CN116226776A - Machine learning-based photovoltaic system abnormal residual current detection method - Google Patents
Machine learning-based photovoltaic system abnormal residual current detection method Download PDFInfo
- Publication number
- CN116226776A CN116226776A CN202310454780.6A CN202310454780A CN116226776A CN 116226776 A CN116226776 A CN 116226776A CN 202310454780 A CN202310454780 A CN 202310454780A CN 116226776 A CN116226776 A CN 116226776A
- Authority
- CN
- China
- Prior art keywords
- residual current
- data
- hypothesis
- photovoltaic system
- value
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 25
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 24
- 238000010801 machine learning Methods 0.000 title claims abstract description 15
- 238000005070 sampling Methods 0.000 claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000004140 cleaning Methods 0.000 claims abstract description 23
- 238000012952 Resampling Methods 0.000 claims abstract description 18
- 238000007405 data analysis Methods 0.000 claims abstract description 5
- 238000000034 method Methods 0.000 claims description 36
- 230000006870 function Effects 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 230000005856 abnormality Effects 0.000 claims description 4
- 238000003064 k means clustering Methods 0.000 claims description 4
- 230000007704 transition Effects 0.000 claims description 4
- 230000001502 supplementing effect Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 230000009467 reduction Effects 0.000 claims description 2
- 238000000926 separation method Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 241001465754 Metazoa Species 0.000 description 1
- 238000002070 Raman circular dichroism spectroscopy Methods 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02S—GENERATION OF ELECTRIC POWER BY CONVERSION OF INFRARED RADIATION, VISIBLE LIGHT OR ULTRAVIOLET LIGHT, e.g. USING PHOTOVOLTAIC [PV] MODULES
- H02S50/00—Monitoring or testing of PV systems, e.g. load balancing or fault identification
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
- G01R19/0092—Arrangements for measuring currents or voltages or for indicating presence or sign thereof measuring current only
-
- 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
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Photovoltaic Devices (AREA)
- Testing Of Individual Semiconductor Devices (AREA)
Abstract
The invention discloses a machine learning-based photovoltaic system abnormal residual current detection method, which comprises the following steps of: s1, sampling residual current of a photovoltaic system to obtain residual current data; s2, converting the residual current data obtained by sampling into discrete digital signals, and storing the discrete digital signals and the sampling time stamp in a database; s3, resampling data and cleaning the data; s4, extracting features of data obtained through resampling and data cleaning, and carrying out hypothesis set acquisition and hypothesis set cleaning based on training data analysis; s5, learning state representation of residual current input data by adopting a variant beta-VAE of a variation automatic encoder; s6, detecting abnormal conditions of residual current. The invention can detect abnormal residual current in the photovoltaic system, and is helpful for detecting the fault or critical state of the system in an early stage.
Description
Technical Field
The invention relates to photovoltaic system anomaly detection, in particular to a machine learning-based photovoltaic system anomaly residual current detection method.
Background
In early power systems, residual current was mainly caused by insulation faults and is therefore also referred to as fault current. The Residual Current Device (RCD) is capable of detecting fault currents in the system and, when a certain threshold is exceeded, of actuating the residual current release to open in order to protect the safe and stable operation of the system. In grounded power systems, the use of RCDs is therefore an effective means of protecting humans, animals and facilities from damage caused by current flowing through the human body and ground. However, the conventional residual current protection method has the defects of low sensitivity and difficult fault location. Moreover, with the development of new power systems and the current access of more and more new energy sources into the power grid, some nonlinear power loads and regulated power generation systems with high-frequency controlled current converters such as Photovoltaic (PV) systems and the like are greatly increased, and the power electronics technology is more widely applied, so that new challenges are brought to the safe and stable operation of the power grid.
