Inverter fault diagnosis method and device, equipment and storage medium

Abstract

The invention is applicable to the field of inverters, and discloses an inverter fault diagnosis method, device and equipment and a storage medium, and the method is used for carrying out online detection on real-time data of inverter equipment so as to timely find whether the inverter equipment has a fault or not. The method comprises the following steps: acquiring historical inverter data; using an improved singular spectrum analysis method to decompose and reconstruct the historical inverter data to obtain inverter data after noise reduction; the inverter data obtained after noise reduction is input into an Angle Transform initial model to be trained, and an inverter fault diagnosis model is obtained; and collecting real-time data of the inverter equipment, and inputting the collected real-time data into the inverter fault diagnosis model to obtain a diagnosis result.

Description

Technical Field

[0001] The present invention relates to the field of inverters, and in particular to an inverter fault diagnosis method, device, equipment and storage medium. Background Art

[0002] In recent years, with the widespread application of inverter equipment, the importance of its operation and maintenance management has become increasingly prominent. However, current fault diagnosis methods have encountered many bottlenecks in actual operation, especially the traditional manual inspection combined with thermal imaging technology, which is not only limited by the narrow detection range, but also consumes a lot of manpower and time resources, is inefficient and difficult to fully cover.

[0003] Therefore, it is particularly important to establish an evaluation model that can evaluate the health status of inverter equipment in real time. This model can effectively identify and predict fault problems of inverter equipment and help optimize operation and maintenance management. Summary of the invention

[0004] The present invention provides an inverter fault diagnosis method, device, equipment and storage medium, which are used for online detection of real-time data of the inverter equipment, so as to timely discover whether the inverter equipment has a fault.

[0005] A first aspect of the present invention provides an inverter fault diagnosis method, which includes: acquiring historical inverter data; decomposing and reconstructing the historical inverter data using an improved singular spectrum analysis method to obtain denoised inverter data; inputting the denoised inverter data into an Anomaly Transformer initial model for training to obtain an inverter fault diagnosis model; collecting real-time data of an inverter device, and inputting the collected real-time data into the inverter fault diagnosis model to obtain a diagnosis result.

[0006] Preferably, the historical inverter data includes the inverter's power generation, AC power, DC power, three-phase voltage, three-phase current, power factor and operating temperature status, and grid efficiency.

[0007] Preferably, before the improved singular spectrum analysis method is used to decompose and reconstruct the historical inverter data, the historical inverter data is also cleaned, missing values ​​are processed, and normalized.

[0008] Preferably, the improved singular spectrum analysis method is used to decompose and reconstruct the historical inverter data to obtain the denoised inverter data, including: converting the historical inverter data into a multidimensional trajectory matrix, each row of the multidimensional trajectory matrix represents the inverter data at a time point, and each column represents a feature dimension; using the ant colony optimization algorithm (ACO) to find the optimal window length for each feature dimension of the multidimensional trajectory matrix, and setting the cumulative contribution rate as the objective function; decomposing each feature dimension according to the found optimal window length to obtain a plurality of feature decomposition vectors corresponding to each feature dimension; from the plurality of feature decomposition vectors corresponding to each feature dimension, selecting the first M feature decomposition vectors as reconstruction elements, and using the reconstruction elements corresponding to the feature dimension to reconstruct each feature dimension one by one to obtain the reconstructed data of each feature dimension; integrating the reconstructed data of all feature dimensions to obtain the denoised inverter data.

[0009] Preferably, the denoised inverter data is input into an Anomaly Transformer initial model for training to obtain an inverter fault diagnosis model, including: dividing the denoised inverter data into a training set, a validation set and a test set; constructing an Anomaly Transformer initial model, and inputting the training set and the validation set into the Anomaly Transformer initial model for training and validation to obtain a trained Anomaly Transformer model; inputting the test set into the trained Anomaly Transformer model to obtain an anomaly score corresponding to each data point in the test set; performing statistics on the distribution of anomaly scores corresponding to each data point, and determining an anomaly score threshold; integrating the trained Anomaly Transformer model and the determined anomaly score threshold to obtain an inverter fault diagnosis model.

