Photovoltaic array fault diagnosis method combined with photovoltaic power prediction and application thereof

Abstract

The invention discloses a photovoltaic array fault diagnosis method combined with photovoltaic power prediction and application thereof. The method comprises the steps of inputting current meteorological variable parameters into a VMD-Crossform model to obtain an ultra-short-term photovoltaic power prediction value; calculating the deviation degree between the predicted value and the real value, judging whether abnormity occurs or not through a threshold value, if yes, entering the next step, and otherwise, returning to the previous step; and obtaining photovoltaic array operation parameters at the abnormal working point, and performing fault type diagnosis by using the CNN-SVM model to obtain a diagnosis result. According to the method, the VMD-Crossform model is used for predicting the photovoltaic power value, an anomaly detection link is introduced before a fault diagnosis link, the data processing amount of a fault diagnosis algorithm is reduced, the resource utilization rate is improved, the computer resource demand is reduced, finally, the CNN-SVM model is used for fault diagnosis, the diagnosis precision is high, safe and stable operation of a photovoltaic array can be guaranteed, and the method is suitable for large-scale popularization and application. The application prospect is good.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic power generation technology, and relates to a photovoltaic array fault diagnosis method combined with photovoltaic power prediction and its application, and in particular to a photovoltaic array fault diagnosis method, device and storage medium combined with photovoltaic power prediction. Background Art

[0002] Photovoltaic arrays are in outdoor environments for a long time and are greatly disturbed by external meteorological and biological factors, making them prone to failure. When a failure occurs, the output power of the photovoltaic array will be reduced, and it may cause damage to the photovoltaic components, or even cause a fire, resulting in large-scale economic losses.

[0003] The fault diagnosis methods of photovoltaic arrays can currently be divided into three categories: traditional fault diagnosis methods, model-based diagnosis methods, and intelligent diagnosis methods.

[0004] Traditional fault diagnosis methods are mainly infrared image detection and multi-sensor methods. The infrared image detection method generally uses drones or other equipment to obtain infrared images of photovoltaic arrays. The temperature anomaly of the panel when a fault occurs is detected by artificial or intelligent algorithms to achieve fault diagnosis of photovoltaic components. Although the infrared image detection method can achieve online and efficient detection, this type of method often requires drones or other camera equipment to obtain infrared images of photovoltaic arrays, which is costly and difficult to find faults that do not cause temperature changes. Therefore, this type of method has certain limitations. The multi-sensor method uses sensors installed inside the photovoltaic array to obtain operating data to diagnose and locate faults. Although this method adds a mechanism of block positioning and layer-by-layer detection, it still requires a large number of sensors for large photovoltaic systems, making it difficult to maintain a low operating cost. In addition, traditional fault diagnosis methods also include ground capacitance detection and time domain reflection analysis. The ground capacitance detection method uses the ground capacitance of photovoltaic components as the basis for fault diagnosis. The time domain reflection law injects a pulse signal into the photovoltaic circuit, calculates and analyzes the delay and shape of the signal when it returns, and then realizes fault detection of the photovoltaic array. These two methods require the suspension of the operation of the entire photovoltaic array when used, so it is difficult to promote them on a large scale in practical applications.

[0005] Model-based diagnostic methods are based on mathematical formulas and build equivalent models for the target system to obtain the theoretical output value of the photovoltaic array. The comparison between the theoretical value and the actual output data of the photovoltaic system is used as the basis for fault diagnosis. However, this type of method is difficult to consider all influencing factors in practical applications, so the error of the diagnostic results is often not ideal.

[0006] With the development of computer technology, intelligent diagnosis methods have become the main direction of current research. Intelligent diagnosis methods use data processing technology or artificial intelligence, such as random forest algorithm (RF), support vector machine, extreme learning machine and other algorithms to analyze and process data, establish a relationship model between input data and output data to diagnose specific fault types. Compared with other methods, intelligent diagnosis algorithms have the advantages of low cost, simple usage and high accuracy. However, the time when photovoltaic arrays fail is much shorter than the time when they operate normally, which not only leads to a waste of computer resources, but also affects the quality of the fault data set.

