CN112787591B - Photovoltaic array fault diagnosis method based on fine-tuning dense connection convolutional neural network - Google Patents
Photovoltaic array fault diagnosis method based on fine-tuning dense connection convolutional neural network Download PDFInfo
- Publication number
- CN112787591B CN112787591B CN202011199284.3A CN202011199284A CN112787591B CN 112787591 B CN112787591 B CN 112787591B CN 202011199284 A CN202011199284 A CN 202011199284A CN 112787591 B CN112787591 B CN 112787591B
- Authority
- CN
- China
- Prior art keywords
- photovoltaic array
- data
- working
- training
- irradiance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000003745 diagnosis Methods 0.000 title claims abstract description 32
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 61
- 238000004088 simulation Methods 0.000 claims abstract description 53
- 238000012360 testing method Methods 0.000 claims abstract description 40
- 238000001514 detection method Methods 0.000 claims abstract description 22
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 21
- 239000011159 matrix material Substances 0.000 claims abstract description 17
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 238000005070 sampling Methods 0.000 claims abstract description 10
- 230000035772 mutation Effects 0.000 claims abstract description 8
- 238000010248 power generation Methods 0.000 claims abstract description 8
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims abstract description 5
- 238000005457 optimization Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 38
- 230000007613 environmental effect Effects 0.000 claims description 21
- 238000000605 extraction Methods 0.000 claims description 20
- 238000012795 verification Methods 0.000 claims description 15
- 238000005259 measurement Methods 0.000 claims description 8
- 230000008014 freezing Effects 0.000 claims description 7
- 238000007710 freezing Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000012546 transfer Methods 0.000 claims description 6
- 230000032683 aging Effects 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 5
- 230000006835 compression Effects 0.000 claims description 4
- 238000007906 compression Methods 0.000 claims description 4
- 239000002131 composite material Substances 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 210000004205 output neuron Anatomy 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims 1
- 230000002159 abnormal effect Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 5
- 238000003491 array Methods 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000000712 assembly Effects 0.000 description 1
- 238000000429 assembly Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013524 data verification Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005401 electroluminescence Methods 0.000 description 1
- 239000002803 fossil fuel Substances 0.000 description 1
- 238000003384 imaging method Methods 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
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000002310 reflectometry Methods 0.000 description 1
- 238000001931 thermography Methods 0.000 description 1
- 238000010792 warming 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
- H02S50/10—Testing of PV devices, e.g. of PV modules or single PV cells
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/045—Combinations of 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
-
- 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
- Y02B—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO BUILDINGS, e.g. HOUSING, HOUSE APPLIANCES OR RELATED END-USER APPLICATIONS
- Y02B10/00—Integration of renewable energy sources in buildings
- Y02B10/10—Photovoltaic [PV]
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Photovoltaic Devices (AREA)
- Testing Of Individual Semiconductor Devices (AREA)
Abstract
The invention relates to a photovoltaic array fault diagnosis method based on a fine-tuning dense connection convolutional neural network, which comprises the steps of firstly, collecting electrical characteristic data and environment data under an actual working condition, then building a model array by using Simulink, and simulating the actual working condition; acquiring simulated electrical characteristic data, then eliminating abnormal data in practice and simulation through a mutation point detection algorithm, acquiring complete electrical waveform data, sampling the electrical waveform data, compressing the characteristics, and splicing the electrical waveform data into a two-dimensional characteristic matrix. Then, designing a dense connection convolution neural network, pre-training the network by using a simulation training set and an Adam optimization algorithm, and then fine-tuning the network by using a small amount of training sets under actual working conditions. And finally, detecting and classifying the photovoltaic power generation array under the working condition test set to be detected by utilizing the FT-DenseNet fault diagnosis network. According to the method, under the condition of a small sample, a classification network with high precision, strong robustness and good generalization capability is obtained, and the accuracy of the fault detection and classification of the photovoltaic array can be effectively improved.
Description
Technical Field
The invention relates to the technical field of photovoltaic power generation string fault detection and classification, in particular to a photovoltaic array fault diagnosis method based on a fine-tuning dense connection convolutional neural network.
Background
In recent years, solar energy has been widely developed as a promising renewable energy source. Photovoltaic energy is a form of solar energy, plays an indispensable role in suppressing global warming, reduces the use and emission of fossil fuels, and has been increasingly increased in global photovoltaic loading and generation according to recent announcements of the world energy organization. With the rapid development of the photovoltaic industry and the rapid increase of the installed photovoltaic capacity, the service life and safety of photovoltaic arrays are receiving more and more attention. Photovoltaic power stations are mostly built in areas with rare people and large areas. Frequent monthly faults not only bring great difficulty to manual inspection, but also increase large-scale operation and maintenance cost and seriously affect the power generation efficiency. The failure of the photovoltaic array seriously affects the efficiency and safe operation of the photovoltaic power generation system and even causes immeasurable consequences such as fire hazard and the like. Therefore, the fault detection and diagnosis of the photovoltaic array are important factors for ensuring the safe and stable operation of the photovoltaic system.
The traditional photovoltaic array fault detection methods include a Time Domain Reflectometry (TDR) method and a ground capacitance measurement (ECM), the methods can realize fault positioning by detection, but the TDR can only detect impedance change, and the ECM can only detect open circuit, so that the method has certain limitation. Still other detection methods are those using thermal imaging, electroluminescence imaging, which require relatively expensive equipment and harsh environmental conditions. Santiago et al evaluates current and voltage indicators, defines thresholds for two parameters that are helpful in identifying system faults, and determines the fault type by comparing the thresholds. In order to determine the type of fault, the dc variable and the deviation from the analog variable, i.e. the ratio of current and voltage, are defined, and the fault is identified by the ratio of current and voltage. These methods of threshold comparison require a large amount of a priori knowledge and are prone to fault types that are difficult to distinguish. In recent years, with the rapid rise of artificial intelligence, a large number of AI fault diagnosis methods are successively applied to fault detection and diagnosis of photovoltaic arrays. For example, traditional machine learning algorithms such as an Artificial Neural Network (ANN), a Decision Tree (DT), a Random Forest (RF), a Probabilistic Neural Network (PNN), a Wavelet Neural Network (WNN), a Support Vector Machine (SVM), and the like, have made significant achievements and breakthroughs in the field of photovoltaic fault diagnosis and detection, but all of them require a large number of training samples, and the fault samples of an actual photovoltaic array are difficult to obtain, and cannot be really applied to a large photovoltaic array. Some unsupervised learning algorithms, such as Density Peak Clustering Algorithm (DPCA), fuzzy C-means algorithm (FCM), etc., have good results in fault diagnosis, but their accuracy is still limited. With the increasingly hot fire, the convolutional neural network and the cyclic neural network which are deeply learned walk into the sight line of photovoltaic fault diagnosis, the strong feature extraction capability of the convolutional neural network and the cyclic neural network can provide high-quality fault features with strong representation capability, the accuracy of classification is further improved, and the support of a large number of samples is still needed.
