A method, device and storage medium for thermal fault identification of photovoltaic inverter

Abstract

The present invention relates to a method, device and storage medium for thermal fault identification of a photovoltaic inverter, and to the technical field of thermal fault identification of photovoltaic inverters. A discrete infrared image and a corresponding RGB image of a photovoltaic inverter are collected; the discrete infrared image is decomposed by a two-dimensional variational modal decomposition method to obtain a modal function with the minimum sum of M two-dimensional analytical signal estimated bandwidths; for the modal function and the corresponding RGB image of the discrete infrared image in a continuous period, the fluctuation amount of the modal function and the RGB image is obtained; the modal function of the discrete infrared image, the RGB image corresponding to the discrete infrared image and the fluctuation amount of the two are alternately interspersed to form a feature matrix F; the obtained feature matrix F is transmitted to a pre-trained thermal fault model to obtain the thermal fault category and trend of the photovoltaic inverter. The present application combines infrared images and RGB images, and further uses the modal function of the infrared image as a feature, combined with the fluctuation amount, to better perform thermal fault classification and trend analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal fault identification of photovoltaic inverters, and in particular to a thermal fault identification method, device and storage medium of photovoltaic inverters. Background Art

[0002] The main component of the inverter is the IGBT, and the thermal condition of the IGBT is closely related to its fault type. The following are some common IGBT thermal fault types and their characteristics: Due to drying of thermal paste, contamination of the heat sink, or poor contact between the heat sink and the IGBT, the thermal resistance between the IGBT module and the heat sink increases, resulting in a decrease in heat dissipation efficiency and a thermal resistance increase fault. The IGBT module experiences frequent start-stop or load fluctuations, resulting in thermal stress caused by temperature changes, resulting in thermal cycle faults; thermal cycle faults may cause fatigue of the solder joints of the IGBT and accelerate material aging. It may be due to local overload, uneven heat dissipation, internal short circuit or poor welding. The temperature of a local area of ​​the IGBT module rises abnormally, forming a hot spot fault; the hot spot may cause local material degradation and even thermal runaway. Thermal runaway faults are usually the result of multiple faults, such as internal short circuits, complete failure of the cooling system, etc. The sharp rise in the temperature of the IGBT module may cause rapid damage to the IGBT module and even cause a fire.

[0003] Diagnosing and resolving thermal failures of IGBTs usually requires a combination of thermal imaging, electrical testing, and maintenance inspections. However, existing thermal imaging-based thermal failure detection mostly uses infrared images to analyze the relationship between temperature and set thresholds to determine whether a thermal failure has occurred, but lacks further analysis of the type of thermal failure and thermal failure trends. Summary of the invention

[0004] In order to solve the above technical problem or at least partially solve the above technical problem, the present invention provides a method, device and storage medium for identifying thermal faults of a photovoltaic inverter.

[0005] In a first aspect, the present invention provides a method for identifying a thermal fault of a photovoltaic inverter, comprising:

[0006] Collecting discrete infrared images of photovoltaic inverters And the corresponding RGB image ;

[0007] Discrete infrared image The modal function with the minimum sum of the estimated bandwidths of M two-dimensional analytical signals is obtained by two-dimensional variational mode decomposition. ;

[0008] For discrete infrared images in continuous time periods The modal function And the corresponding RGB image , obtain the fluctuation of the modal function and RGB image; The modal function , corresponding to discrete infrared image The RGB images and the fluctuations of the two are alternately interspersed to form the feature matrix F;

[0009] The obtained feature matrix F is fed into the pre-trained thermal fault model to obtain the fault category and fault trend of the photovoltaic inverter.

