CN113283580A - Automatic fault detection method for solar cell panel - Google Patents
Automatic fault detection method for solar cell panel Download PDFInfo
- Publication number
- CN113283580A CN113283580A CN202110482858.6A CN202110482858A CN113283580A CN 113283580 A CN113283580 A CN 113283580A CN 202110482858 A CN202110482858 A CN 202110482858A CN 113283580 A CN113283580 A CN 113283580A
- Authority
- CN
- China
- Prior art keywords
- current
- vector
- data set
- layer
- fault
- 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 22
- 238000000034 method Methods 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 36
- 230000006870 function Effects 0.000 claims abstract description 29
- 238000012360 testing method Methods 0.000 claims abstract description 22
- 238000012423 maintenance Methods 0.000 claims abstract description 13
- 238000003062 neural network model Methods 0.000 claims abstract description 8
- 238000013507 mapping Methods 0.000 claims abstract description 5
- 239000013598 vector Substances 0.000 claims description 68
- 230000008859 change Effects 0.000 claims description 16
- 238000005070 sampling Methods 0.000 claims description 15
- 238000004519 manufacturing process Methods 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 9
- 230000032683 aging Effects 0.000 claims description 6
- 230000005540 biological transmission Effects 0.000 claims description 6
- 238000004891 communication Methods 0.000 claims description 6
- 238000010248 power generation Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 230000001788 irregular Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims description 3
- 238000003745 diagnosis Methods 0.000 abstract description 8
- 238000013528 artificial neural network Methods 0.000 abstract description 3
- 230000006872 improvement Effects 0.000 abstract description 2
- 239000000284 extract Substances 0.000 abstract 1
- 238000011161 development Methods 0.000 description 6
- 230000018109 developmental process Effects 0.000 description 6
- 230000007547 defect Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000003912 environmental pollution Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 239000011800 void material Substances 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003631 expected effect Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—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
-
- 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
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Photovoltaic Devices (AREA)
Abstract
The invention relates to an automatic fault detection method for a solar cell panel, belonging to the technical field of fault detection of photovoltaic cell panels; the technical problem to be solved is as follows: the improvement of the automatic detection method for the faults of the solar cell panel is provided; the technical scheme for solving the technical problem is as follows: the method comprises the steps that a training data set and a testing data set suitable for a neural network fault diagnosis model are manufactured according to real-time current data collected by a photovoltaic power plant, the data set is sequentially subjected to three residual modules of a neural network to extract deep features of current, the data set firstly extracts the deep features of current information through expansion convolution of the residual modules in the neural network model, a group of thresholds are automatically learned through a small sub-network according to different samples, soft thresholding is performed on the learned thresholds, finally, residual functions and identity mapping functions learned through the network are weighted, and finally the data set enters a full connection layer to be classified to complete fault detection of a photovoltaic cell panel; the solar photovoltaic panel maintenance device is applied to daily maintenance of the solar photovoltaic panel.
Description
Technical Field
The invention discloses an automatic fault detection method for a solar cell panel, and belongs to the technical field of fault detection of photovoltaic cell panels.
Background
With the development of global economy, the economic development of countries in the world is rapid, the living standard of people is remarkably improved, the demand of energy is increased day by day, the shortage of non-renewable energy becomes an inevitable problem, the solar energy is simple in acquisition mode and not limited by regions, the total amount of solar radiation energy is considerable, the solar energy is relatively long, the use of the solar energy cannot cause environmental pollution, the development of green energy plays a role in solving the problem of energy shortage and environmental pollution caused by non-renewable energy in countries, and the photovoltaic industry is developed along with the development of the solar energy.
The photovoltaic power station is generally large in scale and wide in occupied area, and an energy company generally selects and builds a suburban area with rare people from an economic perspective, so that in order to ensure the smooth operation of the photovoltaic power station, an energy company can configure a few maintenance personnel for the photovoltaic power station; due to the fact that the scale of the photovoltaic power station is large, the inspection and maintenance range of maintenance personnel is wide, the number of photovoltaic modules is large, and the method has extremely important research significance for the maintenance personnel to accurately position the physical positions and fault types of the fault photovoltaic modules.