Disclosure of Invention
In view of this, the present invention provides a photovoltaic array fault diagnosis method based on a fine-tuning dense connection convolutional neural network, which uses a mutation point detection algorithm, and can search for a complete electrical characteristic waveform characteristic at a maximum power point, and can effectively remove abnormal data in original data. The method has high precision and stability, and good robustness and generalization capability.
The invention is realized by adopting the following scheme: a photovoltaic array fault diagnosis method based on a fine-tuning dense connection convolutional neural network comprises the following steps:
step S1: acquiring original measured data: gather photovoltaic electrical characteristic data under various operating mode conditions, specifically include: the photovoltaic array power generation system comprises a photovoltaic array working voltage, a photovoltaic array working current, working currents of three strings, a reference plate open-circuit voltage, a reference plate short-circuit current and photovoltaic array power; recording real-time temperature and irradiance;
step S2: the frequency of the temperature and the irradiance detected by the meteorological station are 1/60HZ respectively, the temperature and the irradiance collected once in one minute are subjected to spline interpolation for three times respectively, and 200 continuous values of data in one second are obtained, namely the environmental sample with the frequency of 200HZ is obtained; building a photovoltaic array simulation model by using Simulink, simulating various working conditions by using the temperature and irradiance, and obtaining photovoltaic electrical characteristic data related to step S1, namely obtaining original simulation data;
step S3: preprocessing original measured data: globally detecting the photovoltaic array working voltage in the measured data by using a jump point detection algorithm, finding out time nodes with sudden changes of the waveform, selecting a complete voltage waveform at the maximum power point among continuous nodes, and obtaining other electrical characteristic data at the maximum power point, wherein the electrical characteristic data comprises the electrical data except the photovoltaic array working voltage in the step S1; performing integral multiple extraction on the electrical characteristic data of the maximum power point, reducing the length of the electrical characteristic data, and obtaining 8 one-dimensional electrical data based on a time sequence; splicing the 8 pieces of electrical data into a two-dimensional characteristic matrix according to rows to obtain an actually measured data set;
step S4: performing the same operation on the simulation data according to the time node obtained by detecting the actual measurement data in the step S3 to obtain a simulation data set; dividing the simulation data set and the measured data set into a training set and a verification set and a test set according to the equal proportion of each working condition; designing a convolution neural network FT-DenseNet based on fine-tuning dense connection, training by using a training set sample in a simulation data set, saving model parameters when a loss function of a model is converged, stopping training, and verifying by using a test set to obtain an optimal pre-training model;
step S5: freezing the model parameters stored in the step S4 to a characteristic extraction layer, retraining a classification layer by using a training set in the actually measured data, and verifying the FT-DenseNet training model on a verification set to obtain an optimal FT-DenseNet fault diagnosis model with most generalization capability;
step S6: and detecting and classifying the electrical characteristic data of the test set under the actual working condition by using the FT-DenseNet fault diagnosis model, and comparing the label information of the test set by using the output result of the classification layer to give the type of the actual working condition.
Further, the various operating conditions in step S1 include normal operation, group string level line fault, array level line fault, aging fault, shadow fault, and open fault: wherein a group string-level line fails, i.e. a component of a single group string is shorted; array level line faults, i.e. nodes with different potential differences in different sets of strings are shorted; aging faults, i.e., array lines age and resistance increases; shadow fault, namely shadow occlusion of the components in the group string; open circuit failure, i.e., a certain group of strings in the array is open circuited; the environment conditions simulating each working condition are consistent with the actually acquired environment conditions.
Further, in step S2, a photovoltaic array simulation model is built by using Simulink, various working conditions are simulated by using the temperature and irradiance, and the specific processing manner of the temperature and irradiance is as follows:
the frequency of the temperature and the frequency of the irradiance detected by the weather station are 1/60HZ respectively, and in order to meet the accuracy of the real-time change of Simulink, an environment sample with the frequency of 200HZ is obtained by using cubic spline interpolation; if the actual collection time is n hours, then 60 × n +1 temperature and irradiance samples are obtained, including 60 × n +1 temperature samples (0, T1), (1, T2), (2, T3), … … (60 × n, T2), (2, T3)60*n) And 60 × n +1 irradiance samples (0, G1), (1, G2), (2, G3), … … (60 × n, G3878)60*n) Finally, (60 × n) × 60 × 200 temperature and irradiance samples are obtained, and the specific formula is as follows:
environmental samples (0, y1), (1, y2), (2, y3), … … (60 × n, y60*n) Step size of 1, miIs a second order differential value;
coefficient of the curve:
ai＝yi
between each two environmental samples:
yi(x)＝ai+bi(x-xi)+ci(x-xi)2+di(x-xi)2
between every two environment samples by yi(x) 12000 continuous environmental samples were obtained, so there were (60 × n) × 60 × 200 temperature samples and irradiance in total; the temperature irradiance sample is used as the SIMULINK input, and a simulation data set is obtained.
