracting time sequence data x corresponding to a fault transformer from a training samplemAnd randomly divided into two segments, respectively data xm,1And data xm,2M is the index of the fault transformer;
(c) data xm,1Respectively input to the shared generators Gen (theta)1) And Gen (θ)2) In the method, the two groups of output parameters are respectively used as the network parameters of a prediction network and a judgment network to construct a prediction model and a judgment model；
(d) Data xm,1Inputting the data into a prediction model to obtain prediction data xm,2' and predict data xm,2' inputting the operation state to a discrimination model to obtain an operation state discrimination result;
(e) computing a predictive model with respect to predicted data xm,2' sum data xm,2The prediction error of the discrimination model, and the discrimination error of the discrimination model about the operation state discrimination result and the real state;
(f) calculating discrimination error and prediction error with respect to shared generator Gen (θ)1) And Gen (θ)2) Network parameter θ of1And theta2And applying a gradient descent method to the network parameter theta1And theta2Updating is carried out;
(g) continuously repeating the steps (b) to (f) until the network parameter theta1And theta2And (6) converging.
In step (4) of one embodiment, the network Net (α) is predictedn) Using and sharing generator Gen (theta)1) The circulating neural networks with the same structure are used for predicting according to the input concentration data of the dissolved gas in the oil and outputting the concentration data of the dissolved gas in the oil at the future time; sharing Generator Gen (θ)1) Estimating the predicted Net (alpha) from the test sample output parametersn) Network parameter alpha ofnThe process of (2), comprising: using shared Generator Gen (θ)1) Test sample y corresponding to each normal transformernFitting is carried out, and a fitting error e is calculated; then calculating the network parameter theta1Gradient with respect to fitting error e
And the network parameter theta is measured by using a gradient descent method1Carrying out primary updating; finally, the gradient descending result is used as a network parameter alpha of the prediction networknThereby, a prediction model is constructed.
In step (4) of one embodiment, the Net (β) is discriminatedn) Adapted to share generator Gen (theta)2) The feedforward neural network with the same structure is used for performing the feedback according to the input concentration data of the dissolved gas in the oilJudging the running state, and outputting a running state judgment result; sharing Generator Gen (θ)2) According to the output parameters of the test samples, the actual estimation of the discrimination network Net (beta)n) Network parameter beta ofnThe process of (2), comprising: using shared Generator Gen (θ)2) Test sample y corresponding to each normal transformernPerforming pooling operation to obtain test sample ynCorresponding feature vector znThe feature vector z is then transformed by a non-linear activation functionnNetwork parameter beta mapped as a discriminating networknThus, a discriminant model is constructed.
In step (5) of one embodiment, predicting the test sample by using a prediction model to obtain the dissolved gas concentration data in the oil at a future time comprises:
first predict sample ynInput to the prediction model Net (alpha)n) In (3), based on the prediction sample ynPredicting the concentration data of the dissolved gas in the oil at the next moment; then the prediction sample ynAnd predicted concentration data of dissolved gas in oil and as new input to prediction model Net (alpha)n) And predicting the concentration data of the dissolved gas in the oil at the next moment, and continuously repeating the process to gradually predict the concentration data of the dissolved gas in the oil at the future time.
In step (5) of an embodiment, the operation state determination result output by the determination model is between 0 and 1, and the operation state determination result is less than 0.5 to indicate that the transformer is in a normal state, and more than 0.5 to indicate that the transformer is in a fault state.