In recent years, the state greatly supports the development of artificial intelligence and big data, the artificial intelligence is applied to the aspects of life, the artificial intelligence also greatly promotes the fault diagnosis research development of a solar photovoltaic module, an infrared image is used for detecting hot spot faults of the photovoltaic module, an infrared image detection method is mainly used for detecting the faults based on the temperature difference of a fault module and a normal module, a sensor is used for detecting the faults, the battery of a photovoltaic system is processed in a blocking mode, a plurality of sensors are respectively placed at proper positions, the current of each branch and the voltage of a battery panel of the photovoltaic module are detected, and the fault occurrence position is comprehensively judged according to the collected current and voltage values; the infrared image detection method and the multi-sensor method require a large number of infrared cameras and sensor equipment for large photovoltaic arrays, so that the economic cost is too high, the method is not suitable and is only suitable for small photovoltaic power stations; in addition, a fault diagnosis method based on electrical measurement carries out fault diagnosis by measuring an I-V curve, a grounding capacitance value, a feedback signal and the like, but the method has the defects that the specific position of the fault can not be accurately positioned, and the type of the fault which can be judged is limited; there is therefore a need for an improved method of fault determination and adjustment of existing photovoltaic panels.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention aims to solve the technical problems that: an improvement of an automatic detection method for solar panel faults is provided.
In order to solve the technical problems, the invention adopts the technical scheme that: a solar cell panel fault automatic detection method comprises the following working parameter detection and adjustment steps:
the method comprises the following steps: detecting parameters of a photovoltaic cell panel in work, collecting a current data set of the photovoltaic cell panel, performing feature extraction on the current data set, manufacturing a training data set and a test data set, and manufacturing a label data set according to visual current data;
step two: performing learning training by using a soft thresholding time sequence convolution neural network model, and adjusting and optimizing the structure and parameters of the model according to a loss function value, a correct rate and a correct rate obtained by verifying a test data set during model training;
step three: and repeatedly verifying and optimizing the model by using current data generated by the actual operation of the photovoltaic power station.
In the first step, a label data set corresponding to the branch is specifically manufactured according to a current curve graph, and the working state of the branch is divided into three levels according to the current data expression form actually acquired during the operation of the branch:
the working state of the grade one is as follows: the method comprises two faults of communication transmission fault, hot spot burn-through or fusing:
the current curve of the communication transmission fault is irregular in change, and a current value point obviously departing from the common sense exists;
the current curve of the hot spot burn-through or fusing fault is suddenly reduced from normal power generation to 0, and the value of 0 is kept unchanged;
the working state of the grade two is as follows: the method comprises three fault types of component aging fault, equipment starting fault and component maintenance fault:
the aging fault of the assembly is specifically that the current value of a certain branch under the same combiner box in operation is lower than the current of other branches under the same combiner box by 0.5A or more than 0.5A;
the equipment starts the trouble, specifically the current is leveled at 0 value or any value keeps unchanged;
the component maintenance fault is characterized in that a current curve shows that the current curve of a certain abnormal branch under the same combiner box is recovered to a normal power generation state from a value of 0;
the working state of grade three is as follows: and (4) normally generating power, wherein the current curve shows that the current changes along with the change of sunlight in the state, and the change trend of the current in one day is approximate to a normal curve.
In the first step, the current data set is collected to extract features, and the specific steps of manufacturing the training data set and the test data set comprise:
the current characteristic data of the input model at the same time step is obtained by down-sampling the current data with the original sampling frequency of 1s into 1-10min for sampling once, setting the sampling interval to be from sunrise in the morning to sunset in the evening, and when each sampling point is extracted by the characteristics, obtaining the current value, the current average value, the maximum value, the variance, the minimum value, the standard deviation and the current change rate of each branch circuit changing along with the change of time and solar irradiation in a data group in a plurality of branch circuits under the same combiner box.
