A multi-zone photovoltaic array fault detection and localization method

Abstract

This invention relates to the field of photovoltaic array fault detection technology, and more particularly to a method for fault detection and location in multi-block photovoltaic arrays. The method includes S1, establishing a single-block photovoltaic array; S2, feature data acquisition; S3, extracting fault features; S4, establishing a CNN-GRU classification prediction model; and S5, fault detection application. This invention divides the entire photovoltaic array cluster into multiple single-block photovoltaic arrays and equips each single-block photovoltaic array with an independent edge node, enabling the CNN-GRU classification prediction model to work independently for the current photovoltaic array. After training, the CNN-GRU classification prediction model can monitor the fault type of the photovoltaic panels online and achieve accurate fault panel location, improving the operation and maintenance cost of the photovoltaic array cluster and enabling convenient maintenance and monitoring.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic array fault detection technology, and in particular to a method for fault detection and location of multi-block photovoltaic arrays. Background Technology

[0002] Centralized photovoltaic (PV) power plant clusters refer to PV power generation systems built and operated using abundant and relatively stable solar radiation resources such as wasteland and lakes, connected to high-voltage transmission systems to supply distant loads. They are characterized by large installed capacity, neat and standardized layout, and a single type of equipment. As the core component of a PV power generation system, the PV array operates in complex outdoor environments for extended periods, inevitably encountering faults such as short circuits, open circuits, shading, and aging. If these faults are not detected promptly and accurately, they will not only reduce power generation efficiency but may also cause safety hazards such as hot spot effects. Therefore, online fault diagnosis of PV arrays is crucial to ensuring the safe and efficient operation of the power plant. However, due to the large number of PV panels and the complex series-parallel design of PV power plant clusters, if precise positioning is not achieved, maintenance personnel need to inspect the PV panels along the entire series branch, a cumbersome and complex process that reduces work efficiency. Summary of the Invention

[0003] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description and other accompanying drawings.

[0004] The purpose of this invention is to overcome the above-mentioned shortcomings and provide a method for fault detection and location of multi-block photovoltaic arrays. By dividing the entire photovoltaic array cluster into multiple single-block photovoltaic arrays and equipping each single-block photovoltaic array with an independent edge node, the CNN-GRU classification prediction model can work independently for the current photovoltaic array. After training, the CNN-GRU classification prediction model can monitor the fault type of the photovoltaic panel online and achieve accurate location of the fault panel, thereby improving the operation and maintenance cost of the photovoltaic array cluster and realizing convenient maintenance and monitoring. This invention provides a method for fault detection and location in a multi-block photovoltaic array, comprising: S1. Establish a single-block photovoltaic array: A single-block photovoltaic array specifically includes M rows and N columns of photovoltaic panels, where M and N are both greater than 0. The M photovoltaic panels in the same column are connected in series sequentially, and the photovoltaic panels in the N columns are connected in parallel. S2. Feature Data Acquisition: A high-precision voltage sensor is connected in parallel at the output end of each photovoltaic panel to acquire the voltage signal of each panel, establishing an MxN voltage matrix; a current sensor is connected in series in each parallel branch to acquire the current of N branches; an irradiance sensor is installed next to the photovoltaic array to acquire the irradiance G; thermocouples or resistance temperature sensors are attached to the backsheet of a representative photovoltaic panel to obtain the backsheet temperature T; the total voltage of the array is acquired. and total array current The voltage matrix of MxN, N-branch current, irradiance G, backplane temperature T, and total array voltage. and total array current All data were collected synchronously in a time-series manner. S3. Extract fault features: Simulate five states: normal, short circuit, open circuit, shaded, and aging, respectively, and obtain feature data under each state to form a fault feature set; S4. Establish a CNN-GRU classification and prediction model: To construct a CNN-GRU classification and prediction model, firstly, construct a multi-channel input map containing fault feature information such as voltage matrix and branch current, and input it into the CNN model for feature extraction. After outputting one-dimensional features, stitch together irradiance G, backplane temperature T, and total array voltage. and total array current Then, the data is input into the GRU model for classification prediction training. The classification prediction results are calculated through the fully connected layer, and softmax is used for classification. Finally, the output layer outputs the data, thereby training a mature CNN-GRU classification prediction model. S5. Fault Detection Application: Establish a multi-block photovoltaic array cluster, which contains multiple M-row N-column photovoltaic arrays. Deploy edge nodes for each M-row N-column photovoltaic array and deploy a well-trained CNN-GRU classification prediction model in each edge node. The management center receives diagnostic information from the edge nodes of each block to achieve visualized management and control.

[0005] In some embodiments, the specific operational steps for simulating the fault state in step S3 include: Under normal conditions, the photovoltaic elements operate normally, and the voltage of each plate and the branch current are balanced and change synchronously with G and T. When simulating a short-circuit fault, the voltage across the preset fault board is clamped to 0. When simulating an open-circuit fault, disconnect the series path of the current column of the preset fault board; When simulating shading conditions, reduce the irradiance G of the preset fault board; When simulating aging conditions, increase the series resistance of the preset fault board; Simultaneously acquire characteristic data under various conditions such as normal operation, short circuit, open circuit, shade, and aging to form a fault feature set.

[0006] In some embodiments, in step S4, the CNN-GRU classification prediction model specifically includes an input layer, a CNN model layer, a GRU model layer, a fully connected classification layer, and an output layer. For the input layer, a multi-channel input map needs to be constructed from the data containing fault feature information. Channel 1 is an MxN voltage matrix, where each voltage value corresponds one-to-one with the actual position of the photovoltaic panel; Channel 2 is a branch current matrix, which fills all rows of the column with the current of each branch to form a branch current matrix of size MxN. Channel 1 and Channel 2 are combined to form a dual-channel feature map of shape (M, N, 2). After adding the temporal feature t, a spatiotemporal feature map of shape (t, M, N, 2) is formed.

[0007] In some embodiments, a spatiotemporal feature map of shape (t, M, N, 2) is input into a CNN model layer. The CNN model layer specifically includes a first convolutional layer and a second convolutional layer. In the first convolutional layer, the number of filters is 32 and the convolutional kernel is 2x2. In the second convolutional layer, the number of filters is 64 and the convolutional kernel is 2x2. The activation function is ReLU. The feature map output by the convolution of the second convolutional layer is flattened into a one-dimensional vector of length L.

[0008] In some embodiments, for the one-dimensional vector output by the second convolutional layer, a splicing node is added at the back end to concatenate the irradiance G, backplane temperature T, and total array voltage. and total array current After sequentially concatenating the one-dimensional vector, the length becomes L+4. After adding the temporal feature t, a spatial feature vector sequence with shape (t, L+4) is generated.

[0009] In some embodiments, a spatial feature vector sequence of shape (t, L+4) is input to the GRU model layer. The GRU model layer specifically includes a first GRU recurrent network and a second GRU recurrent network. The hidden state ht of the last time step is taken as the representative feature of the entire time window. The second GRU recurrent network is followed by a Dropout layer to prevent overfitting.

[0010] In some embodiments, the output ht of the GRU model layer is fed into the fully connected classification layer. The fully connected classification layer specifically includes a fully connected layer and a classification layer. The number of neurons is equal to the total number of categories. The activation function is softmax. The total number of categories = 1 + number of photovoltaic panels x number of fault types, where 1 represents normal operation. Each category is labeled to indicate the fault location and fault type. The output layer outputs the probability of each category, and the maximum probability is taken as the final diagnosis result.

[0011] In some embodiments, in step S5, after receiving fault data from each block edge node, the management center retrains the CNN-GRU classification prediction model and updates it to each edge node via OTA, forming a cyclical iteration.

[0012] By adopting the above technical solution, the beneficial effects of the present invention are: This invention divides the entire photovoltaic array cluster into multiple single-block photovoltaic arrays and equips each single-block photovoltaic array with an independent edge node, enabling the CNN-GRU classification prediction model to work independently for the current photovoltaic array. After training, the CNN-GRU classification prediction model can monitor the fault type of the photovoltaic panel online and achieve accurate location of the fault panel, thereby reducing the operation and maintenance cost of the photovoltaic array cluster and realizing convenient maintenance and monitoring. The CNN-GRU classification prediction model is set up to monitor the voltage changes of each photovoltaic panel and the current changes in the series branches, and together with the total voltage, total current, irradiance G and backsheet temperature T of the entire single-block photovoltaic array, to achieve comprehensive monitoring and positioning.

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0014] Undoubtedly, such and other objects of the present invention will become more apparent after the following detailed description of the preferred embodiments, which are illustrated in various accompanying drawings and figures.

[0015] To make the above and other objects, features and advantages of the present invention more apparent and understandable, one or more preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0017] In the accompanying drawings, the same parts use the same reference numerals, and the drawings are schematic and not necessarily drawn to actual scale.

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only one or more embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on such drawings without creative effort.

[0019] Figure 1This is a schematic diagram of the overall process of the fault detection and location method in some embodiments of the present invention; Figure 2 This is a connection diagram of a 4x4 photovoltaic array under normal operating conditions in some embodiments of the present invention; Figure 3 This is a schematic diagram of the arrangement of multi-block photovoltaic arrays in some embodiments of the present invention; Figure 4 This is a schematic diagram of the overall process of the CNN-GRU classification prediction model in some embodiments of the present invention; Figure 5 This is a schematic diagram illustrating the process of inputting a dual-channel input image into a CNN model for feature extraction in some embodiments of the present invention; Figure 6 This is a schematic diagram of the internal structure of the GRU model in some embodiments of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0021] Furthermore, in the description of this invention, it should be understood that the terms "center," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0022] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral unit; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. However, specifying a direct connection indicates that the two main bodies are not connected through a transitional structure, but rather formed as a whole through a connecting structure. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0023] In this invention, unless otherwise expressly specified and limited, the first feature "on" or "below" the second feature may be in direct contact with the first and second features, or indirect contact through an intermediate medium. In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0024] Reference Figures 1-3 , Figure 1 This is a schematic diagram of the overall process of the fault detection and location method in some embodiments of the present invention; Figure 2 This is a connection diagram of a 4x4 photovoltaic array under normal operating conditions in some embodiments of the present invention; Figure 3 This is a schematic diagram of the arrangement of multi-block photovoltaic arrays in some embodiments of the present invention.

[0025] According to some embodiments of the present invention, the present invention provides a method for fault detection and location of multi-block photovoltaic arrays, including: S1. Establish a single-block photovoltaic array: A single-block photovoltaic array specifically includes M rows and N columns of photovoltaic panels, where M and N are both greater than 0. The M photovoltaic panels in the same column are connected in series sequentially, and the photovoltaic panels in the N columns are connected in parallel. S2, Feature Data Acquisition: such as Figure 3 As shown, a high-precision voltage sensor is connected in parallel at the output terminal of each photovoltaic panel to collect the voltage signal of each photovoltaic panel, establishing an MxN voltage matrix; a current sensor is connected in series in each parallel branch to collect the current of N branches; an irradiance sensor is installed next to the photovoltaic array to collect the irradiance G; thermocouples or resistance temperature sensors are attached to the backplate of a representative photovoltaic panel to obtain the backplate temperature T; and the total voltage of the array is collected. and total array current The voltage matrix of MxN, N-branch current, irradiance G, backplane temperature T, and total array voltage. and total array current All data are collected synchronously in a time-series manner. It is understandable that irradiance sensors and thermocouples for acquiring backsheet temperature are common configurations for photovoltaic power plant-level monitoring. Edge nodes can directly connect to obtain data without the need for additional components for environmental monitoring. S3. Extract fault features: Simulate five states: normal, short circuit, open circuit, shaded, and aging, respectively, and obtain feature data under each state to form a fault feature set; Create a 4x4 photovoltaic array model in MATLAB, such as... Figure 2 As shown; Under normal conditions, the photovoltaic elements operate normally, and the voltage of each plate and the branch current are balanced and change synchronously with G and T. When simulating a short-circuit fault, the voltage across the preset fault board is clamped to 0; the voltage of the short-circuit board is 0, the current in its branch increases slightly, and the voltage of the adjacent board increases slightly. When simulating an open-circuit fault, the series path of the current column of the preset fault board is disconnected; after the series path of the current column of the preset fault board is disconnected, the current is forced to flow through its bypass diode, the current in the branch where the open-circuit board is located drops significantly, and the voltage of other boards in the same series rises abnormally. When simulating shading conditions, reduce the irradiance G of the preset fault board; reduce the irradiance G to 30%-70% of the original data. The voltage of the shading board is significantly lower than that of other boards in the same string. Under severe shading, it may drop to the conduction of the bypass diode, resulting in an imbalance of current between branches. When simulating aging conditions, the series resistance of the preset fault board is increased; under the same G and T conditions, the voltage and output power of the aging board are lower than those of the healthy board in the same series for a long time. The increased internal resistance causes the voltage of the board to gradually decline, but it will not drop to 0 or bypass conduction. The branch current is slightly reduced, and the change is slow and continuous. Simultaneously acquire characteristic data under various conditions such as normal operation, short circuit, open circuit, shade, and aging to form a fault feature set.

[0026] Reference Figures 4-6 , Figure 4 This is a schematic diagram of the overall process of the CNN-GRU classification prediction model in some embodiments of the present invention; Figure 5 This is a schematic diagram illustrating the process of inputting a dual-channel input image into a CNN model for feature extraction in some embodiments of the present invention; Figure 6 This is a schematic diagram of the internal structure of the GRU model in some embodiments of the present invention.

[0027] S4. Establish a CNN-GRU classification and prediction model: To construct a CNN-GRU classification and prediction model, firstly, construct a multi-channel input map containing fault feature information such as voltage matrix and branch current, and input it into the CNN model for feature extraction. After outputting one-dimensional features, stitch together irradiance G, backplane temperature T, and total array voltage. and total array current The data is then input into the GRU model for classification and prediction training. The classification and prediction results are calculated through fully connected layers, classified using softmax, and finally output by the output layer, thus training a mature CNN-GRU classification and prediction model. The CNN-GRU classification and prediction model combines the feature extraction capabilities of CNN with the classification and prediction capabilities of GRU, and can effectively diagnose fault feature information. The CNN-GRU classification prediction model specifically includes an input layer, a CNN model layer, a GRU model layer, a fully connected classification layer, and an output layer. For the input layer, a multi-channel input image needs to be constructed from the data containing fault feature information. Channel 1 is an MxN voltage matrix, where each voltage value corresponds one-to-one with the actual position of the photovoltaic panel; Channel 2 is a branch current matrix, which fills all rows of the column with the current of each branch to form a branch current matrix of size MxN. Taking a 4x4 photovoltaic array as an example, channel 1 corresponds to the actual position of the photovoltaic panel and generates a 4x4 voltage matrix. The current matrix of channel 2 is also 4x4. For the current data in the first column, the current data obtained by branch 1 is filled into all 4 rows. For the current data in the second column, the current data obtained by branch 2 is filled into all 4 rows. And so on, until a complete 4x4 current matrix is ​​formed. In this way, each column of the matrix carries the current information of that branch. The purpose of constructing a multi-channel input image is to improve the spatial resolution of CNN and achieve accurate localization. In the voltage matrix, the anomaly of each pixel (photovoltaic panel) will form a unique spatial pattern. For example, a short circuit in the first photovoltaic panel will cause the pixel value in the upper left corner to be extremely low. The convolution kernel of CNN can capture these local features well and can accurately distinguish even very small sizes. like Figure 5 As shown, channel 1 and channel 2 are combined to form a dual-channel feature map of shape (M, N, 2). After adding the temporal feature t, a spatiotemporal feature map of shape (t, M, N, 2) is formed. A spatiotemporal feature map of shape (t, M, N, 2) is input into a CNN model layer. The CNN model layer specifically includes a first convolutional layer and a second convolutional layer. In the first convolutional layer, the number of filters is 32 and the convolutional kernel is 2x2. In the second convolutional layer, the number of filters is 64 and the convolutional kernel is 2x2. The activation function is ReLU. The feature map output by the second convolutional layer is flattened into a one-dimensional vector of length L. Conventional CNN models follow convolutional layers with pooling layers. However, since the input size of a 4x4 single-block photovoltaic array is very small, in order to retain the position information of the photovoltaic panels in the later output, the traditional pooling layer is removed. Instead, the feature map output by the second convolutional layer is directly flattened into a one-dimensional vector of length L, which is convenient for subsequent GRU processing. For the one-dimensional vector output by the second convolutional layer, a splicing node is added at the back end to concatenate the irradiance G, backplane temperature T, and total array voltage. and total array current After sequentially concatenating the one-dimensional vector, the length becomes L+4. After adding the temporal feature t, a spatial feature vector sequence with shape (t, L+4) is generated. The spatial feature vector sequence of shape (t, L+4) is input into the GRU model layer. The GRU model layer specifically includes a first GRU recurrent network and a second GRU recurrent network. The hidden state ht of the last time step is taken as the representative feature of the entire time window. The second GRU recurrent network is followed by a Dropout layer to prevent overfitting. GRU is a type of recurrent neural network that can solve the gradient problems of CNN, such as the inability to retain information for a long time and the inability to backpropagate. Relying solely on a single frame of voltage graph input to a CNN model can easily lead to misjudgments. Therefore, GRU integrates features from t consecutive time intervals, enabling the model to learn the feature changes of the photovoltaic panel over time, thus achieving both accurate localization and a significant reduction in the false alarm rate. The output ht of the GRU model layer is fed into the fully connected classification layer. The fully connected classification layer includes a fully connected layer and a classification layer. The number of neurons is equal to the total number of categories. The activation function is softmax. The total number of categories = 1 + number of photovoltaic panels x number of fault types, where 1 represents normal operation. Each category is labeled to indicate the fault location and fault type. The output layer outputs the probability of each category. The maximum probability is the final diagnosis result. Taking a 4x4 photovoltaic array as an example, the total number of categories = 1 + number of photovoltaic panels (16) x number of fault types (4) = 65. If panel 5 has a shading fault, it will display "Plate 5, shading". S5. Fault Detection Application: Establish a multi-block photovoltaic array cluster, which contains multiple M-row N-column photovoltaic arrays. Deploy edge nodes for each M-row N-column photovoltaic array and deploy a well-trained CNN-GRU classification prediction model in each edge node. The management center receives diagnostic information from the edge nodes of each block to achieve visualized control. After receiving fault data from each block edge node, the management center retrains the CNN-GRU classification prediction model and updates it to each edge node via OTA, forming a cyclical iteration. By dividing the entire photovoltaic array cluster into multiple single-block photovoltaic arrays and equipping each single-block photovoltaic array with an independent edge node, the CNN-GRU classification prediction model can work independently for the current photovoltaic array. After training, the CNN-GRU classification prediction model can monitor the fault type of the photovoltaic panel online and achieve accurate location of the fault panel, thereby reducing the operation and maintenance cost of the photovoltaic array cluster and enabling convenient maintenance and monitoring. The management center receives two types of data from the edge nodes of each block. One type is diagnostic event snapshots, which automatically upload a time sequence window before and after the fault occurs when the model diagnoses a fault, as well as the fault type and location output by the model. The other type is fuzzy statistics, in which the model periodically uploads 1-2 short time-series segments of normal operation to reflect normal working conditions under different environmental conditions. The uploaded data is cleaned and deduplicated. Staff members then assess the deduplicated data, manually marking the fault location and type. Minor scaling, translation, and Gaussian noise are applied to the data to simulate sensor errors, creating an incremental training set that is both confirmed by on-site faults and manual marking. The incremental and original training sets are then re-input into the CNN-GRU classification prediction model at a 3:7 ratio. The model is retrained, and the weights of each component are adjusted to form an updated CNN-GRU classification prediction model. This updated model is then deployed to each edge node via OTA (Over-The-Air) updates for iterative updates.

[0028] It should be understood that the embodiments disclosed herein are not limited to the specific processing steps or materials disclosed herein, but should be extended to equivalent substitutions of such features as understood by those skilled in the art. It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting.

[0029] The term "embodiment" in this specification refers to a specific feature or characteristic described in connection with an embodiment that is included in at least one embodiment of the invention. Therefore, phrases or "embodiments" appearing in various places throughout the specification do not necessarily refer to the same embodiment.

[0030] Furthermore, the described features or characteristics can be incorporated into one or more embodiments in any other suitable manner. In the above description, specific details, such as thickness, quantity, etc., are provided to provide a comprehensive understanding of embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented without the aforementioned specific details or may be implemented using other methods, components, materials, etc.

Claims

1. A method for fault detection and location in a multi-block photovoltaic array, characterized in that, include S1. Establish a single-block photovoltaic array: A single-block photovoltaic array specifically includes M rows and N columns of photovoltaic panels, where M and N are both greater than 0. The M photovoltaic panels in the same column are connected in series sequentially, and the photovoltaic panels in the N columns are connected in parallel. S2. Feature Data Acquisition: A high-precision voltage sensor is connected in parallel at the output end of each photovoltaic panel to acquire the voltage signal of each panel, establishing an MxN voltage matrix; a current sensor is connected in series in each parallel branch to acquire the current of N branches; an irradiance sensor is installed next to the photovoltaic array to acquire the irradiance G; thermocouples or resistance temperature sensors are attached to the backsheet of a representative photovoltaic panel to obtain the backsheet temperature T; the total voltage of the array is acquired. and total array current The voltage matrix of MxN, N-way branch current, irradiance G, backplane temperature T, and total array voltage. and total array current All data were collected synchronously in a time-series manner. S3. Extract fault features: Simulate five states: normal, short circuit, open circuit, shaded, and aging, respectively, and obtain feature data under each state to form a fault feature set; S4. Establish a CNN-GRU classification and prediction model: To construct a CNN-GRU classification and prediction model, firstly, construct a multi-channel input map containing fault feature information such as voltage matrix and branch current, and input it into the CNN model for feature extraction. After outputting one-dimensional features, stitch together irradiance G, backplane temperature T, and total array voltage. and total array current Then, the data is input into the GRU model for classification prediction training. The classification prediction results are calculated through the fully connected layer, and softmax is used for classification. Finally, the output layer outputs the data, thereby training a mature CNN-GRU classification prediction model. S5. Fault Detection Application: Establish a multi-block photovoltaic array cluster, which contains multiple M-row N-column photovoltaic arrays. Deploy edge nodes for each M-row N-column photovoltaic array and deploy a well-trained CNN-GRU classification prediction model in each edge node. The management center receives diagnostic information from the edge nodes of each block to achieve visualized management and control.

2. The method for fault detection and location of multi-block photovoltaic arrays according to claim 1, characterized in that, The specific operational steps for simulating fault states in step S3 include: Under normal conditions, the photovoltaic elements operate normally, and the voltage of each plate and the branch current are balanced and change synchronously with G and T. When simulating a short-circuit fault, the voltage across the preset fault board is clamped to 0. When simulating an open-circuit fault, disconnect the series path of the current column of the preset fault board; When simulating shading conditions, reduce the irradiance G of the preset fault board; When simulating aging conditions, increase the series resistance of the preset fault board; Simultaneously acquire characteristic data under various conditions such as normal operation, short circuit, open circuit, shade, and aging to form a fault feature set.

3. The method for fault detection and location of multi-block photovoltaic arrays according to claim 1, characterized in that, In step S4, the CNN-GRU classification prediction model specifically includes an input layer, a CNN model layer, a GRU model layer, a fully connected classification layer, and an output layer. For the input layer, a multi-channel input map needs to be constructed from the data containing fault feature information. Channel 1 is an MxN voltage matrix, where each voltage value corresponds one-to-one with the actual position of the photovoltaic panel; Channel 2 is a branch current matrix, which fills all rows of the column with the current of each branch to form a branch current matrix of size MxN. Channel 1 and Channel 2 are combined to form a dual-channel feature map of shape (M, N, 2). After adding the temporal feature t, a spatiotemporal feature map of shape (t, M, N, 2) is formed.

4. The method for fault detection and location of multi-block photovoltaic arrays according to claim 3, characterized in that, A spatiotemporal feature map of shape (t, M, N, 2) is input into the CNN model layer. The CNN model layer specifically includes a first convolutional layer and a second convolutional layer. In the first convolutional layer, the number of filters is 32 and the convolutional kernel is 2x2. In the second convolutional layer, the number of filters is 64 and the convolutional kernel is 2x2. The activation function is ReLU. The feature map output by the second convolutional layer is flattened into a one-dimensional vector of length L.

5. The method for fault detection and location of multi-block photovoltaic arrays according to claim 4, characterized in that, For the one-dimensional vector output by the second convolutional layer, a splicing node is added at the back end to concatenate the irradiance G, backplane temperature T, and total array voltage. and total array current After sequentially concatenating the one-dimensional vector, the length becomes L+4. After adding the temporal feature t, a spatial feature vector sequence with shape (t, L+4) is generated.

6. The method for fault detection and location of multi-block photovoltaic arrays according to claim 5, characterized in that, The spatial feature vector sequence of shape (t, L+4) is input into the GRU model layer. The GRU model layer specifically includes a first GRU recurrent network and a second GRU recurrent network. The hidden state ht of the last time step is taken as the representative feature of the entire time window. The second GRU recurrent network is followed by a Dropout layer to prevent overfitting.

7. The method for fault detection and location of multi-block photovoltaic arrays according to claim 6, characterized in that, The output ht of the GRU model layer is fed into the fully connected classification layer. The fully connected classification layer specifically includes a fully connected layer and a classification layer. The number of neurons is equal to the total number of categories. The activation function is softmax. The total number of categories = 1 + number of photovoltaic panels x number of fault types, where 1 represents normal operation. Each category is labeled to indicate the fault location and fault type. The output layer outputs the probability of each category. The highest probability is taken as the final diagnosis result.

8. The method for fault detection and location of multi-block photovoltaic arrays according to claim 1, characterized in that, In step S5, after receiving fault data from each block edge node, the management center retrains the CNN-GRU classification prediction model and updates it to each edge node via OTA, forming a cyclical iteration.

Images