[0010] Furthermore, the discrete infrared image The modal function with the minimum sum of the estimated bandwidths of M two-dimensional analytical signals is obtained by two-dimensional variational mode decomposition. ,include:

[0011] According to the modal function The definition of the two-dimensional analytical signal in the frequency domain and the Fourier transform characteristics obtain the two-dimensional analytical signal and modal function The connection between:

[0012] Modal Function The two-dimensional analytical signal of is defined in the frequency domain as:

[0013] ;

[0014] in, for The frequency domain representation of for The center frequency, for The corresponding instantaneous frequency in the frequency domain is, represents the symbolic function, represents the inner product;

[0015] According to the Fourier transform characteristics, the two-dimensional analytical signal and modal function are obtained The connection between:

[0016] ;

[0017] in, is the Dirac function, is convolution, for The two-dimensional analytical signal of

[0018] The squared gradient of the Gaussian smoothing offset is used to estimate the bandwidth of the two-dimensional analytical signal. The objective function of minimizing the sum of the bandwidths of the M modal functions is expressed as:

[0019] ;

[0020] in, represents the square of the L2 norm, represents the gradient function, represents the weight coefficient;

[0021] exist Solve the objective function under the constraint of:

[0022] Introducing quadratic penalty terms and Lagrange multipliers into the objective function yields:

[0023] ;

[0024] in, Indicates Discrete infrared images The corresponding Lagrange multiplier is,

[0025] The ADMM algorithm is used to alternately update the mode function, center frequency and Lagrange multiplier in the frequency domain to find the saddle point that satisfies the objective function.

[0026] Furthermore, the ADMM algorithm is used to alternately update the mode function, center frequency and Lagrange multiplier in the frequency domain to find the saddle point that satisfies the objective function. The process includes:

[0027] initialization: , , , , and . Indicates the first iteration. Discrete infrared images No. Modal Function The frequency domain expression of is obtained by fast Fourier transform. Indicates the first iteration, corresponding to Discrete infrared images No. The center frequency of the mode function, Indicates the first iteration. Discrete infrared images The corresponding frequency domain expression of the Lagrange multiplier is, and are the convergence threshold and error threshold respectively;

[0028] No. Discrete infrared images No. The update formula of the frequency domain expression of a modal function is as follows:

[0029] ;

[0030] in, , For the At the iteration step Discrete infrared images No. The frequency domain expression of the modal function is , For the Discrete infrared images The frequency domain expression of For the At the iteration step Discrete infrared images No. The frequency domain expression of the modal function is , Indicates At the iteration step Discrete infrared images The corresponding frequency domain expression of the Lagrange multiplier is, Indicates When the iteration step is Discrete infrared images No. The center frequency of the mode function.

[0031] The update formula of the center frequency in the frequency domain is as follows:

[0032] ;

[0033] in, For the At the iteration step Discrete infrared images No. Frequency domain expression of the modal functions;

[0034] The update formula of the Lagrange multiplier frequency domain expression is as follows:

[0035] ;in,

[0036] in, To control the parameters of the updated Lagrange multiplier step size;

[0037] Iterate until the following convergence conditions are met:

[0038] ,

[0039] and,

[0040] .

[0041] Furthermore, The fluctuation of a modal function:

[0042] ;

[0043] Fluctuation of RGB image: ;

[0044] in, For the Modal Function The average value during the period, For RGB images The average value during the period, , , is the starting point of the continuous time period, The end point of the continuous time period.

[0045] Furthermore, the thermal fault model includes:

[0046] 2D convolution layer and layer normalization, the convolution kernel size of the 2D convolution layer is 3, the stride is 1, and the padding is 1. The features extracted by the 2D convolution layer maintain the spatial dimension of the feature matrix F;

[0047] After the 2D convolutional layers and layer normalization, there are multiple groups of depth scaling layers and ConvNeXt blocks that introduce SENet layers.

[0048] The features are mapped to the fully connected layer for classification through linear mapping, and the Softmax layer is set after the fully connected layer.

[0049] Furthermore, the depth scaling layer consists of layer normalization and several convolutional layers; layer normalization is applied to the channels of features sequentially, and the following convolutional layers gradually expand the channel dimensions.

[0050] Furthermore, the ConvNeXt block introducing the SENet layer includes a depth-wise convolution layer, which processes the channels of each input feature independently, followed by the SENet layer and layer normalization. The SENet layer selectively emphasizes important channels and reduces irrelevant channels to optimize channel features. After layer normalization, two point-by-point convolution layers are adopted: the first point-by-point convolution layer expands the channel dimension of the feature by four times, and the second point-by-point convolution layer restores the channel of the feature to its original size. The two point-by-point convolution layers transform the channel feature representation, and a GELU activation function is set between the two point-by-point convolution layers, and a learnable scaling parameter is also used to adjust the output according to the significance of the feature; the ConvNeXt block introducing the SENet layer also integrates residual connections and dropout to promote information flow and reduce overfitting, respectively.