In the second step, the specific step of performing learning training on the current data by using the soft thresholded time sequence convolution neural network model comprises the following steps:
step 2.1: training a first layer of residual error module of a soft thresholding time sequence convolution neural network model by using a processed four-dimensional data set A, sending the data set into an expansion convolution layer of a residual error module with a first layer of residual error module having a cavity factor of 1 and a filter of 32 for convolution, and extracting deep level current characteristic information to obtain an output vector M;
step 2.2: sequentially inputting the convolved output data into a normalized network layer, an absolute value function layer and a global average pooling network layer to obtain an output vector M;
step 2.3: sequentially inputting the obtained vector into an expansion convolution layer with a cavity factor of 1 and a filter of 32, a normalization network layer, an absolute value function layer and a global average pooling network layer again to extract deep features to obtain an output vector B;
step 2.4: inputting the output vector B into the absolute value function layer again, and solving the absolute values of all elements in the vector to obtain an output vector C;
step 2.5: inputting the output vector C into a global average pooling network layer to obtain an output vector D;
step 2.6: sequentially inputting the output vector D into a full-connection network layer, a normalization network layer, an absolute value function layer, a full-connection network layer and a Sigmoid activation function layer to learn a group of scaling coefficients to obtain a vector E, and performing matrix point-by-point multiplication on the obtained vector D and the vector E to obtain a vector F;
step 2.7: carrying out soft thresholding on the obtained output vector F and the vector B to obtain an output vector G;
step 2.8: inputting the vector A into a convolution network layer with a filter of 32 to learn an identity mapping function to obtain an output vector H;
step 2.9: weighting the vector G and the vector H to obtain a vector I;
step 2.10: and sequentially sending the vector I into a second layer residual error module and a third layer residual error module which have the same model structure and have the cavity factors of 2 and 4 for learning and training, sending the training output result of the third layer residual error module into a full-connection layer for classification to obtain a classification result, and adjusting and optimizing the model according to the loss function value and the accuracy of the test set.
In the third step, the model is repeatedly optimized by using real current data in actual production, and the model is repeatedly verified and optimized by using current data of a photovoltaic array region different from the training set and different seasons.
Compared with the prior art, the invention has the beneficial effects that: the invention provides a solar photovoltaic module fault diagnosis and detection method based on a soft thresholding time sequence convolution neural network (ST-TCN). the method is mainly characterized in that a built ST-TCN deep learning fault diagnosis model is used for training according to the change trend of current data acquired when a photovoltaic module operates as the fault classification basis of the photovoltaic module, the model is repeatedly trained until the model is adjusted to an optimal model, and the trained model is stored; the method uses the current data acquired by the power station to monitor the physical position and the fault level of the solar photovoltaic module, can be directly put into an actual power plant to use in the using process, does not need to add additional equipment, and reduces the expenditure of the operation and maintenance cost of the power plant.
Drawings
The invention is further described below with reference to the accompanying drawings:
FIG. 1 is a diagram of a photovoltaic array topology for inspecting photovoltaic panels in accordance with the present invention;
FIG. 2 is a schematic flow chart illustrating steps of the method for detecting the failure of the solar photovoltaic cell panel according to the present invention;
FIG. 3 is a schematic diagram of a model structure of the method for detecting the fault of the solar photovoltaic cell panel according to the invention;
fig. 4 is a schematic structural diagram of a residual error module in the method for detecting the fault of the solar photovoltaic cell panel.
Detailed Description
As shown in FIGS. 1 to 4, the invention provides a method for detecting, adjusting and optimizing the faults of a solar photovoltaic panel based on a soft-thresholding time sequence convolutional neural network (ST-TCN), which is used for manufacturing a training data set and a testing data set suitable for an ST-TCN fault diagnosis model according to real-time current data acquired by a photovoltaic power plant, and then sequentially passing the data set through three residual modules of the ST-TCN network to extract the deep features of the current, firstly extracting the deep features of the current information through the expansion convolution of the residual modules in the ST-TCN model, then automatically learning a group of thresholds through a small sub-network according to different samples, then carrying out soft thresholding on the learned thresholds, finally weighting the residual functions and the identity mapping functions learned by the network, and finally entering a full connection layer to carry out classification to complete the fault detection of the photovoltaic cell panel.