Compared with the prior art, the invention has the beneficial effects that at least:
by constructing two shared generators to extract the common characteristics and rules in the historical data of the transformers, when the data of the dissolved gas in the oil of different transformers are obtained, the individualized operation characteristics of the transformers can be identified, individualized prediction models and discrimination models for each transformer can be generated, the concentration data of the dissolved gas in the future oil of the transformers and the fault occurrence time can be respectively predicted, and accordingly individualized prediction of the residual life of the transformers is achieved. The method can improve the accuracy of the prediction of the residual life of the oil-immersed transformer and meet the requirement of individualized prediction of the residual life of the transformer.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flowchart of a method for personalized prediction of remaining life of an oil-immersed transformer based on meta-learning according to an embodiment;
FIG. 2 is a statistical graph of the prediction error of the residual life of the transformer according to the embodiment;
fig. 3 is a graph comparing the prediction errors of the remaining life of the transformer provided by the comparative example.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is a flowchart of a method for personalized prediction of remaining life of an oil-immersed transformer based on meta-learning according to an embodiment. As shown in fig. 1, the personalized prediction method for remaining life of an oil-immersed transformer according to the embodiment includes the following steps:
In the embodiment, the mobile phone collects concentration data of dissolved gas in oil generated in the process that the fault transformer is stopped from running and the fault transformer is changed from a normal state to a fault state, and performs data preprocessing on the concentration data of the dissolved gas in the oil to obtain time series data xm,m＝1,...,M. Where M represents the number of failed transformers collected, xmRepresenting the time series data after the m-th transformer preprocessing. These time-series data xmAs training samples for training the shared generator.
Wherein, the gas dissolved in the oil comprises methane, ethane, ethylene, acetylene and hydrogen; the concentration data of the dissolved gas in the oil refers to continuous measurement data of mass concentration or volume concentration of the dissolved gas in the oil within a given time range, and the measurement interval is preferably 1 day; the data preprocessing comprises linear interpolation processing, namely, linear interpolation is adopted to replace missing values in the data.
And 2, constructing two sharing generators with different structures, and training the sharing generators by adopting a meta-learning mode according to the training samples so as to optimize the network parameters of the two sharing generators.
In the embodiment, the two sharing generators with different structures are operation models for describing the mapping relation between the transformer data and the optimal parameter values of the residual life prediction model, and are respectively marked as sharing generators Gen (theta)1) And share generator Gen (θ)2)，θ1And theta2For two shared generators Gen (theta)1) And share generator Gen (θ)2) The network parameter of (2). Wherein generator Gen (θ) is shared1) The method adopts a circulating neural network, takes the dissolved gas data in the oil as input to learn the time sequence characteristics and the rules presented by the dissolved gas data in the oil in the process of changing the normal state into the fault state of different transformers, and outputs a group of parameters as a generator Gen (theta)1) Prediction network Net (alpha) with same structuren) Network parameter alpha ofn(ii) a Sharing Generator Gen (θ)2) Adopting a feedforward neural network, taking the data of dissolved gas in oil as input, learning the personalized fault characteristics of each transformer, and outputting another group of parameters as a generator Gen (theta)2) Discrimination network Net (beta) with same structuren) Network parameter beta ofn。
In an embodiment, the prediction network Net (α)n) Adapted to share generator Gen (theta)1) Circulation spirit with same structureAnd the network is used for predicting according to the input concentration data of the dissolved gas in the oil and outputting the concentration data of the dissolved gas in the oil at the future moment.
In the embodiment, the discrimination network Net (β)n) Adapted to share generator Gen (theta)2) And the front feedback neural networks with the same structure are used for judging the running state according to the input concentration data of the dissolved gas in the oil and outputting a running state judgment result.