[0051] Furthermore, the SENet layer includes: a global average pooling layer, followed by a first fully connected layer with a ReLU activation function and a second fully connected layer with a Sigmoid activation function. The SENet layer is able to dynamically focus on the most relevant features in the feature channel by emphasizing significant features and suppressing smaller features, and introducing nonlinearity through the ReLU activation function and the Sigmoid activation function.

[0052] In a second aspect, the present invention provides a thermal fault identification device for a photovoltaic inverter, comprising: at least one processing unit, the processing unit is connected to a storage unit and a collection unit through a bus unit, the collection unit collects discrete infrared images of the photovoltaic inverter And the corresponding RGB image The storage unit stores a computer program, and when the computer program is executed by the processing unit, the thermal fault identification method of the photovoltaic inverter is implemented.

[0053] In a third aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for identifying thermal faults of a photovoltaic inverter as described above is implemented.

[0054] The above technical solution provided by the embodiment of the present invention has the following advantages compared with the prior art:

[0055] The present application collects the discrete infrared image and the corresponding RGB image of the photovoltaic inverter; decomposes the discrete infrared image through two-dimensional variational mode decomposition to obtain the modal function with the minimum sum of the estimated bandwidth of M two-dimensional analytical signals; for the modal function and the corresponding RGB image of the discrete infrared image in a continuous period, the fluctuation of the modal function and the RGB image is obtained; the modal function of the discrete infrared image, the RGB image corresponding to the discrete infrared image and the fluctuation of the two are alternately interspersed to form a feature matrix F; the obtained feature matrix F is transmitted to the pre-trained thermal fault model to obtain the thermal fault category of the photovoltaic inverter. The present application uses the data of the infrared image and the data of the RGB image to identify thermal faults, and integrates the features of the two images to identify thermal faults, providing richer features for thermal fault identification. And for the infrared image, the present application obtains the modal function with the minimum sum of the estimated bandwidth of M modal functions through two-dimensional variational mode decomposition, and uses the modal function as a partial feature to more finely establish the connection between the thermal fault and the features contained in different modal functions, so as to better classify the thermal fault.

[0056] The characteristic matrix F of the present application also incorporates the fluctuation amount, and the trend of thermal faults can be well reflected through the time series relationship of the fluctuation amount, that is, the relationship between the fluctuation amount channels.

[0057] The fault identification of this application adopts the ConvNeXt block that introduces the SENet layer. The SENet layer selectively emphasizes important channels and reduces irrelevant channels to optimize channel features, so as to better understand the thermal fault and the thermal fault trend through the connection between channel features. The SENet layer can also effectively enhance the model's ability to process and interpret the complexity of multi-channel data sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0060] Figure 1 A flow chart of a method for identifying thermal faults of a photovoltaic inverter provided by an embodiment of the present invention;

[0061] Figure 2 A schematic diagram of a thermal fault model provided by an embodiment of the present invention;

[0062] Figure 3 A flowchart of a ConvNeXt block introducing a SENet layer provided in an embodiment of the present invention;

[0063] Figure 4 A schematic diagram of a SENet layer provided by an embodiment of the present invention;

[0064] Figure 5 A schematic diagram of a thermal fault identification device for a photovoltaic inverter provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0067] Example 1

[0068] like Figure 1 As shown, the technology of the present invention implements a thermal fault identification method for a photovoltaic inverter, including:

[0069] Collecting discrete infrared images of photovoltaic inverters And the corresponding RGB image .

[0070] Discrete infrared image The modal function with the minimum sum of the estimated bandwidths of M two-dimensional analytical signals is obtained by two-dimensional variational mode decomposition. .

[0071] In the specific implementation process, according to the modal function The definition of the two-dimensional analytical signal in the frequency domain and the Fourier transform characteristics obtain the two-dimensional analytical signal and modal function The connection between.