Further, the step S3 specifically includes the following steps:
step S31: n hours of measured data are concentrated with (60 × n) × 60 × 200 samples, and each sample in the original measured data set obtained in step S1 has 8 characteristics of photovoltaic array working voltage, photovoltaic array working current, working current of three strings, reference plate open-circuit voltage, reference plate short-circuit current and photovoltaic array power; independently acquiring the working voltage of the one-dimensional characteristic photovoltaic array, namely the working voltage of the photovoltaic array continuously acquired in n hours; and acquiring the time node at the maximum power point by adopting a mutation point detection algorithm in the following specific acquisition mode:
every 200 continuous photovoltaic array working voltages are used as a set, one point is randomly selected to divide the set into two parts, residual errors of all points on the two sides and the average value of all the parts are calculated, and when the total residual error reaches the minimum value, a catastrophe point is found, and the formula is as follows:
will V1，V2，V3，...，Vk，...V200Sequentially carrying in, when J takes the minimum value, recording Vr＝kThe time node of (2);
step S32: between every two adjacent time nodes, an integral multiple sampling function decimate is used for compressing the characteristic length, the sampling frequency of the photovoltaic array working voltage based on the time sequence is reduced to 1/40, namely 5 voltage data, and the voltage data of 6 adjacent time nodes are spliced into a 1 x 30 one-dimensional characteristic containing a complete voltage waveform;
step S33: according to the operations of the step S31 and the step S32, feature compression is carried out on the photovoltaic array working current, the working current of the three strings, the reference plate open-circuit voltage, the reference plate short-circuit current and the photovoltaic array power by the same time node to obtain 1 × 30 waveform features respectively, finally, the waveform features are spliced into an 8 × 30 two-dimensional feature matrix, and the electrical characteristic features of the simulation data set are obtained into the 8 × 30 two-dimensional feature matrix according to the method of the step S31 to the step S33.
Further, in step S4, the measured data set and the simulation data set are respectively divided into a training set, a verification set and a test set according to different working conditions in equal proportion; specifically, each working condition of the simulation data set is divided into a training set and a testing set according to the proportion of 70 percent to 30 percent. In the measured data, each working condition is divided into a training set, a verification set and a test set according to the proportion of 2%, 10% and 88%.
Further, the designing of the network structure of the fine-tuning dense connection based convolutional neural network FT-DenseNet in step S4 is composed of a feature extraction layer and a classification layer; the feature extraction layer comprises two dense connection modules and a transfer layer, wherein each dense connection module is composed of 5 complex functions (BN-ReLU-Conv2 d); the transfer layer consists of a composite function and a pooling layer of BN-ReLU-Conv2d, one BN-Conv2d and two linear fully-connected layers are used as classification layers, cross entropy is used as a loss function, an Adam optimization algorithm is adopted to train the network until the loss function converges, namely the loss function is less than 0.001, and the calculation formula of the loss function is as follows:
wherein n is the number of output neurons, ykFor the desired output value, σ is the objective function, zkRepresenting the actual output value of the neuron.
Further, the specific implementation manner of obtaining the optimal and most generalized FT-DenseNet fault diagnosis model in step S5 is as follows:
step SA: pre-training FT-DenseNet by using a training set of a simulation data set until a loss function is converged, namely the loss function is less than 0.001, verifying by using a simulation test set to obtain an optimal pre-training network model, obtaining the optimal pre-training network model, and storing network parameters with strong generalization performance;
step SB: freezing a feature extraction layer of the FT-DenseNet, retraining a classification layer of the FT-DenseNet by using an actual measurement data training set and a smaller learning rate range of 0.001-0.0001, stopping training when a loss function is converged, and verifying by using an actual measurement data verification set to obtain an optimal FT-DenseNet network model.
Further, the specific implementation manner of detecting and classifying the test set electrical characteristic data under the actual working condition in step S6 is as follows: inputting the actual working condition test set into FT-DenseNet, outputting the final 1 × 6 class information (x1, x2, x3, x4, x5, x6) through the feature extraction layer and the classification layer, converting the information into class probability through Softmax, wherein the calculation formula is as follows,
and comparing the 1 × 6 label information of the test set, wherein the category of the maximum probability is the current working condition type of the photovoltaic array.
Compared with the prior art, the invention has the following beneficial effects:
by simulating the actual environment, the invention can make up for the problem of lack of actual samples and improve the diagnosis accuracy under the condition of small samples. The invention uses the mutation point detection algorithm, can search the complete electrical characteristic waveform characteristics at the maximum power point, and can effectively eliminate abnormal data in the original data. The method has high precision and stability, and good robustness and generalization capability.
Drawings
Fig. 1 is a flowchart of a fault diagnosis method according to an embodiment of the present invention.
FIG. 2 is a diagram of a photovoltaic array in accordance with an embodiment of the present invention.
Fig. 3 is a Simulink topological diagram of a simulated photovoltaic array according to an embodiment of the present invention, where fig. 3(a) is a 3 × 6 photovoltaic array model diagram, and fig. 3(b) is a 3 × 6 photovoltaic array model fault simulation diagram.
FIG. 4 is a sorted confusion matrix according to an embodiment of the present invention.