Specifically, the invention discloses a photovoltaic cell panel working parameter detection and adjustment method based on a time sequence convolution network, which comprises the following detection steps:
the method comprises the following steps: detecting parameters of a photovoltaic cell panel in work, collecting a current data set of the photovoltaic cell panel, performing feature extraction on the current data set, manufacturing a training data set and a test data set, and manufacturing a label data set according to visual current data;
step two: performing learning training by using a soft thresholding time sequence convolution neural network model, and adjusting and optimizing the structure and parameters of the model according to a loss function value, a correct rate and a correct rate obtained by verifying a test data set during model training;
step three: and repeatedly verifying and optimizing the model by using current data generated by the actual operation of the photovoltaic power station.
In the first step, a label data set corresponding to the branch is specifically manufactured according to a current curve graph, and the working state of the branch is divided into three levels according to the current data expression form actually acquired during the operation of the branch:
the working state of the grade one is as follows: the method comprises two faults of communication transmission fault, hot spot burn-through or fusing:
the current curve of the communication transmission fault is irregular in change, and a current value point obviously departing from the common sense exists;
the current curve of the hot spot burn-through or fusing fault is suddenly reduced from normal power generation to 0, and the value of 0 is kept unchanged;
the working state of the grade two is as follows: the method comprises three fault types of component aging fault, equipment starting fault and component maintenance fault:
the aging fault of the assembly is specifically that the current value of a certain branch under the same combiner box in operation is lower than the current of other branches under the same combiner box by 0.5A or more than 0.5A;
the equipment starts the trouble, specifically the current is leveled at 0 value or any value keeps unchanged;
the component maintenance fault is characterized in that a current curve shows that the current curve of a certain abnormal branch under the same combiner box is recovered to a normal power generation state from a value of 0;
the working state of grade three is as follows: and (4) normally generating power, wherein the current curve shows that the current changes along with the change of sunlight in the state, and the change trend of the current in one day is approximate to a normal curve.
In the first step, in actual operation, the current data set is collected to extract features, and the specific steps of manufacturing the training data set and the test data set comprise: the method comprises the steps of sampling current data with an original sampling frequency of 1s once in 1min, sampling for a sampling interval [ 8:00 in the morning to 6:00 in the evening ], when any sampling point is extracted in a characteristic mode, obtaining current characteristic data of an input model at the same time step by using a current value, a current average value, a maximum value, a variance, a minimum value and a standard deviation of a data group formed by 15 branches under the same combiner box and a current change rate of each branch changing along with the change of time and solar irradiation, using the characteristic current data of a 16-month photovoltaic array area as a training set, and using the characteristic current data of a 17-month photovoltaic array area as a testing set.
In the second step, in the actual operation, in the step of performing learning training on the current data by using the ST-TCN model, the processed four-dimensional data set A is used: (M,600,7,1), enter the three-tier residual module of ST-TCN for training:
firstly, entering an expansion convolutional layer with a residual module of a first layer of which the void factor is 1 and a filter (Filters) of which is 32, carrying out convolution to obtain an output vector (M,600,7,32), sequentially inputting a normalized network layer, an absolute value function layer (Relu activation function layer) and a global average pooling network layer (Dropout layer) which are output after convolution to obtain the output vector (M,600,7,32), sequentially inputting the obtained vector to the expansion convolutional layer with the void factor of 1 and the filter of which is 32, the normalized network layer, the absolute value function layer and the global average pooling network layer to carry out deep level feature extraction to obtain an output vector B: (M,600,7, 32).
And inputting the obtained output vector into the absolute value function layer again, and solving the absolute value of all elements in the vector to obtain an output vector C: (M,32), inputting the vector C into the global average pooling network layer to obtain an output vector D, (M,32), sequentially inputting the obtained output vector into the full-connection network layer, the normalization network layer, the absolute value function layer, the full-connection network layer and the Sigmoid activation function layer to learn a group of scaling coefficients to obtain a vector E, (M,32), and performing matrix point-by-point multiplication on the obtained vector D and the vector E to obtain a vector F, (M,1,1, 32).