[0007] Therefore, it is of great practical significance to develop a photovoltaic array fault diagnosis method with small data processing volume, small computer resource requirements and high fault diagnosis accuracy. Summary of the invention

[0008] Due to the above-mentioned defects in the prior art, the present invention provides a photovoltaic array fault diagnosis method with small data processing volume, small computer resource requirements and high fault diagnosis accuracy. Specifically, it is a photovoltaic array fault diagnosis method combined with photovoltaic power prediction with small data processing volume, small computer resource requirements and high fault diagnosis accuracy, which overcomes the defects of low accuracy and low efficiency of current photovoltaic array fault diagnosis.

[0009] In order to achieve the above object, the present invention provides the following technical solutions:

[0010] A photovoltaic array fault diagnosis method combined with photovoltaic power prediction includes the following steps:

[0011] (1) Input the current meteorological variable parameters into the VMD-Crossformer model to obtain the ultra-short-term photovoltaic power prediction value;

[0012] (2) Calculate the deviation between the predicted value and the true value, and determine whether an abnormality occurs through a threshold. If so, proceed to the next step, otherwise return to step (1). No fault diagnosis is required for normal working points. When an abnormal working point occurs, obtain the corresponding operating data and send it to the fault diagnosis process for specific fault type diagnosis, thereby ensuring the accuracy of abnormal working point identification while reducing the normal working point data that the fault diagnosis algorithm needs to process;

[0013] (3) Obtain the operating parameters of the PV array at the abnormal working point, use the CNN-SVM model to diagnose the specific fault type, and obtain the PV array fault diagnosis results.

[0014] The photovoltaic array fault diagnosis method combined with photovoltaic power prediction of the present invention predicts the photovoltaic power value with the VMD-Crossformer model, introduces an abnormality detection link (based on the comparison between the predicted photovoltaic power value and the threshold value) before the fault diagnosis link, reduces the data processing amount of the fault diagnosis algorithm, improves resource utilization, and reduces computer resource requirements. Finally, the CNN-SVM model is used for fault diagnosis, which has high diagnostic accuracy, can achieve the purpose of ensuring the safe and stable operation of the photovoltaic array, and has good application prospects.

[0015] As the preferred technical solution:

[0016] In the photovoltaic array fault diagnosis method combined with photovoltaic power prediction as described above, the meteorological variable parameters include temperature, relative humidity and wind speed.

[0017] As described above, a photovoltaic array fault diagnosis method combined with photovoltaic power prediction is used. The training process of the VMD-Crossformer model is a process of continuously adjusting the model parameters using the meteorological variable parameters in the training data set I as input and the photovoltaic power corresponding to the meteorological variable parameters as theoretical output. The photovoltaic power corresponding to the meteorological variable parameters of the training data in the training data set I is known.

[0018] In the photovoltaic array fault diagnosis method combined with photovoltaic power prediction as described above, during the training process of the VMD-Crossformer model, the trained VMD-Crossformer model is verified using the verification data set I.

[0019] The acquisition process of the VMD-Crossformer model is as follows:

[0020] 1) Collect raw data (as training dataset I and validation dataset I), including PV power and meteorological variables (temperature, relative humidity, wind speed, etc.). Preprocess the raw data, and then use the Pearson correlation coefficient to measure the correlation between PV power and meteorological variables.

[0021] 2) Use VMD to decompose the PV power time series into multiple IMFs and residuals. The decomposed data are recombined with meteorological variables as the input of the Crossformer model.

[0022] 3) Model training optimization. Since photovoltaic power, temperature and relative humidity have different dimensions and units, direct input will affect the prediction results, so the data is normalized. The normalized data is input into the Crossformer model, and the model is judged to be optimal based on the prediction effect of the validation set.