[0010] Preferably, the Anomaly Transformer initial model includes an anomaly attention module and a feedforward neural network layer, the anomaly attention module and the feedforward neural network layer work together, the anomaly attention module adopts a dual-branch structure, including a main branch and an auxiliary branch, the main branch is used to process time series data, and capture the sequence association in the time series data through a self-attention mechanism, the auxiliary branch is used to calculate the prior association, the feedforward neural network performs dimensionality conversion on the sequence association and prior association output by the anomaly attention module, and uses the ReLU activation function to perform nonlinear conversion on the converted sequence association and prior association to obtain new features, and performs residual connection and layer normalization processing on the new features, and finally outputs the processed features to the anomaly attention module to calculate the anomaly score.

[0011] Preferably, collecting real-time data of the inverter device and inputting the collected real-time data into the inverter fault diagnosis model to obtain a diagnosis result includes:

[0012] Collect real-time data from inverter equipment and perform noise reduction on the collected real-time data;

[0013] The real-time data after noise reduction processing is input into the inverter fault diagnosis model to obtain a diagnosis result, and the diagnosis result includes normal inverter data and abnormal inverter data.

[0014] The second aspect of the present invention provides an inverter fault diagnosis device, including: an acquisition module, used to acquire historical inverter data; a denoising module, used to decompose and reconstruct the historical inverter data using an improved singular spectrum analysis method to obtain denoised inverter data; a training module, used to input the denoised inverter data into an AnomalyTransformer initial model for training to obtain an inverter fault diagnosis model; a diagnosis module, used to collect real-time data of an inverter device, and input the collected real-time data into the inverter fault diagnosis model to obtain a diagnosis result.

[0015] A third aspect of the present invention provides an inverter fault diagnosis device, comprising: a memory and at least one processor, wherein the memory stores computer-readable instructions, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the computer-readable instructions in the memory so that the inverter fault diagnosis device performs the various steps of the inverter fault diagnosis method as described above.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, which, when executed on a computer, enable the computer to execute the various steps of the inverter fault diagnosis method described above.

[0017] In the technical solution provided by the present invention, an improved singular spectrum analysis method is used to decompose and reconstruct the preprocessed inverter data, which can not only effectively remove the noise in the data, but also retain the key features of the data. Moreover, the inverter fault diagnosis model obtained by training the Anomaly Transformer initial model is used to perform fault judgment on the real-time data of the inverter device. The inverter fault diagnosis model combines the advantages of the self-attention mechanism and the anomaly detection algorithm, and can accurately identify abnormal patterns in the inverter data, thereby improving the accuracy and reliability of the diagnosis. In addition, by collecting the real-time data of the inverter device and performing online diagnosis, the inverter fault can be discovered and handled in time to avoid the expansion of the fault and cause a greater impact on the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] 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, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0019] Figure 1 A first flow chart of the inverter fault diagnosis method provided by an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of the structure of an inverter fault diagnosis device provided by an embodiment of the present invention;

[0021] Figure 3 A schematic diagram of the structure of an inverter fault diagnosis device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an inverter fault diagnosis method according to an embodiment of the present invention includes:

[0024] S101, obtaining historical inverter data;

[0025] S102, using an improved singular spectrum analysis method to decompose and reconstruct historical inverter data to obtain inverter data after noise reduction;

[0026] S103, inputting the inverter data after noise reduction into the Anomaly Transformer initial model for training to obtain an inverter fault diagnosis model;

[0027] S104: Collect real-time data of the inverter device, and input the collected real-time data into the inverter fault diagnosis model to obtain a diagnosis result.

