CN115951257A - Inverter state monitoring and health diagnosis method based on electromagnetic radiation signals - Google Patents
Inverter state monitoring and health diagnosis method based on electromagnetic radiation signals Download PDFInfo
- Publication number
- CN115951257A CN115951257A CN202211644825.8A CN202211644825A CN115951257A CN 115951257 A CN115951257 A CN 115951257A CN 202211644825 A CN202211644825 A CN 202211644825A CN 115951257 A CN115951257 A CN 115951257A
- Authority
- CN
- China
- Prior art keywords
- domain
- electromagnetic radiation
- inverter
- monitoring
- layer
- 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 70
- 230000005670 electromagnetic radiation Effects 0.000 title claims abstract description 62
- 238000012544 monitoring process Methods 0.000 title claims abstract description 58
- 238000003745 diagnosis Methods 0.000 title claims abstract description 39
- 230000036541 health Effects 0.000 title claims abstract description 39
- 238000012549 training Methods 0.000 claims abstract description 31
- 230000002159 abnormal effect Effects 0.000 claims abstract description 16
- 238000009826 distribution Methods 0.000 claims abstract description 16
- 210000002569 neuron Anatomy 0.000 claims description 29
- 230000007547 defect Effects 0.000 claims description 23
- 238000013508 migration Methods 0.000 claims description 23
- 230000005012 migration Effects 0.000 claims description 21
- 238000013527 convolutional neural network Methods 0.000 claims description 19
- 230000006978 adaptation Effects 0.000 claims description 16
- 238000012512 characterization method Methods 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 16
- 238000005457 optimization Methods 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 3
- 238000013178 mathematical model Methods 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 2
- 239000000758 substrate Substances 0.000 claims 1
- 238000012732 spatial analysis Methods 0.000 abstract 1
- 238000010248 power generation Methods 0.000 description 12
- 238000011161 development Methods 0.000 description 7
- 230000018109 developmental process Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 7
- 238000000605 extraction Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 208000001613 Gambling Diseases 0.000 description 3
- 241000282414 Homo sapiens Species 0.000 description 3
- 230000003042 antagnostic effect Effects 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 238000010792 warming Methods 0.000 description 1
Images
Classifications
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/50—Photovoltaic [PV] energy
Landscapes
- Measurement Of Radiation (AREA)
Abstract
The invention discloses an inverter state monitoring and health diagnosis method based on electromagnetic radiation signals, which comprises the steps of adopting a non-invasive and non-contact data acquisition method to take three-dimensional electromagnetic radiation space distribution intensity signals generated by high-frequency on-off of a switching tube of a photovoltaic platform area inverter as monitoring quantities; extracting and diagnosing the optimal distribution of electromagnetic radiation space and abnormal data as a training target; and carrying out real-time monitoring and health diagnosis on the running state of the photovoltaic grid-connected inverter. The method avoids the difference of the internal circuits of the power electronics. The non-electric quantity is used as the input quantity of health diagnosis, including the electromagnetic radiation intensity and the spatial distribution thereof, so that the complicated electromagnetic spatial analysis problem is avoided.
Description
Technical Field
The invention belongs to the technical field of power electronic equipment monitoring, and particularly relates to an inverter state monitoring and health diagnosis method based on electromagnetic radiation signals.
Background
The excessive consumption of traditional non-renewable fossil energy and the consequent problems of global warming and deterioration of ecological environment bring a great survival crisis to human beings. Therefore, the development of renewable energy sources is greatly promoted, and the change of the existing energy source structure is a necessary choice for solving the problem of sustainable development of energy sources and environments. Among them, solar energy is the most promising new energy in the world today, and has the advantages of large total amount, cleanness, no pollution, easy acquisition and utilization, sustainable supply, etc., and thus has been continuously focused and developed worldwide in recent years. The photovoltaic power generation technology is in strategic position in the current energy policy by vigorously promoting development. Expert predictions indicate that the photovoltaic power generation industry will remain at 30% annual growth rate in the next decade; by 2050 years, the total power generation of a photovoltaic system accounts for 5% -20% of all energy sources in the world, and solar energy is about to be developed into basic energy sources in human production and life.