[0023] 4) Model testing: input the test set into the optimal model to obtain the prediction sequence of each IMF and residual, and obtain the final photovoltaic power prediction sequence through reorganization calculation.

[0024] ,As described above, a photovoltaic array fault diagnosis method combined with photovoltaic power prediction, the calculation formula of the deviation is as follows:

[0025]

[0026] Among them, x true is the true value, x pred is the predicted value;

[0027] The threshold is determined by selecting data of the photovoltaic array under normal working conditions, calculating its deviation and using the Laida law.

[0028] The deviation between the true value and the predicted value is used as the judgment standard, and the Laida rule is used as the basis for setting the threshold. The judgment standard is

[0029]

[0030] Where μ is the mean of the judgment standard and σ is its standard deviation.

[0031] In the photovoltaic array fault diagnosis method combined with photovoltaic power prediction as described above, the photovoltaic array operating parameters include the maximum power point power, current, voltage, open circuit voltage, short circuit current, irradiance and temperature output by the photovoltaic array.

[0032] In the photovoltaic array fault diagnosis method combined with photovoltaic power prediction as described above, the CNN-SVM model extracts feature information from input data using CNN, inputs the feature information from the input data into SVM, and obtains the fault type corresponding to the input data;

[0033] The training process of the CNN-SVM model is a process of continuously adjusting the model parameters by taking the photovoltaic array operating parameters in the training data set II as input and the fault types corresponding to the photovoltaic array operating parameters as theoretical output. The fault types corresponding to the photovoltaic array operating parameters of the training data in the training data set II are known.

[0034] The operation process of the CNN-SVM model is as follows:

[0035] 1) Data acquisition and preprocessing, obtaining the output parameters of the photovoltaic array at the time when the abnormal working point is determined in step (2).

[0036] 2) Data feature extraction: input the original fault feature data into CNN, and extract the feature information in the data through CNN.

[0037] 3) Fault type diagnosis: The feature information extracted by CNN is input into SVM for diagnosis to obtain the specific fault type of the working point corresponding to the data.

[0038] The present invention also provides a computer device, the computer device comprising:

[0039] at least one processor; and,

[0040] a memory communicatively connected to the at least one processor; wherein,

[0041] The memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the photovoltaic array fault diagnosis method combined with photovoltaic power prediction as described above is implemented.

[0042] In addition, the present invention also provides a computer-readable storage medium, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by a processor, the photovoltaic array fault diagnosis method combined with photovoltaic power prediction as described above is implemented. The photovoltaic array fault diagnosis method combined with photovoltaic power prediction of the present invention can be run on a computer or similar device in the form of software.

[0043] The above technical solution is only a feasible technical solution of the present invention, and the protection scope of the present invention is not limited thereto. Those skilled in the art can reasonably adjust the specific design according to actual needs.

[0044] The above invention has the following advantages or beneficial effects:

[0045] (1) The photovoltaic array fault diagnosis method combined with photovoltaic power prediction of the present invention can effectively predict the photovoltaic power value using the VMD-Crossformer model. VMD reduces the impact of the nonlinearity and non-stationarity of photovoltaic power data on the prediction accuracy. Crossformer can simultaneously extract the time and dimension information of the photovoltaic power time series to achieve accurate prediction of photovoltaic power.

[0046] (2) In the photovoltaic array fault diagnosis method combined with photovoltaic power prediction of the present invention, the operation time of the photovoltaic array in a normal state is much longer than that in a fault state, so in most of the actual time, the general fault diagnosis algorithm is performing useless work. The present invention adds an abnormality detection link before the fault diagnosis link, reduces the data processing amount of the fault diagnosis algorithm, and improves resource utilization;

[0047] (3) The photovoltaic array fault diagnosis method combined with photovoltaic power prediction of the present invention, the CNN-SVM model can perform high-precision diagnosis of different types of fault conditions, thereby achieving the purpose of ensuring the safe and stable operation of the photovoltaic array, and has good application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention and its features, appearance and advantages will become more apparent by reading the detailed description of non-limiting embodiments with reference to the following drawings. The same reference numerals indicate the same parts in all the drawings. The drawings are not drawn to scale, and emphasis is placed on illustrating the subject matter of the present invention.