FIG. 5 is FT-DenseNet according to an embodiment of the present invention.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
As shown in fig. 1, the present embodiment provides a method for diagnosing a fault of a photovoltaic array based on a fine tuning dense connection convolutional neural network, including the following steps:
step S1: acquiring original measured data: gather photovoltaic electrical characteristic data under various operating mode conditions, specifically include: the photovoltaic array power generation system comprises a photovoltaic array working voltage, a photovoltaic array working current, working currents of three strings, a reference plate open-circuit voltage, a reference plate short-circuit current and photovoltaic array power; recording real-time temperature and irradiance;
step S2: the frequency of the temperature and the irradiance detected by the meteorological station are 1/60HZ respectively, the temperature and the irradiance collected once in one minute are subjected to spline interpolation for three times respectively, and 200 continuous values of data in one second are obtained, namely the environmental sample with the frequency of 200HZ is obtained; building a photovoltaic array simulation model by using Simulink, simulating various working conditions by using the temperature and irradiance, and obtaining photovoltaic electrical characteristic data related to step S1, namely obtaining original simulation data;
step S3: preprocessing original measured data: globally detecting the photovoltaic array working voltage in the measured data by using a jump point detection algorithm, finding out time nodes with sudden changes of the waveform, selecting a complete voltage waveform at the maximum power point among continuous nodes, and obtaining other electrical characteristic data at the maximum power point, wherein the electrical characteristic data comprises the electrical data except the photovoltaic array working voltage in the step S1; performing integral multiple extraction on the electrical characteristic data of the maximum power point, reducing the length of the electrical characteristic data, and obtaining 8 one-dimensional electrical data based on a time sequence; splicing the 8 pieces of electrical data into a two-dimensional characteristic matrix according to rows to obtain an actually measured data set;
step S4: performing the same operation on the simulation data according to the time node obtained by detecting the actual measurement data in the step S3 to obtain a simulation data set; dividing the simulation data set and the measured data set into a training set and a verification set and a test set according to the equal proportion of each working condition; designing a convolution neural network FT-DenseNet based on fine-tuning dense connection, training by using a training set sample in a simulation data set, saving model parameters when a loss function of a model is converged, stopping training, and verifying by using a test set to obtain an optimal pre-training model;
step S5: freezing the model parameters stored in the step S4 to a characteristic extraction layer, retraining a classification layer by using a training set in the actually measured data, and verifying the FT-DenseNet training model on a verification set to obtain an optimal FT-DenseNet fault diagnosis model with most generalization capability;
step S6: and detecting and classifying the electrical characteristic data of the test set under the actual working condition by using the FT-DenseNet fault diagnosis model, and comparing the label information of the test set by using the output result of the classification layer to give the type of the actual working condition.
In this embodiment, the various operating conditions in step S1 include normal operation, group string level line fault, array level line fault, aging fault, shadow fault, and open fault: wherein a group string-level line fails, i.e. components of a single group string are shorted; array level line faults, i.e. nodes with different potential differences in different sets of strings are shorted; burn-in failure, i.e., the array lines burn in and the resistance increases; shadow fault, namely shadow occlusion of the components in the group string; open circuit fault, i.e. a certain group of strings in the array is open circuited; the environment conditions simulating each working condition are consistent with the actually acquired environment conditions.
In this embodiment, in step S2, a photovoltaic array simulation model is built by using Simulink, and various working conditions are simulated by using the temperature and irradiance, and the specific processing manner of the temperature and irradiance is as follows:
the frequency of the temperature and the frequency of the irradiance detected by the weather station are 1/60HZ respectively, and in order to meet the accuracy of real-time change of simulink, an environment sample with the frequency of 200HZ is obtained by using cubic spline interpolation; if the actual collection time is n hours, then 60 × n +1 temperature and irradiance samples are obtained, including 60 × n +1 temperature samples (0, T1), (1, T2), (2, T3), … … (60 × n, T2), (2, T3)60*n) And 60 × n +1 irradiance samples (0, G1), (1, G2), (2, G3), … … (60 × n, G3878)60*n) Finally, (60 × n) × 60 × 200 temperature and irradiance samples are obtained, and the specific formula is as follows:
environmental samples (0, y1), (1, y2), (2, y3), … … (60 × n, y60*n) Step size of 1, miIs a quadratic differential value;
coefficient of the curve:
ai＝yi
between each two environmental samples:
yi(x)＝ai+bi(x-xi)+ci(x-xi)2+di(x-xi)2
between every two environment samples by yi(x) 12000 continuous environmental samples were obtained, so there were (60 × n) × 60 × 200 temperature samples and irradiance in total; the temperature irradiance samples are used as the SIMULINK input to obtain a simulation data set.
Preferably, in the embodiment, the actual acquisition time is 3 hours, 181 temperature and irradiance samples can be obtained, and 2160000 environmental samples can be obtained finally, and the specific formula is as follows: 181 environmental samples (0, y1), (1, y2), (2, y3), … … (180, y180) with a step size of 1, miIs a second order differential value;
coefficient of the curve:
ai＝yi
between each two environmental samples:
yi(x)＝ai+bi(x-xi)+ci(x-xi)2+di(x-xi)2
between every two environment samples by yi(x) 12000 continuous environmental samples were obtained, so there were 2160000 temperature samples and irradiance in total; the temperature irradiance samples are used as the SIMULINK input to obtain a simulation data set.
In this embodiment, the step S3 specifically includes the following steps:
step S31: each sample in the original measured data set obtained in step S1 has 8 characteristics of photovoltaic array operating voltage, photovoltaic array operating current, operating current of three string groups, reference plate open-circuit voltage, reference plate short-circuit current and photovoltaic array power; independently acquiring the working voltage of the one-dimensional characteristic photovoltaic array, namely the working voltage of the photovoltaic array continuously acquired in n hours; the method comprises the following steps of (namely, continuously acquiring the working voltage of the photovoltaic array for 3 hours) acquiring a time node at a Maximum Power Point (MPP) by adopting a catastrophe point detection algorithm, wherein the specific acquisition mode is as follows:
every 200 continuous photovoltaic array working voltages are used as a set, one point is randomly selected to divide the set into two parts, residual errors of all points on the two sides and the average value of all the parts are calculated, and when the total residual error reaches the minimum value, a catastrophe point is found, and the formula is as follows:
will V1，V2，V3，...，Vk，...V200Sequentially carrying in, when J obtains the minimum value, recording Vr＝kThe time node of (2);
step S32: between every two adjacent time nodes, an integral multiple sampling function decimate is used for compressing the characteristic length, the sampling frequency of the photovoltaic array working voltage based on the time sequence is reduced to 1/40, namely 5 voltage data, and the voltage data of 6 adjacent time nodes are spliced into a 1 x 30 one-dimensional characteristic containing a complete voltage waveform;
step S33: according to the operations of the step S31 and the step S32, feature compression is carried out on the photovoltaic array working current, the working current of the three strings, the reference plate open-circuit voltage, the reference plate short-circuit current and the photovoltaic array power by the same time node to obtain 1 × 30 waveform features respectively, finally, the waveform features are spliced into an 8 × 30 two-dimensional feature matrix, and the electrical characteristic features of the simulation data set are obtained into the 8 × 30 two-dimensional feature matrix according to the method of the step S31 to the step S33.