Generator Gen (theta) sharing based on the above1) And share generator Gen (θ)2) According to the training sample and adopting meta learning mode to share generator Gen (theta)1) And share generator Gen (θ)2) Performing training, including:
(a) initializing the shared Generator Gen (θ)1) And share Generator Gen (θ)2) Network parameter θ of1And theta2；
(b) Extracting time sequence data x corresponding to a fault transformer from a training samplemAnd randomly divided into two segments, respectively data xm,1And data xm,2；
(c) Data xm,1Respectively input to the shared generators Gen (theta)1) And Gen (θ)2) In (2), the two output parameters are respectively used as the prediction network Net (alpha)n) And discriminating network Net (beta)n) To construct a prediction model and a discrimination model;
(d) data xm,1Inputting the data into a prediction model to obtain prediction data xm,2' and predict data xm,2' inputting the operation state to a discrimination model to obtain an operation state discrimination result;
(e) computing a predictive model with respect to predicted data xm,2' sum data xm,2The prediction error of the discrimination model, and the discrimination error of the discrimination model about the operation state discrimination result and the real state;
(f) calculating discrimination error and prediction error with respect to shared generator Gen (θ)1) And Gen (θ)2) Network parameter θ of1And theta2And applying a gradient descent method to the network parameter theta1And theta2Updating is carried out;
(g) continuously repeating the steps (b) to (f) until the network parameter theta1And theta2And (6) converging.
Trained shared generator Gen (theta)1) And Gen (θ)2) The method hides the common characteristics and rules in the historical data of the transformers, and provides an accurate basis for the follow-up generation of the personalized prediction model and the discrimination model for each transformer.
And 3, collecting concentration data of dissolved gas in oil at the current time generated in the normal operation process of each normal transformer, and performing data preprocessing to obtain time sequence data as a test sample.
In the embodiment, the collected concentration data of the dissolved gas in the oil at the current time generated in the normal operation process of each normal transformer is subjected to data preprocessing to form time series data ynN1, N represents the number of transformers that are in operation. Note that the same method as that in step 1, i.e., a linear interpolation processing method, is used for data preprocessing.
And 4, respectively inputting the test samples into two sharing generators for parameter optimization, and respectively taking the two groups of output parameters as network parameters of a prediction network and a judgment network so as to construct a prediction model and a judgment model for each normal transformer.
In an embodiment, generator Gen (θ) is shared1) Estimating the predicted Net (alpha) from the test sample output parametersn) Network parameter alpha ofnThe process of (2), comprising: using shared Generator Gen (θ)1) Test sample y corresponding to each normal transformernFitting is carried out, and a fitting error e is calculated; then calculating the network parameter theta1Gradient with respect to fitting error e
And the gradient descent method is utilized to carry out on the network parameter theta1Carrying out primary updating; finally, the gradient descending result is used as a network parameter alpha of the prediction networknConstructing a prediction model by the method, wherein the updating rate of the parameters in the gradient descent method is set asρ, predicted network parameter set αnThe calculation process of (a) can be expressed as:
in an embodiment, generator Gen (θ) is shared2) According to the output parameters of the test samples, the actual estimation of the discrimination network Net (beta)n) Network parameter beta ofnThe process of (2), comprising: using shared Generator Gen (θ)2) Test sample y corresponding to each normal transformernPerforming pooling operation to obtain test sample ynCorresponding feature vector znThe feature vector z is then transformed by a non-linear activation functionnNetwork parameter beta mapped as a discriminating networknThus, a discriminant model is constructed. Specifically, the discrimination network Net (β)n) Network parameter beta ofnThe calculation process of (a) can be expressed as:
zn＝pool(θ2,1,yn)
βn＝φ(θ2,2,zn)
wherein, theta2,1And theta2,2Representative shared Generator Gen (θ)2) Parameter of (a), theta2＝{θ2,1,θ2,2}; pool (·) represents a pooling function, returning a matrix θ2,1×ynThe maximum value of each column; phi (-) represents a non-linear activation function of the functional form
When the prediction model and the prediction model are constructed, parameters are independently generated for sample data of each transformer to construct the prediction model and the prediction model for each transformer, so that the individualized residual life prediction of each transformer can be realized.
The process of constructing the prediction model and the discriminant model in step (c) is the same as that in step 4.
And 5, predicting the test sample by using the prediction model to obtain concentration data of the dissolved gas in the oil at the future time, and judging the concentration data of the dissolved gas in the oil at the future time by using the judgment model to obtain an operation state judgment result.