Carrying out soft thresholding on the obtained output vector F and the vector B to obtain an output vector G: (M,600,7, 32).
And (A): (M,600,7,1) learning an identity mapping function for a filter of 32 convolutional network layers to obtain an output vector H: (M,600,7, 32).
Weighting the vector G and the vector H to obtain a vector I: (M,600,7, 32).
And sequentially sending the vectors into a second layer residual error module and a third layer residual error module which have the same model structure and the cavity factors of the expansion convolution of 2 and 4 respectively for learning training, sending the training output result of the third layer residual error module into a full-link layer for classification to obtain a classification result, and optimizing the model according to the loss function value and the accuracy of the test set.
And repeatedly adjusting and optimizing the model by using real current data in actual production, and repeatedly verifying and optimizing the model by using current data of other photovoltaic array regions different from the training set and different seasons.
Based on the fault detection and adjustment method, the following embodiments are provided for operating specific cases:
the experimental data come from a certain photovoltaic energy company. The basic information of the experimental data is as follows:
photovoltaic power plant contains 60 photovoltaic battle array districts, and every photovoltaic battle array district contains 2 inverters, contains 1 to 7 conflux case under 1 inverter, contains 8 to 14 conflux cases under 2 inverters, contains 15 photovoltaic branch roads under 1 conflux case, contains 21 panels and establishes ties under 1 branch road and constitutes, and data acquisition equipment can accurately gather the real-time current data of every subassembly branch road.
The data of 2 photovoltaic array areas belonging to a certain photovoltaic power station 2018 from 1 month to 2018 from 10 months are used for test verification. The data distribution is that 16 photovoltaic array area data are used as a training data set, and 17 photovoltaic array area data are used as a testing data set. The data set was trained for 63840 data days with 14 combiner boxes, 210 branches, and 304 zones. The training environment is Linux 6, the display card is 8 TeslaT4 display cards, the experimental environment is Tensorflow-GPU1.12.0, the training iteration times are 30 times, and model parameter tuning is combined with grid search by using experience tuning. For comparison experiments, the GRU model, the LSTM model, the CNN-LSTM model, the TCN model and the ST-TCN model are respectively used for comparison experiments, and the experiments adopt the same experiment environment and iteration times.
Table 1 shows the experimental results, and the optimal model structure is the ST-TCN model. And the cavity factors of the expansion convolution are respectively 1,2 and 4.
After the test data set is used for testing, the accuracy rate reaches 92.45 percent. Test results show that the fault diagnosis effect of the ST-TCN model is superior to that of other models, and the expected effect is achieved.
TABLE 1 test results
The invention provides an ST-TCN-based solar photovoltaic cell panel fault detection and adjustment method, wherein a current expression form acquired when a photovoltaic module operates is used as a basis for classifying faults of the photovoltaic module, and a deep learning model structure of the ST-TCN is used for training and learning. After verification testing is carried out on a large sample data set, the trained model is stored, so that the time for model training is saved, the model can be better put into use in the detection of the fault problem of the actual power station, equipment does not need to be additionally erected in the power station, the running speed is high, the defect that additional data acquisition equipment needs to be added in the power station in the existing methods is overcome, the defect that the existing methods only can detect a plurality of faults is overcome, and the daily use requirement of the power station is met.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (5)
1. A solar cell panel fault automatic detection method is characterized in that: the method comprises the following working parameter detection and adjustment steps:
the method comprises the following steps: detecting parameters of a photovoltaic cell panel in work, collecting a current data set of the photovoltaic cell panel, performing feature extraction on the current data set, manufacturing a training data set and a test data set, and manufacturing a label data set according to visual current data;
step two: performing learning training by using a soft thresholding time sequence convolution neural network model, and adjusting and optimizing the structure and parameters of the model according to a loss function value, a correct rate and a correct rate obtained by verifying a test data set during model training;
step three: and repeatedly verifying and optimizing the model by using current data generated by the actual operation of the photovoltaic power station.
