CN109873610B - Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network - Google Patents
Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network Download PDFInfo
- Publication number
- CN109873610B CN109873610B CN201910206962.5A CN201910206962A CN109873610B CN 109873610 B CN109873610 B CN 109873610B CN 201910206962 A CN201910206962 A CN 201910206962A CN 109873610 B CN109873610 B CN 109873610B
- Authority
- CN
- China
- Prior art keywords
- training
- dimensional
- array
- resnet
- 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.)
- Expired - Fee Related
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- 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
- Testing Of Individual Semiconductor Devices (AREA)
- Photovoltaic Devices (AREA)
Abstract
The invention relates to a photovoltaic array fault diagnosis method based on IV characteristics and a deep residual error network. Firstly, building a model array by using Simulink, and collecting electrical data and environmental data under various working conditions; secondly, eliminating abnormal data in the original simulated data, acquiring an original I-V curve for down-sampling, and splicing the one-dimensional characteristics into two-dimensional characteristics as the overall characteristics of the fault; then, dividing sample data into a training set, a verification set and a test set, designing a network structure of a dimension-transformed residual convolutional neural network and training parameters of a training algorithm Adam thereof, and carrying out sample training to obtain a DT-ResNet fault diagnosis training model; and finally, detecting and classifying the photovoltaic power generation array under the working condition test set to be detected by utilizing the DT-ResNet fault diagnosis training model, and diagnosing the fault type. The method has the advantages of high accuracy, fast convergence, strong robustness, good generalization capability and the like, and can effectively improve the accuracy of the fault detection and classification of the photovoltaic power generation array.
Description
Technical Field
The invention relates to a photovoltaic power generation group string fault detection and classification technology, in particular to a photovoltaic array fault diagnosis method based on IV characteristics and a deep residual error network.
Background
Solar energy has been widely noticed and used in recent years as a clean and renewable new energy source. According to the latest announcements of the world energy organization, the global photovoltaic loading capacity and the power generation capacity are increased in the year, and by the end of 2017, the loading capacity of a global photovoltaic power station reaches 399,613MW, and the global photovoltaic power generation capacity is increased to 442,618 GW. However, problems of violent installation or irregular installation often occur in the existing installation and deployment of the photovoltaic power station. In addition, the photovoltaic power station works in a severe outdoor environment all the year round, and is easily influenced by various environmental factors such as thermal cycle, humidity, ultraviolet rays, wind excitation and the like during working, so that faults such as local aging, performance reduction, cracks and the like of materials are caused, energy loss is caused, and the power generation efficiency of the photovoltaic power station is further influenced. If the detection is not carried out in time, the damage of the photovoltaic module is likely to be caused, and even the disaster of array burning is caused. It is therefore extremely important and meaningful to provide an accurate, efficient and reliable method of diagnosing faults in a photovoltaic module/array photovoltaic array.
In recent years, various offline fault diagnosis methods and techniques have been proposed in succession. Kang proposes a method of adjusting IV output characteristics through kalman filtering, thereby enabling accurate diagnosis of shadow faults. However, this method cannot accurately classify other various faults. Platon et al propose a fault diagnosis algorithm based on power loss between model prediction and actual measurement. However, this algorithm requires the development of simulation models for a variety of different irradiance ranges. AliMH proposes a real-time I-V characteristic-based detection method, which judges the type of error by quantitatively comparing the difference between a normal working model and various abnormal models. However, the method has low accuracy and high requirements on the discriminant model. The above method cannot give good consideration to both the accuracy and the generalization of the detection. According to the characteristics of multiple elements, multiple layers, fuzzy state and the like of the photovoltaic array faults, in order to more accurately identify the faults, an artificial intelligence algorithm is introduced into the photovoltaic array faults. In recent years, methods such as Artificial Neural Networks (ANN), Decision Trees (DT), Support Vector Machines (SVM), kernel function-based extreme learning machines (KELM), Random Forests (RF), and the like have been introduced into the field of photovoltaic array fault diagnosis. Although the classification accuracy can be greatly improved by the methods, the problems of poor nonlinear feature extraction capability, poor generalization capability and the like still exist, for this reason, a photovoltaic array fault diagnosis method based on the IV characteristic and the deep residual error network is introduced, although the residual error network based on the two-dimensional CNN is widely applied to the image recognition target tracking field, the 1-dimensional CNN is widely applied to the word meaning analysis field and the like, and the effect is greatly superior to that of the traditional fully-connected ANN in terms of training speed and training effect. However, the methods are not yet applied to the field of photovoltaic array fault diagnosis.