Today, there are still a large number of operation-related leakage current protection methods that are the same as conventional fault current protection. However, the conventional protection method based on the residual current amplitude cannot meet the requirements of the novel power system on rapid and accurate detection of the leakage fault. Although some new protection methods have been applied in electrical power systems, there are still some problems. For example, in the actual operation process, the current pulse protection method has more phenomena of refusal and misoperation, so that the current pulse protection method is not popularized and applied in a large area. Compared with the current pulse type and amplitude comparison type RCD, the amplitude and phase discrimination type RCD has the advantages that the action dead zone is reduced greatly, but the structure and the setting method are complex, and the phenomenon of refusal action still exists, so that the actual operation effect is not ideal. The current separation type RCD is limited by the accuracy of the separation algorithm, so that the current separation type RCD is also very limited in practical application. In the grid-connected photovoltaic system, the residual current is changed frequently due to some faults and unexpected conditions, and the grid-connected photovoltaic system not only contains a sinusoidal alternating current component and a pulsating direct current component, but also contains a smooth direct current component and even a high-frequency component, so that the residual current waveform is more complex, and the equipment is difficult to detect and inaccurate in judgment. In addition, some measures to ensure electromagnetic compatibility (EMC) often interfere with current diversion to ground potential and increase the operating leakage current, which can be up to 150 kHz, and most of the existing protection methods do not take into account these high frequency residual current components.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a machine learning-based method for detecting abnormal residual current of a photovoltaic system, which can detect the abnormal residual current in the photovoltaic system and is beneficial to detecting the fault or critical state of the system in an early stage.
The aim of the invention is realized by the following technical scheme: a photovoltaic system abnormal residual current detection method based on machine learning comprises the following steps:
s1, sampling residual current of a photovoltaic system to obtain residual current data;
s2, converting the residual current data obtained by sampling into discrete digital signals S= [ S ] 1 ,S 2 ,S 3 ,...,S M ]And stored in a database along with a sampling time stamp, M representing the number of discrete digital signal samples, where the jth signal sample S j =[s 1 ,s 2 ,s 3 ,...,s n ]N represents the number of discrete sampling points;
s3, resampling data and cleaning the data:
converting the data stored in the database into data with constant sampling period, deleting the repeated value, and then supplementing the missing value by alternately utilizing forward filling and reverse filling;
the forward filling refers to filling the current missing value with the previous non-missing value, and the backward filling refers to filling the current missing value with the next non-missing value;
s4, extracting features of data obtained through resampling and data cleaning, and carrying out hypothesis set acquisition and hypothesis set cleaning based on training data analysis;
s5, learning state representation of residual current input data by adopting a variant beta-VAE of a variation automatic encoder;
s6, detecting abnormal conditions of residual current.
The beneficial effects of the invention are as follows: the method for detecting abnormal residual current in the photovoltaic system can be used for detecting the fault or critical state of the system in an early stage, and expands the traditional detection method based on the threshold value.
Drawings
FIG. 1 is a process flow of the present invention;
FIG. 2 is a schematic diagram of a model training process in accordance with the present invention;
FIG. 3 is a flow chart of k-means clustering;
FIG. 4 is a flow chart of the model detection of the present invention.
Detailed Description
The technical solution of the present invention will be described in further detail with reference to the accompanying drawings, but the scope of the present invention is not limited to the following description.
The invention analyzes the residual current of the alternating current side of the grid-connected photovoltaic system in real time within a wide frequency range, and the intelligent RCM sensor system is provided with a plurality of magnetic coupling circuits for different frequency bands, can detect the residual current signals within different frequency ranges besides the total effective signals, and extracts the discrete frequency and the effective value of the frequency range from the residual current signals. According to the technical scheme, an intelligent residual current detector (RCM) sensor system is combined with a mode identification and variation automatic encoder to automatically adapt to different environmental conditions, faults or critical states in a grid-connected photovoltaic system can be detected in advance, and functions of enhancing the existing detection method are finally realized, in particular:
as shown in fig. 1, a method for detecting abnormal residual current of a photovoltaic system based on machine learning includes the following steps:
s1, sampling residual current of a photovoltaic system to obtain residual current data;
the residual current data comprises M types of sampling data;
the sampled data includes, but is not limited to:
the method comprises the following steps of a direct current component of residual current, a sum of all alternating current components, a residual current signal of a 50 Hz frequency band, a residual current signal of a frequency band below 100Hz frequency bands, a residual current signal of a 150 Hz frequency band, a residual current signal in a 100Hz-1kHz frequency band, a residual current signal of a 1kHz frequency band and a high-frequency band residual current signal, wherein the high-frequency band refers to a frequency band larger than 10 kHz.