2. The method for automatically detecting the faults of the solar panels as claimed in claim 1, wherein the method comprises the following steps: in the first step, a label data set corresponding to the branch is specifically manufactured according to a current curve graph, and the working state of the branch is divided into three levels according to the current data expression form actually acquired during the operation of the branch:
the working state of the grade one is as follows: the method comprises two faults of communication transmission fault, hot spot burn-through or fusing:
the current curve of the communication transmission fault is irregular in change, and a current value point obviously departing from the common sense exists;
the current curve of the hot spot burn-through or fusing fault is suddenly reduced from normal power generation to 0, and the value of 0 is kept unchanged;
the working state of the grade two is as follows: the method comprises three fault types of component aging fault, equipment starting fault and component maintenance fault:
the aging fault of the assembly is specifically that the current value of a certain branch under the same combiner box in operation is lower than the current of other branches under the same combiner box by 0.5A or more than 0.5A;
the equipment starts the trouble, specifically the current is leveled at 0 value or any value keeps unchanged;
the component maintenance fault is characterized in that a current curve shows that the current curve of a certain abnormal branch under the same combiner box is recovered to a normal power generation state from a value of 0;
the working state of grade three is as follows: and (4) normally generating power, wherein the current curve shows that the current changes along with the change of sunlight in the state, and the change trend of the current in one day is approximate to a normal curve.
3. The method for automatically detecting the faults of the solar panels as claimed in claim 2, wherein the method comprises the following steps: in the first step, the current data set is collected to extract features, and the specific steps of manufacturing the training data set and the test data set comprise:
the current characteristic data of the input model at the same time step is obtained by down-sampling the current data with the original sampling frequency of 1s into 1-10min for sampling once, setting the sampling interval to be from sunrise in the morning to sunset in the evening, and when each sampling point is extracted by the characteristics, obtaining the current value, the current average value, the maximum value, the variance, the minimum value, the standard deviation and the current change rate of each branch circuit changing along with the change of time and solar irradiation in a data group in a plurality of branch circuits under the same combiner box.
4. The method for automatically detecting the faults of the solar panels as claimed in claim 3, wherein the method comprises the following steps: in the second step, the specific step of performing learning training on the current data by using the soft thresholded time sequence convolution neural network model comprises the following steps:
step 2.1: training a first layer of residual error module of a soft thresholding time sequence convolution neural network model by using a processed four-dimensional data set A, sending the data set into an expansion convolution layer of a residual error module with a first layer of residual error module having a cavity factor of 1 and a filter of 32 for convolution, and extracting deep level current characteristic information to obtain an output vector M;
step 2.2: sequentially inputting the convolved output data into a normalized network layer, an absolute value function layer and a global average pooling network layer to obtain an output vector M;
step 2.3: sequentially inputting the obtained vector into an expansion convolution layer with a cavity factor of 1 and a filter of 32, a normalization network layer, an absolute value function layer and a global average pooling network layer again to extract deep features to obtain an output vector B;
step 2.4: inputting the output vector B into the absolute value function layer again, and solving the absolute values of all elements in the vector to obtain an output vector C;
step 2.5: inputting the output vector C into a global average pooling network layer to obtain an output vector D;
step 2.6: sequentially inputting the output vector D into a full-connection network layer, a normalization network layer, an absolute value function layer, a full-connection network layer and a Sigmoid activation function layer to learn a group of scaling coefficients to obtain a vector E, and performing matrix point-by-point multiplication on the obtained vector D and the vector E to obtain a vector F;
step 2.7: carrying out soft thresholding on the obtained output vector F and the vector B to obtain an output vector G;
step 2.8: inputting the vector A into a convolution network layer with a filter of 32 to learn an identity mapping function to obtain an output vector H;
step 2.9: weighting the vector G and the vector H to obtain a vector I;
step 2.10: and sequentially sending the vector I into a second layer residual error module and a third layer residual error module which have the same model structure and have the cavity factors of 2 and 4 for learning and training, sending the training output result of the third layer residual error module into a full-connection layer for classification to obtain a classification result, and adjusting and optimizing the model according to the loss function value and the accuracy of the test set.