[0049] Figure 1 A step diagram of a photovoltaic array fault diagnosis method combined with photovoltaic power prediction according to the present invention;

[0050] Figure 2 Schematic diagram of the acquisition of VMD-Crossformer photovoltaic power prediction model;

[0051] Figure 3 Schematic diagram of the anomaly detection process;

[0052] Figure 4 It is a schematic diagram of the CNN-SVM fault diagnosis process;

[0053] Figure 5 This is the anomaly detection result diagram;

[0054] Figure 6 This is the fault diagnosis result diagram. DETAILED DESCRIPTION

[0055] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0056] Example 1

[0057] A photovoltaic array fault diagnosis method combined with photovoltaic power prediction includes the following steps (such as Figure 1 shown):

[0058] (1) Input the current meteorological variable parameters (including temperature, relative humidity and wind speed) into the VMD-Crossformer model to obtain the ultra-short-term photovoltaic power prediction value;

[0059] The training process of the VMD-Crossformer model is a process of continuously adjusting the model parameters by taking the meteorological variable parameters in the training data set I as input and the photovoltaic power corresponding to the meteorological variable parameters as the theoretical output. The photovoltaic power corresponding to the meteorological variable parameters of the training data in the training data set I is known. During the training process of the VMD-Crossformer model, the test data set I is used to verify the trained VMD-Crossformer model.

[0060] The acquisition process of the VMD-Crossformer model is as follows (e.g. Figure 2 shown):

[0061] 1.1) Collect raw data (as training data set I and test data set I), including PV power and meteorological variables (temperature, relative humidity, wind speed, etc.), preprocess the raw data, and then use the Pearson correlation coefficient to measure the correlation between PV power and meteorological variables;

[0062] 1.2) Use VMD to decompose the PV power time series into multiple IMFs and residuals, and recombine the decomposed data with meteorological variables as the input of the Crossformer model;

[0063] 1.3) Model training optimization. Since photovoltaic power, temperature and relative humidity have different dimensions and units, direct input will affect the prediction results. Therefore, the data is normalized and input into the Crossformer model. The model is judged to be optimal based on the prediction effect of the validation set.

[0064] 1.4) Model testing: input the test set into the optimal model to obtain the prediction sequence of each IMF and residual, and obtain the final photovoltaic power prediction sequence through reorganization calculation;

[0065] (2) Calculate the deviation between the predicted value and the true value, and determine whether an abnormality occurs by using a threshold. If so, proceed to the next step, otherwise return to step (1). The process of steps (1) to (2) is as follows: Figure 3 As shown; the calculation formula of deviation is as follows:

[0066]

[0067] Among them, x true is the true value, x pred is the predicted value;

[0068] The threshold is determined by selecting the data of the photovoltaic array under normal working conditions and calculating its deviation and using the Laida law;

[0069] The deviation between the true value and the predicted value is used as the judgment standard, and the Laida rule is used as the basis for setting the threshold. The judgment standard is

[0070]

[0071] In the formula, μ is the mean of the judgment standard; σ is its standard deviation;

[0072] (3) Obtain the operating parameters of the photovoltaic array at the abnormal working point (including the maximum power point power, current, voltage, open circuit voltage, short circuit current, irradiance and temperature output by the photovoltaic array), use the CNN-SVM model to diagnose the specific fault type, and obtain the photovoltaic array fault diagnosis results;

[0073] The CNN-SVM model uses CNN to extract feature information from the input data, and after inputting the feature information from the input data into the SVM, the fault type corresponding to the input data is obtained;

[0074] The training process of the CNN-SVM model is a process of continuously adjusting the model parameters by taking the photovoltaic array operating parameters in the training data set II as input and the fault types corresponding to the photovoltaic array operating parameters as theoretical output. The fault types corresponding to the photovoltaic array operating parameters of the training data in the training data set II are known.