However, with the rapid increase of the photovoltaic loading capacity, the problems of cost control, loss, aging, etc. of the photovoltaic power generation system become a significant problem for the photovoltaic industry. These problems are becoming more serious as the operating time of photovoltaic power generation systems reaches a certain age. In order to reduce the fault loss, timely find and process faults and avoid unnecessary loss, real-time state monitoring and fault diagnosis of a photovoltaic power generation system become a research focus, wherein the real-time monitoring of an inverter is a key for the operation of the photovoltaic power generation system. The operation condition is subjected to data monitoring through the photovoltaic power generation monitoring system, and the operation condition is diagnosed through information acquisition and processing, so that normal and continuous power supply of the photovoltaic power generation system is ensured.
Because the starting time of the photovoltaic system in China is relatively late, the application research of the monitoring technology in the system is not perfect. However, with the development of industry and the increasing popularization of internet technology and sensing communication technology in China, photovoltaic monitoring systems are also under vigorous development. In the early days, monitoring of photovoltaic systems relied primarily on field monitoring and manual inspection. With the development of network technology, sensing technology and communication technology, the main work of monitoring of the photovoltaic power generation system is gradually changed from manual work to automatic network monitoring, and the situation monitoring and control of the photovoltaic power generation system are completed through data acquisition, transmission and communication, so that the remote monitoring of the photovoltaic power generation system is really realized, the intelligent operation and maintenance industrial information development trend is met, and the operation stability of the inverter and the operation reliability of the system are effectively improved.
Disclosure of Invention
This section is for the purpose of summarizing some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. In this section, as well as in the abstract and the title of the invention of this application, simplifications or omissions may be made to avoid obscuring the purpose of the section, the abstract and the title, and such simplifications or omissions are not intended to limit the scope of the invention.
The present invention has been made in view of the above-mentioned conventional problems.
Therefore, an object of the present invention is to provide a method for monitoring and diagnosing the state of an inverter based on electromagnetic radiation signals, which avoids the complicated electromagnetic space analysis problem and avoids the variability of the internal circuit of power electronics. In actual operation, electromagnetic radiation signals are sampled and transmitted to a trained domain anti-migration convolutional neural network, so that the state of the photovoltaic grid-connected inverter can be monitored in real time and diagnosed healthily, and an abnormal inverter is isolated in time, so that faults are avoided, and the operation reliability of a photovoltaic platform area is guaranteed.
In order to achieve the technical purpose, the invention provides the following technical scheme that the method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals comprises the following steps:
a non-invasive and non-contact data acquisition method is adopted to take a three-dimensional electromagnetic radiation spatial distribution intensity signal generated by high-frequency on-off of a switching tube of a photovoltaic platform area inverter as a monitoring quantity;
extracting and diagnosing the optimal distribution of the electromagnetic radiation space and abnormal data as a training target;
and carrying out real-time monitoring and health diagnosis on the running state of the photovoltaic grid-connected inverter.
The inverter state monitoring and health diagnosis method based on the electromagnetic radiation signals is characterized in that a non-invasive and non-contact data acquisition method is adopted, and comprises the steps of learning transferable inverter defect characterization features from defect samples through a convolutional neural network, introducing domain antagonistic transfer learning, realizing the transfer of training models under mass data to complex working conditions and small samples so as to improve the diagnosis accuracy, and learning class boundary characterization features and domain space characterization features through an antagonistic training method, so that the transfer of diagnosis knowledge is realized.
The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals is characterized by comprising the following steps: the monitoring quantity is that aiming at electromagnetic radiation abnormity caused by abnormal switching-on or switching-off of the inverter, non-electrical quantity is used as input quantity of health diagnosis, wherein the input quantity comprises electromagnetic radiation intensity and spatial distribution thereof, and an electromagnetic radiation signal is specific to the inverter.