In this embodiment, in step S4, the measured data set and the simulation data set are respectively divided into a training set, a verification set, and a test set according to different working conditions in equal proportion; specifically, each working condition of the simulation data set is divided into a training set and a testing set according to the proportion of 70 percent to 30 percent. In the measured data, the working conditions are divided into a training set, a verification set and a test set according to the proportion of 2 percent, 10 percent and 88 percent.
In the present embodiment, the network structure for designing the fine-tuning dense connection based convolutional neural network FT-DenseNet in step S4 is composed of a feature extraction layer and a classification layer; the feature extraction layer comprises two dense connection modules and a transfer layer, wherein each dense connection module is composed of 5 complex functions (BN-ReLU-Conv2 d); the transfer layer consists of a composite function and a pooling layer of BN-ReLU-Conv2d, one BN-Conv2d and two linear fully-connected layers are used as classification layers, cross entropy is used as a loss function, an Adam optimization algorithm is adopted to train the network until the loss function converges, namely the loss function is less than 0.001, and the calculation formula of the loss function is as follows:
wherein n is the number of output neurons, ykFor the desired output value, σ is the objective function, zkRepresenting the actual output value of the neuron.
In this embodiment, the specific implementation manner of obtaining the optimal and most generalized FT-DenseNet fault diagnosis model in step S5 is as follows:
step SA: pre-training FT-DenseNet by using a training set of a simulation data set until a loss function is converged, namely the loss function is less than 0.001, verifying by using a simulation test set to obtain an optimal pre-training network model, and storing network parameters with strong generalization performance; step SB: freezing a feature extraction layer of the FT-DenseNet, retraining a classification layer of the FT-DenseNet by using an actually measured data training set and a smaller learning rate (0.001-0.0001), and stopping training when a loss function is converged to obtain an optimal FT-DenseNet model.
In this embodiment, the specific implementation manner of detecting and classifying the electrical characteristic data of the test set under the actual working condition in step S6 is as follows: inputting the actual working condition test set into FT-DenseNet, outputting the final 1 × 6 category information (x1, x2, x3, x4, x5, x6) through the feature extraction layer and the classification layer, converting the information into category probability through Softmax, wherein the calculation formula is as follows,
and comparing the 1 × 6 label information of the test set, wherein the category of the maximum probability is the current working condition type of the photovoltaic array.
Firstly, acquiring electrical characteristic data and environmental data under an actual working condition, and then building a model array by using Simulink to simulate the actual working condition; acquiring simulated electrical characteristic data, then eliminating abnormal data in practice and simulation through a mutation point detection algorithm, acquiring complete electrical waveform data, sampling the electrical waveform data, compressing the characteristics, and splicing the electrical waveform data into a two-dimensional characteristic matrix. Then, designing a dense connection convolution neural network, pre-training the network by using a simulation training set and an Adam optimization algorithm, and then finely tuning the network by using a small amount of training sets under actual working conditions. And finally, detecting and classifying the photovoltaic power generation array under the working condition test set to be detected by utilizing the FT-DenseNet fault diagnosis network.
Preferably, the embodiment obtains complete and abnormal-free one-dimensional electrical waveform characteristics through mutation point detection, and splices multiple types of characteristics into a two-dimensional characteristic matrix. And simulating the actual working condition by acquiring the real-time environment temperature. The FT-DenseNet trains a good network by utilizing a simulation data set, freezes a part of the network, finely adjusts the network by using a small amount of actual data, overcomes the overfitting problem caused by a small amount of samples, and is an effective photovoltaic fault diagnosis alternative scheme when a classification network with high accuracy, good robustness and good generalization capability is obtained.
Preferably, the specific examples of the present embodiment are as follows:
fig. 2 is a physical diagram of a photovoltaic array for obtaining an actual sample in the present embodiment, wherein the photovoltaic array employs 20 solar modules with model numbers of GL-M100, two reference modules, and a 3 × 6 array. Fig. 3(a) is a Simulink photovoltaic array simulation topology of the present embodiment example, which is a 3 × 6 array and two reference assemblies, as with an actual photovoltaic array. The fault diagnosis and classification method comprises the following specific implementation processes:
step S1: gather photovoltaic electrical characteristic data under various operating mode conditions, specifically include: the photovoltaic array working voltage, the photovoltaic array working current, the working current of the three strings, the reference plate open-circuit voltage, the reference plate short-circuit current and the photovoltaic array power. Recording real-time temperature and irradiance;
step S2: and (3) respectively carrying out cubic spline interpolation on the temperature and irradiance acquired once in one minute to obtain continuous values of 200 data in one second. Building a photovoltaic array simulation model by using Simulink, simulating various working conditions by using the temperature and irradiance, and obtaining photovoltaic electrical characteristic data related to step S1;
step S3: globally detecting the photovoltaic array working voltage of the actually measured data set by using a jump point detection algorithm, finding out time nodes with sudden changes of the waveform of the photovoltaic array working voltage, selecting a complete voltage waveform at the maximum power point among continuous nodes, and obtaining other electrical characteristic data at the maximum power point, wherein the electrical characteristic data comprises the electrical data except the photovoltaic array working voltage in the step S1; the electrical characteristic data of the maximum power point is extracted by integral multiple, the length of the electrical characteristic data is reduced, and 8 one-dimensional electrical data based on the time series are obtained. Splicing the 8 pieces of electrical data into a two-dimensional characteristic matrix according to rows to obtain an actually measured data set;
step S4: performing the same operation on the simulation data according to the maximum power point detected from the actual measurement data in the step S3 to obtain a simulation data set; and respectively dividing the simulation data set and the measured data set into a training set and a verification set and a test set according to the equal proportion of each working condition. Designing a convolution neural network FT-DenseNet based on fine-tuning dense connection, training by using a training set sample in a simulation data set, saving model parameters when a loss function of a model is converged, stopping training, and verifying by using a test set to obtain an optimal pre-training model;
step S5: freezing the model parameters stored in the step S4 to a characteristic extraction layer, retraining a classification layer by using a training set in the actually measured data, and verifying the FT-DenseNet training model on a verification set to obtain an optimal FT-DenseNet fault diagnosis model with most generalization capability;
step S6: and detecting and classifying the electrical characteristic data of the test set under the actual working condition by using the FT-DenseNet fault diagnosis model, and comparing the label information of the test set by using the output result of the classification layer to give the type of the actual working condition.