Sampling the residual current by a sensor system of an intelligent residual current detector at a rate of 200 kHz;
s2, converting the residual current data obtained by sampling into discrete digital signals S= [ S ] 1 ,S 2 ,S 3 ,...,S M ]And stored in a database along with a sampling time stamp, M representing the number of discrete digital signal samples, where the jth signal sample S j =[s 1 ,s 2 ,s 3 ,...,s n ]N represents the number of discrete sampling points;
the residual current data are collected in M groups, and M discrete digital signal samples S are obtained after the collected data of the M groups are converted into discrete digital signals 1 ,S 2 ,S 3 ,...,S M The signal of each discrete digital signal at different discrete sampling points is a residual current value;
jth signal sample S j =[s 1 ,s 2 ,s 3 ,...,s n ]Wherein s is i Representing the value of the residual current at the i-th discrete sample point.
The residual current data includes, but is not limited to, the following categories:
the method comprises the following steps of (1) a direct current component of residual current, a sum of all alternating current components, a residual current signal of a 50 Hz frequency band, a residual current signal of a frequency band below 100Hz frequency band, a residual current signal of a 150 Hz frequency band, a residual current signal in a 100Hz-1kHz frequency band, a residual current signal of a 1kHz frequency band and a high-frequency band residual current signal, wherein the high-frequency band refers to a frequency band larger than 10 kHz;
one or more sets of residual current data need to be collected for each category.
S3, resampling data and cleaning the data:
converting the data stored in the database into data with constant sampling period, deleting the repeated value, and then supplementing the missing value by using forward filling and reverse filling alternately and practically;
the forward filling refers to filling the current missing value with the previous non-missing value, and the backward filling refers to filling the current missing value with the next non-missing value;
the constant sampling period in step S3 is a period of 1 minute, and resampling and data cleaning are required in step S3 for each signal sample.
Step S2 and step S3 mainly finish the processing of the sampled data;
s4, extracting features of data obtained through resampling and data cleaning, and carrying out hypothesis set acquisition and hypothesis set cleaning based on training data analysis;
s401, mainly performing training data analysis and feature extraction: let the data obtained by resampling and data cleaning be
Wherein->
Representing signal samples obtained by resampling and data cleaning of M signal samples; wherein the j-th signal sample obtained by resampling and data cleaning->
The sample length of the N signal samples, the resampled and data processed signal samples are used as training data to form a training data set:
feature extraction requires that each training data:
constructing a hypothesis corresponding to each training data: the hypothesis is defined by the detected residual current characteristics and the period of time during which the residual current characteristics appear for each component X j The extracted residual current features include root mean square
Skewness->
Form factor->
Peak indexPPulse indexIThe calculation formula is as follows:
in the method, in the process of the invention,
represents the i-th residual current value, +.>
Represents absolute average value +.>
Represents a maximum value;
s402, carrying out hypothesis set acquisition based on feature extraction results:
within each time period, a set of assumptions is constructed
K represents the number of hypotheses in the hypothesis set, and each hypothesis is defined by a set of feature data, as shown in equation (6)
Wherein the method comprises the steps of
Representing the ith hypothesis, T representing the time period;
s403, cleaning the hypothesis set:
comparing each hypothesis with other hypotheses of the same set in the same time period, calculating the deviation of a plurality of characteristic values between every two hypotheses to obtain an abnormality factor, and defining the maximum value of the deviation of each abnormality factor as a deviation threshold e max The calculation formula is shown as formula (7):
where i represents the ith eigenvalue in the hypothesis, j, k represent two different hypotheses in the same set of hypotheses, and a hypothesis exceeding the deviation threshold will be defined as an anomaly and deleted from the set of hypotheses;
s404, repeatedly executing the step S403 for the hypothesis set in each time period, and completing the hypothesis set cleaning in each time period.
S5, learning state representation of residual current input data by adopting a variant beta-VAE of a variation automatic encoder;
model learning the state of an electrical system, such as a photovoltaic inverter, indicates that a mapping function needs to be learned from measured residual current data to estimate the operating mode of the system. To obtain control of the state mapping function targets, a variant beta-VAE of a variant automatic encoder is employed to learn a state representation of the residual current input data. The VAE consists of two parts, an encoder for identifying the model and a decoder for generating the model. The beta-VAE better controls the objectives of the mapping function by introducing a super-parameter beta to balance reconstruction errors and latent space normalization.
Figure 2 shows the training flow of the model. After the residual current data are converted, cleaned and resampled to a period of 1 minute, the residual current data are input into the beta-VAE for training, and the specific process is as follows:
s501, inputting the training data set and the cleaned hypothesis set data into the beta-VAE for training;
within the potential space of the beta-VAE, the encoder calculates a low-dimensional mean value for each training dataμSum of variances
Then, the low-dimensional hidden variable coding vector z is obtained by utilizing the reparameterization technique, whereinμ、/>
Andzthe calculation formula of (2) is shown as formulas (8), (9) and (10)>
Representing a set of variables that conform to a normal distribution;
s502, through probability decoder
Mapping reconstruction yields a new, low-dimensional state representation +.>
At this time, the hypotheses in the hypothesis set are converted into labels corresponding to the state representations, and the performance of the β -VAE is evaluated by calculating the loss function, so as to update the parameters of the model:
the loss function is shown in equation (11):
wherein the method comprises the steps of
Is a variation parameter; />
Is a regularization coefficient; the first term on the right of the equal sign represents the error between the reconstructed output and input, i.e., the reconstructed error; the second term represents the prior probability distribution +.>
And posterior probability distribution->
KL divergence between;
s503, performing k-means clustering on the generated state representation values to determine state representations frequently appearing in training data, wherein the k-means clustering process is shown in FIG. 3. The resulting cluster result will be used to classify the new state representation values;
since the placement of states in the state representation space is random when using β -VAEs, it is also necessary to reorder the proximity between states according to the state representation, and finally form a state transition table for state transitions between the state representations and their corresponding labels;
s504, storing the clustering result and the state conversion table in the model for subsequent detection.
S6, detecting abnormal conditions of residual current.
The beta-VAE can be used for detecting the abnormal condition of the residual current after the training. Fig. 4 shows a detection flow. The real-time data detected by the RCM is converted and resampled and then input into the beta-VAE. And obtaining low-dimensional state representation after dimension reduction reconstruction in the potential space, judging whether the detected residual current data is a normal value or an abnormal value according to the state representation by the beta-VAE model, and then outputting the result as a state estimation result after conversion of a state conversion table, namely a detection result. And meanwhile, calculating relative reconstruction errors, wherein the relative reconstruction errors are used for judging the accuracy of detection judgment made by the model, and the smaller the error value is, the more accurate the judgment made is.
Although exemplary embodiments of the present invention have been described for illustrative purposes, those skilled in the art will appreciate that various modifications, additions and substitutions, and the like, can be made in the form and detail without departing from the scope and spirit of the invention as disclosed in the accompanying claims, all such modifications are intended to be within the scope of the invention as disclosed in the accompanying claims, and the various steps of the claimed method can be combined together in any combination. Therefore, the description of the embodiments disclosed in the present invention is not intended to limit the scope of the present invention, but is used to describe the present invention. Accordingly, the scope of the invention is not limited by the above embodiments, but is defined by the claims or equivalents thereof.
Claims (8)
1. A machine learning-based photovoltaic system abnormal residual current detection method is characterized by comprising the following steps of: the method comprises the following steps:
s1, sampling residual current of a photovoltaic system to obtain residual current data;
s2, converting the residual current data obtained by sampling into discrete digital signals S= [ S ] 1 ,S 2 ,S 3 ,...,S M ]And stored in a database along with a sampling time stamp, M representing the number of discrete digital signal samples, where the jth signal sample S j =[s 1 ,s 2 ,s 3 ,...,s n ]N represents the number of discrete sampling points;
s3, resampling data and cleaning the data:
converting the data stored in the database into data with constant sampling period, deleting the repeated value, and then supplementing the missing value by alternately utilizing forward filling and reverse filling;
the forward filling refers to filling the current missing value with the previous non-missing value, and the backward filling refers to filling the current missing value with the next non-missing value;
s4, extracting features of data obtained through resampling and data cleaning, and carrying out hypothesis set acquisition and hypothesis set cleaning based on training data analysis;
s5, learning state representation of residual current input data by adopting a variant beta-VAE of a variation automatic encoder;
s6, detecting abnormal conditions of residual current.
2. The machine learning-based photovoltaic system abnormal residual current detection method according to claim 1, wherein the method comprises the following steps: the residual current data are collected in M groups, and M discrete digital signal samples S are obtained after the collected data of the M groups are converted into discrete digital signals 1 ,S 2 ,S 3 ,...,S M The signal of each discrete digital signal at different discrete sampling points is a residual current value;
jth signal sample S j =[s 1 ,s 2 ,s 3 ,...,s n ]Wherein s is i Representing the value of the residual current at the i-th discrete sample point.
3. The machine learning-based photovoltaic system abnormal residual current detection method according to claim 2, wherein: the residual current data includes, but is not limited to, the following categories:
the method comprises the following steps of (1) a direct current component of residual current, a sum of all alternating current components, a residual current signal of a 50 Hz frequency band, a residual current signal of a frequency band below 100Hz frequency band, a residual current signal of a 150 Hz frequency band, a residual current signal in a 100Hz-1kHz frequency band, a residual current signal of a 1kHz frequency band and a high-frequency band residual current signal, wherein the high-frequency band refers to a frequency band larger than 10 kHz;
one or more sets of residual current data need to be collected for each category.
4. The machine learning-based photovoltaic system abnormal residual current detection method according to claim 1, wherein the method comprises the following steps: the sampling of the residual current data in the step S1 is carried out at a rate of 200kHz through a sensor system of an intelligent residual current detector;
the constant sampling period in step S3 is a period of 1 minute, and resampling and data cleaning are required in step S3 for each signal sample.
5. The machine learning-based photovoltaic system abnormal residual current detection method according to claim 1, wherein the method comprises the following steps: the step S4 includes:
s401, setting data obtained by resampling and data cleaning as
Wherein X is 1 ~X M Representing signal samples obtained by resampling and data cleaning of M signal samples; wherein the jth signal sample X obtained by resampling and data cleaning j =[x 1 ,x 2 ,x 3 ,...,x N ]N signal samplesThe sample length of the method is taken as training data of the signal sample after resampling and data processing, and a training data set is formed:
feature extraction requires that each training data:
constructing a hypothesis corresponding to each training data: the hypothesis is defined by the detected residual current characteristics and the period of time during which the residual current characteristics appear for each component X j The extracted residual current features include root mean square
Skewness->
Form factor->
Peak indexPPulse indexIThe calculation formula is as follows:
in the method, in the process of the invention,
represents the i-th residual current value, +.>
Represents absolute average value +.>
Represents a maximum value;
s402, carrying out hypothesis set acquisition based on feature extraction results:
within each time period, a Hypothesis set hypothesis= [ H ] is constructed 1 ,H 2 ,H 3 ,...,H K ]K represents the number of hypotheses in the hypothesis set, and each hypothesis is defined by a set of feature data, as shown in equation (6)
Wherein the method comprises the steps of
Representing the ith hypothesis, T representing the time period;
s403, cleaning the hypothesis set:
comparing each hypothesis with other hypotheses of the same set in the same time period, calculating the deviation of a plurality of characteristic values between every two hypotheses to obtain an abnormality factor, and defining the maximum value of the deviation of each abnormality factor as a deviation threshold e max The calculation formula is shown as formula (7):
where i represents the ith eigenvalue in the hypothesis, j, k represent two different hypotheses in the same set of hypotheses, and a hypothesis exceeding the deviation threshold will be defined as an anomaly and deleted from the set of hypotheses;
s404, repeatedly executing the step S403 for the hypothesis set in each time period, and completing the hypothesis set cleaning in each time period.
6. The machine learning-based photovoltaic system abnormal residual current detection method according to claim 1, wherein the method comprises the following steps: in the step S5, the VAE is composed of two parts, namely an encoder and a decoder, the encoder is used for identifying the model, the decoder is used for generating the model, and the β -VAE is used for reconstructing errors and normalization of potential space by introducing a super parameter β, so as to control the target of the mapping function.
7. The machine learning-based photovoltaic system abnormal residual current detection method according to claim 6, wherein: the step S5 includes:
s501, inputting the training data set and the cleaned hypothesis set data into the beta-VAE for training;
within the potential space of the beta-VAE, the encoder calculates a low-dimensional mean value for each training dataμSum of variances
Then, the low-dimensional hidden variable coding vector z is obtained by utilizing the reparameterization technique, whereinμ、/>
Andzthe calculation formula of (2) is shown as formulas (8), (9) and (10)>
Representing a set of variables that conform to a normal distribution;
s502, through probability decoder
Mapping reconstruction yields a new, low-dimensional state representation +.>
At this time, the hypotheses in the hypothesis set are converted into labels corresponding to the state representations, and the performance of the β -VAE is evaluated by calculating the loss function, so as to update the parameters of the model:
the loss function is shown in equation (11):
wherein the method comprises the steps of
Is a variation parameter; />
Is a regularization coefficient; the first term on the right of the equal sign represents the error between the reconstructed output and input, i.e., the reconstructed error; the second term represents the prior probability distribution +.>
And posterior probability distribution->
KL divergence between;
s503, performing k-means clustering on the generated state representation values to determine state representations frequently appearing in training data, wherein the obtained clustering result is used for classifying new state representation values;
since the placement of states in the state representation space is random when using β -VAEs, it is also necessary to reorder the proximity between states according to the state representation, and finally form a state transition table for state transitions between the state representations and their corresponding labels;
s504, storing the clustering result and the state conversion table in the model for subsequent detection.
8. The machine learning-based photovoltaic system abnormal residual current detection method according to claim 7, wherein: the step S6 includes:
converting and resampling the detected real-time residual current data and inputting the converted and resampled real-time residual current data into the beta-VAE;
obtaining low-dimensional state representation after dimension reduction reconstruction in a potential space, judging whether the detected residual current data is a normal value or an abnormal value according to the state representation by a beta-VAE model, and then outputting the result as a state estimation result after conversion of a state conversion table, namely a detection result;
and meanwhile, calculating relative reconstruction errors, wherein the relative reconstruction errors are used for judging the accuracy of detection judgment made by the model, and the smaller the error value is, the more accurate the judgment made is.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310454780.6A CN116226776A (en) | 2023-04-25 | 2023-04-25 | Machine learning-based photovoltaic system abnormal residual current detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310454780.6A CN116226776A (en) | 2023-04-25 | 2023-04-25 | Machine learning-based photovoltaic system abnormal residual current detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116226776A true CN116226776A (en) | 2023-06-06 |
Family
ID=86589517
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310454780.6A Pending CN116226776A (en) | 2023-04-25 | 2023-04-25 | Machine learning-based photovoltaic system abnormal residual current detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116226776A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103718055A (en) * | 2011-07-28 | 2014-04-09 | 伊顿公司 | Systems and apparatus for fault detection in dc power sources using ac residual current detection |
| CN106560718A (en) * | 2016-05-06 | 2017-04-12 | 艾思玛新能源技术（上海）有限公司 | Grounding fault current detection protection method and device |
| JP2019191006A (en) * | 2018-04-25 | 2019-10-31 | 三洋電機株式会社 | Current detection system and power storage system |