5. The method for automatically detecting the faults of the solar panels as claimed in claim 4, wherein the method comprises the following steps: in the third step, the model is repeatedly optimized by using real current data in actual production, and the model is repeatedly verified and optimized by using current data of a photovoltaic array region different from the training set and different seasons.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110482858.6A CN113283580A (en) | 2021-04-30 | 2021-04-30 | Automatic fault detection method for solar cell panel |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110482858.6A CN113283580A (en) | 2021-04-30 | 2021-04-30 | Automatic fault detection method for solar cell panel |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113283580A true CN113283580A (en) | 2021-08-20 |
Family
ID=77277997
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110482858.6A Pending CN113283580A (en) | 2021-04-30 | 2021-04-30 | Automatic fault detection method for solar cell panel |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113283580A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113625094A (en) * | 2021-09-11 | 2021-11-09 | 沈阳工业大学 | Subway platform screen door fault diagnosis system |
| CN113688987A (en) * | 2021-08-30 | 2021-11-23 | 远景智能国际私人投资有限公司 | Training method of photovoltaic monitoring model, monitoring method of photovoltaic equipment and equipment |
| CN114298219A (en) * | 2021-12-27 | 2022-04-08 | 江苏国科智能电气有限公司 | Wind driven generator fault diagnosis method based on deep space-time feature extraction |
| CN116699297A (en) * | 2023-08-08 | 2023-09-05 | 江西驴充充充电技术有限公司 | Charging pile detection system and method thereof |
| CN117743930A (en) * | 2023-12-20 | 2024-03-22 | 北京智盟信通科技有限公司 | Sensing data processing method and server applied to intelligent photovoltaic |
| CN117792279A (en) * | 2024-02-27 | 2024-03-29 | 鑫琪(苏州)新能源科技有限公司 | Distributed photovoltaic monitoring system based on neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109215042A (en) * | 2018-09-28 | 2019-01-15 | 吉林电力股份有限公司科技开发分公司 | A kind of photovoltaic battery panel hot spot effect detection system based on computer vision and its calculation method |
| CN110224673A (en) * | 2019-05-14 | 2019-09-10 | 太原理工大学 | A kind of solar photovoltaic cell panel fault detection method based on deep learning |
| CN112163551A (en) * | 2020-10-14 | 2021-01-01 | 黄振海 | Photovoltaic cell panel deformation prediction method based on image processing and multi-dimensional perception |
| CN112578089A (en) * | 2020-12-24 | 2021-03-30 | 河北工业大学 | Air pollutant concentration prediction method based on improved TCN |
| GB202102016D0 (en) * | 2020-02-14 | 2021-03-31 | Nvidia Corp | Robotic control using deep learning |
-
2021
- 2021-04-30 CN CN202110482858.6A patent/CN113283580A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109215042A (en) * | 2018-09-28 | 2019-01-15 | 吉林电力股份有限公司科技开发分公司 | A kind of photovoltaic battery panel hot spot effect detection system based on computer vision and its calculation method |
| CN110224673A (en) * | 2019-05-14 | 2019-09-10 | 太原理工大学 | A kind of solar photovoltaic cell panel fault detection method based on deep learning |
| GB202102016D0 (en) * | 2020-02-14 | 2021-03-31 | Nvidia Corp | Robotic control using deep learning |
| CN112163551A (en) * | 2020-10-14 | 2021-01-01 | 黄振海 | Photovoltaic cell panel deformation prediction method based on image processing and multi-dimensional perception |
| CN112578089A (en) * | 2020-12-24 | 2021-03-30 | 河北工业大学 | Air pollutant concentration prediction method based on improved TCN |
Non-Patent Citations (2)
| Title |
|---|
| 程起泽等: "基于CNN-LSTM的太阳能光伏组件故障诊断研究", 《计算机技术与应用》 * |
| 高永成: "GEO卫星电源系统故障仿真及诊断研究", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113688987A (en) * | 2021-08-30 | 2021-11-23 | 远景智能国际私人投资有限公司 | Training method of photovoltaic monitoring model, monitoring method of photovoltaic equipment and equipment |