Preferably, in the present embodiment, in step S2, the temperature and irradiance in the actual environment are used as input parameters to simulate the actual operating condition. The specific treatment of temperature and irradiance is as follows:
the frequency of the temperature and the frequency of the irradiance detected by the weather station are 1/60HZ respectively, and in order to meet the accuracy of real-time change of simulink, an environment sample with the frequency of 200HZ is obtained by using cubic spline interpolation; if the actual collection time is n hours, then 60 × n +1 temperature and irradiance samples are obtained, including 60 × n +1 temperature samples (0, T1), (1, T2), (2, T3), … … (60 × n, T2), (2, T3)60*n) And 60 × n +1 irradiance samples (0, G1), (1, G2), (2, G3), … … (60 × n, G3878)60*n) Finally, (60 × n) × 60 × 200 temperature and irradiance samples are obtained, and the specific formula is as follows:
environmental samples (0, y1), (1, y2), (2, y3), … … (60 × n, y60*n) Step size of 1, miIs a second order differential value;
coefficient of the curve:
ai＝yi
between each two environmental samples:
yi(x)＝ai+bi(x-xi)+ci(x-xi)2+di(x-xi)2
between every two environment samples by yi(x) 12000 continuous environmental samples were obtained, so there were (60 × n) × 60 × 200 temperature samples and irradiance in total.
Preferably, in step S3 of the present embodiment,
step S31: 2160000 samples exist in 3-hour simulation data, and each sample has 8 characteristics of photovoltaic array working voltage, photovoltaic array working current, working current of three strings, reference plate open-circuit voltage, reference plate short-circuit current and photovoltaic array power. And (3) independently obtaining the working voltage of the one-dimensional characteristic photovoltaic array, namely the working voltage of the photovoltaic array continuously collected for 3 hours. Acquiring a time node at a Maximum Power Point (MPP) by adopting a mutation point detection algorithm, wherein the specific acquisition mode is as follows:
taking every continuous 20 photovoltaic array working voltages as a set, randomly selecting a point to divide the set into two parts, calculating residual errors of all points at two sides and the average value of all parts, and finding a catastrophe point when the total residual error reaches a minimum value, wherein the formula is as follows:
will V1，V2，V3，...，Vk，...V200Sequentially carrying in, when J takes the minimum value, recording Vr＝kThe time node of (a) is,
step S32: between every two adjacent time nodes, an integral multiple sampling function decimate is used to compress the characteristic length, the sampling frequency of the photovoltaic array working voltage based on the time sequence is reduced to 1/4, namely 5 voltage data, and the voltage data of the 6 adjacent time nodes are spliced into a 1 × 30 one-dimensional characteristic containing a complete voltage waveform:
step S33: according to the operation of S31 and S32, feature compression is carried out on the photovoltaic array working current, the working current of the three strings, the reference plate open-circuit voltage, the reference plate short-circuit current and the photovoltaic array power by the same time node to obtain 1 x 30 waveform features respectively, finally, the waveform features are spliced into an 8 x 30 two-dimensional feature matrix, and the electrical characteristic features of the measured data set obtain the 8 x 30 two-dimensional feature matrix by the same method.
In the present embodiment, the Simulink component topology is shown in fig. 3(b), where R _ oc ∞, R _ de ∞ E1, E2, E3, E4 and E5 are flip-flops, and their operating conditions correspond to those shown in table 1.
In this embodiment, the fault detection accuracy of the FT-DenseNet fault diagnosis training model can reach 100% in the verification set and 99.3% in the test set, and the classification fault diagnosis rate for each subclass is shown in table 2. The confusion matrix is shown in fig. 4, and the structure of FT-DenseNet is shown in fig. 5.
TABLE 2
The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.
Claims (8)
1. A photovoltaic array fault diagnosis method based on a fine-tuning dense connection convolutional neural network is characterized by comprising the following steps: the method comprises the following steps:
step S1: acquiring original measured data: gather photovoltaic electrical characteristic data under various operating mode conditions, specifically include: the photovoltaic array power generation system comprises a photovoltaic array working voltage, a photovoltaic array working current, working currents of three strings, a reference plate open-circuit voltage, a reference plate short-circuit current and photovoltaic array power; recording real-time temperature and irradiance;
step S2: the frequency of the temperature and the irradiance detected by the meteorological station are 1/60HZ respectively, the temperature and the irradiance collected once in one minute are subjected to spline interpolation for three times respectively, and 200 continuous values of data in one second are obtained, namely the environmental sample with the frequency of 200HZ is obtained; building a photovoltaic array simulation model by using Simulink, simulating various working conditions by using the temperature and irradiance, and obtaining photovoltaic electrical characteristic data related to step S1, namely obtaining original simulation data;
step S3: preprocessing original measured data: globally detecting the working voltage of the photovoltaic array in the measured data by using a jump point detection algorithm, finding out time nodes with sudden changes of the waveform of the photovoltaic array, selecting a complete voltage waveform at the maximum power point among continuous nodes, and obtaining other electrical characteristic data at the maximum power point, wherein the electrical characteristic data comprises electrical data except the working voltage of the photovoltaic array in the step S1; performing integral multiple extraction on the electrical characteristic data of the maximum power point, reducing the length of the electrical characteristic data, and obtaining 8 one-dimensional electrical data based on a time sequence; splicing the 8 pieces of electrical data into a two-dimensional characteristic matrix according to rows to obtain an actually measured data set;
step S4: performing the same operation on the simulation data according to the time node obtained by detecting the actual measurement data in the step S3 to obtain a simulation data set; dividing the simulation data set and the measured data set into a training set and a verification set and a test set according to the equal proportion of each working condition; designing a convolution neural network FT-DenseNet based on fine-tuning dense connection, training by using a training set sample in a simulation data set, saving model parameters when a loss function of a model is converged, stopping training, and verifying by using a test set to obtain an optimal pre-training model;
step S5: freezing the model parameters stored in the step S4 to a characteristic extraction layer, retraining a classification layer by using a training set in the actually measured data, and verifying the FT-DenseNet training model on a verification set to obtain an optimal FT-DenseNet fault diagnosis model with most generalization capability;
step S6: and detecting and classifying the electrical characteristic data of the test set under the actual working condition by using the FT-DenseNet fault diagnosis model, and comparing the label information of the test set by using the output result of the classification layer to give the type of the actual working condition.