In an embodiment, predicting the test sample by using the prediction model to obtain the concentration data of the dissolved gas in the oil at a future time includes:
first predict sample ynInput to the prediction model Net (alpha)n) In (3), based on the prediction sample ynPredicting the concentration data of the dissolved gas in the oil at the next moment; then the prediction sample ynAnd predicted concentration data of dissolved gas in oil and as new input to prediction model Net (alpha)n) And predicting the concentration data of the dissolved gas in the oil at the next moment, and continuously repeating the process to gradually predict the concentration data of the dissolved gas in the oil at the future time.
In the embodiment, when the concentration data of the dissolved gas in the oil at the future time is discriminated by the discrimination model, the prediction model Net (α) is first determinedn) Inputting the future dissolved gas data prediction result of the transformer into a discrimination model Net (beta)n) Then, nonlinear mapping is carried out on input prediction data through a forward algorithm of a feedforward neural network to obtain a discrimination model Net (beta)n) And outputting the operation state judgment result. Net (beta)n) The output operation state judgment result is between 0 and 1, the operation state judgment result is smaller than 0.5 to represent that the transformer is in a normal state, and the operation state judgment result is larger than 0.5 to represent that the transformer is in a fault state.
And 6, when the operation state judgment result of the judgment model is a fault state, subtracting the current time from the time of the fault state to obtain the prediction result of the residual life of the oil-immersed transformer.
In the embodiment, the corresponding prediction time is recorded by using the concentration data of the dissolved gas in the oil at the future time of the prediction model, so that when the operation state judgment result of the judgment model is in the fault state, the current time is subtracted from the time in the fault state to obtain the prediction result of the residual life of the oil-immersed transformer.
Examples of the experiments
In this embodiment, 10 failed 220kv oil-immersed transformers are selected, and the concentration data of dissolved gas in the oil in the whole process from normal state to failure state of these devices is collected, where the gas indexes include six types of gas, methane, ethane, ethylene, acetylene, and total hydrocarbon. According to step 1, the embodiment performs linear interpolation preprocessing on the collected data, and obtains 7348 pieces of data after preprocessing.
Then, according to steps 2 and 3, two generators are built by using the preprocessed data, an individualized prediction model and a discrimination model are generated by combining data of each transformer, the remaining life of each transformer is predicted according to step S04, and the prediction result is verified according to the actual failure time of the transformer. In order to test the reliability of the method of the present invention, the present embodiment verifies the device prediction result in a leave-one-verify manner. And (3) a verification method is reserved for testing each transformer once, the selected equipment in each test is used as a test sample for predicting the residual service life, and the rest transformers are used as training samples for training the network generator model. The verification was performed 10 times, and 100 pieces of data were available for remaining life prediction for each tested transformer. The error of the prediction result is evaluated by an average percentage error index, and the specific calculation mode is as follows:
err＝|Tpre–Treal|/Treal
where err represents the average percent error of the time to failure prediction, TrealAnd TpreRespectively representing the real result and the predicted result of the residual life of the transformer.
Fig. 2 shows the statistical result of the prediction error of the method provided by the present invention in 10 verification tests, and it can be seen that the personalized prediction method for the remaining life of the transformer described in this embodiment has a higher prediction accuracy, the average of the prediction errors for the remaining life of 10 transformer devices is 0.167, and the variance is 0.085, so that the accurate and reliable prediction can be made for the remaining life of the transformer.
Comparative example
In order to illustrate the advantages of the method for predicting the residual life of the transformer, compared with the traditional method, the comparative example adopts the traditional feedforward neural network model and the recurrent neural network model to respectively build a model for predicting the residual life of the transformer, and uses the same transformer data as the embodiment to carry out verification. And the verification is carried out by adopting a leave-one-verification method, the verification is carried out for 10 times totally, the residual life of one device is predicted every time, and the rest transformers are used for building a model. 100 pieces of data of the tested transformer can be used for residual life prediction each time, and the error of the prediction result is evaluated by the average percentage error index. The comparison results of the average percentage errors of the conventional feedforward neural network, the recurrent neural network and the method provided by the invention are shown in fig. 3.
As can be seen from fig. 3, the average prediction error of the proposed method is 16.7 in 10 verifications, which is significantly lower than the average prediction errors (38.6% and 24.4%) of the conventional feedforward neural network and recurrent neural network methods. According to the method provided by the invention, the personalized difference of the degradation process of the health state of the transformer is considered, and the prior art such as meta-learning and generator is adopted to optimize and adjust the traditional model, so that higher prediction precision can be obtained.
The above-mentioned embodiments are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions, equivalents, etc. made within the scope of the principles of the present invention should be included in the scope of the present invention.
Claims (8)
1. The oil-immersed transformer residual life personalized prediction method based on meta-learning is characterized by comprising the following steps:
(1) collecting concentration data of dissolved gas in oil generated in the process that a fault transformer is changed from a normal state to a fault state, and carrying out data preprocessing to obtain time sequence data serving as a training sample;
(2) constructing two sharing generators with different structures, and training the sharing generators by adopting a meta-learning mode according to training samples so as to optimize parameters of the two sharing generators;
(3) collecting concentration data of dissolved gas in oil at the current time generated in the normal operation process of each normal transformer, and performing data preprocessing to obtain time sequence data as a test sample;
(4) respectively inputting the test samples into two sharing generators for parameter optimization, and respectively taking the two groups of output parameters as network parameters of a prediction network and a judgment network to construct a prediction model and a judgment model for each normal transformer;
(5) predicting the test sample by using a prediction model to obtain concentration data of dissolved gas in oil at a future time, and distinguishing the concentration data of the dissolved gas in the oil at the future time by using a distinguishing model to obtain an operation state distinguishing result;
(6) and when the operation state judgment result of the judgment model is a fault state, subtracting the current time from the time of the fault state to obtain the prediction result of the residual life of the oil-immersed transformer.
2. The oil-immersed transformer residual life personalized prediction method based on meta-learning according to claim 1, characterized in that dissolved gas in oil comprises methane, ethane, ethylene, acetylene, hydrogen; the dissolved gas concentration data in the oil comprises continuous measurement data of mass concentration or volume concentration of the dissolved gas in the oil within a given time range; the data preprocessing comprises linear interpolation processing, namely linear interpolation is adopted to replace missing values in the data.
3. Oil-immersed transformer residual life personalized prediction method based on meta-learning according to claim 1, characterized in that two structurally different shared generators are marked as shared generator Gen (θ)1) And share generator Gen (θ)2) Wherein generator Gen (θ) is shared1) A cyclic neural network is adopted, the data of the dissolved gas in the oil is used as input to learn the time sequence characteristics and the law of the data of the dissolved gas in the oil in the process that different transformers are changed from a normal state to a fault state, and a group of parameters are output and recorded as alphanRepresentative and shared Generator Gen (θ)1) Prediction network Net (alpha) with same structuren) The network parameter of (2); sharing Generator Gen (θ)2) Adopting a feedforward neural network, taking the dissolved gas data in the oil as input to learn the personalized fault characteristics of each transformer, and outputting another group of parameters to be recorded as betanRepresentative and shared Generator Gen (θ)2) Discrimination network Net (beta) with same structuren) Of network parameters of, wherein θ1And theta2For two shared generators Gen (theta)1) And share generator Gen (θ)2) The parameter (c) of (c).