[0072] Modal Function The two-dimensional analytical signal of is defined in the frequency domain as:

[0073] ;

[0074] in, for The frequency domain representation of for The center frequency, for The corresponding instantaneous frequency in the frequency domain is, represents the symbolic function, represents the inner product;

[0075] According to the Fourier transform characteristics, the two-dimensional analytical signal and modal function are obtained The connection between:

[0076] ;

[0077] in, is the Dirac function, is convolution, for The two-dimensional analytical signal.

[0078] The squared gradient of the Gaussian smoothing offset is used to estimate the bandwidth of the two-dimensional analytical signal. The objective function of minimizing the sum of the bandwidths of the M modal functions is expressed as:

[0079] ;

[0080] in, represents the square of the L2 norm, represents the gradient function, represents the weight coefficient;

[0081] Need to Solve the objective function under the constraint of:

[0082] Introducing quadratic penalty terms and Lagrange multipliers into the objective function yields:

[0083] ;

[0084] in, Indicates Discrete infrared images The corresponding Lagrange multiplier is,

[0085] The ADMM algorithm is used to alternately update the mode function, center frequency, and Lagrange multiplier in the frequency domain. The process includes:

[0086] initialization: , , , , and . Indicates the first iteration. Discrete infrared images No. Modal Function The frequency domain expression of is obtained by fast Fourier transform. Indicates the first iteration, corresponding to Discrete infrared images No. The center frequency of the mode function, Indicates the first iteration. Discrete infrared images The corresponding frequency domain expression of the Lagrange multiplier is, and are the convergence threshold and error threshold respectively.

[0087] No. Discrete infrared images No. The update formula of the frequency domain expression of a modal function is as follows:

[0088] ;

[0089] in, , For the At the iteration step Discrete infrared images No. The frequency domain expression of the modal function is , For the Discrete infrared images The frequency domain expression of For the At the iteration step Discrete infrared images No. The frequency domain expression of the modal function is , Indicates At the iteration step Discrete infrared images The corresponding frequency domain expression of the Lagrange multiplier is, Indicates When the iteration step is Discrete infrared images No. The center frequency of the mode function.

[0090] The update formula of the center frequency in the frequency domain is as follows:

[0091] ;

[0092] in, For the At the iteration step Discrete infrared images No. Frequency domain expression of the modal functions;

[0093] The update formula of the Lagrange multiplier frequency domain expression is as follows:

[0094] ;in,

[0095] in, Parameter for controlling the step size of updating the Lagrange multipliers.

[0096] Iterate until the following convergence conditions are met:

[0097] ,

[0098] and,

[0099] .

[0100] For discrete infrared images in continuous time periods The modal function And the corresponding RGB image , , obtain the fluctuation of the mode function and RGB image;

[0101] No. The fluctuation of a modal function: ;

[0102] Fluctuation of RGB image: ;

[0103] in, For the Modal Function The average value during the period, For RGB images The average value during the period, , , is the starting point of the continuous time period, The end point of the continuous time period.

[0104] Discrete infrared image The modal function , corresponding to discrete infrared image The RGB images and the fluctuations of the two are alternately interspersed to form the feature matrix F.

[0105] Discrete infrared image The modal function And the corresponding RGB image The first characteristic matrix , the discrete infrared image The modal function And the corresponding RGB image The fluctuations of the second characteristic matrix , then the feature matrix , is the first characteristic matrix and the second characteristic matrix channel.

[0106] This application uses infrared image data and RGB image data to identify thermal faults, and integrates the features of the two images to identify thermal faults, providing richer features for thermal fault identification. For infrared images, this application obtains the modal function with the smallest sum of M estimated bandwidths through two-dimensional variational modal decomposition, and uses the modal function as a partial feature to more finely establish the connection between thermal faults and the features contained in different modal functions, so as to better classify thermal faults.

[0107] The obtained feature matrix F is fed into the pre-trained thermal fault model to obtain the thermal fault category of the photovoltaic inverter.

[0108] In the specific implementation process, Figure 2 As shown, the thermal fault model includes:

[0109] 2D convolution layer and layer normalization,The convolution kernel size of the 2D convolution layer is 3, the stride is 1, and the padding is 1. The features extracted by the 2D convolution layer maintain the spatial dimension of the feature matrix F. After the convolution operation in the 2D convolution layer, the layer normalization is used to normalize the extracted features.