2. The method for diagnosing the faults of the photovoltaic array based on the fine-tuning dense connection convolutional neural network as claimed in claim 1, wherein the method comprises the following steps: the various operating conditions in step S1 include normal operation, group serial line fault, array level line fault, aging fault, shadow fault, and open circuit fault: wherein a group string-level line fails, i.e. components of a single group string are shorted; array level line faults, i.e. nodes with different potential differences in different sets of strings are shorted; aging faults, i.e., array lines age and resistance increases; shadow fault, namely shadow occlusion of the components in the group string; open circuit failure, i.e., a certain group of strings in the array is open circuited; the environment conditions simulating each working condition are consistent with the actually acquired environment conditions.
3. The method for diagnosing the faults of the photovoltaic array based on the fine-tuning dense connection convolutional neural network as claimed in claim 1, wherein the method comprises the following steps: in step S2, a photovoltaic array simulation model is built by using Simulink, various working conditions are simulated by using the temperature and irradiance, and the specific processing method for the temperature and irradiance is as follows:
the frequency of the temperature and the frequency of the irradiance detected by the weather station are 1/60HZ respectively, and in order to meet the accuracy of the real-time change of Simulink, an environment sample with the frequency of 200HZ is obtained by using cubic spline interpolation; if the actual collection time is n hours, then 60 x n +1 temperature and irradiance samples are obtained, including 60 x n +1 temperature samples (0, T1), (1, T2), (2, T3), … … (60 x n, T2), (2, T3), and so on60*n) And 60 × n +1 irradiance samples (0, G1), (1, G2), (2, G3), … … (60 × n, G)60*n) Finally, (60 × n) × 60 × 200 temperature and irradiance samples are obtained, and the specific formula is as follows:
environmental samples (0, y1), (1, y2), (2, y3), … … (60 × n, y60*n) Which isStep size of 1, miIs a second order differential value;
coefficient of the curve:
ai＝yi
between each two environmental samples:
yi(x)＝ai+bi(x-xi)+ci(x-xi)2+di(x-xi)2
between every two environment samples by yi(x) 12000 continuous environmental samples were obtained, so there were (60 × n) × 60 × 200 temperature samples and irradiance in total; the temperature irradiance samples are used as the SIMULINK input to obtain a simulation data set.
4. The method for diagnosing the faults of the photovoltaic array based on the fine-tuning dense connection convolutional neural network as claimed in claim 1, wherein the method comprises the following steps: the step S3 specifically includes the following steps:
step S31: n hours of measured data are concentrated with (60 × n) × 60 × 200 samples, and each sample in the original measured data set obtained in step S1 has 6 characteristics of photovoltaic array working voltage, photovoltaic array working current, working current of three strings, reference plate open-circuit voltage, reference plate short-circuit current and photovoltaic array power; independently acquiring the working voltage of the one-dimensional characteristic photovoltaic array, namely the working voltage of the photovoltaic array continuously acquired in n hours; and acquiring the time node at the maximum power point by adopting a mutation point detection algorithm, wherein the specific acquisition mode is as follows:
every 200 continuous photovoltaic array working voltages are used as a set, one point is randomly selected to divide the set into two parts, residual errors of all points on the two sides and the average value of all parts are calculated, and when the total residual error reaches the minimum value, a catastrophe point is found, and the formula is as follows:
will V1，V2，V3，…，Vk，...V200Sequentially carrying in, when J takes the minimum value, recording Vr＝kA time node;
step S32: between every two adjacent time nodes, an integral multiple sampling function decimate is used for compressing the characteristic length, the sampling frequency of the photovoltaic array working voltage based on the time sequence is reduced to 1/40, namely 5 voltage data, and the voltage data of 6 adjacent time nodes are spliced into a 1 x 30 one-dimensional characteristic containing a complete voltage waveform;
step S33: according to the operations of the step S31 and the step S32, feature compression is carried out on the photovoltaic array working current, the working current of the three strings, the reference plate open-circuit voltage, the reference plate short-circuit current and the photovoltaic array power by the same time node to obtain 1 × 30 waveform features respectively, finally, the waveform features are spliced into an 8 × 30 two-dimensional feature matrix, and the electrical characteristic features of the simulation data set are obtained into the 8 × 30 two-dimensional feature matrix according to the method of the step S31 to the step S33.
5. The method for diagnosing the faults of the photovoltaic array based on the fine-tuning dense connection convolutional neural network as claimed in claim 1, wherein the method comprises the following steps: in the step S4, dividing the actual measurement data set and the simulation data set into a training set, a verification set and a test set according to different working conditions in equal proportion; dividing each working condition of a simulation data set into a training set and a testing set according to the proportion of 70 percent and 30 percent; in the measured data, each working condition is divided into a training set, a verification set and a test set according to the proportion of 2%, 10% and 88%.
6. The method for diagnosing the fault of the photovoltaic array based on the fine-tuning dense connection convolutional neural network as claimed in claim 1, wherein: the network structure of the convolutional neural network FT-DenseNet based on fine tuning dense connection designed in the step S4 is composed of a feature extraction layer and a classification layer; the feature extraction layer comprises two dense connection modules and a transfer layer, and each dense connection module consists of 5 complex functions; the transfer layer consists of a composite function and a pooling layer of BN-ReLU-Conv2d, one BN-Conv2d and two linear fully-connected layers are used as classification layers, cross entropy is used as a loss function, an Adam optimization algorithm is adopted to train the network until the loss function converges, namely the loss function is less than 0.001, and the calculation formula of the loss function is as follows:
wherein n is the number of output neurons, ykFor the desired output value, σ is the objective function, zkRepresenting the actual output value of the neuron.