[0075] The operation process of the CNN-SVM model is as follows (such as Figure 4 shown):

[0076] 1) Data acquisition and preprocessing, obtaining the output parameters of the photovoltaic array at the time of the abnormal working point determined in step (2);

[0077] 2) Data feature extraction: input the original fault feature data into CNN and extract the feature information in the data through CNN;

[0078] 3) Fault type diagnosis: The feature information extracted by CNN is input into SVM for diagnosis to obtain the specific fault type of the working point corresponding to the data.

[0079] The present invention uses VMD algorithm to decompose photovoltaic power sequence, so as to avoid the problems of modal aliasing and pseudo-modal when decomposing photovoltaic power sequence. The decomposition of photovoltaic power includes: establishing and solving the variational model based on photovoltaic power sequence.

[0080] The variational problem in the VMD algorithm can be described as: finding K intrinsic mode functions g k (t), where the sum of the functions is equal to the input signal (PV power sequence) G k (t) under the constraint condition that each g k The sum of the estimated bandwidths of (t) is minimized.

[0081]

[0082] In the formula, {g k}(k=1,2,…,K) is the intrinsic mode function, {ω k} is {g k} corresponds to the center frequency, δ(t) represents the Dirichlet function, Yes k (t) The analytical signal obtained after Hilbert transform.

[0083] The above variational model belongs to a constrained optimization problem, which is transformed into an unconstrained optimization problem by introducing the equilibrium constraint parameter α (quadratic penalty factor) and the Lagrange multiplication operator λ(t).

[0084]

[0085] Alternating update using the multiplication operator alternating direction method until

[0086]

[0087] Among them, τ is the update parameter. Perform inverse Fourier transform, the real part is g k (t). ε represents the convergence stop condition, which is usually 1e-6. The selection of the number of modes K is very important. If K is too small, it will lead to under-decomposition, which will not achieve the purpose of reducing the nonlinearity and non-stationarity of the photovoltaic power series; if K is too large, it will cause over-decomposition, and too many modes will increase the complexity of prediction. The optimal value of K is 7 according to the center frequency of the mode.

[0088] The Crossformer model mainly includes the following key contents:

[0089] Dimension-Segment (DSW) embedding: PV power, temperature, and relative humidity data are divided into segments of length L along each dimension. seg The segments are then embedded into a vector. This method preserves the time and dimensional information of the MTS data, which helps the model capture information better.

[0090]

[0091] In the formula, T is the total length of the sequence, is a d-dimensional seg Then, each segment is embedded into a vector using linear projection plus position embedding. i,d Represent a univariate time series segment:

[0092]

[0093] In the formula, represents the learnable projection matrix, The position (i, d) is a learnable position embedding. After embedding, a two-dimensional vector group is obtained:

[0094]

[0095] Two-Stage Attention (TSA) Layer: Crossformer uses TSA layers to effectively capture dependencies between time and different dimensions. Both dependencies are important for MTS prediction.

[0096] Cross-temporal stage: Directly apply multi-head attention mechanism to each dimension.

[0097]

[0098] In the formula, Z :,d represents all time steps in d dimensions, and Z time is the output of the multi-head self-attention layer (MSA) and the multi-layer feedforward network (MLP). time The cross-temporal dependencies in the same dimension are captured and serve as the input to the cross-dimensional stage.

[0099] Cross-dimensional stage: For each time step i, a fixed number of learnable vectors are set as routers, so that the information is aggregated with a small number of intermediate routers before information transmission, reducing the complexity of attention calculation.

[0100]

[0101] In the formula, represents a set of learnable vectors, Represents the aggregated message of all dimensions, is the output of the router, Z i,: A vector of all dimensions representing time step i. Z dim is the output of MLP, capturing both cross-time and cross-dimensional dependencies.

