A fault diagnosis method and device suitable for an inverter
Patent Information
- Application Number
- CN202210791402.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-07-06
- Publication Date
- 2026-09-29
- Estimated Expiration
- 2042-07-06
AI Technical Summary
[0002]相关技术中，逆变器发生故障将影响其所在的整个新能源发电系统的发电效率，因为逆变器故障类型复杂，故障特征区分度不明显，所以，逆变器的故障诊断精度不高
[0026]本申请实施例采用深度学习与迁移学习结合的方式，获取多个样本数据对，并将样本数据对输入到故障诊断模型，再通过训练和迁移学习，用少量的实测数据样本对迁移后的模型进行训练，能够保证再实测样本数据较少的情况下具有较高的分类精度，提高了故障诊断模型的准确度，从而促进新能源发电系统稳定可靠运行。
Smart Images
Figure CN115618707B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of inverter fault diagnosis technology, and in particular to a fault diagnosis method and apparatus applicable to inverters. Background Technology
[0002] In related technologies, inverter failures will affect the power generation efficiency of the entire new energy power generation system. Because inverter fault types are complex and fault characteristics are not easily distinguishable, the accuracy of inverter fault diagnosis is not high. Therefore, how to accurately determine the type and location of inverter faults in real time, so as to quickly eliminate faults and restore the stable and reliable operation of new energy power generation systems in a timely manner, has become one of the important research directions. Summary of the Invention
[0003] This application aims to at least partially address one of the technical problems in the related art. Therefore, one objective of this application is to provide a fault diagnosis method and apparatus suitable for inverters.
[0004] The first aspect of this application proposes a training method for a fault diagnosis model suitable for inverters, comprising:
[0005] Acquire multiple sample data pairs, each of which includes the three-phase current data of the inverter and the corresponding reference diagnostic results;
[0006] Input the three-phase current data into the initial fault diagnosis model to obtain the predictive diagnosis results of the inverter.
[0007] The initial fault diagnosis model is trained based on the predicted diagnosis results and the reference diagnosis results to obtain candidate fault diagnosis models;
[0008] Transfer learning is performed on the candidate fault diagnosis model to obtain the target fault diagnosis model.
[0009] A second aspect of this application provides a fault diagnosis method for an inverter, including:
[0010] Acquire the target three-phase current data of the inverter under test;
[0011] The target three-phase current data is input into the target fault diagnosis model to obtain the target diagnosis result of the inverter under test. The target fault diagnosis model is obtained according to the training method of the first aspect embodiment.
[0012] A third aspect of this application provides a training device for a fault diagnosis model suitable for inverters, comprising:
[0013] The first acquisition module is used to acquire multiple sample data pairs, each sample data pair including the three-phase current data of the inverter and the corresponding reference diagnostic results;
[0014] The second acquisition module is used to input the three-phase current data into the initial fault diagnosis model in order to obtain the predictive diagnosis results of the inverter.
[0015] The training module is used to train the initial fault diagnosis model based on the predicted diagnosis results and the reference diagnosis results to obtain candidate fault diagnosis models.
[0016] The transfer learning module is used to perform transfer learning on candidate fault diagnosis models to obtain the target fault diagnosis model.
[0017] A fourth aspect of this application provides a fault diagnosis device for an inverter, characterized in that it includes:
[0018] The first acquisition module is used to acquire the target three-phase current data of the inverter under test;
[0019] The second acquisition module is used to input the target three-phase current data into the target fault diagnosis model in order to obtain the target diagnosis result of the inverter under test. The target fault diagnosis model is obtained according to the training device as described in the third aspect embodiment.
[0020] A third aspect of this application provides an electronic device, comprising:
[0021] At least one processor; and
[0022] A memory that is communicatively connected to at least one processor; wherein,
[0023] The memory stores instructions that can be executed by at least one processor, which enables the at least one processor to execute the training method for a fault diagnosis model applicable to an inverter provided in the first aspect embodiment of this application or the fault diagnosis method for an inverter provided in the second aspect embodiment.
[0024] A fourth aspect of this application provides a computer-readable storage medium storing computer instructions thereon, wherein the computer instructions are used to cause a computer to execute a training method for a fault diagnosis model applicable to an inverter provided according to a first aspect of this application or a fault diagnosis method for an inverter provided according to a second aspect of this application.
[0025] The fifth aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the training method for a fault diagnosis model applicable to an inverter provided in the first aspect of this application or the fault diagnosis method for an inverter provided in the second aspect of this application.
[0026] This application adopts a combination of deep learning and transfer learning to obtain multiple sample data pairs, input the sample data pairs into the fault diagnosis model, and then train the transferred model with a small amount of measured data samples through training and transfer learning. This can ensure high classification accuracy even with a small amount of measured sample data, improve the accuracy of the fault diagnosis model, and thus promote the stable and reliable operation of the new energy power generation system. Attached Figure Description
[0027] Figure 1 This is a flowchart of a training method for a fault diagnosis model applicable to an inverter, according to one embodiment of this application;
[0028] Figure 2 This is a flowchart of a training method for a fault diagnosis model applicable to an inverter, according to one embodiment of this application;
[0029] Figure 3 This is a flowchart of a training method for a fault diagnosis model applicable to an inverter, according to one embodiment of this application;
[0030] Figure 4 This is a flowchart of a training method for a fault diagnosis model applicable to an inverter, according to one embodiment of this application;
[0031] Figure 5 This is a schematic diagram of the structure of a fault diagnosis model according to an embodiment of this application;
[0032] Figure 6 This is a schematic diagram illustrating the acquisition of fused feature representation according to an embodiment of this application;
[0033] Figure 7 This is a flowchart of a training method for a fault diagnosis model applicable to an inverter, according to one embodiment of this application;
[0034] Figure 8 This is a flowchart of a fault diagnosis method for an inverter according to an embodiment of this application;
[0035] Figure 9 This is a structural diagram of a training device for a fault diagnosis model applicable to an inverter, according to one embodiment of this application.
[0036] Figure 10 This is a structural diagram of a fault diagnosis device for an inverter according to an embodiment of this application;
[0037] Figure 11 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation
[0038] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.
[0039] The following describes, with reference to the accompanying drawings, a fault diagnosis method and apparatus applicable to inverters according to embodiments of this application.
[0040] Figure 1 This is a flowchart of a training method for a fault diagnosis model applicable to inverters according to one embodiment of this application, as shown below. Figure 1 As shown, the method includes the following steps:
[0041] S101, acquire multiple sample data pairs, each sample data pair including the three-phase current data of the inverter and the corresponding reference diagnostic results.
[0042] An inverter is a converter that transforms direct current (DC) energy (from batteries or storage batteries) into alternating current (AC) with fixed frequency and voltage or adjustable frequency and voltage (typically 220V, 50Hz sine wave). It consists of an inverter bridge, control logic, and filter circuitry.
[0043] An inverter contains multiple insulated gate bipolar transistors (IGBTs). As a switching device, the IGBT is in a high-voltage, high-current, high-frequency switching state for a long time. Therefore, in the embodiments of this application, the fault type of the inverter can be IGBT failure, and the corresponding reference diagnostic result is that the switching device at a certain location is damaged.
[0044] In some implementations, the sample data can be historical diagnostic data of an inverter that has already failed. For example, the output signal of an inverter whose reference diagnostic result is that the switching device at position A has failed can be detected to obtain three-phase current data.
[0045] In some implementations, sample data pairs can be obtained from simulation models. For example, a simulation model of the inverter can be built based on a simulation platform, and then the simulation system can control the inverter simulation model to simulate and diagnose faults. The reference diagnostic results of the inverter can be obtained based on the simulated fault diagnosis, and the three-phase current data can be obtained based on the output signal of the inverter under the simulated fault diagnosis.
[0046] In this embodiment of the application, the sample data pairs can be divided into a training set and a test set. Taking 500 sample data pairs as an example, 400 of them can be used as the training set and the remaining 100 as the test set.
[0047] Optionally, before inputting the initial fault diagnosis model, K-fold cross-validation can be used to process the sample data pairs in the training set.
[0048] S102, input the three-phase current data into the initial fault diagnosis model to obtain the predictive diagnosis results of the inverter.
[0049] In this embodiment of the application, the three-phase current data in the training set can be preprocessed and then input into the initial fault diagnosis model.
[0050] In implementation, the three-phase current data is current data containing three components, with each component being a channel. In this application embodiment, the three-phase current data within a preset period can be extracted. In some implementations, the three-phase current data is standardized, and in some implementations, the three-phase current data is noise-added. Optionally, the three-phase current data can also be differentially amplified.
[0051] In this embodiment, the initial fault diagnosis model can be a neural network containing convolutional layers, pooling layers, fully connected layers and an output layer. Based on the initial fault model, features are extracted from the three-phase current data in the training set to obtain the predictive diagnosis results of the inverter.
[0052] S103, Train the initial fault diagnosis model based on the predicted diagnosis results and the reference diagnosis results to obtain candidate fault diagnosis models.
[0053] In some implementations, the predicted diagnostic results and the reference diagnostic results are matched to obtain a matching degree. This matching degree is then used to adjust the image dehazing model. For example, it can be determined whether the matching degree meets a preset condition. If the matching degree does not meet the preset condition, the parameters in the model, such as the parameters of the convolutional kernels in the convolutional layers and the parameters of the fully connected layers, are adjusted until the matching degree meets the preset condition. Optionally, the preset condition can be that both the first matching degree and the second matching degree are greater than a preset matching degree threshold.
[0054] In some implementations, a loss function is obtained based on the predicted diagnosis results and the reference diagnosis results. The initial fault diagnosis model is then adjusted using a reverse gradient based on the loss function to obtain a candidate fault diagnosis model.
[0055] Optionally, to improve the accuracy of the model, after obtaining the candidate fault diagnosis model, the candidate fault diagnosis model is evaluated based on the sample data of the test set, and the evaluation result is obtained. That is, the three-phase current data of the test set is input into the candidate fault diagnosis model to obtain the test diagnosis result. If the test diagnosis result is consistent with the reference diagnosis result of the test set, the evaluation result of the candidate fault diagnosis model is passed.
[0056] S104, perform transfer learning on the candidate fault diagnosis model to obtain the target fault diagnosis model.
[0057] If the evaluation result of the candidate fault diagnosis model is unsuccessful, the candidate fault diagnosis model will be adjusted again based on the sample data. If the evaluation result of the candidate fault diagnosis model is successful, the specified layer structure of the candidate fault diagnosis model will be fixed, and the candidate fault diagnosis model will be trained based on the measured inverter sample data to obtain the target fault diagnosis model after transfer learning.
[0058] This application adopts a combination of deep learning and transfer learning to obtain multiple sample data pairs, input the sample data pairs into the fault diagnosis model, and then train the transferred model with a small amount of measured data samples through training and transfer learning. This can ensure high classification accuracy even with a small amount of measured sample data, improve the accuracy of the fault diagnosis model, and thus promote the stable and reliable operation of the new energy power generation system.
[0059] Figure 2 This is a flowchart of a training method for a fault diagnosis model applicable to inverters according to one embodiment of this application, as shown below. Figure 2 As shown, the method includes the following steps:
[0060] S201, obtain the main circuit topology and modulation type of the inverter.
[0061] Circuit topology, also known as a circuit diagram or circuit structure, is a set of branches and nodes that is a further abstraction of the circuit diagram. It can represent the connection relationships and properties of the circuit, that is, the connection relationships between branches and nodes.
[0062] In this embodiment of the application, the inverter type can be pulse width modulation (PWM), sinusoidal pulse width modulation (SPWM), or space vector pulse width modulation (SVPWM).
[0063] In this embodiment, the main circuit topology and modulation type of the inverter are obtained to facilitate the subsequent construction of the inverter simulation model.
[0064] S202, based on the main circuit topology and modulation type, calls the simulation system to build an inverter simulation model.
[0065] Optionally, to facilitate data analysis and signal processing, in this embodiment of the application, an interactive application can be used as the simulation system, that is, the inverter simulation model can be built based on mathematical software MATLAB or visual simulation tool Simulink.
[0066] S203, obtain fault simulation information based on reference diagnostic results.
[0067] In this embodiment of the application, the fault simulation information can be the simulation information of the switching device in the simulation model. Optionally, a reference diagnostic result can be obtained according to the fault type. Taking the reference diagnostic result indicating that the switching device at position A is damaged as an example, the fault simulation information obtained based on the reference diagnostic result can instruct the simulation system to generate a simulated diagnostic fault of the switching device at position A.
[0068] S204 sends fault simulation information to the simulation system. The fault simulation information is used to instruct the simulation system to generate a simulated fault diagnosis of the inverter simulation model based on the fault simulation information. The three-phase current data is acquired under the simulated fault diagnosis.
[0069] The fault simulation information is sent to the simulation system, which generates a simulated fault diagnosis of the inverter simulation model based on the fault simulation information. The three-phase current data is obtained under the simulated fault diagnosis.
[0070] This application embodiment uses a built fault simulation system to simulate fault data, obtain multiple sample data pairs, input the sample data pairs into the fault diagnosis model, and then train the transferred model with a small amount of measured data samples through training and transfer learning. This can ensure high classification accuracy even with a small amount of measured sample data, improve the accuracy of the fault diagnosis model, and also improve the flexibility and efficiency of inverter fault diagnosis.
[0071] Figure 3 This is a flowchart of a training method for a fault diagnosis model applicable to inverters according to one embodiment of this application, as shown below. Figure 3 As shown, the method includes the following steps:
[0072] S301, acquire multiple sample data pairs, each sample data pair including the three-phase current data of the inverter and the corresponding reference diagnostic results.
[0073] For a description of step S301, please refer to the relevant content in the above embodiments, which will not be repeated here.
[0074] S302 standardizes the current data of any channel in the three-phase current data to obtain the standardized current data of any channel.
[0075] The three-phase current data within a specified period is extracted, combined into one-dimensional three-channel data, and the current data of each channel is standardized. For example, the current data of each channel is normalized by Z-Score processing. Z-Score processing scales the data proportionally to make it fall into a specific range. Z-Score standardization is a common data processing method that can convert data of different magnitudes into Z-Score values of a uniform metric for comparison.
[0076] S303 adds noise to the standardized current data of any channel based on Gaussian random noise to obtain the noise-added three-phase current data.
[0077] By adding Gaussian random noise to the standardized current data, the embodiments of this application can process the data into the input data format required for the initial fault diagnosis model, making the simulation data closer to the actual data, and reducing the impact of the data distribution range.
[0078] S304 inputs the three-phase current data into the initial fault diagnosis model to obtain the predicted diagnosis results.
[0079] S305, Train the initial fault diagnosis model based on the predicted diagnosis results and the reference diagnosis results to obtain candidate fault diagnosis models.
[0080] S306, perform transfer learning on the candidate fault diagnosis model to obtain the target fault diagnosis model.
[0081] The descriptions of steps S304 to S306 can be found in the relevant content of the above embodiments, and will not be repeated here.
[0082] This application standardizes the current data of any channel in the three-phase current data to obtain standardized current data for any channel. Then, it adds Gaussian random noise to the standardized current data of any channel to obtain noisy three-phase current data. This application makes the simulation data closer to the actual data, reduces the influence of data distribution range, ensures high classification accuracy even with limited measured sample data, improves the accuracy of the fault diagnosis model, and enhances the flexibility of inverter fault diagnosis.
[0083] Figure 4 This is a flowchart of a training method for a fault diagnosis model applicable to inverters according to one embodiment of this application, as shown below. Figure 4 As shown, the method includes the following steps:
[0084] S401, Based on the first convolutional layer, feature extraction is performed on the three-phase current data to obtain the first feature representation.
[0085] like Figure 5As shown, the initial fault diagnosis model of the inverter consists of an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a second pooling layer, a fully connected layer, and an output layer.
[0086] The input layer inputs the three-phase current data of the three channels into the first convolutional layer for feature extraction to obtain the first feature representation.
[0087] The first convolutional layer uses multiple one-dimensional convolutional kernels. Optionally, the feature extraction process can be represented by the following formula:
[0088]
[0089] in, Let f be the activation function, d be the number of channels, and m be the number of convolutional kernels. Input the feature to the i-th convolutional kernel of the j-th channel in the l-th layer. The weights of the i-th convolutional kernel in the j-th channel of the l-th layer are: This is the bias unit for the i-th convolutional kernel in the l-th layer.
[0090] S402, perform pooling operation on the first feature based on the first pooling layer to obtain the first high-frequency feature representation and the first low-frequency feature representation.
[0091] like Figure 5 As shown in the embodiments of this application, the first pooling layer uses one-dimensional discrete wavelet transform for pooling operation. Optionally, the pooling operation process can be represented by the following formula:
[0092]
[0093] in, For the output signal of layer l, x l-1 (2n-k) is the input signal, g i (x) is the filtering function, i=1 corresponds to the high-pass filter, i=2 corresponds to the low-pass filter, and ↓2 is the downsampling.
[0094] The first pooling layer uses a one-dimensional discrete wavelet transform pooling formula to pool the input features, obtaining high-frequency features H and low-frequency features L.
[0095] S403, obtain the fused feature representation based on the first high-frequency feature representation and the first low-frequency feature representation.