4. The oil-immersed transformer residual life personalized prediction method based on meta-learning according to claim 3, wherein in the step (2), the training of the shared generator is performed by adopting a meta-learning mode according to the training samples, and the method comprises the following steps:
(a) random initialization shared generator Gen (theta)1) And share generator Gen (θ)2) Parameter theta of1And theta2；
(b) Extracting time sequence data x corresponding to a fault transformer from a training samplemAnd randomly divided into two segments, respectively data xm,1And data xm,2M is the index of the fault transformer;
(c) data xm,1Respectively input to the shared generators Gen (theta)1) And Gen (θ)2) The two output groups of parameters are respectively used as network parameters of a prediction network and a judgment network to construct a prediction model and a judgment model;
(d) data xm,1Inputting the data into a prediction model to obtain prediction data xm,2' and predict data xm,2' inputting the operation state to a discrimination model to obtain an operation state discrimination result;
(e) computing a predictive model with respect to predicted data xm,2' sum data xm,2The prediction error of the discrimination model, and the discrimination error of the discrimination model about the operation state discrimination result and the real state;
(f) calculating discrimination error and prediction error with respect to shared generator Gen (θ)1) And Gen (θ)2) Network parameter θ of1And theta2And applying a gradient descent method to the network parameter theta1And theta2Updating is carried out;
(g) continuously repeating the steps (b) to (f) until the network parameter theta1And theta2And (6) converging.
5. The oil-filled transformer residual life personalized prediction method based on meta-learning as claimed in claim 1, characterized in that in step (4), the prediction network Net (α) is predictedn) Using and sharing generator Gen (theta)1) The circulating neural networks with the same structure are used for predicting according to the input concentration data of the dissolved gas in the oil and outputting the concentration data of the dissolved gas in the oil at the future time; sharing Generator Gen (θ)1) Estimating the predicted Net (alpha) from the test sample output parametersn) Network parameter alpha ofnThe process of (2), comprising: using shared Generator Gen (θ)1) Test sample y corresponding to each normal transformernFitting is carried out, and a fitting error e is calculated; then calculating the network parameter theta1Gradient with respect to fitting error e
And the network parameter theta is measured by using a gradient descent method1Carrying out primary updating; finally, the gradient descending result is used as a network parameter alpha of the prediction networknThus, a prediction model is constructed.
6. The oil-filled transformer residual life personalized prediction method based on meta-learning according to claim 1, characterized in that in step (4), a network Net (β) is judgedn) Adapted to share generator Gen (theta)2) The front feedback neural networks with the same structure are used for judging the operation state according to the input concentration data of the dissolved gas in the oil and outputting an operation state judgment result; sharing Generator Gen (θ)2) According to the output parameters of the test samples, the actual estimation of the discrimination network Net (beta)n) Network parameter beta ofnThe process of (2), comprising: using shared Generator Gen (θ)2) For each normalTest sample y corresponding to transformernPerforming pooling operation to obtain test sample ynCorresponding feature vector znThe feature vector z is then transformed by a non-linear activation functionnNetwork parameter beta mapped as a discriminating networknThus, a discriminant model is constructed.
7. The oil-immersed transformer residual life personalized prediction method based on meta-learning according to claim 1, wherein in the step (5), predicting the test sample by using a prediction model to obtain the concentration data of the dissolved gas in the oil at a future time comprises:
first predict sample ynInput to the prediction model Net (alpha)n) In accordance with the prediction sample ynPredicting the concentration data of the dissolved gas in the oil at the next moment; then the prediction sample ynAnd predicted concentration data of dissolved gas in oil and as new input, input to prediction model Net (alpha)n) And predicting the concentration data of the dissolved gas in the oil at the next moment, and continuously repeating the process to gradually predict the concentration data of the dissolved gas in the oil at the future time.