The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals is characterized by comprising the following steps: the convolutional neural network comprises a convolutional layer, a pooling layer and a full-link layer which are sequentially arranged, and a softmax classifier, wherein the expression of the convolutional layer is as follows:
wherein,
is the b-th neuron output characteristic of the l-th layer, f is a nonlinear activation function, M is the number of neuron input characteristics of the current layer, and>
for convolution kernel, <' > based on>
For the input features of the a-th neuron at layer l-1, B l Biasing the l layer;
the maximum pooling is selected by the pooling layer, and the expression is as follows:
wherein,
outputs characteristic for the b-th neuron on the l +1 th layer>
Inputting features for the mth neuron of the l-th layer;
the expression of the full connection layer is as follows:
wherein, y k+2 Is the k +2 th neuron output of the full connection layer, k is the number of the neurons of the full connection layer, omega k+2 The weight ratio of the k +2 th neuron of the full connection layer,
is the b-th neuron output characteristic of the (l + 1) -th convolutional layer.
The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals is characterized by comprising the following steps: the output model of the softmax classifier is as follows:
wherein p is m The probability that the inverter switch tube is in the mth defect state is shown, and m belongs to [1, z ]]，y k+2 Exp is an exponential function with e as the base for the k +2 th neuron output of the fully connected layer.
The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals is characterized by comprising the following steps of: countermeasure training is introduced behind a full connection layer of the convolutional neural network and consists of two parts, namely domain space countermeasure adaptation and decision boundary countermeasure adaptation, so that matching of a source domain and a target domain is achieved, and decision boundary countermeasure adaptation is achieved by transforming classification boundary decisions of the source domain and the target domain, and approximation unification of the decisions of the source domain and the target domain is achieved.
The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals is characterized by comprising the following steps of: the domain-space countermeasure adaptation includes: adding domain discriminators to neural networks to form a two-player gambling game, wherein the discriminator G d The method has the functions of judging whether the extracted features come from a source domain or a target domain, training two classifiers to respectively carry out a maximum-minimum game to realize low-error classification of electromagnetic radiation signals of the small sample inverter, and a feature extractor G f The features between the source domain and the target domain are matched.
The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals is characterized by comprising the following steps of: the optimization objective function of the domain-space countermeasure adaptation can be described as:
wherein, C 1 And C 2 Are respectively two classifiers, theta C1 And theta C2 Two label predictor parameters, respectively, and λ is a trade-off parameter between two losses.
The invention relates to an inverter state monitoring and health diagnosis method based on electromagnetic radiation signals, wherein the method comprises the following steps: the method is characterized in that: in the max-min game, the loss is a function of:
and then obtaining the optimal parameters of the model through the following optimization:
wherein, E dis As a difference between classifiers, L dis Is the cross entropy loss between the two classifiers.
The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals is characterized by comprising the following steps: the domain discriminator is added into a neural network to form a two-player game, the feature extractor is continuously subjected to back propagation optimization, and the parameter updating in each batch of training of the network can be expressed by a mathematical model as follows:
reduce electromagnetic radiation signal abnormal classification loss E to the maximum extent c While maximizing domain classification loss E d Domain discrimination aiming at minimizing E d Updating the classifier to minimize E c 。
The invention has the beneficial effects that: a non-invasive and non-contact data acquisition method is adopted, three-dimensional electromagnetic radiation space distribution intensity signals generated by high-frequency on-off of a photovoltaic platform inverter switching tube are used as monitoring quantities, electromagnetic radiation space optimal distribution and abnormal data extraction diagnosis are used as training targets, and inverter states are monitored in a targeted mode. The method realizes the migration of the training model under mass data to complex working conditions and small samples, and realizes the migration of diagnosis knowledge by learning class boundary characterization features and domain space characterization features through a confrontation training method. The method avoids the difference of the internal circuit of the power electronics. The non-electrical quantity is used as the input quantity of the health diagnosis, wherein the input quantity comprises the electromagnetic radiation intensity and the spatial distribution thereof, and the complicated electromagnetic space analysis problem is avoided.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the description below are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without inventive labor. Wherein:
fig. 1 is a flowchart illustrating an overall training process of a method for monitoring the status and diagnosing the health of an inverter based on electromagnetic radiation signals according to an embodiment of the present invention;
fig. 2 is a domain deconvolution neural network diagram of an electromagnetic radiation signal-based inverter state monitoring and health diagnosis method according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of an inverter distributed power supply of a method for monitoring an inverter status and diagnosing health based on electromagnetic radiation signals according to an embodiment of the present invention;
fig. 4 is a grid-connected structure diagram of an inverter distributed power source of the inverter state monitoring and health diagnosis method based on electromagnetic radiation signals according to an embodiment of the present invention;
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, specific embodiments accompanied with figures are described in detail below, and it is apparent that the described embodiments are a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, shall fall within the protection scope of the present invention.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than those specifically described and will be readily apparent to those of ordinary skill in the art without departing from the spirit of the present invention, and therefore the present invention is not limited to the specific embodiments disclosed below.