[0028] In this embodiment, the diagnosis result includes normal inverter data and abnormal inverter data. When the diagnosis result is that the inverter data is normal, it means that the inverter is in a normal state. When the diagnosis result is that the inverter data is abnormal, it means that the inverter is in an abnormal state, that is, the inverter may have a fault and needs further processing.

[0029] In this embodiment, an improved singular spectrum analysis method is used to decompose and reconstruct the preprocessed inverter data. This method can not only effectively remove noise in the data, but also retain the key features of the data. Moreover, the inverter fault diagnosis model obtained by training the Anomaly Transformer initial model is used to perform fault judgment on the real-time data of the inverter device. The inverter fault diagnosis model combines the advantages of the self-attention mechanism and the anomaly detection algorithm, and can accurately identify abnormal patterns in the inverter data, thereby improving the accuracy and reliability of the diagnosis. In addition, by collecting the real-time data of the inverter device and performing online diagnosis, the inverter fault can be discovered and handled in time to avoid the expansion of the fault and cause a greater impact on the system.

[0030] In this embodiment, in step S101, the historical inverter data includes the inverter's power generation, AC power, DC power, three-phase voltage, three-phase current, power factor, operating temperature status, and grid efficiency.

[0031] In this embodiment, data generated by the inverter device during production or operation and in accordance with its normal operating state is obtained as historical inverter data.

[0032] In this embodiment, before using the improved singular spectrum analysis method to decompose and reconstruct the historical inverter data, the historical inverter data is also preprocessed, specifically including data cleaning, processing missing values ​​and normalizing the data. The preprocessing improves the quality of the data and the reliability of the analysis results.

[0033] Data cleaning can be done by identifying and processing outliers using statistical methods (such as standard deviation and box plots) and deciding whether to delete, replace or retain them. On the other hand, it can be done by checking and deleting duplicate records in the data set to ensure that each piece of data is unique.

[0034] When dealing with missing values, first determine the location and type of missing values ​​in the data set (such as numerical or categorical), and then choose to delete, fill, or interpolate to deal with missing values ​​based on the characteristics of the data and the distribution of missing values. If there are few missing values, you can directly delete the records with missing values. If there are many missing values, you can use the mean, median, mode, or prediction model to fill in the missing values. For time series data, you can use linear interpolation or other time series interpolation methods.

[0035] When normalizing data, first scale the data to a uniform scale, usually using Z-score normalization, and then scale the data to the range of [0,1]. Minimum-maximum normalization can be used.

[0036] In this embodiment, in step S102, an improved singular spectrum analysis method is used to decompose and reconstruct the historical inverter data to obtain the inverter data after noise reduction, including:

[0037] S1021, converting historical inverter data into a multidimensional trajectory matrix, where each row of the multidimensional trajectory matrix represents inverter data at a time point, and each column represents a feature dimension;

[0038] S1022, using the ant colony optimization algorithm (ACO) to find the optimal window length for each characteristic dimension of the multidimensional trajectory matrix, and setting the cumulative contribution rate as the objective function;

[0039] S1023, decomposing each feature dimension according to the found optimal window length to obtain multiple feature decomposition vectors corresponding to each feature dimension;

[0040] S1024. Select the first M feature decomposition vectors from the multiple feature decomposition vectors corresponding to each feature dimension as reconstruction elements, and reconstruct each feature dimension one by one using the reconstruction elements corresponding to the feature dimension to obtain reconstructed data for each feature dimension;

[0041] S1025. Integrate the reconstructed data of all feature dimensions to obtain the inverter data after noise reduction.

[0042] In this embodiment, in step S1021, the historical inverter data is represented by F N =(f1, f2, ..., f N ), when converting the historical inverter data into a multidimensional trajectory matrix, the window length is selected as L, and the multidimensional trajectory matrix is ​​defined to have L rows and K columns, where K = N-L+1. Therefore, the multidimensional trajectory matrix X is expressed as:

[0043]

[0044] In the multidimensional trajectory matrix X, each row represents the inverter data at a time point, and each column represents a feature dimension.