| CN113721070A (en) * | 2020-05-26 | 2021-11-30 | 天津首瑞智能电气有限公司 | Residual current detection device and residual current detection method |
| CN114397494A (en) * | 2021-12-06 | 2022-04-26 | 深圳供电局有限公司 | Residual current fault detection method and system |
| CN115618261A (en) * | 2022-09-22 | 2023-01-17 | 国网江苏省电力有限公司南京供电分公司 | Photovoltaic access power distribution network electric leakage identification method based on NCA and SSA-KELM |
-
2023
- 2023-04-25 CN CN202310454780.6A patent/CN116226776A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103718055A (en) * | 2011-07-28 | 2014-04-09 | 伊顿公司 | Systems and apparatus for fault detection in dc power sources using ac residual current detection |
| CN106560718A (en) * | 2016-05-06 | 2017-04-12 | 艾思玛新能源技术（上海）有限公司 | Grounding fault current detection protection method and device |
| JP2019191006A (en) * | 2018-04-25 | 2019-10-31 | 三洋電機株式会社 | Current detection system and power storage system |
| CN113721070A (en) * | 2020-05-26 | 2021-11-30 | 天津首瑞智能电气有限公司 | Residual current detection device and residual current detection method |
| CN114397494A (en) * | 2021-12-06 | 2022-04-26 | 深圳供电局有限公司 | Residual current fault detection method and system |
| CN115618261A (en) * | 2022-09-22 | 2023-01-17 | 国网江苏省电力有限公司南京供电分公司 | Photovoltaic access power distribution network electric leakage identification method based on NCA and SSA-KELM |
Non-Patent Citations (2)
| Title |
|---|
| BEHRENDS, H.; MILLINGER, D.; WEIHS-SEDIVY, W.; JAVORNIK, A.; ROOLFS, G.; GEISSENDÖRFER, S.: "Analysis of Residual Current Flows in Inverter Based Energy Systems Using Machine Learning Approaches.", 《ENERGIES 2022》, pages 1 - 17 * |
| 有小偷翻跟头: "笔记: beta VAE", 《HTTPS://ZHUANLAN.ZHIHU.COM/P/434152309》, pages 1 - 6 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO1997008627A1 (en) | A neural network based system for equipment surveillance | |
| CN108535572B (en) | Metering system secondary circuit monitoring method and device based on fundamental wave zero sequence characteristics | |
| CN112327104B (en) | Fault detection and positioning method for power distribution network with distributed power supply | |
| CN111426905B (en) | Power distribution network common bus transformation relation abnormity diagnosis method, device and system | |
| CN111537853A (en) | Intelligent detection method for partial discharge of switch cabinet based on multi-source heterogeneous data analysis | |
| CN110927527A (en) | Transient fault sensing method based on transient energy double channels and information fusion | |
| CN112147474A (en) | XLPE power cable typical defect partial discharge type identification system and method | |
| CN117113180A (en) | Direct current distribution network high-resistance fault identification method based on VMD and convolutional neural network | |
| CN115144702A (en) | Ground fault type identification method based on self-organizing competitive neural network | |
| CN115688880A (en) | Open-circuit fault diagnosis method and device for switching device of modular multilevel converter | |
| CN113514743B (en) | Method for constructing GIS partial discharge pattern recognition system based on multidimensional features | |
| CN113702767B (en) | Island direct-current microgrid fault diagnosis method based on wavelet sliding window energy | |
| CN106019043B (en) | Power grid fault diagnosis method based on fuzzy integral multi-source information fusion theory | |
| CN113610119B (en) | Method for identifying power transmission line development faults based on convolutional neural network | |
| Xue et al. | A data-driven diagnosis method of open-circuit switch faults for PMSG-based wind generation system | |
| CN109635430A (en) | Grid power transmission route transient signal monitoring method and system | |
| CN116226776A (en) | Machine learning-based photovoltaic system abnormal residual current detection method | |
| JP2023502659A (en) | Machine Learning Based Method and Apparatus for Transmission Line Disturbance Classification | |
| CN116819382A (en) | DC-DC converter state monitoring method based on digital twin | |
| CN114089181B (en) | XG-Boost-based MMC switching tube open-circuit fault detection method | |
| CN115409335A (en) | Electric power system disturbance identification method based on deep learning and considering unknown disturbance types | |
| CN111781467B (en) | Power grid fault intelligent discrimination method based on multidimensional panoramic data | |
| CN113746132A (en) | Photovoltaic power station based on cloud edge cooperation and control method thereof | |
| Mumtaz et al. | Ultra High-Speed Fault Diagnosis Scheme for DC Distribution Systems based on Discrete Median filter and Mathematical Morphology | |
| CN111398810A (en) | Slip ring electric spark detection and diagnosis system and detection and diagnosis method |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20230606 |