Furthermore, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
The present invention will be described in detail with reference to the drawings, wherein the cross-sectional views illustrating the structure of the device are not enlarged partially in general scale for convenience of illustration, and the drawings are only exemplary and should not be construed as limiting the scope of the present invention. In addition, the three-dimensional dimensions of length, width and depth should be included in the actual fabrication.
Also in the description of the present invention, it should be noted that the terms "upper, lower, inner and outer" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, which are only for convenience of description and simplification of description, but do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus, should not be construed as limiting the present invention. Furthermore, the terms first, second, or third are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
The terms "mounted, connected and connected" in the present invention are to be understood broadly, unless otherwise explicitly specified or limited, for example: can be fixedly connected, detachably connected or integrally connected; they may be mechanically, electrically, or directly connected, or indirectly connected through intervening media, or may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
Example 1
Referring to fig. 1-4, a first embodiment of the present invention provides an inverter status monitoring and health diagnosing method based on electromagnetic radiation signals, including:
s1: a non-invasive and non-contact data acquisition method is adopted to take a three-dimensional electromagnetic radiation spatial distribution intensity signal generated by high-frequency on-off of a switching tube of a photovoltaic platform area inverter as a monitoring quantity;
learning migratable inverter defect characterization features from defect samples by using a Convolutional Neural Network (CNN), wherein countermeasure training is introduced in convolution, and category difference (representing minimum target) and domain invariant (representing maximum target) features are learned through a minimum-maximum two-person game. Then, domain confrontation migration learning is introduced, so that the migration of a training model under mass data (source domain) to a complex working condition and a small sample (target domain) is realized, and the diagnosis accuracy is improved; finally, the class boundary characterization features and the domain space characterization features are learned through a countertraining method, and the migration of the diagnosis knowledge is achieved (the knowledge migration method is firstly proposed in the field of power electronic equipment monitoring). The method avoids the problem of complicated electromagnetic space analysis and avoids the difference of internal circuits of power electronics.
Furthermore, for electromagnetic radiation abnormality caused by inverter on or off abnormality, a non-electrical quantity (electromagnetic radiation signal) is used as an input quantity for health diagnosis, wherein the input quantity comprises electromagnetic radiation intensity and spatial distribution thereof (the intensity is combined with the space, and other signal characteristics are ignored, and the patent is firstly proposed). The electromagnetic radiation signal is specific to the inverter, so that the method can effectively diagnose the health of the inverter, and the problem of abnormal defect positioning of the photovoltaic system caused by the fact that the traditional electric quantity is used as diagnosis input is solved.
Furthermore, in actual operation of the photovoltaic platform area, electromagnetic radiation signals leaked to the outside of the box body under normal conditions are small and stable, so that sample data is small in amplitude and has no obvious difference in space; however, when the internal switch state of the inverter is abnormal, a strong radiation signal leaks out of the box body.