[0110] After the 2D convolutional layers and layer normalization, there are multiple sets of deep scaling layers and ConvNeXt blocks that introduce SENet layers. The deep scaling layers contain layer normalization and several convolutional layers; layer normalization is applied to the channels of the features in sequence, followed by convolutional layers that change the channel dimension of the features, and the convolutional layers gradually expand the channel dimension. The convolutional layers have a kernel size of 3, a stride of 1, and are configured to retain the padding of the original input size to maintain the spatial dimension of the features.

[0111] like Figure 3 As shown in the figure, the ConvNeXt block introduced into the SENet layer includes a depth-wise convolution layer, the convolution kernel size of the depth-wise convolution layer is 7, and the padding width is 3. The depth-wise convolution layer processes the channel of each input feature independently to improve the efficiency of spatial feature extraction. Subsequently, there is a SENet layer and layer normalization, which selectively emphasizes important channels and reduces irrelevant channels through the SENet layer to optimize channel features, so as to better understand the thermal fault trend while understanding the thermal fault through the connection between channel features.

[0112] SENet layers can also effectively enhance the model's ability to handle and interpret the complexity of multi-channel datasets. Figure 4 As shown in the figure, the SENet layer includes: a global average pooling layer, followed by the first fully connected layer with a ReLU activation function and the second fully connected layer with a Sigmoid activation function. The SENet layer is able to dynamically focus on the most relevant features in the feature channel, by emphasizing significant features and suppressing smaller features, and introducing nonlinearity through the ReLU activation function and the Sigmoid activation function. SENet enhances the ability to understand and predict complex thermal fault types.

[0113] After layer normalization, two point-wise convolutional layers are adopted: the first point-wise convolutional layer expands the channel dimension of the feature by four times. The second point-wise convolutional layer restores the channel of the feature to its original size, effectively transforming the channel feature representation. A GELU activation function is set between the two point-wise convolutional layers, and a learnable scaling parameter is also utilized to adjust the output according to the saliency of the feature. The ConvNeXt block introduced into the SENet layer also integrates residual connections and dropout to promote information flow and alleviate overfitting, respectively. In this application, point-wise convolution is used for efficient channel integration and dimensionality refinement.

[0114] In the thermal fault model, features are mapped to a fully connected layer for classification through linear mapping, and a Softmax layer is set after the fully connected layer. The output dimension of the fully connected layer covers all identifiable thermal fault types and their trends to meet the classification and trend analysis of all thermal fault types.

[0115] During the model training process, for photovoltaic inverters with different thermal fault types, discrete infrared images and corresponding RGB images are obtained, and the feature matrix F is extracted in the above manner. The feature matrix F is input into the thermal fault model. The thermal fault model uses the feature matrix F to obtain the possibility and trend of all thermal fault types, and the weighted root mean square error (RMSE) of the real thermal fault type and the predicted thermal fault type is used to evaluate the quality of the prediction. The gradient descent algorithm is used to update the model parameters.

[0116] Example 2

[0117] See also Figure 5 As shown, an embodiment of the present invention provides a thermal fault identification device for a photovoltaic inverter, comprising: at least one processing unit, the processing unit is connected to a storage unit and a collection unit through a bus unit, the collection unit collects discrete infrared images of the photovoltaic inverter And the corresponding RGB image The storage unit is a computer-readable storage medium that can be used to store software programs, computer executable programs, and modules, such as the software programs, computer executable programs, and modules corresponding to the thermal fault identification method of a photovoltaic inverter in an embodiment of the present invention. The processing unit implements the thermal fault identification method of a photovoltaic inverter by running the software programs, computer executable programs, and modules stored in the storage unit, including:

[0118] Collecting discrete infrared images of photovoltaic inverters And the corresponding RGB image ;

[0119] Discrete infrared image The modal function with the minimum sum of the estimated bandwidths of M two-dimensional analytical signals is obtained by two-dimensional variational mode decomposition. ;

[0120] For discrete infrared images in continuous time periods The modal function And the corresponding RGB image , obtain the fluctuation of the modal function and RGB image; The modal function , corresponding to discrete infrared image The RGB images and the fluctuations of the two are alternately interspersed to form the feature matrix F;

[0121] The obtained feature matrix F is fed into the pre-trained thermal fault model to obtain the thermal fault categories and trends of the PV inverter.