[0045] In this embodiment, step S1022 uses the ant colony optimization algorithm (ACO) to find the optimal window length for each feature dimension, and sets the cumulative contribution rate as the objective function. The specific formula is as follows:

[0046]

[0047] In the formula, CR(k) represents the cumulative contribution rate of the first k eigendecomposition vectors, represents the variance of the i-th singular value; n represents the total number of singular values; k represents the number of the first k singular values ​​considered.

[0048] In this embodiment, the ant colony continuously updates the pheromone through continuous iteration until an optimal window length is found so that the decomposed components reach the set cumulative contribution rate.

[0049] In this embodiment, in step S1023, the decomposition formula is expressed as:

[0050] X=USV T

[0051] Where U represents the left singular vector matrix; V T represents the right singular vector matrix; S is a diagonal matrix.

[0052] In this embodiment, each feature dimension is decomposed according to the decomposition formula to obtain L feature decomposition vectors corresponding to each feature dimension.

[0053] In this embodiment, in step S1024, singular entropy is introduced to help select an appropriate number of eigendecomposition vectors to reconstruct each feature dimension. If the increment of singular entropy is reduced to a stable value at the Mth order, it means that the singular values ​​from 1 to Mth order contain valid information. Based on this, the first M eigendecomposition vectors are selected to reconstruct each feature dimension to obtain the reconstructed data of each feature dimension. The formula of singular entropy increment is expressed as:

[0054]

[0055] Where k represents the order of singular entropy, ΔE k represents the increment of singular entropy at the kth order, λ k represents the kth singular value, It represents the sum of all singular values ​​and represents the total energy of the original sequence.

[0056] In this embodiment, the above decomposition and reconstruction are performed on each characteristic dimension sequence of the multi-dimensional trajectory matrix to achieve the purpose of noise reduction.

[0057] In this embodiment, the improved singular spectrum analysis method combined with the ant colony optimization algorithm is used to decompose and reconstruct the historical inverter data to obtain the inverter data after noise reduction. On the one hand, the ant colony optimization algorithm is used to find the optimal window length for each feature dimension instead of using a fixed window size, which makes the method more flexible and adaptable. Different feature dimensions may have different dynamic characteristics. The adaptive window length can capture these characteristics more accurately, thereby improving the accuracy of decomposition and reconstruction. On the other hand, by selecting the top M feature decomposition vectors with the highest contribution rate in each feature dimension for reconstruction, noise and redundant information can be effectively removed, and the part that contributes most to the change of the original data is retained. This method can significantly improve the signal-to-noise ratio of the data and provide a more reliable basis for subsequent data analysis and fault diagnosis.

[0058] In this embodiment, in step S103, the denoised inverter data is input into the Anomaly Transformer initial model for training to obtain an inverter fault diagnosis model, including: dividing the denoised inverter data into a training set, a validation set and a test set; constructing the Anomaly Transformer initial model, and inputting the training set and the validation set into the Anomaly Transformer initial model for training and verification to obtain a trained Anomaly Transformer model; inputting the test set into the trained Anomaly Transformer model to obtain an anomaly score corresponding to each data point of the test set; performing statistics on the distribution of the anomaly score corresponding to each data point, and determining an anomaly score threshold; integrating the trained Anomaly Transformer model and the determined anomaly score threshold to obtain an inverter fault diagnosis model.

[0059] In this embodiment, the training set and the validation set are input into the Anomaly Transformer initial model for training and validation, the characteristics of the data in the normal mode are learned, and the trained model weight files are saved. These weight files contain the optimal parameter configuration learned by the model after a large amount of data training. These pre-trained model weights can be used directly to predict and analyze new data in the future, reducing time consumption.

[0060] In this embodiment, the abnormal score threshold is determined by statistical methods (such as ROC curve analysis, quantiles, etc.).