The photovoltaic array fault diagnosis method based on the IV characteristic and the depth residual error network comprises the steps of building a model array through Simulink, simulating various working conditions, and collecting electrical data and environmental data under various conditions, wherein the I-V characteristic curve and the corresponding illuminance and irradiance are obtained by scanning the model array under the working conditions; and then, removing abnormal data in the original simulated data, acquiring an original I-V curve for down sampling, performing feature splicing on the four obtained one-dimensional features of current, voltage, temperature and irradiance, splicing the one-dimensional features into two-dimensional features, taking the one-dimensional features as the overall features of the fault, dividing the sample data into a training set, a verification set and a test set according to different working conditions in equal proportion. Training a training set by utilizing a designed network structure of a residual convolutional neural network (DT-ResNet) based on dimension transformation, verifying a training model on a verification set to obtain the optimal training model with the most generalization capability of DT-ResNet on the training set, and finally, further verifying the accuracy and the generalization capability of the obtained optimal model on a test set. The photovoltaic array fault diagnosis method based on the dimension transformation residual convolutional neural network can effectively improve the accuracy of photovoltaic power generation array fault detection and classification.
At present, no study on application of a dimension-transformed residual convolutional neural network algorithm to fault diagnosis and classification of a photovoltaic power generation array is found in published documents and patents.
Disclosure of Invention
The invention aims to provide a photovoltaic array fault diagnosis method based on IV characteristics and a deep residual error network, so as to overcome the defects in the prior art.
In order to achieve the purpose, the technical scheme of the invention is as follows: a photovoltaic array fault diagnosis method based on IV characteristics and a depth residual error network comprises the following steps:
step S1, building a model array by using Simulink, simulating various working conditions, and acquiring electrical data and environmental data under various working conditions, wherein the simulation specifically comprises the following steps: scanning the model array under corresponding working conditions to obtain an I-V characteristic curve and corresponding illuminance and irradiance;
s2, removing abnormal data in the original simulated data, down-sampling the collected original I-V curve, performing feature splicing on the four obtained one-dimensional features of current, voltage, temperature and irradiance to obtain a two-dimensional feature, and taking the two-dimensional feature as the overall feature of the fault;
step S3, dividing the sample data obtained in the step S1 into a training set, a verification set and a test set according to different working conditions in equal proportion, and designing a network structure of a dimension-transformed residual convolution neural network DT-ResNet and training parameters of a training algorithm Adam of the network structure;
step S4, training samples in a training set according to the training parameters of the training algorithm Adam set in the step S3 and the residual convolution neural network DT-ResNet, and verifying a DT-ResNet training model on a verification set to obtain an optimal DT-ResNet fault diagnosis training model with the most generalization capability of the residual convolution neural network DT-ResNet on the training set;
step S5: and (4) detecting and classifying the photovoltaic power generation array under the working condition test set to be detected by using the DT-ResNet fault diagnosis training model, scanning the I-V characteristic curve of the photovoltaic power generation array, setting the characteristics according to the step S2, and giving the fault type if the fault state occurs.
In an embodiment of the present invention, in step S1, the operating conditions include: normal operation, shadow fault, open circuit fault, aging fault, group serial line fault; wherein, the shadow fault comprises that shadow shielding is respectively carried out on one, two and three components in the group string; open circuit faults, including open circuits within a string; aging faults including string aging in the array and array aging; a group of serial line faults, including one or two assemblies in the assembly being abnormally short-circuited; the environment conditions corresponding to the simulated working conditions are as follows: the irradiance range is 50-1000W/m2The temperature range is 10-70 ℃.
In an embodiment of the present invention, in step S2, the down-sampling is performed on the acquired original I-V curve, and the four obtained one-dimensional features of current, voltage, temperature, and irradiance are subjected to feature stitching, so as to obtain a two-dimensional feature, which is specifically implemented as follows:
step S21, the scanning current value obtained by down sampling the collected original I-V curve is in [0, Isc]20 points are taken between the points and divided at equal intervals, and is marked as [ I ]R1,IR2,...,IR20]Obtaining the corresponding voltage value of the 20 points by using a linear interpolation method, wherein a specific voltage down-sampling formula is as follows;
wherein, I1And I2Is the sum of I in the data of the original simulationRXTwo current values, V, which are closest to each other1And V2Are respectively I1And I2The corresponding voltage value;
step S22, the scanning voltage value obtained by down sampling the collected original I-V curve is [0, V ]oc]Points divided at 20 equal intervals are marked as VR1,VR2,...,VR20]In which V isRXFor the voltage values corresponding to the 20 down-sampling points to be extracted, a linear interpolation method is used to obtain the corresponding current values of the 20 points, and the specific current sampling formula is as follows:
wherein, V1And V2Is the sum of V in the data of the original simulationRXTwo voltage values closest to each other, I1And I2Are each V1And V2The corresponding current value;
step S23, splicing 40 pairs of electrical data obtained by respectively down-sampling according to current and voltage to form a 40 multiplied by 2 array matrix, and arranging the array matrix in ascending order according to voltage as input sample data after down-sampling;
step S24, tiling an array matrix with the corresponding temperature and irradiance of 40 multiplied by 2, and performing column splicing on the array matrix and the electrical parameter matrix to form a new input array matrix;
and step S25, removing the simulated partial singular value samples.
In an embodiment of the present invention, in step S3, the sample data obtained in step S1 is proportionally divided into a training set, a verification set, and a test set according to different working conditions, specifically, 70% of samples under various working conditions are used as the training set, the remaining 30% of samples are used as the test set, and 10% of samples in the training set are used as the verification set.
In an embodiment of the present invention, in step S3, the specific implementation manner of the network structure of the design dimension-transformed residual convolutional neural network DT-ResNet and the training parameters of the training algorithm Adam thereof is as follows: the method comprises the following steps of extracting features of sample data by adopting a two-dimensional residual error module, converting the two-dimensional data into one-dimensional data by utilizing a two-dimensional volume base layer Conv2d with a convolution kernel of 4, performing classified prediction on the extracted one-dimensional features by utilizing a one-dimensional residual error network, and taking the minimum cross entropy as a target by a training algorithm Adam, wherein a calculation formula of the cross entropy 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, zkIn the forward propagation process, when passing through the linear fully-connected layer, its value can be expressed as:
z(l)=W(l)*a(l-1)+b(l-1)
wherein Wl is the weight of the l layer, a is the activation value of the l-1 layer, and bl is the deviation value of the l layer; if passing through the two-dimensional convolutional layer, the output of the neuron can be expressed as:
z(l-1)=conv2d(a(l-1))+b(l-1)
wherein, conv2d represents a two-dimensional convolution operation process; if the forward output passes through the one-dimensional convolutional layer, the output of the neuron can be expressed as:
z(l-1)=conv1d(a(l-1))+b(l-1)
wherein, conv1d represents a one-dimensional convolution operation process;
in addition, when the input and output dimensions are the same, the forward propagation process of the residual module is as follows:
y=ReLU(BN(z(l))+x)
the output y and the input x have the same dimensionality, and the BN is a batch normalization layer and is used for keeping the interlayer distribution to keep normal distribution and improving the training speed and the training performance of the network;
when the input and output dimensions are different, a dimension adjustment factor needs to be added:
y=ReLU(BN(z(l))+Wsx)
wherein Ws is a dimension adjustment factor;
from this, the output function of the one-dimensional residual module is:
y=ReLU(BN(conv1d(a(l-1))+b(l-1))+x)
y=ReLU(BN(conv1d(a(l-1))+b(l-1))+Wsx)
the output function of the two-dimensional residual module is:
y=ReLU(BN(conv2d(a(l-1))+b(l-1))+x)
y=ReLU(BN(conv2d(a(l-1))+b(l-1))+Wsx)
and finally, iteratively updating Ws, b and internal parameters of conv1d and conv2d in the feedback to obtain an optimal residual convolution neural network DT-ResNet.
In an embodiment of the present invention, in step S3, the specific implementation manner of obtaining the optimal and most generalized DT-ResNet fault diagnosis training model of the residual convolutional neural network DT-ResNet on the training set is as follows:
step S41, calculating the performance of the residual convolutional neural network DT-ResNet on a verification set obtained under the current iteration, calculating the total correct rate, comparing the correct rates of the current optimal residual convolutional neural network DT-ResNet and the current residual convolutional neural network DT-ResNet, and keeping the network parameters with more generalization;
step S42, if the verification set has the same precision, the precision of the training set is further compared, and the network parameters with better performance are stored;
and step S43, entering the next iteration, updating the network parameters, and returning to step S41 until the iteration is finished.
Compared with the prior art, the invention has the following beneficial effects: according to the photovoltaic array fault diagnosis method based on the IV characteristic and the depth residual error network, simulation data are verified and analyzed, I-V equal-interval down sampling is utilized, the simulation I-V data are preprocessed, a uniform and smooth characteristic splicing process of an I-V curve is obtained, and the photovoltaic array fault diagnosis is carried out by utilizing the residual error network based on the dimensionality transformation. The method can accurately identify nine different fault types, namely normal, one short circuit, two short circuits, open circuit of the component, open circuit of the array, aging of the component, aging of the array, shadow faults of three different degrees and the like. From the experimental result, the method can be efficiently and accurately applied to the field of fault diagnosis of the distributed photovoltaic array.
Drawings
Fig. 1 is a flowchart of a photovoltaic array fault diagnosis method based on a dimension transformation residual convolutional neural network in the present invention.
Fig. 2 is a topological diagram of the Simulink component of the analog photovoltaic array in an embodiment of the present invention, where fig. 2(a), fig. 2(b), and fig. 2(c) respectively show topological diagrams of the Simulink component of the analog photovoltaic array when one, two, and three components in the string are shaded.
Fig. 3 is a schematic diagram of a one-dimensional and two-dimensional residual convolution module applied in an embodiment of the present invention, where fig. 3(a) and fig. 3(b) respectively show schematic diagrams of the one-dimensional and two-dimensional residual convolution modules.
Fig. 4 is a block diagram of a residual convolutional neural network based on dimension transformation according to an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is specifically explained below with reference to the accompanying drawings.
The invention provides a photovoltaic array fault diagnosis method based on IV characteristics and a deep residual error network, which comprises the following steps:
step S1, building a model array by using Simulink, simulating various working conditions, and acquiring electrical data and environmental data under various working conditions, wherein the simulation specifically comprises the following steps: scanning the model array under corresponding working conditions to obtain an I-V characteristic curve and corresponding illuminance and irradiance;
s2, removing abnormal data in the original simulated data, down-sampling the collected original I-V curve, performing feature splicing on the four obtained one-dimensional features of current, voltage, temperature and irradiance to obtain a two-dimensional feature, and taking the two-dimensional feature as the overall feature of the fault;
step S3, dividing the sample data obtained in the step S1 into a training set, a verification set and a test set according to different working conditions in equal proportion, and designing a network structure of a dimension-transformed residual convolution neural network DT-ResNet and training parameters of a training algorithm Adam of the network structure;
step S4, training samples in a training set according to the training parameters of the training algorithm Adam set in the step S3 and the residual convolution neural network DT-ResNet, and verifying a DT-ResNet training model on a verification set to obtain an optimal DT-ResNet fault diagnosis training model with the most generalization capability of the residual convolution neural network DT-ResNet on the training set;
step S5: and (4) detecting and classifying the photovoltaic power generation array under the working condition test set to be detected by using the DT-ResNet fault diagnosis training model, scanning the I-V characteristic curve of the photovoltaic power generation array, setting the characteristics according to the step S2, and giving the fault type if the fault state occurs.
The following are specific embodiments of the present invention.
The invention provides a photovoltaic array fault diagnosis method based on IV characteristics and a depth residual error network, and a flow chart is shown in figure 1. Fig. 2 is a Simulink photovoltaic array simulation topological diagram of the embodiment, the system is composed of 3 string groups, 6 solar modules in each string group, and 18 modules in total, different fault conditions of the photovoltaic array, such as open circuit, short circuit, local shadow and other working states, are simulated, different time periods are selected under different weather conditions, and a large amount of simulation data are acquired for each fault condition. The specific implementation process of the fault diagnosis and classification method comprises the following steps:
s1, building a model array by using Simulink, simulating various working conditions, and collecting electrical data and environmental data under various conditions, wherein the simulation specifically comprises an I-V characteristic curve obtained by scanning the model array under the working conditions and corresponding illuminance and irradiance;
and S2, removing abnormal data in the original simulated data, collecting an original I-V curve for down sampling, performing feature splicing on the four obtained one-dimensional features of current, voltage, temperature and irradiance, splicing the one-dimensional features into two-dimensional features, and taking the features as the overall features of the fault.
Step S3, dividing the sample data into training sets, verification sets and test sets according to different working conditions in equal proportion, and designing a network structure of a dimension-transformed residual convolution neural network (DT-ResNet) and training parameters of a training algorithm Adam thereof;
step S4, training samples in a training set according to the training parameters set in the step S3 and the model of DT-ResNet, and verifying the training model on a verification set to obtain the optimal training model with the most generalization capability of DT-ResNet on the training set;
and S5, detecting and classifying the photovoltaic power generation array under the working condition test set to be tested by using the DT-ResNet fault diagnosis training model, scanning an I-V characteristic curve of the photovoltaic power generation array, and setting the characteristics according to the step S2, wherein if a fault state occurs, the fault type is given.
Further, in this embodiment, in the step S1, the Simulink component topology is as shown in fig. 2 to 4, wherein each group string is composed of 6 photovoltaic components, and one array is composed of 3 photovoltaic groups. Wherein irradiance Gain factors preset by the Gain simulation shadow fault (fig. 2(a), fig. 2(b) and fig. 2(c) respectively show that shadow shielding occurs to one, two and three components in a string), and the normal value is 1; r _ oc1 is a preset resistance value for simulating a string open circuit, and the normal value is 0.001 omega; r _ de1 is a resistance value preset for simulating string aging, and its normal value is 0.001 Ω; r _ de2 is a preset resistance value for simulating array aging, and the normal value is 0.001 omega; r _ sc1 and R _ sc2 are resistance values preset for simulating string short circuit respectively, and the normal values are 100000 omega; the specific settings for simulating the fault are shown in table 1.
Table 1 simulation of the simulation data set to obtain the number and distribution of various faults
Further, in this embodiment, in step S1, the operating conditions include: normal operation, shadow fault, open circuit fault, aging fault, group serial line fault; wherein, the shadow fault comprises that one, two and three components in the string are respectively shaded by shadow; open circuit faults, including open circuits within a string; aging faults including string aging in the array and array aging; a group string line fault includes one or both of the components being abnormally shorted. In addition, the environment conditions corresponding to the simulated working conditions are as follows: irradiance ranges from 50-1000W/m 2, temperatures range from 10-70 c, set intervals for temperatures of 2 c, irradiance ranges from 10 c, and specific simulated sample numbers and distribution ratios are shown in table 1, for example.
Further, in this embodiment, in the step S2, the down-sampling and feature stitching method includes:
step S21, the scanning current value obtained by down sampling the collected original I-V curve is in [0, Isc]20 points are taken between the points and divided at equal intervals, and is marked as [ I ]R1,IR2,...,IR20]Obtaining the corresponding voltage value of the 20 points by using a linear interpolation method, wherein a specific voltage down-sampling formula is as follows;
wherein, I1And I2Is the sum of I in the data of the original simulationRXTwo current values, V, which are closest to each other1And V2Are respectively I1And I2The corresponding voltage value;
step S22, the scanning voltage value obtained by down sampling the collected original I-V curve is [0, V ]oc]Points divided at 20 equal intervals are marked as VR1,VR2,...,VR20]In which V isRXFor the voltage values corresponding to the 20 down-sampling points to be extracted, a linear interpolation method is used to obtain the corresponding current values of the 20 points, and the specific current sampling formula is as follows:
wherein, V1And V2Is the sum of V in the data of the original simulationRXTwo voltage values closest to each other, I1And I2Are each V1And V2The corresponding current value;
step S23, splicing 40 pairs of electrical data obtained by respectively down-sampling according to current and voltage to form a 40 multiplied by 2 array matrix, and arranging the array matrix in ascending order according to voltage as input sample data after down-sampling;
step S24, tiling an array matrix with the corresponding temperature and irradiance of 40 multiplied by 2, and performing column splicing on the array matrix and the electrical parameter matrix to form a new input array matrix;
and step S25, after removing the simulated partial singular value samples, taking 70% of samples under various working conditions as a training set, taking the remaining 30% of samples as a test set, and taking 10% of samples in the training set samples as a verification set.
In this embodiment, in the step S3, a structure of DT-ResNet is designed, in which a specific structure of a residual network is used as shown in fig. 3, in which a residual module based on one-dimensional convolution (fig. 3(a)) and a residual module based on two-dimensional convolution (fig. 3(b)) are utilized to complete the extraction of features. The batch normalization layer is used for optimizing distribution of interlayer training parameters, improving network effect and accelerating network training speed, and ReLU is used as an interlayer activation function. Wherein the solid line represents that dimension conversion is performed between layers, and dimension matching is performed through a supplementary convolutional layer; the dashed lines indicate that the dimensions of the inputs and outputs are the same and can be added directly. The output function of one-dimensional residual module is:
y=ReLU(BN(conv1d(a(l-1))+b(l-1))+x)
y=ReLU(BN(conv1d(a(l-1))+b(l-1))+Wsx)
the output function of the two-dimensional residual module is:
y=ReLU(BN(conv2d(a(l-1))+b(l-1))+x)
y=ReLU(BN(conv2d(a(l-1))+b(l-1))+Wsx)
and finally, iteratively updating the W, b and the internal parameters of the conv1d and the conv2d in the back propagation to finally obtain the trained optimal model.
In this embodiment, in step S3, a structure based on the IV characteristic and the depth residual error network is designed, and its specific implementation manner is: the method comprises the steps of performing feature pre-extraction on original input data by using a two-dimensional residual error module, converting the two-dimensional data into one-dimensional data by using a two-dimensional volume base layer Conv2d with a convolution kernel of 4, performing classified prediction on the extracted one-dimensional features by using a one-dimensional residual error network, and performing training on a training set by using an Adam optimization algorithm with the minimum cross entropy as a target to finally obtain an optimal model. The specific structure of the proposed DT-ResNet is shown in FIG. 4. Wherein the first parameter of each layer represents the size of the convolution kernel; in the second parameter, conv1d represents a one-dimensional convolutional layer, conv2d represents a two-dimensional convolutional layer, and AvgPool represents an average pooling layer; the third parameter represents the number of output channels, the last layer of optional parameters represents the step size of convolution kernel movement, and the default value is 1.
In this embodiment, in the step S4, the selection process of selecting the network with the best performance and the strongest generalization capability includes the following steps of S41, calculating the performance of the network obtained under the current iteration on the verification set, calculating the total accuracy, comparing the accuracy of the current optimal network with the accuracy of the current network, and retaining the network parameters with the greatest generalization capability. Step S42: if the same precision is obtained on the verification set, the precision of the training set is further compared, and the network parameters with better performance are stored. Step S43: and entering the next iteration, updating the network parameters, and returning to S41 until the iteration is finished.
Further, in this embodiment, the fault detection accuracy of the DT-ResNet fault diagnosis training model can reach 100%, specifically, from the training set, the total accuracy is 100% (15834/15834), the total detection rate on the verification set is 100% (1858/1858) training, and the fault detection rate under the test set is 100% (7545/7545), and the classification fault diagnosis rates for each subclass are shown in table 2.
TABLE 2 Classification accuracy of the method of the present invention for each condition
The above are preferred embodiments of the present invention, and all changes made according to the technical scheme of the present invention that produce functional effects do not exceed the scope of the technical scheme of the present invention belong to the protection scope of the present invention.
Claims (6)
1. A photovoltaic array fault diagnosis method based on IV characteristics and a depth residual error network is characterized by comprising the following steps:
step S1, building a model array by using Simulink, simulating various working conditions, and acquiring electrical data and environmental data under various working conditions, wherein the simulation specifically comprises the following steps: scanning the model array under corresponding working conditions to obtain an I-V characteristic curve and corresponding temperature and irradiance;
s2, removing abnormal data in the original simulated data, down-sampling the collected original I-V curve, performing feature splicing on the four obtained one-dimensional features of current, voltage, temperature and irradiance to obtain a two-dimensional feature, and taking the two-dimensional feature as the overall feature of the fault;
step S3, dividing the sample data obtained in the step S1 into a training set, a verification set and a test set according to different working conditions in equal proportion, and designing a network structure of a dimension-transformed residual convolution neural network DT-ResNet and training parameters of a training algorithm Adam of the network structure;
step S4, training samples in a training set according to the training parameters of the training algorithm Adam set in the step S3 and the residual convolution neural network DT-ResNet, and verifying a DT-ResNet training model on a verification set to obtain an optimal DT-ResNet fault diagnosis training model with the most generalization capability of the residual convolution neural network DT-ResNet on the training set;
step S5: and (4) detecting and classifying the photovoltaic power generation array under the working condition test set to be detected by using the DT-ResNet fault diagnosis training model, scanning the I-V characteristic curve of the photovoltaic power generation array, setting the characteristics according to the step S2, and giving the fault type if the fault state occurs.
2. The photovoltaic array fault diagnosis method based on the IV characteristic and the deep residual error network as claimed in claim 1, wherein in step S1, the working conditions comprise: normal operation, shadow fault, open circuit fault, aging fault, group serial line fault; wherein, the shadow fault comprises that shadow shielding is respectively carried out on one, two and three components in the group string; open circuit faults, including open circuits within a string; aging faults including string aging in the array and array aging; a group of serial line faults, including one or two assemblies in the assembly being abnormally short-circuited; the environment conditions corresponding to the simulated working conditions are as follows: the irradiance range is 50-1000W/m2The temperature range is 10-70 ℃.
3. The photovoltaic array fault diagnosis method based on the IV characteristic and the depth residual error network according to claim 1, wherein in step S2, the collected original I-V curve is downsampled, and four one-dimensional characteristics of the obtained current, voltage, temperature and irradiance are subjected to feature splicing, and a specific implementation manner of obtaining the two-dimensional characteristics is as follows:
step S21, the scanning current value obtained by down sampling the collected original I-V curve is in [0, Isc]20 points are taken between the points and divided at equal intervals, and is marked as [ I ]R1,IR2,...,IR20]Obtaining the corresponding voltage value of the 20 points by using a linear interpolation method, wherein a specific voltage down-sampling formula is as follows;
wherein, I1And I2Is the sum of I in the data of the original simulationRXTwo current values, V, which are closest to each other1And V2Are respectively I1And I2The corresponding voltage value;
step S22, the scanning voltage value obtained by down sampling the collected original I-V curve is [0, V ]oc]Points divided at 20 equal intervals are marked as VR1,VR2,...,VR20]In which V isRXFor the voltage values corresponding to the 20 down-sampling points to be extracted, a linear interpolation method is used to obtain the corresponding current values of the 20 points, and the specific current sampling formula is as follows:
wherein, V1And V2Is the sum of V in the data of the original simulationRXTwo voltage values closest to each other, I1And I2Are each V1And V2The corresponding current value;
step S23, splicing 40 pairs of electrical data obtained by respectively down-sampling according to current and voltage to form a 40 multiplied by 2 array matrix, and arranging the array matrix in ascending order according to voltage as input sample data after down-sampling;
step S24, tiling an array matrix with the corresponding temperature and irradiance of 40 multiplied by 2, and performing column splicing on the array matrix and the electrical parameter matrix to form a new input array matrix;
and step S25, removing the simulated partial singular value samples.
4. The photovoltaic array fault diagnosis method based on the IV characteristic and the deep residual error network according to claim 1, wherein in step S3, the sample data obtained in step S1 is proportionally divided into a training set, a verification set, and a test set according to different working conditions, specifically, 70% of samples under various working conditions are used as the training set, the remaining 30% of samples are used as the test set, and 10% of samples in the training set are used as the verification set.
5. The photovoltaic array fault diagnosis method based on the IV characteristic and the depth residual error network, according to claim 1, wherein in step S3, the network structure of the design dimension transformed residual error convolution neural network DT-ResNet and the training parameters of the training algorithm Adam thereof are realized in a manner that: the method comprises the following steps of extracting features of sample data by adopting a two-dimensional residual error module, converting the two-dimensional data into one-dimensional data by utilizing a two-dimensional volume base layer Conv2d with a convolution kernel of 4, performing classified prediction on the extracted one-dimensional features by utilizing a one-dimensional residual error network, and taking the minimum cross entropy as a target by a training algorithm Adam, wherein a calculation formula of the cross entropy 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, zkIn the forward propagation process, when passing through the linear fully-connected layer, its value can be expressed as:
z(l)=W(l)*a(l-1)+b(l-1)
wherein Wl is the weight of the l layer, a is the activation value of the l-1 layer, and bl is the deviation value of the l layer; if passing through the two-dimensional convolutional layer, the output of the neuron can be expressed as:
z(l-1)=conv2d(a(l-1))+b(l-1)
wherein, conv2d represents a two-dimensional convolution operation process; if the forward output passes through the one-dimensional convolutional layer, the output of the neuron can be expressed as:
z(l-1)=conv1d(a(l-1))+b(l-1)
wherein, conv1d represents a one-dimensional convolution operation process;
in addition, when the input and output dimensions are the same, the forward propagation process of the residual module is as follows:
y=ReLU(BN(z(l))+x)
the output y and the input x have the same dimensionality, and the BN is a batch normalization layer and is used for keeping the interlayer distribution to keep normal distribution and improving the training speed and the training performance of the network;
when the input and output dimensions are different, a dimension adjustment factor needs to be added:
y=ReLU(BN(z(l))+Wsx)
wherein Ws is a dimension adjustment factor;
from this, the output function of the one-dimensional residual module is:
y=ReLU(BN(conv1d(a(l-1))+b(l-1))+x)
y=ReLU(BN(conv1d(a(l-1))+b(l-1))+Wsx)
the output function of the two-dimensional residual module is:
y=ReLU(BN(conv2d(a(l-1))+b(l-1))+x)
y=ReLU(BN(conv2d(a(l-1))+b(l-1))+Wsx)
and finally, iteratively updating Ws, b and internal parameters of conv1d and conv2d in the feedback to obtain an optimal residual convolution neural network DT-ResNet.
6. The photovoltaic array fault diagnosis method based on the IV characteristic and the deep residual error network according to claim 1, wherein in step S3, the specific implementation manner of the DT-ResNet fault diagnosis training model with the optimal and most generalized capability on the training set of the residual convolutional neural network DT-ResNet is:
step S41, calculating the performance of the residual convolutional neural network DT-ResNet on a verification set obtained under the current iteration, calculating the total correct rate, comparing the correct rates of the current optimal residual convolutional neural network DT-ResNet and the current residual convolutional neural network DT-ResNet, and keeping the network parameters with more generalization;
step S42, if the verification set has the same precision, the precision of the training set is further compared, and the network parameters with better performance are stored;
and step S43, entering the next iteration, updating the network parameters, and returning to step S41 until the iteration is finished.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910206962.5A CN109873610B (en) | 2019-03-19 | 2019-03-19 | Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910206962.5A CN109873610B (en) | 2019-03-19 | 2019-03-19 | Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109873610A CN109873610A (en) | 2019-06-11 |
CN109873610B true CN109873610B (en) | 2020-04-10 |
Family
ID=66920788
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910206962.5A Expired - Fee Related CN109873610B (en) | 2019-03-19 | 2019-03-19 | Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109873610B (en) |
Families Citing this family (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110398348B (en) * | 2019-06-18 | 2021-02-09 | 清华大学 | Memory, and method and device for diagnosing mechanical fault of high-voltage circuit breaker |
CN110414601A (en) * | 2019-07-30 | 2019-11-05 | 南京工业大学 | Photovoltaic module method for diagnosing faults, system and equipment based on depth convolution confrontation network |
CN110619479B (en) * | 2019-09-23 | 2022-05-24 | 阳光电源股份有限公司 | Fault judgment model modeling method and fault judgment method for photovoltaic module |
CN111049477A (en) * | 2019-10-24 | 2020-04-21 | 国网甘肃省电力公司电力科学研究院 | Photovoltaic module output characteristic analysis method under power supply side fault |
CN110907826B (en) * | 2019-11-14 | 2022-04-08 | 中车株洲电力机车研究所有限公司 | Motor fault diagnosis method and system based on convolutional neural network filtering |
CN111291814B (en) * | 2020-02-15 | 2023-06-02 | 河北工业大学 | Crack identification algorithm based on convolutional neural network and information entropy data fusion strategy |
CN111562110B (en) * | 2020-05-25 | 2021-02-26 | 南京理工大学 | Fault diagnosis model based on convolutional neural network and cross-component fault diagnosis method |
CN111810124B (en) * | 2020-06-24 | 2023-09-22 | 中国石油大学(华东) | Oil pumping well fault diagnosis method based on characteristic recalibration residual convolutional neural network model |
CN112199888B (en) * | 2020-09-30 | 2021-08-27 | 太原理工大学 | Rotary equipment fault diagnosis method and system based on deep residual error network and readable storage medium |
CN112787591B (en) * | 2020-10-30 | 2022-05-13 | 福州大学 | Photovoltaic array fault diagnosis method based on fine-tuning dense connection convolutional neural network |
CN112949198B (en) * | 2021-03-10 | 2024-04-12 | 阳光电源股份有限公司 | Method and device for detecting faults of photovoltaic string and computer storage medium |
CN113114105B (en) * | 2021-03-10 | 2022-08-09 | 上海工程技术大学 | Dynamic measurement method for output characteristics of photovoltaic cell assembly |
CN113326667B (en) * | 2021-08-03 | 2021-12-10 | 深圳格瑞特新能源有限公司 | Solar cell panel fault diagnosis method and system |
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 |
CN114019921A (en) * | 2021-11-01 | 2022-02-08 | 碳纪科技(北京)有限公司 | Production line equipment control method, device, equipment and medium based on combined model |
CN114259684B (en) * | 2021-12-22 | 2022-10-04 | 三一汽车制造有限公司 | Fire fighting truck waterway fault detection method and device |
CN114997745B (en) * | 2022-07-28 | 2022-10-04 | 华能国际电力江苏能源开发有限公司清洁能源分公司 | Photovoltaic fault diagnosis tracing method based on depth feature extraction |
CN115498956A (en) * | 2022-09-22 | 2022-12-20 | 福州大学 | Photovoltaic array series arc fault diagnosis method |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108170736B (en) * | 2017-12-15 | 2020-05-05 | 南瑞集团有限公司 | Document rapid scanning qualitative method based on cyclic attention mechanism |
CN109389104A (en) * | 2018-11-30 | 2019-02-26 | 浙江碳银互联网科技有限公司 | A kind of family photovoltaic plant fault of converter prediction technique |
CN109409024B (en) * | 2018-12-25 | 2022-09-06 | 福州大学 | Photovoltaic module voltage and current characteristic modeling method based on one-dimensional depth residual error network |
-
2019
- 2019-03-19 CN CN201910206962.5A patent/CN109873610B/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
CN109873610A (en) | 2019-06-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109873610B (en) | Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network | |
CN109842373B (en) | Photovoltaic array fault diagnosis method and device based on space-time distribution characteristics | |
CN108647716B (en) | Photovoltaic array fault diagnosis method based on composite information | |
Liu et al. | Fault diagnosis approach for photovoltaic array based on the stacked auto-encoder and clustering with IV curves | |
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 | |
CN111444615B (en) | Photovoltaic array fault diagnosis method based on K nearest neighbor and IV curve | |
CN109165774A (en) | A kind of short-term photovoltaic power prediction technique | |
CN111754026B (en) | Photovoltaic power station group power prediction method and device, computer equipment and storage medium | |
CN109409024B (en) | Photovoltaic module voltage and current characteristic modeling method based on one-dimensional depth residual error network | |
CN109992911B (en) | Photovoltaic module rapid modeling method based on extreme learning machine and IV characteristics | |
CN110503153B (en) | Photovoltaic system fault diagnosis method based on differential evolution algorithm and support vector machine | |
CN111245365B (en) | Photovoltaic module fault diagnosis method based on dynamic weighted depth forest | |
CN108336739B (en) | RBF neural network-based probability load flow online calculation method | |
CN109672406B (en) | Photovoltaic power generation array fault diagnosis and classification method based on sparse representation and SVM | |
CN110726898B (en) | Power distribution network fault type identification method | |
CN110672905A (en) | CNN-based self-supervision voltage sag source identification method | |
CN110287540B (en) | Photovoltaic cell parameter identification method based on elite crowd differential evolution algorithm | |
CN112787591B (en) | Photovoltaic array fault diagnosis method based on fine-tuning dense connection convolutional neural network | |
WO2024051524A1 (en) | Joint prediction method and apparatus for hydraulic, wind and photovoltaic generation power | |
CN113283580A (en) | Automatic fault detection method for solar cell panel | |
Omer et al. | Comparison of decision tree based ensemble methods for prediction of photovoltaic maximum current | |
CN116229380A (en) | Method for identifying bird species related to bird-related faults of transformer substation | |
CN114640304B (en) | Photovoltaic module current mismatch fault diagnosis method based on I-V curve | |
CN114117921A (en) | Intelligent diagnosis method for faults of photovoltaic array |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200410 |