8. The oil-immersed transformer residual life personalized prediction method based on meta-learning according to claim 1, characterized in that in step (5), the operation state discrimination result output by the discrimination model is between 0 and 1, and the operation state discrimination result is less than 0.5 to represent that the transformer is in a normal state, and more than 0.5 to represent that the transformer is in a fault state.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210217435.6A CN114595883A (en) | 2022-03-07 | 2022-03-07 | Oil-immersed transformer residual life personalized dynamic prediction method based on meta-learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210217435.6A CN114595883A (en) | 2022-03-07 | 2022-03-07 | Oil-immersed transformer residual life personalized dynamic prediction method based on meta-learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114595883A true CN114595883A (en) | 2022-06-07 |
Family
ID=81815835
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210217435.6A Pending CN114595883A (en) | 2022-03-07 | 2022-03-07 | Oil-immersed transformer residual life personalized dynamic prediction method based on meta-learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114595883A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115982976A (en) * | 2022-12-19 | 2023-04-18 | 南京航空航天大学 | Method for predicting residual life of oil-immersed power transformer |
| CN117390520A (en) * | 2023-12-08 | 2024-01-12 | 惠州市宝惠电子科技有限公司 | Transformer state monitoring method and system |
-
2022
- 2022-03-07 CN CN202210217435.6A patent/CN114595883A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115982976A (en) * | 2022-12-19 | 2023-04-18 | 南京航空航天大学 | Method for predicting residual life of oil-immersed power transformer |
| CN115982976B (en) * | 2022-12-19 | 2024-01-05 | 南京航空航天大学 | Residual life prediction method for oil-immersed power transformer |
| CN117390520A (en) * | 2023-12-08 | 2024-01-12 | 惠州市宝惠电子科技有限公司 | Transformer state monitoring method and system |
| CN117390520B (en) * | 2023-12-08 | 2024-04-16 | 惠州市宝惠电子科技有限公司 | Transformer state monitoring method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109543737B (en) | Information system health degree evaluation method based on FAHP-FCA combined empowerment | |
| CN110542819B (en) | Transformer fault type diagnosis method based on semi-supervised DBNC | |
| CN114595883A (en) | Oil-immersed transformer residual life personalized dynamic prediction method based on meta-learning | |
| CN109740859A (en) | Transformer condition evaluation and system based on Principal Component Analysis and support vector machines | |
| CN110530650B (en) | Method for monitoring performance state of heavy-duty gas turbine based on generalized regression neural network and box diagram analysis | |
| CN104020401A (en) | Cloud-model-theory-based method for evaluating insulation thermal ageing states of transformer | |
| WO2012029154A1 (en) | Internal abnormality diagnosis method, internal abnormality diagnosis system, and decision tree generating method for internal abnormality diagnosis of oil-filled electric apparatus utilizing gas concentration in oil | |
| CN111460727A (en) | Method for predicting service life of transformer by using multiple parameters | |
| Mao et al. | Polymer electrolyte membrane fuel cell fault diagnosis and sensor abnormality identification using sensor selection method | |
| CN115563563A (en) | Fault diagnosis method and device based on transformer oil chromatographic analysis | |
| CN112288147B (en) | Method for predicting insulation state of generator stator by BP-Adaboost strong predictor | |
| CN109164248A (en) | A kind of predicting model for dissolved gas in transformer oil method | |
| CN116842337A (en) | Transformer fault diagnosis method based on LightGBM (gallium nitride based) optimal characteristics and COA-CNN (chip on board) model | |
| CN117113166A (en) | Industrial boiler fault detection method based on improved integrated learning | |
| Mao et al. | Comparative study on prediction of fuel cell performance using machine learning approaches | |
| CN105741184B (en) | Transformer state evaluation method and device | |
| CN114492559A (en) | Power equipment fault diagnosis method based on data time-frequency domain modeling | |
| Khalyasmaa et al. | Training samples construction for energy utilities operational assets management | |
| CN116664098A (en) | Abnormality detection method and system for photovoltaic power station | |
| CN116204825A (en) | Production line equipment fault detection method based on data driving | |
| CN115907198A (en) | Long-distance heat supply load intelligent prediction system | |
| CN115828744A (en) | White light LED fault on-line diagnosis and service life prediction method | |
| CN115327081A (en) | Transformer fault diagnosis method for improved goblet sea squirt group optimization support vector machine | |
| CN115310653A (en) | Photovoltaic array residual life prediction method based on performance degradation data | |
| CN114487643A (en) | On-spot handing-over of extra-high voltage GIL equipment is accepted and is synthesized test platform |
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 |