Furthermore, the convolutional neural network comprises a convolutional layer, a pooling layer and a full-connection layer which are arranged in sequence; the expression of the convolutional layer is as follows:
in the formula:
outputting features for the b-th neuron of the l-th layer; f is a nonlinear activation function; m is the input characteristic quantity of the neuron in the current layer; />
Is a convolution kernel; />
Inputting features for the a-th neuron of the l-1 layer; b is l Biasing the l layer;
the maximum pooling is selected by the pooling layer, and the expression is as follows:
in the formula
Outputting the characteristic for the b-th neuron of the l + 1-th layer; />
Inputting features for the b-th neuron of the l-th layer;
the expression of the full connection layer is as follows:
in the formula: y is k+2 Outputting the k +2 th neuron of the full connection layer; k is the number of neurons in the full connecting layer; omega k+2 The weight ratio of the k +2 th neuron of the full connection layer is obtained;
is the b-th neuron output characteristic of the (l + 1) -th convolutional layer.
The output model of the softmax classifier is as follows:
in the formula, p m Is the probability that the inverter switch tube is in the m-th defect state, and m belongs to [1, z ]]，y k+2 Exp is an exponential function with e as the base for the k +2 th neuron output of the fully connected layer.
It should be noted that the sample data has a large amplitude and presents the characteristic of uneven spatial distribution, and a health monitoring and diagnosis method based on electromagnetic radiation signals is provided for the sample data.
S2: extracting and diagnosing the optimal distribution of electromagnetic radiation space and abnormal data as a training target;
furthermore, countermeasure training (suitable for inverter abnormal small sample data training) is introduced into convolution, and countermeasure training is introduced behind a convolution neural network full connection layer and consists of two parts, namely domain space countermeasure adaptation and decision boundary countermeasure adaptation. The domain space countermeasure adaptation is to realize the matching connection of the inverter characteristic space between the source domain and the target domain, so as to realize the matching of the source domain and the target domain. The decision boundary confrontation adaptation is to transform the classification boundary decisions of the source domain and the target domain to realize the approximation unification of the decisions of the source domain and the target domain. The structure diagram of the domain deconvolution neural network is shown in the attached figure (2).
Further, in the domain space confrontation adaptation, a two-player gambling game is formed by adding a domain discriminator to a neural network. Wherein the discriminator G d The method has the functions of judging whether the extracted features come from a source domain or a target domain, training two classifiers to respectively carry out a maximum-minimum game to realize low-error classification of electromagnetic radiation signals (source data) of the small-sample inverter, and a feature extractor G f The features between the source domain and the target domain are matched.
Further, the optimization objective function of the domain-space confrontation adaptation is:
in, C 1 And C 2 Two classifiers are respectively provided; theta C1 And theta C2 Two label predictor parameters respectively; λ is a trade-off parameter between the two losses.
In playing the max-min game, the loss function is:
in the formula, E dis Is the difference between classifiers; l is dis Is the cross entropy loss between the two classifiers;
based on the objective function, the optimal parameters of the model are obtained through the following optimization:
furthermore, by adopting the anti-migration training scheme, the feature extractor is continuously subjected to back propagation optimization, so that the electromagnetic radiation signal abnormal classification loss E is reduced to the maximum extent c While maximizing domain classification loss E d . Domain discrimination aims at minimizing E d Updating the classifier to minimize E c . Therefore, the parameter update in each batch of training of the network can be expressed by a mathematical model as follows:
s3: monitoring and diagnosing the running state of the photovoltaic grid-connected inverter in real time;
in this embodiment, migratable inverter defect characterization features are first learned from defect samples with a Convolutional Neural Network (CNN), where countermeasure training is introduced in the convolution, and class differences and domain invariant features are learned by min-max two-person gambling. The automatic optimization construction method reduces human intervention in the network construction process, and effectively improves various performances such as network precision and the like; then, domain confrontation migration learning is introduced, so that the migration of a training model under mass data (source domain) to a complex working condition and a small sample (target domain) is realized, and the diagnosis accuracy is improved; finally, the class boundary characterization features and the domain space characterization features are learned through a countervailing training method, and migration of diagnosis knowledge is achieved. The method avoids the problem of complicated electromagnetic space analysis and avoids the difference of internal circuits of power electronics. In actual operation, electromagnetic radiation signals are sampled and transmitted to a trained domain-confrontation migration convolutional neural network, so that the state of the photovoltaic grid-connected inverter can be monitored in real time and health diagnosis can be performed, an abnormal inverter can be isolated in advance, faults are avoided, and the operation reliability of a photovoltaic platform area is guaranteed.
Example 2
Another embodiment of the invention: the system of the embodiment is realized by the method and specifically comprises a signal acquisition module, a feature extraction module, a migration module and a diagnosis module, wherein,
the signal acquisition module is used for acquiring and learning the characteristic features of the defects of the transferable inverter from the defect samples by adopting a Convolutional Neural Network (CNN);
the characteristic extraction module is used for extracting common defect characteristics of the source domain and the target domain;
the migration module migrates the module trained by the source domain data to the target domain;
the diagnosis module is used for monitoring and diagnosing the running state of the photovoltaic grid-connected inverter in real time;
the method comprises the steps that a signal acquisition module is used for acquiring and learning the characteristic features of defects of a migratable inverter from defect samples by adopting a Convolutional Neural Network (CNN), a feature extraction module is used for extracting common defect features of a source domain and a target domain, a migration module is used for migrating the common defect features to the target domain through a module trained by source domain data, and finally, a diagnosis module is used for carrying out real-time monitoring and health diagnosis on the running state of the photovoltaic grid-connected inverter.
In the system for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals, the signal acquisition module acquires and learns the characteristic features of the defects of the transferable inverter from defect samples by adopting a Convolutional Neural Network (CNN); the characteristic extraction module is used for extracting common defect characteristics of a source domain and a target domain, wherein a target domain sample is obtained by acquiring an average value of electromagnetic radiation intensity data generated when 6 IGBT (insulated gate bipolar transistor) (namely a switching tube) in a three-phase inverter bridge circuit operates every 5 min. The source domain sample is that aiming at the typical defect state of the switching tube, 20 fundamental wave periods of inverter output current data are collected at the sampling frequency of 1KHz, and the switching tube is uniformly distributed at each moment of the data collecting time period by changing a series of parameters of the switching tube and continuously translating the occurrence time of the abnormal state.
The method is obtained according to the method, the accuracy of electromagnetic radiation intensity data abnormity detection and switch tube positioning can reach 0.98, the recall rate can reach 0.6, and the accuracy of a characteristic extraction diagnosis method for carrying out frequency spectrum analysis and the like by utilizing an electric signal can only reach about 0.88, so that the method has a remarkable difference with the method.
The method is characterized in that confrontation training is introduced in convolution, and learning categories are extracted for the minimum-maximum two-person game; heterogeneous (embodying minimum objective) and domain invariant (embodying maximum objective) features; the migration module migrates the module trained by the source domain data to the target domain, and is characterized in that domain confrontation migration learning is introduced to realize the migration of a training model under mass data (source domain) to a complex working condition and a small sample (target domain) so as to improve the diagnosis accuracy; the diagnosis module is used for monitoring and diagnosing the running state of the photovoltaic grid-connected inverter in real time and healthily, and learning the boundary characterization characteristics and the domain space characterization characteristics by an antagonistic training method, so that the migration of diagnosis knowledge is realized (the knowledge migration method is firstly proposed in the field of power electronic equipment monitoring).
Claims (10)
1. An inverter state monitoring and health diagnosis method based on electromagnetic radiation signals is characterized in that: comprises the steps of (a) preparing a substrate,
a non-invasive and non-contact data acquisition method is adopted to take a three-dimensional electromagnetic radiation spatial distribution intensity signal generated by high-frequency on-off of a switching tube of a photovoltaic platform area inverter as a monitoring quantity;
extracting and diagnosing the optimal distribution of the electromagnetic radiation space and abnormal data as a training target;
and carrying out real-time monitoring and health diagnosis on the running state of the photovoltaic grid-connected inverter.
2. The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals as claimed in claim 1, wherein: the non-invasive and non-contact data acquisition method comprises the steps of learning migratable inverter defect characterization features from defect samples through a convolutional neural network, introducing domain confrontation migration learning, achieving migration of a training model under mass data to complex working conditions and small samples, improving diagnosis accuracy, learning class boundary characterization features and domain space characterization features through a confrontation training method, and achieving migration of diagnosis knowledge.
3. The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals as claimed in any one of claims 1 or 2, wherein: the monitoring quantity is that aiming at electromagnetic radiation abnormity caused by abnormal switching-on or switching-off of the inverter, non-electrical quantity is used as input quantity of health diagnosis, wherein the input quantity comprises electromagnetic radiation intensity and spatial distribution thereof, and an electromagnetic radiation signal is specific to the inverter.
4. The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals as claimed in claim 2, wherein: the convolutional neural network comprises a convolutional layer, a pooling layer and a full-link layer which are sequentially arranged, and a softmax classifier, wherein the expression of the convolutional layer is as follows:
wherein,
is the b-th neuron output characteristic of the l-th layer, f is a non-linear activation function, and->
Is a convolution kernel->
For the a-th neuron input feature of layer l-1, B l Biasing the l layer;
the maximum pooling is selected by the pooling layer, and the expression of the maximum pooling is as follows:
wherein,
outputting features for the b-th neuron in level l +1>
Inputting features for the mth neuron of the l-th layer;
the expression of the full connection layer is as follows:
wherein, y k+2 For the k +2 th neuron of the full junction layerOut, k is the number of neurons in the full junction layer, ω k+2 The weight ratio of the k +2 th neuron of the full connection layer,
is the b-th neuron output characteristic of the l +1 th convolutional layer.
5. The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals as claimed in claim 4, wherein: the output model of the softmax classifier is as follows:
wherein p is m The probability that the inverter switch tube is in the mth defect state is shown, and m belongs to [1, z ]]，y k+2 Exp is an exponential function with e as the base for the k +2 th neuron output of the fully connected layer.
6. The method of claim 4, wherein the method comprises the following steps: countermeasure training is introduced behind a full connection layer of the convolutional neural network and consists of two parts, namely domain space countermeasure adaptation and decision boundary countermeasure adaptation, so that matching of a source domain and a target domain is achieved, and decision boundary countermeasure adaptation is achieved by transforming classification boundary decisions of the source domain and the target domain, and approximation unification of the decisions of the source domain and the target domain is achieved.
7. The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals as claimed in any one of claims 1 to 6, wherein: the domain-space countermeasure adaptation includes: adding domain discriminator to neural network to form two-person game, wherein discriminator G d The two classifiers are trained to respectively carry out maximum-minimum game to realize low-error classification of electromagnetic radiation signals of the small-sample inverter, and the characteristicsExtractor G f The features between the source domain and the target domain are matched.
8. The method of claim 6, wherein the method comprises the steps of: the optimization objective function of the domain-space countermeasure adaptation can be described as:
wherein E is d As a domain space objective function, θ f As a feature extractor parameter, θ d For the domain discriminator parameter, n s And n t Number of samples in source domain and target domain, respectively, D s And D t Respectively source domain and target domain sample sets, G f As a feature extractor, G d Is a domain discriminator, L y And L d Are each G y And G d Cross entropy loss of (1), x i For the ith sample, d i Is a prediction tag of the ith sample, C 1 And C 2 Are respectively two classifiers, theta C1 And theta C2 Two label predictor parameters, respectively, and λ is a trade-off parameter between two losses.
9. The method for monitoring the state and diagnosing the health of the inverter based on the electromagnetic radiation signals as claimed in claim 8, wherein in the max-min game, the function of the loss is as follows:
and then obtaining the optimal parameters of the model through the following optimization:
wherein E is dis As a difference between classifiers, L dis Is the cross entropy loss between the two classifiers.
10. The method of claim 9, wherein the method comprises the steps of: the domain discriminator is added into a neural network to form a two-player game, the feature extractor is continuously subjected to back propagation optimization, and the parameter updating in each batch of training of the network can be expressed by a mathematical model as follows:
reduce electromagnetic radiation signal abnormal classification loss E to the maximum extent c While maximizing domain classification loss E d Domain discrimination aiming at minimizing E d Updating the classifier to minimize E c ，α c ，α dis ，α d Are respectively E c ，E dis ，E d The penalty factor of (1), δ, is the learning rate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211644825.8A CN115951257A (en) | 2022-12-20 | 2022-12-20 | Inverter state monitoring and health diagnosis method based on electromagnetic radiation signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211644825.8A CN115951257A (en) | 2022-12-20 | 2022-12-20 | Inverter state monitoring and health diagnosis method based on electromagnetic radiation signals |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115951257A true CN115951257A (en) | 2023-04-11 |
Family
ID=87282066
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211644825.8A Pending CN115951257A (en) | 2022-12-20 | 2022-12-20 | Inverter state monitoring and health diagnosis method based on electromagnetic radiation signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115951257A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117728588A (en) * | 2024-02-07 | 2024-03-19 | 华能江苏综合能源服务有限公司 | Inverter state monitoring method and system |
-
2022
- 2022-12-20 CN CN202211644825.8A patent/CN115951257A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117728588A (en) * | 2024-02-07 | 2024-03-19 | 华能江苏综合能源服务有限公司 | Inverter state monitoring method and system |
| CN117728588B (en) * | 2024-02-07 | 2024-05-10 | 华能江苏综合能源服务有限公司 | Inverter state monitoring method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108647716B (en) | Photovoltaic array fault diagnosis method based on composite information | |
| CN108614548B (en) | Intelligent fault diagnosis method based on multi-mode fusion deep learning | |
| CN111563611A (en) | Cloud data center renewable energy space-time prediction method for orientation graph convolutional network | |
| Haris et al. | Degradation curve prediction of lithium-ion batteries based on knee point detection algorithm and convolutional neural network | |
| CN104408562A (en) | Photovoltaic system generating efficiency comprehensive evaluation method based on BP (back propagation) neural network | |
| CN115951257A (en) | Inverter state monitoring and health diagnosis method based on electromagnetic radiation signals | |
| CN112884008B (en) | Prediction evaluation method and device for running state of power information acquisition system | |
| Liu et al. | Fault diagnosis of PV array based on optimised BP neural network by improved adaptive genetic algorithm | |
| CN116308304B (en) | New energy intelligent operation and maintenance method and system based on meta learning concept drift detection | |
| CN116050888A (en) | Method applied to intelligent high-voltage switch cabinet sensor health state assessment | |
| CN114722879A (en) | Bearing cross-working-condition fault prediction method based on anti-migration learning | |
| Zhang et al. | Remaining useful life prediction of lithium-ion batteries based on TCN-DCN fusion model combined with IRRS filtering | |
| CN114118592B (en) | Smart power grids power consumption end short-term energy consumption prediction system | |
| CN116401610A (en) | Inverter parameter fault diagnosis method based on depth residual error network and joint distribution | |
| CN117491872A (en) | Reconfigurable battery module fault multistage diagnosis method | |
| CN116317937A (en) | Distributed photovoltaic power station operation fault diagnosis method | |
| CN116304918A (en) | Substation equipment fault identification method and system based on depth forest algorithm | |
| CN112801815B (en) | Power communication network fault early warning method based on federal learning | |
| Liu et al. | Battery fault diagnosis scheme based on improved RBF neural network | |
| CN114710116A (en) | Actual measurement modeling method and system of photovoltaic cell assembly based on fuzzy model | |
| Ma et al. | Fault diagnosis of microbial fuel cell based on wavelet packet and SOM neural network | |
| Chen et al. | A transformer vibration amplitude prediction method via fusion of multi-signals | |
| Zhang et al. | A Hybrid Knowledge-Based and Data-Driven Method for Photovoltaic System Fault Diagnosis | |
| Lai et al. | Power data quality inspection and repair model based on neural network | |
| CN117394409B (en) | Intelligent assessment method and system for equipment state of energy storage power station |
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 |