[0102] Hierarchical Encoder-Decoder (HED) Architecture: The model uses HED to leverage information at different scales for prediction. This hierarchical approach helps understand and predict MTS data more efficiently.

[0103] Encoder: In each layer of the encoder (except the first layer), every two vectors that are adjacent in time domain are merged to get a coarser representation. Then TSA layer is applied to capture dependencies at this scale.

[0104]

[0105] H is the two-dimensional vector group obtained by DSW embedding operation, Z enc,l represents the output of the encoder at layer l, M represents the learnable matrix of segmented merging, [·] represents the connection operation, and L l-1 represents the number of segments in each dimension in the l-1th layer, Represents the array after the l-th layer segments are merged.

[0106] Decoder: Take the output of the encoder as the input of the decoder and output the decoded array of each layer:

[0107]

[0108] E (dec) represents the learnable position embedding of the decoder, is the output of TSA. The MSA layer is As a query, As key and value, establish the connection between encoder and decoder. is the output of MSA, Z dec,l is the output of MLP.

[0109] Apply linear projection to get the prediction results of each layer output of the decoder. Add the results of each layer to get the final prediction result:

[0110]

[0111] W l is a learnable matrix used to project vectors into time series segments. represents the prediction of the ith segment in dimension d, Represents the layer prediction of the lth layer, which is added to get the final prediction result

[0112] The model parameters in step (1) are set as follows: the division ratio of the training dataset I, the validation dataset I, and the test dataset I is 7:1:2; the learning rate is 0.0001; the batch size is 72; and the epoch is 10.

[0113] The internal structure and parameters of the CNN-SVM model are as follows: The CNN part includes convolutional layer, batch normalization layer, activation layer and pooling layer. The convolution kernel size of the convolution layer is 3*1, the sliding step is 1, the number of convolution kernels in the first convolution layer is 16, and the number of convolution kernels in the second convolution layer is 32. The activation layer uses ReLu as the activation function, and the pooling layer uses the maximum pooling strategy. CNN uses the Adam optimizer for parameter optimization, and the initial learning rate is set to 0.01. The nonlinear SVM uses RBF as the kernel function, and the penalty factor C is determined to be 55.897 using the grid search method, and the kernel function parameter g is 0.5.

[0114] The actual operation data of a photovoltaic power station from DKASC is used to verify the effectiveness of the present invention. The data contains the corresponding time, meteorological parameters and output power, and each piece of data is separated by 10 minutes. Due to the scarcity of photovoltaic array fault data sets, the present invention obtains photovoltaic array fault data by processing normal operation data. 10% of the 5000 real photovoltaic power data in the prediction model test data set are randomly selected as abnormal working points where faults occur, and their power is randomly reduced by 10-30% to simulate the impact of different faults on photovoltaic power. The irradiance in the test set is selected to be greater than 300W / m 2 The data with stable weather conditions is used as the source of anomaly detection thresholds. By building multiple 3*3 photovoltaic array simulation models in MATLAB / Simulink, a data set for verifying the fault diagnosis effect is obtained. The models include normal model, partial shading fault model, component degradation fault model, open circuit fault model and short circuit fault model. Each fault model includes two fault states of different degrees. Through simulation, a total of 9,000 data points of 9 operating states are collected, with 1,000 data points for each operating state.

[0115] Through experimental results (such as Figure 5 As shown in Figure 2, it can be proved that the anomaly detection method used in the present invention can reduce the 79.43% of normal working points that need to be processed in the fault diagnosis link while ensuring the average false negative rate is less than 10%, thereby achieving the purpose of improving the efficiency of the fault diagnosis link. The diagnostic accuracy of the CNN-SVM fault diagnosis model used in the present invention is 99.833% (as shown in Figure 2, Figure 6 As shown in the figure, the single SVM and CNN diagnosis results have an accuracy rate of 98.945% and 99.056% respectively. This result verifies the effectiveness and superiority of the fault diagnosis method used in the present invention.