[0061] In this embodiment, the Anomaly Transformer model is trained using the training set data. During the training process, the model parameters are optimized by minimizing the loss function (such as mean square error, cross entropy, etc., depending on the definition of the anomaly score).

[0062] In this embodiment, the validation set data is used to evaluate the model performance and adjust the parameters as needed. The purpose of the validation set is to help find the best model configuration and avoid overfitting.

[0063] In this embodiment, the Anomaly Transformer initial model includes an anomaly attention module and a feedforward neural network layer, which mine potential correlation patterns from deeper time series features by alternately stacking.

[0064] In this embodiment, the abnormal attention module adopts a dual-branch structure, including a main branch and an auxiliary branch. The main branch is used to process time series data and capture the sequence association in the time series data through the self-attention mechanism. The auxiliary branch is used to calculate the prior association, which is usually estimated based on the statistical properties of the time series (such as Gaussian distribution). The prior association provides a baseline for comparison with the sequence association calculated by the main branch.

[0065] In this embodiment, the core mechanism of the abnormal attention module is to calculate the association difference, which is formalized as the asymmetric KL divergence between the prior association and the serial association. The prior association is calculated based on the relative time distance through a learnable Gaussian kernel, and the serial association comes from the association weight distribution of the entire original time series (i.e., the inverter data after denoising). The formulas of the prior association and the serial association are respectively expressed as:

[0066]

[0067]

[0068]

[0069] Where P l represents the prior association matrix of the lth layer, S l represents the sequence association matrix of the lth layer, Q, K, V, σ are the query, key, value and learning scale in the self-attention mechanism, Q, K, V ∈ R N*d ,σ∈R N*1 , represents the parameter matrix of the lth layer Q, K, V, σ, P l ∈R N*N .

[0070] In addition, the reconstruction loss can guide the sequence association to find the most informative association. The formula is:

[0071]

[0072] In this embodiment, the information gain between the prior association and the sequence association is represented by the asymmetric KL divergence, and the formula of the information gain AssDis(P, S; X) is as follows:

[0073]

[0074] In this embodiment, in order to amplify the difference between normal and abnormal time points, an additional loss is introduced to enlarge the association difference. The loss function is defined as:

[0075]

[0076] In the formula, is the reconstruction of X; ||·| F represents the Euclidean norm, ||·|| k represents the k-norm; λ represents the trade-off loss term. When λ>0, it means that we hope to expand the associated difference through the optimization process.

[0077] Since directly maximizing the associated difference will cause the scale parameter of the Gaussian kernel to decrease rapidly, the pre-order sequence is meaningless. Therefore, the maximum and minimum associated learning strategy is used for optimization. The formula for optimizing the loss function is:

[0078] Minimum:

[0079] Maximi ze:

[0080] Where λ>0 and *detach means stopping the associated gradient backpropagation.

[0081] In this embodiment, the Anomaly Transformer initial model includes an anomaly attention module and a feedforward neural network layer. The anomaly attention module and the feedforward neural network layer work together. The anomaly attention module adopts a dual-branch structure, including a main branch and an auxiliary branch. The main branch is used to process time series data and capture the sequence association in the time series data through a self-attention mechanism. The auxiliary branch is used to calculate the prior association. The feedforward neural network performs dimensionality conversion on the sequence association and prior association output by the anomaly attention module, and uses the ReLU activation function to perform nonlinear conversion on the converted sequence association and prior association to obtain new features, and performs residual connection and layer normalization processing on the new features. Finally, the processed features are output to the anomaly attention module to calculate the anomaly score.

[0082] Specifically, the feed-forward neural network layer uses two 1D convolutional layers to map the dimensions of sequence association and prior association to the hidden layer size, and then further to the model dimension.