7. The method for diagnosing the faults of the photovoltaic array based on the fine-tuning dense connection convolutional neural network as claimed in claim 1, wherein the method comprises the following steps: the specific implementation manner of obtaining the optimal FT-DenseNet fault diagnosis model with the most generalization capability in step S5 is as follows:
step SA: pre-training FT-DenseNet by using a training set of a simulation data set until a loss function is converged, namely the loss function is less than 0.001, verifying by using a simulation test set to obtain an optimal pre-training network model, and storing network parameters with strong generalization performance;
step SB: freezing a feature extraction layer of the FT-DenseNet, retraining a classification layer of the FT-DenseNet by using an actually measured data training set and a smaller learning rate range of 0.001-0.0001, and stopping training when a loss function is converged to obtain an optimal FT-DenseNet model.
8. The method for diagnosing the fault of the photovoltaic array based on the fine-tuning dense connection convolutional neural network as claimed in claim 1, is characterized in that: the specific implementation manner of detecting and classifying the test set electrical characteristic data under the actual working condition in the step S6 is as follows: inputting the actual working condition test set into FT-DenseNet, outputting the final 1 × 6 class information (x1, x2, x3, x4, x5, x6) through the feature extraction layer and the classification layer, converting the information into class probability through Softmax, wherein the calculation formula is as follows,
and comparing the 1 × 6 label information of the test set, wherein the category of the maximum probability is the current working condition type of the photovoltaic array.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011199284.3A CN112787591B (en) | 2020-10-30 | 2020-10-30 | Photovoltaic array fault diagnosis method based on fine-tuning dense connection convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011199284.3A CN112787591B (en) | 2020-10-30 | 2020-10-30 | Photovoltaic array fault diagnosis method based on fine-tuning dense connection convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112787591A CN112787591A (en) | 2021-05-11 |
| CN112787591B true CN112787591B (en) | 2022-05-13 |
Family
ID=75750150
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011199284.3A Active CN112787591B (en) | 2020-10-30 | 2020-10-30 | Photovoltaic array fault diagnosis method based on fine-tuning dense connection convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112787591B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113283175B (en) * | 2021-06-09 | 2023-02-03 | 上海交通大学 | Photovoltaic power station joint fault diagnosis method based on asynchronous decentralized federal learning |
| CN113538167A (en) * | 2021-07-06 | 2021-10-22 | 江阴长仪集团有限公司 | Photovoltaic system state identification method and system based on deep neural network |
| CN113848078B (en) * | 2021-08-25 | 2023-12-08 | 祥博传热科技股份有限公司 | Detection method for judging failure of heat pipe in heat pipe radiator |
| CN115659799B (en) * | 2022-10-24 | 2023-05-16 | 国网浙江省电力有限公司电力科学研究院 | Lithium battery energy storage power station fault diagnosis method with threshold self-adaption function |
| CN117851887A (en) * | 2024-01-10 | 2024-04-09 | 浙江大学 | Deep learning-based open-circuit fault diagnosis method for inverter power device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109873610A (en) * | 2019-03-19 | 2019-06-11 | 福州大学 | Diagnosing failure of photovoltaic array method based on IV characteristic and depth residual error network |
| US20190386611A1 (en) * | 2018-08-10 | 2019-12-19 | Jiangnan University | Photovoltaic Array Fault Diagnosis Method Based on Random Forest Algorithm |
-
2020
- 2020-10-30 CN CN202011199284.3A patent/CN112787591B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190386611A1 (en) * | 2018-08-10 | 2019-12-19 | Jiangnan University | Photovoltaic Array Fault Diagnosis Method Based on Random Forest Algorithm |
| CN109873610A (en) * | 2019-03-19 | 2019-06-11 | 福州大学 | Diagnosing failure of photovoltaic array method based on IV characteristic and depth residual error network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112787591A (en) | 2021-05-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112787591B (en) | Photovoltaic array fault diagnosis method based on fine-tuning dense connection convolutional neural network | |
| CN109873610B (en) | Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network | |
| CN108062571B (en) | Photovoltaic array fault diagnosis method based on differential evolution random forest classifier | |
| CN110398663B (en) | Flexible direct current power grid fault identification method based on convolutional neural network | |
| CN109660206B (en) | Wasserstein GAN-based photovoltaic array fault diagnosis method | |
| CN109766952B (en) | Photovoltaic array fault detection method based on partial least square method and extreme learning machine | |
| CN109459669B (en) | 10kV power distribution network single-phase earth fault section positioning method | |
| CN110008628B (en) | Photovoltaic array fault parameter identification method | |
| CN109670553B (en) | Photovoltaic array fault diagnosis method based on adaptive neural fuzzy inference system | |
| CN109034220A (en) | A kind of intelligent photovoltaic array method for diagnosing faults based on optimal rotation forest | |
| CN110570122A (en) | Offshore wind power plant reliability assessment method considering wind speed seasonal characteristics and current collection system element faults | |
| CN108805256A (en) | Photovoltaic Module Fault Diagnosis Method Based on Cuckoo Algorithm and BP Neural Network | |
| CN111999591B (en) | Method for identifying abnormal state of primary equipment of power distribution network | |
| CN112485587B (en) | Layered positioning method for fault section of distribution network containing distributed photovoltaic | |
| CN114528883A (en) | Wind power converter IGBT fault identification method | |
| Fu et al. | Fault diagnosis of photovoltaic array based on SE-ResNet | |
| Sima et al. | Edge–cloud collaboration detection approach for small-sample imbalanced faults in power lines | |
| CN118130968A (en) | Data-driven non-invasive anti-island detection method based on deep learning | |
| Han et al. | An intelligent fault diagnosis method for PV arrays based on an improved rotation forest algorithm | |
| CN113595132A (en) | Photovoltaic online parameter identification method based on global maximum power point tracking and hybrid optimization algorithm | |
| CN117150220A (en) | Photovoltaic array fault diagnosis method based on machine learning | |
| CN110348489B (en) | Transformer partial discharge mode identification method based on self-coding network | |
| Dehghani et al. | Distribution feeder classification based on self organized maps (case study: Lorestan province, Iran) | |
| Vahidi et al. | Fault detection and classification in PV arrays using machine learning algorithms in the presence of noisy data | |
| CN114157552A (en) | Distribution network fault detection method based on twin timing diagram network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |