Photovoltaic power generation array fault diagnosis method and device based on multi-channel feature fusion and regularized deep model, equipment, storage medium and program product

Abstract

The application discloses a photovoltaic array fault diagnosis method and device based on multi-channel feature fusion and a regularized deep model, equipment, a storage medium and a program product. First, irradiance, temperature and current-voltage curve historical data of the photovoltaic array in normal and various fault state operations are acquired, and initial training sample sets are obtained after data preprocessing. Multi-channel feature fusion calculation operations are performed on the initial training sample sets to obtain two-dimensional image training sample sets. A regularized deep model is constructed, and the two-dimensional image training sample sets are used for training to obtain a photovoltaic array fault diagnosis model. New data samples of the photovoltaic array in operation are acquired, and two-dimensional images of new samples are obtained after data preprocessing and multi-channel feature fusion calculation operations. The two-dimensional images of the new samples are input into the photovoltaic array fault diagnosis model to obtain final diagnosis results. The application has the advantage of improving fault diagnosis accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of new energy power equipment state monitoring, in particular to a photovoltaic power generation array fault diagnosis method, device, equipment, storage medium and program product based on multi-channel feature fusion and regularized deep model. BACKGROUND

[0002] As a clean and pollution-free new energy power generation technology with abundant resources, photovoltaic power generation has been rapidly developed and popularized. However, due to the long-term operation of massive photovoltaic power generation arrays and their components in complex and harsh outdoor environments, such as sand, stains, shadows, high temperatures, insulation degradation, etc., photovoltaic power generation arrays are prone to short circuit, open circuit, shading, aging and their composite types. These faults not only reduce the power generation efficiency and economic benefits of photovoltaic power stations, but also easily induce safety accidents such as fire and explosion. Therefore, in order to ensure the safe, stable and economic operation of photovoltaic power stations, it is necessary to accurately diagnose the faults of photovoltaic power generation arrays in a timely manner.

[0003] Currently, the photovoltaic power generation array fault diagnosis technology based on current-voltage (IV) curve electrical signal data and irradiance, temperature environmental sensor data can quickly and accurately diagnose faults, effectively save manual inspection costs and greatly improve work efficiency. This fault diagnosis technology first uses the maximum power point tracker in the photovoltaic inverter to collect current-voltage curve data of the photovoltaic array under normal and various fault conditions, and records the irradiance and temperature environmental sensor values at the corresponding time; then establishes a photovoltaic power generation array fault diagnosis model based on the above collected historical data; finally, new data samples are collected online and input into the fault diagnosis model to obtain the current operation state of the photovoltaic array (normal or a certain type of fault). Compared with the photovoltaic power generation array fault diagnosis technology based on visible light and infrared photos, this technology has the advantages of good privacy protection and fast fault diagnosis speed, and is a promising photovoltaic intelligent operation and maintenance technology.

[0004] For the photovoltaic array fault diagnosis technology based on current-voltage (IV) curve electrical signal data and irradiance, temperature environmental sensor data, document CN111444615A discloses a photovoltaic array fault diagnosis method based on K-nearest neighbor and IV curve, which calculates the feature vector similarity of historical and measured IV curves by K-nearest neighbor method to diagnose the fault state of the current photovoltaic array, but this method only uses part of the feature parameters of the IV curve instead of the complete IV curve, the distinguishable fault types are limited and the fault diagnosis accuracy is insufficient. Document CN109873610A discloses a photovoltaic array fault diagnosis method based on IV characteristics and deep residual network, which diagnoses the fault state of the current photovoltaic array by splicing and inputting the current, voltage, temperature and irradiance four one-dimensional features into the deep residual network, but the fault diagnosis accuracy of this method is poor under the interference of irradiance and temperature environmental changes. Chinese patent document CN114048620A discloses an intelligent fault diagnosis method for photovoltaic array, which uses five feature values on the IV curve to input into a multivariate prediction model for photovoltaic array fault diagnosis, but cannot solve the problem of different state category feature aliasing and difficulty in distinguishing caused by different fault degrees.

[0005] In summary, the existing photovoltaic array fault diagnosis method still cannot effectively reduce the interference caused by irradiance and temperature environmental changes and different fault degrees when using complete IV curve for fault diagnosis, and has the disadvantages of easy aliasing between different state categories and low fault diagnosis accuracy. The main reason for the above-mentioned disadvantages is that some existing methods do not extract more distinguishable and anti-interference features of complete IV curve from multiple channels.

[0006] This is the place where the present application needs to be improved. SUMMARY

[0007] The technical problem to be solved by the present application is to provide a photovoltaic array fault diagnosis method, device, equipment, storage medium and program product based on multi-channel feature fusion and regularized deep model, which effectively improves the fault diagnosis accuracy.

[0008] In order to solve the above technical problems, the present application provides a photovoltaic array fault diagnosis method based on multi-channel feature fusion and regularized deep model, which comprises the following steps:

[0009] Step S1, obtaining irradiance , temperature and current-voltage (IV) curve historical data samples of the photovoltaic array in normal and various fault states, and obtaining initial training sample set after data preprocessing;

[0010] Step S2, performing multi-channel feature fusion calculation operation on the initial training sample set obtained in step S1 to obtain a two-dimensional image training sample set of the photovoltaic power generation array in normal state and various fault states;

[0011] Step S3, using the two-dimensional image training sample set obtained in step S2, after marking the state class label, building a regularized deep model architecture, training the regularized deep model and its parameters, and taking it as a photovoltaic power generation array fault diagnosis model;

[0012] Step S4, obtaining new data samples of irradiance , temperature and current-voltage IV curve of the photovoltaic power generation array in operation, performing data preprocessing and multi-channel feature fusion calculation operation to obtain a two-dimensional image of the new sample;

[0013] Step S5, inputting the two-dimensional image of the new sample into the photovoltaic power generation array fault diagnosis model trained in step S3 to obtain the final diagnosis result.

[0014] The step S1 includes the following steps:

[0015] Step S11, obtaining historical data of irradiance , temperature and current-voltage curve of the photovoltaic power generation array in normal and various fault states;

[0016] Obtain IV curve data samples of the photovoltaic power generation array in normal and fault states in the actual irradiance and temperature variation range, and collect samples for each state, wherein each sample includes irradiance , temperature and 128-point sampling corresponding current-voltage IV curve data, a total of 258 feature values ;

[0017] Step S12, performing data preprocessing calculation operation on the IV curve data obtained in step S11 to obtain an initial training sample set;

[0018] For the current channel, the IV curve data obtained in step S11 is uniformly down-sampled by 64 voltage values in the voltage direction , and then linear interpolation is performed on the current to obtain the current value after data preprocessing, and the calculation formula is as follows:

[0019] (1);

[0020] In the formula: This is the current value after data preprocessing. It is the voltage value after uniform downsampling. and These are the closest to the original IV curves. The voltage values ​​on the left and right sides, and These are the original IV curves and The corresponding current value;

[0021] For the voltage channel, the IV curve data obtained in step S11 64 current values ​​were uniformly downsampled along the current direction. Then, linear interpolation is performed on the voltage to obtain the voltage value after data preprocessing. The calculation formula is as follows:

[0022] (2);

[0023] In the formula: This is the voltage value after data preprocessing. It is the current value of the uniformly downsampled data points. and These are the closest to the original IV curves. The left and right current values, and These are the original IV curves and The corresponding voltage value.

[0024] Step S2 includes the following steps:

[0025] Step S21: The data obtained after preprocessing in step S1 Initial training samples for each current channel Performing Gram angle and field GASF and Gram angle difference field GADF calculations yields results indicating that the photovoltaic array is in a normal state. Training sample set of two-dimensional images of current-Gram angle and field and current-Gram angle difference field during operation under various fault conditions;

[0026] For the current path First, normalization calculations are performed to obtain the normalized current value. The calculation formula is as follows:

[0027] (3);

[0028] In the formula: is the new current value obtained after normalization, is corresponding irradiance and temperature at the short-circuit current value in the normal state;

[0029] Then the new current value obtained after normalization is converted to polar coordinate form, and the calculation formula is as follows:

[0030] (4) ;

[0031] In the formula: is polar angle after arccosine function transformation;

[0032] Finally, the Gram angle of the current channel and the field GASF image feature matrix are calculated, and the image feature matrix is a 64x64 numerical square matrix, and the calculation formula is as follows:

[0033] (5) ;

[0034] At the same time, the Gram angle difference field GADF image feature matrix of the current channel is calculated, and the image feature matrix is a 64x64 numerical square matrix, and the calculation formula is as follows:

[0035] (6) ;

[0036] Step S22: Perform Gram angle and field GASF and Gram angle difference field GADF calculation operation on the initial training sample of the voltage channel obtained after data preprocessing in step S1 , to obtain the current-Gram angle and field and current-Gram angle difference field two-dimensional image training sample set when the photovoltaic power generation array is in normal state and kind of fault state operation;

[0037] For the voltage channel , first, the normalization calculation is performed, and finally the normalized voltage value is obtained, and the calculation formula is as follows:

[0038] (7) ;

[0039] In the formula: is the new voltage value obtained after normalization, is The corresponding irradiance and temperature in the normal state open-circuit voltage value, open-circuit voltage value refers to the IV curve in the current is 0 voltage value;

[0040] The new voltage value obtained by normalization Convert to polar form, the calculation formula is as follows:

[0041] (8);

[0042] In the formula: The polar angle after the inverse cosine function transformation;

[0043] Finally, the voltage channel Gram angle and field GASF image feature matrix of The image feature matrix is a 64x64 numerical square matrix, and the calculation formula is: (9);

[0044] At the same time, the voltage channel Gram angle difference field GADF image feature matrix of The image feature matrix is a 64x64 numerical square matrix, and the calculation formula is as follows:

[0045] (10);

[0046] Step S23, the current channel GASF image feature matrix And GADF image feature matrix And the voltage channel GASF image feature matrix And GADF image feature matrix Calculated in step S22, the values of the corresponding positions of the matrix are weighted and added, so as to complete the multi-channel feature fusion calculation operation, and finally obtain the two-dimensional image training sample set when the photovoltaic power generation array is in normal state and various fault state operation The calculation of each sample As shown in the following formula:

[0047] (11);

[0048] In the formula: The weight value set by man, and satisfies .

[0049] The step S3 includes the following steps:

[0050] ​Step S31, obtaining the two-dimensional image training sample set of the photovoltaic power generation array in normal state and a fault state , and marking the state category label;

[0051] a two-dimensional image training sample set containing normal state and a fault state, each state collects training samples, each training sample is calculated by formula (11) image matrix; the normal state is marked as label 1, and the labels of the fault states are ;

[0052] Step S32, building a regularized deep model architecture;

[0053] The front part of the model architecture contains three layers of nested local feature extraction modules, and the back part of the model architecture contains four layers of nested global feature extraction modules;

[0054] The calculation formula of the three layers of nested local feature extraction modules is as follows:

[0055] (12) ;

[0056] In the formula, is the two-dimensional image training sample calculated by formula (11), and are the features and module parameters extracted by the first layer of local feature extraction modules, and are the features and module parameters extracted by the second layer of local feature extraction modules, and are the features and module parameters extracted by the third layer of local feature extraction modules, , , , are the two-dimensional convolution, batch normalization, activation function and two-dimensional maximum pooling calculation operations commonly used in deep learning, respectively;

[0057] The calculation formula of the four layers of nested global feature extraction modules is as follows:

[0058] (13) ;

[0059] In the formula, and are the features and module parameters extracted by the first layer of global feature extraction modules, and and and and and and , and

[0060] Step S33, constructing a loss function for training the regularized deep model parameters;

[0061] The loss function includes a cross-entropy loss term and a regularization loss term:

[0062] Minimizing the cross-entropy loss term makes the angle between the feature and the classification weight vector corresponding to its true label smaller, enhancing the intra-class compactness of the same state category feature, and its calculation method is shown in the following formula:

[0063] (14);

[0064] In the formula: are the classification weights corresponding to the labels , and and are both L2 norm normalized;

[0065] Minimizing the regularization loss term makes the angle between the classification weight vectors corresponding to different true labels overall maximum, enhancing the inter-class distinguishability of different state category features, and its calculation method is shown in the following formula:

[0066] (15);

[0067] In the formula: is the maximum value in the set of angle cosine values between each label and the classification weight vectors of other labels; the explanations of the remaining parameters are given in the above formula;

[0068] Finally, the loss function of the regularized deep model is expressed as the following formula:

[0069] (16);

[0070] In the formula: is the regularization loss term trade-off coefficient set by man.

[0071] The preferred The number of iterations of model training is dynamically set, and its calculation formula is as follows:

[0072] (17);

[0073] In the formula: is the current iteration number of model training, is the total iteration number of model training, i.e. the maximum iteration number;

[0074] Step S34, initialize the module parameters of the regularized deep model built in step S32 and the label corresponding classification weight vector ;

[0075] For and vector parameters, randomly assign them a value ranging from [-1, 1] as the initial value of the model parameters;

[0076] Step S35, the loss function described in formula (16) is minimized as the goal, and the parameters of the regularized deep model are updated step by step by gradient descent method and When the preset maximum iteration number is reached, the regularized deep model and its parameters are used as the final photovoltaic array fault diagnosis model;

[0077] The calculation formula of updating the parameters at each step is as follows:

[0078] (18);

[0079] In the formula, represents , the parameter value at the first iteration, represents , the parameter value at the first iteration, represents the learning rate, i.e. the iteration step size, represents the derivative of the loss function at the first iteration to .

[0080] The step S4 includes the following steps:

[0081] Step S41, obtain the irradiance, temperature and current-voltage IV curve data samples of the new photovoltaic array in operation, which includes irradiance , temperature and 128-point sampled corresponding current-voltage IV curve data, a total of 258 feature values ;

[0082] Step S42, after performing the data preprocessing in formula (1) and formula (2) in step S1 and the multi-channel feature fusion calculation operation shown in formula (3)-(11) in step S2, the two-dimensional image of the new sample is obtained ;

[0083] In calculating the two-dimensional image of the new sample , the short-circuit current value and the open-circuit voltage value in formula (3) and formula (7) need to be re-determined; the determination method is: first find the historical sample closest to the irradiance and temperature of the new sample in the historical data sample library, and then take the short-circuit current value and the open-circuit voltage value of the historical sample as the short-circuit current value and the open-circuit voltage value in formula (3) and formula (7), so as to calculate the two-dimensional image of the new sample .

[0084] The step S5 includes the following steps:

[0085] Step S51, input the two-dimensional image of the new sample to the photovoltaic power generation array fault diagnosis model obtained by formula (12) and formula (13) in step S3 to obtain the features extracted by the fourth layer global feature extraction module of the model ;

[0086] For , perform formula (12) and formula (13) in step S32, and sequentially calculate the three-layer nested local feature extraction module and the four-layer nested global feature extraction module to obtain the feature output of the new sample ;

[0087] Step S52, perform Soft-max normalization calculation and maximum logic judgment on the feature output of the new sample to obtain the final diagnosis label, and obtain the final diagnosis result by one-to-one correspondence between the label and the photovoltaic power generation array state;

[0088] The calculation formula of Soft-max normalization is as follows:

[0089] (19);

[0090] In the formula: are the labels corresponding classification weight, L2 norm normalization processing is performed;

[0091] The calculation formula of the maximum logical judgment is as follows:

[0092] (20);

[0093] In the formula: is the diagnostic label of the new sample, is the maximum value in The subscript corresponding to the maximum value;

[0094] If is the maximum value in The diagnostic label of the new sample is The first class.

[0095] Specifically, , the labels respectively represent that the photovoltaic power generation array is in a normal state, an open circuit fault, a short circuit fault, a shadow fault, an aging fault, a shadow fault and an open circuit fault, a shadow fault and a short circuit fault, a shadow fault and an aging fault, an aging fault and an open circuit fault, and an aging fault and a short circuit fault. If , it is diagnosed that the new sample is in a normal state; if , it is diagnosed that the new sample is in an aging fault and a short circuit fault.

[0096] In addition, the application also provides a photovoltaic power generation array fault diagnosis device based on multi-channel feature fusion and regularized deep model, comprising:

[0097] A data acquisition module is configured to acquire data samples of irradiance, temperature and current-voltage curve of the photovoltaic power generation array in normal and various fault states; and acquire data samples of irradiance, temperature and current-voltage IV curve of the new photovoltaic power generation array in operation;

[0098] The fault diagnosis module obtains an initial training sample set after pre-processing data samples of the initial training sample set, performs a multi-channel feature fusion calculation operation on the initial training sample set, obtains a two-dimensional image training sample set when the photovoltaic power generation array is in a normal state and various fault states, and after marking a state category label, builds a regularized deep model architecture, trains the regularized deep model and parameters thereof, and takes the regularized deep model and parameters thereof as a photovoltaic power generation array fault diagnosis model; after data pre-processing and the multi-channel feature fusion calculation operation are performed on the obtained irradiance, temperature, and current-voltage IV curve data samples of the photovoltaic power generation array in operation, a two-dimensional image of a new sample is obtained; the two-dimensional image of the new sample is input into the trained photovoltaic power generation array fault diagnosis model, and a final diagnosis result is obtained.

[0099] In addition, the application further provides a photovoltaic power generation array fault diagnosis device based on multi-channel feature fusion and a regularized deep model, comprising a data acquisition module, a data processing and storage module, a fault diagnosis module, and a communication module.

[0100] The data acquisition module is used to acquire irradiance, temperature, and current-voltage curve data when the photovoltaic power generation array is in a normal state and various fault states, and transmit the data to the data processing and storage module.

[0101] The data processing and storage module is used to perform a multi-channel feature fusion calculation operation on the data, build a regularized deep model, and obtain the data in the photovoltaic power generation array fault diagnosis method as described above.

[0102] The fault diagnosis module is used to store and run a computer program; wherein the computer program is configured to implement the steps of the photovoltaic power generation array fault diagnosis method as described above.

[0103] The communication module is used to transmit the data and the fault diagnosis result to an upper computer host module.

[0104] In addition, the application further provides a storage medium, which is a computer readable storage medium, and a computer program is stored on the storage medium, and the computer program is executed by a processor to implement the steps of the photovoltaic power generation array fault diagnosis method as described above.

[0105] In addition, the application further provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the steps of the photovoltaic power generation array fault diagnosis method as described above.

[0106] The one or more technical solutions of the application have at least the following beneficial effects:

[0107] 1) The technical solution of the present application extracts features of the complete IV curve that are more distinguishable and resistant to interference through a multi-channel feature fusion method, effectively improving the adaptability of the fault diagnosis model under irradiance and temperature variation interference;

[0108] 2) The technical solution of the present application constructs a regularized deep model to enhance the inter-class distinguishability and intra-class compactness of the extracted features, effectively reducing the degree of different class mixing caused by different fault levels, and thereby improving the fault diagnosis accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0109] The drawings accompanying the specification of this application serve to provide further understanding of the present application, the illustrative embodiments of the present application and the description thereof serve to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0110] Figure 1 is a workflow diagram of a specific embodiment of the present application;

[0111] Figure 2 is a schematic diagram of the current-voltage IV curve data of a specific embodiment of the present application;

[0112] Figure 3 is a flowchart of the multi-channel feature fusion calculation operation of a specific embodiment of the present application;

[0113] Figure 4 is an architecture diagram of the regularized deep model of a specific embodiment of the present application. DETAILED DESCRIPTION

[0114] The embodiments of the present application will be described in detail below with reference to the accompanying drawings. It should be noted that the execution subject of the specific embodiments of the present application is a computing service device with data processing, network communication and program running functions, such as a personal computer, an embedded computing platform, etc., or an electronic device, a photovoltaic array fault diagnosis device, etc. that realizes the above functions.

[0115] As shown in Figure 1 , the present application provides a photovoltaic power generation array fault diagnosis method based on multi-channel feature fusion and regularized deep model, comprising the following steps:

[0116] Step S1, obtaining irradiance , temperature and current-voltage IV curve historical data samples of the photovoltaic power generation array in normal and various fault states, obtaining initial training sample set after data preprocessing;

[0117] Step S2, performing multi-channel feature fusion calculation operation on the initial training sample set obtained in step S1, obtaining two-dimensional image training sample set of the photovoltaic power generation array in normal and various fault states;

[0118] Step S3, using the two-dimensional image sample set obtained in step S2, after marking the state category label, building a regularized deep model architecture, training the regularized deep model and its parameters, and taking them as a photovoltaic power generation array fault diagnosis model;

[0119] Step S4, obtaining new photovoltaic power generation array running irradiance , temperature and current-voltage IV curve data samples, performing data preprocessing and multi-channel feature fusion calculation operations to obtain new sample two-dimensional images;

[0120] Step S5, inputting the new sample two-dimensional image into the photovoltaic power generation array fault diagnosis model trained in step S3 to obtain the final diagnosis result.

[0121] The step S1 includes the following steps:

[0122] Step S11, obtaining photovoltaic power generation array normal and various fault state running irradiance , temperature and current-voltage IV curve historical data, specifically as follows:

[0123] Obtain photovoltaic power generation array normal and various fault state IV curve data samples within the actual irradiance and temperature variation range, and collect samples for each state, wherein each sample includes irradiance , temperature and 128-point sampling corresponding current-voltage IV curve data, a total of 258 feature values ;

[0124] The embodiment of the present application is to install a photovoltaic inverter with IV curve scanning function for the photovoltaic power generation array, and install an irradiance and temperature environment sensor to collect the irradiance , temperature and current-voltage (IV) curve historical data of the photovoltaic power generation array in normal and various fault states. The preferred historical data samples should be collected within a temperature range of 10 to 40 degrees Celsius and an irradiance range of 600 to 1200 watts per square meter for nearly a month. An embodiment of the photovoltaic power generation array normal and various fault states is , label These represent the following conditions for photovoltaic (PV) arrays: normal operation, open circuit fault, short circuit fault, shading fault, aging fault, simultaneous occurrence of shading and open circuit faults, simultaneous occurrence of shading and short circuit faults, simultaneous occurrence of shading and aging faults, simultaneous occurrence of aging and open circuit faults, and simultaneous occurrence of aging and short circuit faults. The sample sets for open circuit, short circuit, shading, and aging faults should include samples of different fault severity. For example, for aging faults, the severity is categorized as mild aging, moderate aging, and severe aging, and samples should be collected for each condition. One sample, The value should be at least 400 to meet the requirements for training deep learning models. The IV curve is sampled at 128 points, corresponding to the current and voltage values, which are used as the IV curve data. Figure 2 This is a schematic diagram of current-voltage (IV) curve data, showing the short-circuit current. It is the current value on the IV curve when the voltage is 0, and the open-circuit voltage. It is the voltage value on the IV curve when the current is 0.

[0125] Step S12: Process the IV curve data obtained in step S11. Perform data preprocessing and computation operations to obtain the initial training sample set;

[0126] For the current path, the IV curve data obtained in step S11 64 voltage values ​​were uniformly downsampled along the voltage direction. Then, linear interpolation is performed on the current to obtain the preprocessed current value. The calculation formula is as follows:

[0127] (1);

[0128] In the formula: This is the current value after data preprocessing. It is the voltage value after uniform downsampling. and These are the closest to the original IV curves. The voltage values ​​on the left and right sides, and These are the original IV curves and The corresponding current value;

[0129] For the voltage channel, the IV curve data obtained in step S11 64 current values ​​were uniformly downsampled along the current direction. Then, linear interpolation is performed on the voltage to obtain the voltage value after data preprocessing. The calculation formula is as follows:

[0130] (2);

[0131] In the formula: This is the voltage value after data preprocessing. It is the current value of the uniformly downsampled data points. and These are the closest to the original IV curves. The left and right current values, and These are the original IV curves and The corresponding voltage value.

[0132] Step S2 includes the following steps (e.g.) Figure 3 (as shown)

[0133] Step S21: The data obtained after preprocessing in step S1 Initial training samples for each current channel Performing Gram angle and field GASF and Gram angle difference field GADF calculations yields results indicating that the photovoltaic array is in a normal state. Training sample set of two-dimensional images of current-Gram angle and field and current-Gram angle difference field during operation under various fault conditions;

[0134] For the current path First, normalization calculations are performed to obtain the normalized current value. The calculation formula is as follows:

[0135] (3);

[0136] In the formula: It is the new current value obtained after normalization. yes The corresponding short-circuit current value under normal irradiance and temperature conditions. The short-circuit current value refers to the current value when the voltage is 0 in the IV curve.

[0137] Then, the normalized new current value is obtained using the following formula. The formula for calculating polar coordinates is as follows:

[0138] (4);

[0139] In the formula: for Polar angle after inverse cosine function transformation;

[0140] Finally, the current path is calculated. Gram angle and field GASF image feature matrix The image feature matrix is ​​a 64-row, 64-column numerical matrix, and its calculation formula is as follows: (5);

[0141] At the same time, the current path is calculated. Gram angular difference field GADF image feature matrix The image feature matrix is ​​a 64-row, 64-column numerical matrix, and its calculation formula is as follows: (6).

[0142] Step S22: The data obtained after preprocessing in step S1 Initial training samples for each voltage channel Performing Gram angle and field GASF and Gram angle difference field GADF calculations yields results indicating that the photovoltaic array is in a normal state. Training sample set of two-dimensional images of current-Gram angle and field and current-Gram angle difference field during operation under various fault conditions;

[0143] For voltage channel First, a normalization calculation is performed to obtain the normalized voltage value. The calculation formula is as follows:

[0144] (7);

[0145] In the formula: It is the new voltage value obtained after normalization. yes The corresponding open-circuit voltage values ​​under normal irradiance and temperature conditions. The open-circuit voltage value refers to the voltage value when the current is 0 in the IV curve.

[0146] The new voltage value obtained after normalization is obtained by the following formula. The formula for calculating polar coordinates is as follows:

[0147] (8);

[0148] In the formula: for Polar angle after inverse cosine function transformation;

[0149] Finally, the voltage channel is calculated and obtained. Gram angle and field GASF image feature matrix The image feature matrix is ​​a 64-row, 64-column numerical matrix, and its calculation formula is as follows: (9);

[0150] At the same time, obtain the voltage channel Gram angular difference field GADF image feature matrix The image feature matrix is ​​a 64-row, 64-column numerical matrix, and its calculation formula is as follows: (10).

[0151] Step S23: The current path calculated in step S21 is... GASF image feature matrix and GADF image feature matrix and the voltage channel calculated in step S22 GASF image feature matrix and GADF image feature matrix The values ​​at corresponding positions in the matrix are weighted and summed to complete the multi-channel feature fusion calculation, ultimately obtaining a two-dimensional image training sample set of the photovoltaic power generation array under normal and various fault conditions. Each sample The calculation is shown in the following formula:

[0152] (11);

[0153] In the formula: Weights set manually, and satisfying .

[0154] The preferred embodiment is .

[0155] Step S3 includes the following steps:

[0156] Step S31: Obtain the photovoltaic power generation array obtained in step S23 as being in a normal state and Two-dimensional image training sample set during operation under various fault conditions and for Label the status category;

[0157] Two-dimensional image training sample set Includes normal state and Types of fault states, each state collects data. Each training sample It is calculated by formula (11) Image matrix; label the normal state as label 1. The labels for the various fault states are as follows: ;

[0158] Preferred The table below shows the corresponding labels and the status categories of the photovoltaic power generation array;

[0159] .

[0160] Step S32, a regularized deep model architecture is built, as shown in Figure 4 ;

[0161] The model front architecture contains three layers of nested local feature extraction modules and the model back architecture contains four layers of nested global feature extraction modules;

[0162] The formula for calculating the three layers of nested local feature extraction modules is as follows:

[0163] (12);

[0164] In the formula: is the two-dimensional image training sample calculated by formula (11), and are the features and module parameters extracted by the first layer of local feature extraction modules, and are the features and module parameters extracted by the second layer of local feature extraction modules, and are the features and module parameters extracted by the third layer of local feature extraction modules, , , , are the two-dimensional convolution, batch normalization, activation function and two-dimensional maximum pooling calculation operations commonly used in deep learning;

[0165] The formula for calculating the four layers of nested global feature extraction modules is as follows:

[0166] (13);

[0167] In the formula: and are the features and module parameters extracted by the first layer of global feature extraction modules, and are the features and module parameters extracted by the second layer of global feature extraction modules, and are the features and module parameters extracted by the third layer of global feature extraction modules, and are the features and module parameters extracted by the fourth layer of global feature extraction modules, , are the activation function and fully connected calculation operations commonly used in deep learning;

[0168] Preferably, the parameters of the first two-dimensional convolution are set as , the parameter setting of the first layer max-pooling is , the parameter setting of the second layer two-dimensional convolution is , the parameter setting of the second layer max-pooling is , the parameter setting of the third layer two-dimensional convolution is , the parameter setting of the third layer max-pooling is ;

[0169] wherein: represents a two-dimensional sliding window size of , the number of feature maps of the previous layer is 1, the sliding window and the number of feature maps of the current layer are both 10, and the two-dimensional sliding step is , , the symbol explanation of represents a two-dimensional sliding window size of , and the two-dimensional sliding step is ;

[0170] Preferably, the number of neurons of the first layer full connection is 64, the number of neurons of the second layer full connection is 32, the number of neurons of the third layer full connection is 16, and the number of neurons of the fourth layer full connection is 3.

[0171] Step S33, constructing a loss function for training the regularization deep model parameters;

[0172] The loss function includes a cross-entropy loss term and a regularization loss term;

[0173] Minimizing the cross-entropy loss term makes the angle between the feature and the classification weight vector corresponding to its true label smaller, and enhances the intra-class compactness of the same state category feature, and the calculation method is as shown in the following formula:

[0174] (14);

[0175] wherein: are the classification weights corresponding to the labels , and and are both L2 norm normalized;

[0176] Minimizing the regularization loss term makes the angle between the classification weight vectors corresponding to different true labels keep overall maximum, and enhances the inter-class distinguishability of different state category features, and the calculation method is as shown in the following formula:

[0177] (15);

[0178] wherein: is the maximum value in the set of angle cosine values between each label and other label classification weight vectors; the explanations of the rest of the parameters have been given in the above formula;

[0179] Finally, the loss function of the regularized deep model is expressed as follows:

[0180] (16) ;

[0181] wherein: is the artificially set regularization loss term trade-off coefficient;

[0182] The preferred is dynamically set according to the number of iterations of model training, and the calculation formula is as follows:

[0183] (17) ;

[0184] wherein: is the current iteration number of model training, is the total iteration number of model training, i.e. the maximum iteration number.

[0185] Step S34, initialize the module parameters of the regularized deep model built in step S32 and the label corresponding classification weight vector ;

[0186] For and vector parameters, randomly assign them a value ranging from [-1, 1] as the initial value of the model parameters.

[0187] Step S35, the loss function described in formula (16) is taken as the goal to minimize, and the parameters of the regularized deep model are updated step by step by the gradient descent method and When the preset maximum iteration number is reached, the regularized deep model and its parameters are taken as the final photovoltaic array fault diagnosis model;

[0188] The calculation formula of updating the parameters at each step is as follows:

[0189] (18) ;

[0190] wherein, represents , the parameter value at the first iteration, denotes , the parameter value at the i-th iteration, denotes the learning rate, i.e. the iteration step size, denotes the derivative of the loss function with respect to at the i-th iteration.

[0191] The step S4 comprises the following steps:

[0192] Step S41, obtaining the irradiance, temperature and current-voltage (IV) curve data sample of the new photovoltaic power generation array in operation, the new sample comprising irradiance , temperature and 128-point sampled corresponding current-voltage (IV) curve data, a total of 258 feature values ;

[0193] The preferred embodiment is to install a photovoltaic inverter with IV curve scanning function for the photovoltaic power generation array, and install irradiance and temperature environment sensors, to collect the irradiance, temperature and current-voltage (IV) curve data sample of the new photovoltaic power generation array in normal and various fault states.

[0194] Step S42, after performing the data preprocessing of formula (1) and formula (2) in step S1 and the multi-channel feature fusion calculation operation shown in formula (3)-(11) in step S2, obtaining the two-dimensional image of the new sample;

[0195] In the calculation of the two-dimensional image of the new sample, the short-circuit current value and the open-circuit voltage value in formula (3) and formula (7) need to be determined again; the determination method is: first find the historical sample closest to the irradiance and temperature of the new sample in the historical data sample library, and then take the short-circuit current value and the open-circuit voltage value of the historical sample as the short-circuit current value and the open-circuit voltage value in formula (3) and formula (7), so as to calculate the two-dimensional image of the new sample.

[0196] The step S5 comprises the following steps:

[0197] Step S51, inputting the two-dimensional image of the new sample into the photovoltaic power generation array fault diagnosis model obtained from formula (12) and formula (13) in step S3 to obtain the feature extracted by the fourth layer global feature extraction module of the model ;

[0198] To Performing step S32 formula (12) and formula (13) in formula (13), in turn, the three layers of nested local feature extraction module and four layers of nested global feature extraction module, the feature output of new sample .

[0199] Step S52, the feature output of new sample Soft-max normalization calculation and maximum logic judgment to get the final diagnostic label, and the label and photovoltaic array state one by one to get the final diagnosis result;

[0200] The calculation formula of Soft-max normalization is as follows:

[0201] (19);

[0202] In the formula: The label Corresponding classification weight, Has carried out L2 norm normalization processing;

[0203] The calculation formula of maximum logic judgment is as follows:

[0204] (20);

[0205] In the formula: The diagnosis label of new sample, Is The subscript corresponding to the maximum value;

[0206] If Is The maximum value in, the diagnosis label of new sample Is the first Class;

[0207] Specifically, , label Respectively represent photovoltaic array in normal state, open circuit fault, short circuit fault, shading fault, aging fault, shading fault and open circuit fault, shading fault and short circuit fault, shading fault and aging fault, aging fault and open circuit fault, aging fault and short circuit fault occurs simultaneously. If , the new sample is diagnosed in normal state; if , the new sample is diagnosed in aging fault and short circuit fault occurs simultaneously.

[0208] Further, the application also provides a photovoltaic array fault diagnosis device based on multi-channel feature fusion and regularized deep model, comprising:

[0209] a data acquisition module, configured to acquire data samples of irradiance, temperature and current-voltage curve of the photovoltaic array in the initial training sample set when the photovoltaic array is in normal and various fault states, and acquire data samples of irradiance, temperature and current-voltage IV curve of the photovoltaic array when the photovoltaic array is in operation;

[0210] a fault diagnosis module, configured to obtain an initial training sample set after pre-processing the data samples of the initial training sample set, perform multi-channel feature fusion calculation operation on the initial training sample set, obtain a two-dimensional image training sample set of the photovoltaic array in normal and various fault states, and after marking the state category label, build a regularized deep model architecture, train the regularized deep model and parameters thereof, and take the regularized deep model and parameters thereof as a photovoltaic array fault diagnosis model; after data pre-processing and multi-channel feature fusion calculation operation on the acquired irradiance, temperature and current-voltage IV curve data samples of the photovoltaic array in operation, obtain a two-dimensional image of a new sample; input the two-dimensional image of the new sample into the trained photovoltaic array fault diagnosis model to obtain a final diagnosis result.

[0211] The photovoltaic array fault diagnosis device based on multi-channel feature fusion and regularized deep model provided by the application adopts the photovoltaic array fault diagnosis method based on multi-channel feature fusion and regularized deep model in the above embodiment, and solves the problems of different state category feature aliasing and difficulty in distinguishing caused by irradiance and temperature environment change interference and different fault degrees. Compared with the prior art, the photovoltaic array fault diagnosis device based on multi-channel feature fusion and regularized deep model provided by the application has the same beneficial effects as the photovoltaic array fault diagnosis method based on multi-channel feature fusion and regularized deep model provided by the above embodiment, and other technical features of the photovoltaic array fault diagnosis device based on multi-channel feature fusion and regularized deep model are the same as the features disclosed in the above method, which will not be repeated here.

[0212] The application also provides a photovoltaic array fault diagnosis device based on multi-channel feature fusion and regularized deep model, comprising: a data acquisition module, a data processing and storage module, a fault diagnosis module and a communication module;

[0213] The data acquisition module is configured to acquire irradiance, temperature and current-voltage curve data of the photovoltaic array in normal and various fault states, and transmit the data to the data processing and storage module;

[0214] The data processing storage module is configured to perform multi-channel feature fusion calculation operation on the data, construct a regularized deep model, and obtain the data in the photovoltaic power generation array fault diagnosis method as described above.

[0215] The fault diagnosis module is configured to store and run a computer program, wherein the computer program is configured to implement the steps of the photovoltaic power generation array fault diagnosis method as described above.

[0216] The communication module is configured to transmit the data and the fault diagnosis result to the host computer main module.

[0217] The photovoltaic power generation array fault diagnosis device based on multi-channel feature fusion and regularized deep model in the embodiment of the application includes but is not limited to mobile terminals such as notebook computers and fixed terminals such as desktop computers.

[0218] A photovoltaic power generation array fault diagnosis device based on multi-channel feature fusion and regularized deep model, comprising a data acquisition module, a data processing storage module, a fault diagnosis module and a communication module. The data acquisition module is a photovoltaic inverter with IV curve scanning function installed on the photovoltaic power generation array, and an irradiance and temperature environment sensor is installed to acquire data samples of irradiance, temperature and current-voltage (IV) curve when the photovoltaic power generation array is running in normal and various fault states. The data acquisition module is electrically connected to the photovoltaic power generation array to obtain data samples of the photovoltaic power generation array. The data processing storage module and the fault diagnosis module are modules integrated with central processors, graphic processors and other processors, as well as RAM, ROM and other memories. The communication module allows the photovoltaic power generation array fault diagnosis device to communicate with other devices wirelessly or by wire to exchange data. For example, the communication module allows the photovoltaic power generation array fault diagnosis device to communicate with the host computer main module through the RS485 bus to exchange data.

[0219] The application further provides a computer readable storage medium having computer readable program instructions stored thereon, the computer readable program instructions being used to execute the photovoltaic power generation array fault diagnosis method in the above embodiment.

[0220] The computer readable storage medium provided in the application is, for example, a U disk, but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination of the above. More specific examples of the computer readable storage medium include, but are not limited to, an electrical connection with one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer readable storage medium is any tangible medium containing or storing a program used or combined with an instruction execution system, system, or device. The program code contained on the computer readable storage medium is transmitted by any suitable medium, including but not limited to an electrical wire, an optical cable, a radio frequency (RF), or any suitable combination of the above.

[0221] The application also provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the photovoltaic power generation array fault diagnosis method as described.

[0222] The computer program product provided by the application solves the problem of feature aliasing and difficulty in distinguishing different state categories caused by different degrees of interference and faults due to changes in irradiance and temperature environment. Compared with the prior art, the computer program product provided by the application has the same beneficial effects as the photovoltaic power generation array fault diagnosis method based on multi-channel feature fusion and regularized deep model provided by the above-mentioned embodiments, and will not be described here.

[0223] The application solves the problem of feature aliasing and difficulty in distinguishing different state categories caused by different degrees of interference and faults due to changes in irradiance and temperature environment when using complete IV curves for photovoltaic power generation array fault diagnosis, enhances the inter-class distinguishability and intra-class compactness of the features extracted by the photovoltaic power generation array fault diagnosis model, and improves the fault diagnosis accuracy.

[0224] The above only describes the preferred embodiments of the application and is not intended to limit the application. Those skilled in the art can make various modifications and changes to the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A photovoltaic power generation array fault diagnosis method based on multi-channel feature fusion and regularized deep model, characterized in that, Comprising: Step S1: Obtain the irradiance of the photovoltaic power generation array under normal and various fault conditions. ,temperature The initial training sample set is obtained by preprocessing historical data samples of current-voltage IV curves. Step S2, performing multi-channel feature fusion calculation operation on the initial training sample set to obtain a two-dimensional image training sample set of the photovoltaic array in normal state and various fault states; Step S3, using the above two-dimensional image training sample set, after marking the state category label, building a regularized deep model architecture, training the regularized deep model and its parameters, and taking it as a photovoltaic array fault diagnosis model; Step S4, obtaining data samples of irradiance, temperature and current-voltage IV curve of the new photovoltaic array in operation , and executing data preprocessing and multi-channel feature fusion calculation operation to obtain a two-dimensional image of the new sample; ​ Step S5, inputting the two-dimensional image of the new sample into the photovoltaic array fault diagnosis model trained in step S3 to obtain the final diagnosis result; The step S1 comprises the following steps: Step S11, acquire irradiance when the photovoltaic power generation array is in normal and various fault state operation , temperature and current-voltage IV curve history data; Obtaining normal and faulty state IV curve data samples for a range of actual irradiance and temperature variations, each state collected over a number of samples, wherein each sample includes irradiance , temperature and 128 point sampled corresponding current-voltage IV curve data, a total of 258 feature values Step S12, processing the IV curve data obtained in step S11 performing data preprocessing calculation operation to obtain initial training sample set; For the current channel, the IV curve data obtained in step S11 is fitted with a linear function , and the data is uniformly down-sampled by 64 voltage values in the voltage direction , and then linear interpolation is performed on the current to obtain the current value after data preprocessing , and the calculation formula is as follows: （1）； In the formula: is the current value after data preprocessing, is the voltage value after uniform down-sampling, and are the left and right voltage values closest to in the original IV curve, respectively, and are the current values corresponding to and in the original IV curve, respectively. For the voltage channel, the IV curve data obtained in step S11 is preprocessed as follows , and the data is uniformly down-sampled by 64 current values in the current direction , and then linear interpolation is performed on the voltage to obtain the preprocessed voltage value , and the calculation formula is as follows: （2）； In the formula: is the voltage value after data preprocessing, is the current value of uniformly down-sampled data points, and are the left and right current values closest to in the original IV curve, respectively, and are the corresponding voltage values of and , respectively. The step S2 comprises the following steps: Step S21: obtaining the initial training sample of the current channel from the data obtained after the data preprocessing in step S1 Step S21: obtaining the initial training sample of the current channel from the data obtained after the data preprocessing in step S1 performing Gram angle and field GASF and Gram angle difference field GADF calculation operations to obtain the current-Gram angle and field and current-Gram angle difference field two-dimensional image training sample set when the photovoltaic power generation array is in a normal state and performing Gram angle and field GASF and Gram angle difference field GADF calculation operations to obtain the current-Gram angle and field and current-Gram angle difference field two-dimensional image training sample set when the photovoltaic power generation array is in a normal state and For the current channel , first, a normalization calculation is performed to obtain a normalized current value , the calculation formula of which is as follows: （3）； wherein: is the new current value obtained by normalization, is corresponding to the irradiance and temperature at normal conditions. The normalized new current value is then converted to polar form by the following equation The conversion to polar form is calculated as follows: （4）； In the formula: is polar angle after arccosine function transformation; Finally , the current channel is calculated gram angle and field GASF image feature matrix of the image The image matrix is a 64x64 numerical square matrix, and its calculation formula is as follows: (5); At the same time, the current channel is calculated Gram angle difference field (GADF) image feature matrix The image matrix is a 64x64 numerical square matrix, and its calculation formula is as follows: (6) Step S22: The data obtained after preprocessing in step S1 Initial training samples for each voltage channel Performing Gram angle and field GASF and Gram angle difference field GADF calculations yields results indicating that the photovoltaic array is in a normal state. Training sample set of two-dimensional images of current-Gram angle and field and current-Gram angle difference field during operation under various fault conditions; For the voltage channel First, a normalization calculation is performed to obtain a normalized voltage value The calculation formula is as follows: （7）； In the formula: is the new voltage value obtained by normalization, is corresponding to the irradiance and temperature in the normal state open circuit voltage value; The new voltage value obtained after normalization is converted into polar form by the following formula The conversion into polar form is calculated by the following formula: （8）； In the formula: is polar angle after arccosine function transformation; Finally, the voltage channel is calculated and obtained. Gram angle and field GASF image feature matrix The image matrix is ​​a 64-row, 64-column numerical matrix, and its calculation formula is as follows: (9); At the same time, the voltage channel is obtained Gram angle difference field (GADF) image feature matrix The image matrix is a 64x64 numerical square matrix, and the calculation formula is as follows: (10) Step S23, the current channel calculated in step S21 GASF image feature matrix and GADF image feature matrix and the voltage channel calculated in step S22 GASF image feature matrix and GADF image feature matrix , the value of the matrix corresponding position is weighted and added, so as to complete the multi-channel feature fusion calculation operation, and finally obtain the two-dimensional image training sample set when the photovoltaic power generation array is in normal state and various fault states The calculation of each sample is as follows: （11）； In the formula: is a weight set by man and satisfies ; The step S3 comprises the following steps: Step S31, obtaining the two-dimensional image training sample set of the photovoltaic power generation array in the normal state and a fault state and for marking the state category label on the image A two-dimensional image training sample set includes normal state and a kind of fault state, each state is collected One training sample, each training sample Is calculated by formula (11) Image matrix; the normal state is marked as label 1, The label of one kind of fault state is ; Step S32, building a regularized deep model architecture; The front part of the model architecture contains three layers of nested local feature extraction modules, and the back part of the model architecture contains four layers of nested global feature extraction modules; The calculation formula of the three layers of nested local feature extraction modules is as follows: （12）； In the formula: is a two-dimensional image training sample calculated by formula (11), and are features and module parameters extracted by the first layer local feature extraction module, and are features and module parameters extracted by the second layer local feature extraction module, and are features and module parameters extracted by the third layer local feature extraction module, , , , are common two-dimensional convolution, batch normalization, activation function and two-dimensional maximum pooling calculation operations in deep learning, respectively. The calculation formula of the four layers of nested global feature extraction modules is as follows: （13）； In the formula: and are respectively the features and module parameters extracted by the first layer global feature extraction module, and are respectively the features and module parameters extracted by the second layer global feature extraction module, and are respectively the features and module parameters extracted by the third layer global feature extraction module, and are respectively the features and module parameters extracted by the fourth layer global feature extraction module, , are respectively a general activation function and a full connection calculation operation in deep learning. Step S33, constructing a loss function for training the parameters of the regularized deep model; The loss function includes a cross-entropy loss term and a regularization loss term: Minimize the cross-entropy loss term Make features The angle between the classification weight vectors corresponding to their true labels is smaller, which enhances the intra-class compactness of features of the same state category. The calculation method is shown in the following formula: (14)； In the formula: respectively are label corresponding classification weights, and and are subjected to L2 norm normalization processing; minimizing a regularization loss term The angle between the classification weight vectors corresponding to different real labels is maximized, which enhances the class distinction of different state categories. The calculation method is as follows: (15)； In the formula: is the maximum value in the set of angle cosine values between each label and the classification weight vector of other labels; Finally, the loss function of the regularized deep model is expressed as expressed as: (16)； In the formula: is a regularization loss term set artificially trade-off coefficient; The number of iterations for model training is dynamically set according to the formula: (17)； In the formula: is the current iteration number of model training, is the total iteration number of model training, i.e. the maximum iteration number; Step S34, initialize the module parameters of the regularized deep model built in step S32 and the label corresponding classification weight vector ; For and Vector parameters are randomly assigned values ranging between [-1, 1] as initial values for the model parameters; Step S35, the loss function described in the minimum formula (16) is taken as the goal, and the parameters of the regularized deep model are updated step by step by the gradient descent method and When the preset maximum iteration number is reached , the regularized deep model and its parameters are taken as the final photovoltaic power generation array fault diagnosis model The calculation formula of each step of updating the parameters is as follows: （18）； wherein denotes , at the th iteration, denotes , at the th iteration, denotes the learning rate, i.e. the iteration step size, denotes the derivative of the loss function with respect to at the th iteration; The step S4 comprises the following steps: Step S41, obtain the irradiance, temperature and current-voltage IV curve data sample of the new photovoltaic power generation array in operation, the new sample includes irradiance , temperature and 128-point sampled corresponding current-voltage IV curve data, a total of 258 characteristic values ; Step S42, after performing the data preprocessing of formula (1) and formula (2) in step S1 and the multi-channel feature fusion calculation operation shown in formula (3)-(11) in step 2, the two-dimensional image of the new sample is obtained ; When calculating the two-dimensional image of the new sample , the short-circuit current value in formula (3) and formula (7) and the open-circuit voltage value need to be re-determined; the determination method is: first find the historical sample closest to the irradiance and temperature of the new sample in the historical data sample library, and then take the short-circuit current value and the open-circuit voltage value of the historical sample as the short-circuit current value in formula (3) and formula (7) and the open-circuit voltage value , so as to calculate the two-dimensional image of the new sample ; The step S5 comprises the following steps: Step S51: Obtain the two-dimensional image of the new sample. The photovoltaic array fault diagnosis model obtained by inputting formulas (12) and (13) in step S3 is used to obtain the features extracted by the global feature extraction module of the fourth layer of the model. ; To The formula (12) and formula (13) in step S32 are executed, and the three-layer nested local feature extraction module and the four-layer nested global feature extraction module are sequentially calculated to obtain the feature output of the new sample ; Step S52, output the features of the new sample The final diagnosis label is obtained by performing Soft-max normalization calculation and maximum logic judgment, and the final diagnosis result is obtained by one-to-one correspondence between the label and the photovoltaic power generation array state. The calculation formula of the Soft-max normalization is as follows: (19)； In the formula: respectively, the label corresponding classification weight, L2 norm normalization processing is performed; The calculation formula of the maximum logical judgment is as follows: (20)； wherein: is the diagnostic label of the new sample, is is the index corresponding to the maximum value in like yes The maximum value in the range is the diagnostic label of the new sample. It is the first kind. 2.The device for diagnosing based on the photovoltaic power generation array fault diagnosis method based on multi-channel feature fusion and regularization deep model according to claim 1, characterized in that: Comprising: The data acquisition module is used to acquire data samples of irradiance, temperature and current-voltage curve of the photovoltaic array in normal and various fault states in the initial training sample set; and is used to acquire data samples of irradiance, temperature and current-voltage IV curve of the new photovoltaic array in operation; The fault diagnosis module is used to obtain the initial training sample set after pre-processing the data samples in the initial training sample set, perform multi-channel feature fusion calculation operation on the initial training sample set, obtain a two-dimensional image training sample set of the photovoltaic array in normal state and various fault states, and after marking the state category label, build a regularized deep model architecture, train the regularized deep model and its parameters, and take it as a photovoltaic array fault diagnosis model; after data pre-processing and multi-channel feature fusion calculation operation on the acquired irradiance, temperature and current-voltage IV curve data samples of the new photovoltaic array in operation, obtain the two-dimensional image of the new sample; input the two-dimensional image of the new sample into the trained photovoltaic array fault diagnosis model to obtain the final diagnosis result.

3. A photovoltaic power generation array fault diagnosis device based on multi-channel feature fusion and regularized deep model, characterized by: Comprising a data acquisition module, a data processing and storage module, a fault diagnosis module and a communication module; The data acquisition module is used to acquire irradiance, temperature and current-voltage curve data of the photovoltaic array in normal and various fault states, and transmit the data to the data processing and storage module; The data processing and storage module is used to perform multi-channel feature fusion calculation operation on the data, build a regularized deep model, and obtain the data in the photovoltaic array fault diagnosis method according to claim 1. The fault diagnosis module is used for storing and running a computer program; wherein the computer program is configured to realize the steps of the photovoltaic power generation array fault diagnosis method according to claim 1. The communication module is used for transmitting the data and the fault diagnosis result to an upper computer host module.

4. A storage medium characterized by: The storage medium is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by the processor to realize the steps of the photovoltaic power generation array fault diagnosis method according to claim 1.

5. A computer program product, characterised in that: The computer program is executed by the processor to realize the steps of the photovoltaic power generation array fault diagnosis method according to claim 1.

Images