[0122] Of course, the computer program stored in the storage unit of the device for implementing the thermal fault identification method of a photovoltaic inverter provided by an embodiment of the present invention is not limited to the method operations described above, and can also execute related operations in the thermal fault identification method of a photovoltaic inverter provided by any embodiment of the present invention.

[0123] Example 3

[0124] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. When the computer program is executed, the method for identifying thermal faults of a photovoltaic inverter is implemented, including:

[0125] Collecting discrete infrared images of photovoltaic inverters And the corresponding RGB image ;

[0126] Discrete infrared image The modal function with the minimum sum of the estimated bandwidths of M two-dimensional analytical signals is obtained by two-dimensional variational mode decomposition. ;

[0127] For discrete infrared images in continuous time periods The modal function And the corresponding RGB image , obtain the fluctuation of the modal function and RGB image; The modal function , corresponding to discrete infrared image The RGB images and the fluctuations of the two are alternately interspersed to form the feature matrix F;

[0128] The obtained feature matrix F is fed into the pre-trained thermal fault model to obtain the thermal fault categories and trends of the PV inverter.

[0129] A computer-readable storage medium provided in an embodiment of the present invention stores a computer program which is not limited to the method operations described above, but can also execute related operations in a thermal fault identification method for a photovoltaic inverter provided in any embodiment of the present invention.

[0130] In the embodiments provided by the present invention, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the structural embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, structures or units, which can be electrical, mechanical or other forms.

[0131] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0132] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0133] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for identifying thermal faults of photovoltaic inverters, characterized in that: include: Collecting discrete infrared images of photovoltaic inverters And the corresponding RGB image ; Discrete infrared image The modal function with the minimum sum of the estimated bandwidths of M two-dimensional analytical signals is decomposed by the two-dimensional variational mode decomposition method. ; For discrete infrared images in continuous time periods The modal function And the corresponding RGB image , obtain the fluctuation of the modal function and RGB image; The modal function , corresponding to discrete infrared image The RGB images and the fluctuations of the two are alternately interspersed to form the feature matrix F; The obtained feature matrix F is transmitted to the pre-trained thermal fault model to obtain the thermal fault category and thermal fault trend of the photovoltaic inverter; Wherein, the thermal fault model includes: 2D convolution layer and layer normalization, the convolution kernel size of the 2D convolution layer is 3, the stride is 1, and the padding is 1. The features extracted by the 2D convolution layer maintain the spatial dimension of the feature matrix F; After the 2D convolution layer and layer normalization, there are multiple groups of depth scaling layers and ConvNeXt blocks that introduce SENet layers. Layer normalization is set after multiple groups of depth scaling layers and ConvNeXt blocks that introduce SENet layers. The depth scaling layer contains layer normalization and several convolution layers. Layer normalization is applied to the channels of the features in sequence, and the following convolution layers gradually expand the channel dimensions. The ConvNeXt block that introduces the SENet layer includes a depth-wise convolution layer that processes the channels of each input feature independently, followed by the SENet layer and layer normalization. The SENet layer selectively emphasizes important channels and reduces irrelevant channels to optimize For channel features, after layer normalization, two point-by-point convolution layers are used: the first point-by-point convolution layer expands the channel dimension of the feature by four times, and the second point-by-point convolution layer restores the channel of the feature to its original size. The two point-by-point convolution layers convert the channel feature representation. A GELU activation function is set between the two point-by-point convolution layers, and a learnable scaling parameter is also used to adjust the output according to the significance of the feature. After layer normalization, the features are mapped to a fully connected layer for classification through linear mapping. A Softmax layer is set after the fully connected layer. The output dimension of the fully connected layer covers all identifiable thermal fault types and their trends to meet the classification and trend analysis of all thermal fault types.