| CN113688987B (en) * | 2021-08-30 | 2023-08-01 | 远景智能国际私人投资有限公司 | Training method of photovoltaic monitoring model, monitoring method and device of photovoltaic device |
| CN113625094A (en) * | 2021-09-11 | 2021-11-09 | 沈阳工业大学 | Subway platform screen door fault diagnosis system |
| CN114298219A (en) * | 2021-12-27 | 2022-04-08 | 江苏国科智能电气有限公司 | Wind driven generator fault diagnosis method based on deep space-time feature extraction |
| CN116699297A (en) * | 2023-08-08 | 2023-09-05 | 江西驴充充充电技术有限公司 | Charging pile detection system and method thereof |
| CN116699297B (en) * | 2023-08-08 | 2023-10-27 | 江西驴充充充电技术有限公司 | Charging pile detection system and method thereof |
| CN117743930A (en) * | 2023-12-20 | 2024-03-22 | 北京智盟信通科技有限公司 | Sensing data processing method and server applied to intelligent photovoltaic |
| CN117792279A (en) * | 2024-02-27 | 2024-03-29 | 鑫琪(苏州)新能源科技有限公司 | Distributed photovoltaic monitoring system based on neural network |
| CN117792279B (en) * | 2024-02-27 | 2024-05-17 | 鑫琪(苏州)新能源科技有限公司 | Distributed photovoltaic monitoring system based on neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113283580A (en) | Automatic fault detection method for solar cell panel | |
| 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 | |
| CN109934423B (en) | Photovoltaic power station power prediction method and system based on grid-connected inverter operation data | |
| CN109842373B (en) | Photovoltaic array fault diagnosis method and device based on space-time distribution characteristics | |
| Liu et al. | Fault diagnosis approach for photovoltaic array based on the stacked auto-encoder and clustering with IV curves | |
| Natarajan et al. | Fault detection of solar PV system using SVM and thermal image processing | |
| CN110224673B (en) | Deep learning-based solar photovoltaic cell panel fault detection method | |
| CN109660206B (en) | Wasserstein GAN-based photovoltaic array fault diagnosis method | |
| CN108983749B (en) | Photovoltaic array fault diagnosis method based on K-SVD training sparse dictionary | |
| CN111245365B (en) | Photovoltaic module fault diagnosis method based on dynamic weighted depth forest | |
| CN106777984A (en) | A kind of method that photovoltaic array Working state analysis and fault diagnosis are realized based on density clustering algorithm | |
| Wang et al. | Performance assessment of photovoltaic modules based on daily energy generation estimation | |
| Liu et al. | A dilation and erosion-based clustering approach for fault diagnosis of photovoltaic arrays | |
| Chen et al. | A novel fault diagnosis method of PV based-on power loss and IV characteristics | |
| Djalab et al. | Robust method for diagnosis and detection of faults in photovoltaic systems using artificial neural networks | |
| CN109992911A (en) | Photovoltaic module fast modeling method based on extreme learning machine and IV characteristic | |
| Yun et al. | Research on fault diagnosis of photovoltaic array based on random forest algorithm | |
| CN116054740A (en) | Fault detection and fault detection model processing method and device for photovoltaic power generation equipment | |
| CN113326667B (en) | Solar cell panel fault diagnosis method and system | |
| Yang et al. | Fast simulation modeling and multiple-PS fault diagnosis of the PV array based on I–V curve conversion | |
| Zhou et al. | A shadow fault diagnosis method based on the quantitative analysis of photovoltaic output prediction error | |
| Gong et al. | Fault diagnosis method of photovoltaic array based on random forest algorithm | |
| CN116317937A (en) | Distributed photovoltaic power station operation fault diagnosis method | |
| CN117764547A (en) | Photovoltaic string fault diagnosis method and system |
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: 20210820 |