[0083] In this embodiment, in step S104, real-time data of the inverter device is collected, and the collected real-time data is input into the inverter fault diagnosis model to obtain a diagnosis result, including: collecting real-time data of the inverter device, and performing noise reduction processing on the collected real-time data; inputting the real-time data after noise reduction processing into the inverter fault diagnosis model to obtain a diagnosis result.

[0084] In this embodiment, the collected real-time data can be denoised using the method in step S102, or by smoothing denoising, wavelet transform, deep learning denoising, filter denoising and other methods.

[0085] In this implementation, the inverter fault diagnosis model receives the collected real-time data, first calculates the reconstruction loss and KL divergence loss of each data point in the real-time data, then generates the anomaly score of each data point based on the reconstruction loss and KL divergence loss of each data point, and finally compares the anomaly score of each data point with the anomaly threshold. When the anomaly score is higher than the anomaly threshold, the data point is considered to be abnormal, and the output diagnosis result is that the inverter data is abnormal.

[0086] It can be understood that the abnormal threshold can be obtained by testing the inverter fault diagnosis model through the test set.

[0087] The inverter fault diagnosis method in the embodiment of the present invention is described above. The device in the embodiment of the present invention is described below. Figure 2 , the implementation of the inverter fault diagnosis device in the embodiment of the present invention includes:

[0088] An acquisition module 201 is used to acquire historical inverter data;

[0089] A noise reduction module 202 is used to decompose and reconstruct the historical inverter data using an improved singular spectrum analysis method to obtain inverter data after noise reduction;

[0090] A training module 203 is used to input the de-noised inverter data Anomaly Transformer into an initial model for training to obtain an inverter fault diagnosis model;

[0091] The diagnosis module 204 is used to collect real-time data of the inverter device and input the collected real-time data into the inverter fault diagnosis model to obtain a diagnosis result.

[0092] In this embodiment, an improved singular spectrum analysis method is used to decompose and reconstruct the preprocessed inverter data, which can not only effectively remove the noise in the data, but also retain the key features of the data. Moreover, the inverter fault diagnosis model obtained by training the AnomalyTransformer initial model is used to perform fault judgment on the real-time data of the inverter device. The inverter fault diagnosis model combines the advantages of the self-attention mechanism and the anomaly detection algorithm, and can accurately identify abnormal patterns in the inverter data, thereby improving the accuracy and reliability of the diagnosis. In addition, by collecting the real-time data of the inverter device and performing online diagnosis, the inverter fault can be discovered and handled in time to avoid the expansion of the fault and cause a greater impact on the system.

[0093] above Figure 2 The inverter fault diagnosis device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The inverter fault diagnosis device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0094] Figure 3 3 is a schematic diagram of the structure of an inverter fault diagnosis device provided by an embodiment of the present invention. The device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage devices) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown), and each module may include a series of instruction operations in the device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330 to execute a series of instruction operations in the storage medium on the device 300.

[0095] The device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc.

[0096] An embodiment of the present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the inverter fault diagnosis method.

[0097] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0098] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0099] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing inverter faults, It is characterized in that The inverter fault diagnosis method comprises: Get historical inverter data; Decomposing and reconstructing the historical inverter data by using an improved singular spectrum analysis method to obtain inverter data after noise reduction; Inputting the de-noised inverter data into the Anomaly Transformer initial model for training to obtain an inverter fault diagnosis model; The real-time data of the inverter device is collected, and the collected real-time data is input into the inverter fault diagnosis model to obtain a diagnosis result.

2. The inverter fault diagnosis method according to claim 1, It is characterized in that The historical inverter data includes the inverter's power generation, AC power, DC power, three-phase voltage, three-phase current, power factor, operating temperature status, and grid efficiency.

3. The inverter fault diagnosis method according to claim 1, It is characterized in that Before the improved singular spectrum analysis method is used to decompose and reconstruct the historical inverter data, the historical inverter data is also cleaned, missing values ​​are processed, and normalized.

4. The inverter fault diagnosis method according to claim 1, It is characterized in that The improved singular spectrum analysis method is used to decompose and reconstruct the historical inverter data to obtain the inverter data after noise reduction, including: Converting the historical inverter data into a multidimensional trajectory matrix, wherein each row of the multidimensional trajectory matrix represents inverter data at a time point, and each column represents a feature dimension; The ant colony optimization algorithm (ACO) is used to find the optimal window length for each characteristic dimension of the multidimensional trajectory matrix, and the cumulative contribution rate is set as the objective function; Decompose each feature dimension according to the found optimal window length to obtain multiple feature decomposition vectors corresponding to each feature dimension; From the multiple feature decomposition vectors corresponding to each feature dimension, select the first M feature decomposition vectors as reconstruction elements, and use the reconstruction elements corresponding to the feature dimension to reconstruct each feature dimension one by one to obtain the reconstructed data of each feature dimension; The reconstructed data of all feature dimensions are integrated to obtain the denoised inverter data.

5. The inverter fault diagnosis method according to claim 1, It is characterized in that The inverter data after noise reduction is input into the Anomaly Transformer initial model for training to obtain an inverter fault diagnosis model, including: Divide the denoised inverter data into a training set, a validation set, and a test set; Constructing an Anomaly Transformer initial model, and inputting the training set and the validation set into the Anomaly Transformer initial model for training and validation, to obtain a trained Anomaly Transformer model; Input the test set into the trained Anomaly Transformer model to obtain the anomaly score corresponding to each data point in the test set; Count the distribution of anomaly scores corresponding to each data point and determine the anomaly score threshold; The trained Anomaly Transformer model and the determined anomaly score threshold are integrated to obtain the inverter fault diagnosis model.

6. The inverter fault diagnosis method according to claim 5, It is characterized in that The Anomaly Transformer initial model includes an anomaly attention module and a feedforward neural network layer. The anomaly attention module and the feedforward neural network layer work together. The anomaly attention module adopts a dual-branch structure, including a main branch and an auxiliary branch. The main branch is used to process time series data and capture the sequence association in the time series data through a self-attention mechanism. The auxiliary branch is used to calculate the prior association. The feedforward neural network performs dimensionality conversion on the sequence association and prior association output by the anomaly attention module, and uses the ReLU activation function to perform nonlinear conversion on the converted sequence association and prior association to obtain new features, and performs residual connection and layer normalization processing on the new features. Finally, the processed features are output to the anomaly attention module to calculate the anomaly score.

7. The inverter fault diagnosis method according to claim 1, It is characterized in that The collecting of real-time data of the inverter device and inputting the collected real-time data into the inverter fault diagnosis model to obtain a diagnosis result include: Collect real-time data from inverter equipment and perform noise reduction on the collected real-time data; The real-time data after noise reduction processing is input into the inverter fault diagnosis model to obtain a diagnosis result, and the diagnosis result includes normal inverter data and abnormal inverter data.

8. An inverter fault diagnosis device, It is characterized in that include: An acquisition module is used to obtain historical inverter data; A noise reduction module, used for decomposing and reconstructing the historical inverter data by using an improved singular spectrum analysis method to obtain inverter data after noise reduction; A training module, used for inputting the inverter data after noise reduction Anomaly Transformer into the initial model for training to obtain an inverter fault diagnosis model; The diagnosis module is used to collect real-time data of the inverter device and input the collected real-time data into the inverter fault diagnosis model to obtain a diagnosis result.

9. An inverter fault diagnosis device, It is characterized in that comprising a memory and at least one processor, wherein the memory has computer-readable instructions stored therein; The at least one processor calls the computer-readable instructions in the memory to execute each step of the inverter fault diagnosis method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer-readable instructions stored thereon, It is characterized in that When the computer-readable instructions are executed by a processor, the steps of the inverter fault diagnosis method according to any one of claims 1 to 7 are implemented.

Images