2. The method for identifying thermal faults of photovoltaic inverters according to claim 1, characterized in that: Discrete infrared image The modal function with the minimum sum of the estimated bandwidths of M two-dimensional analytical signals is decomposed by the two-dimensional variational mode decomposition method. ,include: According to the modal function The definition of the two-dimensional analytical signal in the frequency domain and the Fourier transform characteristics obtain the two-dimensional analytical signal and modal function The connection between: Modal Function The two-dimensional analytical signal of is defined in the frequency domain as: ； in, for The frequency domain representation of for The center frequency, for The corresponding instantaneous frequency in the frequency domain is, represents the symbolic function, represents the inner product; According to the Fourier transform characteristics, the two-dimensional analytical signal and modal function are obtained The connection between: ； in, is the Dirac function, is convolution, for The two-dimensional analytical signal of The squared gradient of the Gaussian smoothing offset is used to estimate the bandwidth of the two-dimensional analytical signal. The objective function of minimizing the sum of the bandwidths of the M modal functions is expressed as: ； in, represents the square of the L2 norm, represents the gradient function, represents the weight coefficient; exist Solve the objective function under the constraint of: Introducing quadratic penalty terms and Lagrange multipliers into the objective function yields: ； in, Indicates Discrete infrared images The corresponding Lagrange multiplier is, The ADMM algorithm is used to alternately update the mode function, center frequency and Lagrange multiplier in the frequency domain to find the saddle point that satisfies the objective function.

3. The method for identifying thermal faults of photovoltaic inverters according to claim 2, characterized in that: The ADMM algorithm is used to alternately update the modal function, center frequency, and Lagrange multiplier in the frequency domain to find the saddle point that satisfies the objective function. The process includes: initialization: , , , , and , Indicates the first iteration. Discrete infrared images No. Modal Function The frequency domain expression of is obtained by fast Fourier transform; Indicates the first iteration, corresponding to Discrete infrared images No. The center frequency of the mode function, Indicates the first iteration. Discrete infrared images The corresponding frequency domain expression of the Lagrange multiplier is, and are the convergence threshold and error threshold respectively; No. Discrete infrared images No. The update formula of the frequency domain expression of a modal function is as follows: ； in, , For the At the iteration step Discrete infrared images No. The frequency domain expression of the modal function is For the Discrete infrared images The frequency domain expression of For the At the iteration step Discrete infrared images No. The frequency domain expression of the modal function is , Indicates At the iteration step Discrete infrared images The corresponding frequency domain expression of the Lagrange multiplier is, Indicates When the iteration step is Discrete infrared images No. The center frequency of the modal function; The update formula of the center frequency in the frequency domain is as follows: ； in, For the At the iteration step Discrete infrared images No. Frequency domain expression of the modal functions; The update formula of the Lagrange multiplier frequency domain expression is as follows: ;in, in, To control the parameters of the updated Lagrange multiplier step size; Iterate until the following convergence conditions are met: ， and, 。 4. The method for identifying thermal faults of photovoltaic inverters according to claim 1, characterized in that: No. The fluctuation of a modal function: ; Fluctuation of RGB image: ; in, For the Modal Function The average value during the period, For RGB images The average value during the period, , , is the starting point of the continuous time period, The end point of the continuous time period.

5. The method for identifying thermal faults of photovoltaic inverters according to claim 1, characterized in that: The ConvNeXt block introduced into the SENet layer also integrates residual connections and dropout to promote information flow and alleviate overfitting, respectively.

6. The method for identifying thermal faults of photovoltaic inverters according to claim 1, characterized in that: The SENet layer includes: a global average pooling layer, followed by the first fully connected layer with ReLU activation function and the second fully connected layer with Sigmoid activation function. The SENet layer is able to dynamically focus on the most relevant features in the feature channel by emphasizing significant features and suppressing smaller features, and introducing nonlinearity through ReLU activation function and Sigmoid activation function.

7. A thermal fault identification device for a photovoltaic inverter, characterized in that: include: At least one processing unit, the processing unit is connected to the storage unit and the collection unit through the bus unit, the collection unit collects the discrete infrared image of the photovoltaic inverter And the corresponding RGB image The storage unit stores a computer program, and when the computer program is executed by the processing unit, the thermal fault identification method of the photovoltaic inverter as described in any one of claims 1-6 is implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the thermal fault identification method for a photovoltaic inverter according to any one of claims 1 to 6 is implemented.

Images