[0116] Example 2

[0117] A computer device, comprising: at least one processor and a memory communicatively connected to the at least one processor;

[0118] The memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the photovoltaic array fault diagnosis method combined with photovoltaic power prediction as described in Example 1 is implemented.

[0119] Example 3

[0120] A computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the photovoltaic array fault diagnosis method combined with photovoltaic power prediction as described in Example 1.

[0121] Those skilled in the art should understand that those skilled in the art can implement variations by combining the prior art and the above embodiments, which will not be described in detail here. Such variations do not affect the essential content of the present invention, and will not be described in detail here.

[0122] The above describes the preferred embodiments of the present invention. It should be understood that the present invention is not limited to the above-mentioned specific embodiments, and the devices and structures that are not described in detail should be understood to be implemented in a common manner in the art; any technician familiar with the art can use the above-disclosed methods and technical contents to make many possible changes and modifications to the technical solutions of the present invention without departing from the scope of the technical solutions of the present invention, or modify them into equivalent embodiments of equivalent changes, which does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention are still within the scope of protection of the technical solutions of the present invention.

Claims

1. A photovoltaic array fault diagnosis method combined with photovoltaic power prediction, characterized in that: The following steps are involved: (1) Input the current meteorological variable parameters into the VMD-Crossformer model to obtain the ultra-short-term photovoltaic power prediction value; (2) Calculate the deviation between the predicted value and the true value, and determine whether an abnormality occurs by using a threshold. If so, proceed to the next step, otherwise return to step (1); (3) Obtain the operating parameters of the PV array at the abnormal working point, use the CNN-SVM model to diagnose the specific fault type, and obtain the PV array fault diagnosis results.

2. A photovoltaic array fault diagnosis method combined with photovoltaic power prediction according to claim 1, characterized in that: The meteorological variable parameters include temperature, relative humidity and wind speed.

3. A photovoltaic array fault diagnosis method combined with photovoltaic power prediction according to claim 1, characterized in that: The training process of the VMD-Crossformer model is a process of continuously adjusting the model parameters by taking the meteorological variable parameters in the training data set I as input and the photovoltaic power corresponding to the meteorological variable parameters as theoretical output. The photovoltaic power corresponding to the meteorological variable parameters of the training data in the training data set I is known.

4. A photovoltaic array fault diagnosis method combined with photovoltaic power prediction according to claim 3, characterized in that: During the training process of the VMD-Crossformer model, the validation dataset I is used to validate the trained VMD-Crossformer model.

5. A photovoltaic array fault diagnosis method combined with photovoltaic power prediction according to claim 1, characterized in that: The calculation formula of the deviation is as follows: Among them, x true is the true value, x pred is the predicted value; The threshold is determined by selecting data of the photovoltaic array under normal working conditions, calculating its deviation and using the Laida law.

6. A photovoltaic array fault diagnosis method combined with photovoltaic power prediction according to claim 1, characterized in that: The photovoltaic array operating parameters include the maximum power point power, current, voltage, open circuit voltage, short circuit current, irradiance and temperature output by the photovoltaic array.

7. A photovoltaic array fault diagnosis method combined with photovoltaic power prediction according to claim 6, characterized in that: The CNN-SVM model extracts feature information from input data using CNN, inputs the feature information from the input data into SVM, and obtains the fault type corresponding to the input data; The training process of the CNN-SVM model is a process of continuously adjusting the model parameters by taking the photovoltaic array operating parameters in the training data set II as input and the fault types corresponding to the photovoltaic array operating parameters as theoretical output. The fault types corresponding to the photovoltaic array operating parameters of the training data in the training data set II are known.

8. A computer device, characterized in that: The computer device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the photovoltaic array fault diagnosis method combined with photovoltaic power prediction according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the photovoltaic array fault diagnosis method combined with photovoltaic power prediction according to any one of claims 1 to 7 is implemented.

Images