[0096] like Figure 5 , Figure 6As shown, a one-dimensional discrete wavelet transform is performed on the first high-frequency feature representation and the first low-frequency feature representation to obtain the second low-frequency feature representation LL and multiple second high-frequency feature representations. In this embodiment, the second high-frequency feature representation includes three representations: second high-frequency feature representation HH, second high-frequency feature representation HL, and second high-frequency feature representation LH. Where x(n) represents the input signal, g... 1 (n) represents the high-pass filter function, g 2 (n) represents the low-pass filter function.
[0097] Based on the second convolutional layer, feature extraction is performed on multiple second high-frequency feature representations to obtain the third feature representation. That is, the second high-frequency feature representations HH, HL, and LH are concatenated by channels and then input into the second convolutional layer for feature extraction to obtain the third feature representation.
[0098] Feature extraction is performed on the second low-frequency feature representation based on the third convolutional layer to obtain the fourth feature representation. The third and fourth feature representations are then concatenated to obtain the fused feature representation. In this embodiment, the feature extraction process of the second and third convolutional layers can be referred to the feature extraction process of the first convolutional layer, and will not be repeated here.
[0099] S404, based on the second pooling, performs dimensionality reduction on the fused feature representation to obtain the target feature representation.
[0100] In this embodiment, the second pooling layer uses max pooling to reduce the dimensionality of the fused feature representation to obtain the target feature representation.
[0101] S405 maps the target feature representation based on a fully connected layer to obtain the prediction and diagnosis results.
[0102] A fully connected layer is a layer in which each node is connected to all nodes in the previous layer, used to synthesize the features extracted earlier. Based on the fully connected layer, the target feature representation is mapped to obtain the predicted diagnostic results, which are then output through the output layer.
[0103] This application's embodiments obtain a fused feature representation based on a first high-frequency feature representation and a first low-frequency feature representation. The fused feature representation is then dimensionality-reduced using a second pooling method to obtain a target feature representation. Finally, the target feature representation is mapped using a fully connected layer to obtain a predicted diagnostic result. This application ensures high classification accuracy even with limited measured sample data, improving the accuracy of the fault diagnosis model and enhancing the flexibility of inverter fault diagnosis. The use of wavelet transform for pooling operations provides high interpretability in the feature extraction process. Furthermore, the utilization of frequency domain information during wavelet transform improves fault diagnosis and localization effectiveness.
[0104] Figure 7 This is a flowchart of a training method for a fault diagnosis model applicable to inverters according to one embodiment of this application, as shown below. Figure 7 As shown, the method includes the following steps:
[0105] S701, acquire multiple sample data pairs, each sample data pair includes the three-phase current data of the inverter and the corresponding reference diagnostic results.
[0106] S702 inputs the three-phase current data into the initial fault diagnosis model to obtain the predictive diagnosis results of the inverter.
[0107] The descriptions of steps S701 to S702 can be found in the relevant content of the above embodiments, and will not be repeated here.
[0108] S703, based on the predicted diagnostic results and the reference diagnostic results, determines the loss function of the initial fault diagnosis model.
[0109] In this embodiment of the application, a loss function is determined based on the predicted diagnosis results and the reference diagnosis results, and the initial fault diagnosis model is trained in a supervised manner. Optionally, information-based cross-entropy loss calculation can be performed based on the predicted diagnosis results and the reference diagnosis results to obtain the loss function of the initial fault diagnosis model.
[0110] S704, adjust the initial fault diagnosis model in reverse according to the loss function, return the adjusted initial fault diagnosis model for the next training, until the training ends and a candidate fault diagnosis model is generated.
[0111] The initial fault diagnosis model is adjusted using a backpropagation gradient based on the loss function. For example, the model parameters of the initial fault diagnosis model can be tuned. Then, the initial fault diagnosis model is trained again based on the next set of sample data until the training ends when a set condition is met, thus obtaining the target text generation model. Optionally, the set condition can be that the loss of the initial fault diagnosis model converges to a preset value.
[0112] Optionally, the model parameters are the parameters of the convolution kernel in the convolutional layer and the parameters of the fully connected layer. For example, if the convolutional layer uses a 3*3 convolution kernel, then the convolutional layer has 9 model parameters that need to be adjusted.
[0113] S705, acquire historical data pairs of the inverter. Any historical data pair includes the measured three-phase current data of the inverter and the measured diagnostic results.
[0114] In some implementations, the sample data can be historical diagnostic data of an inverter that has already failed. For example, the output signal of an inverter whose reference diagnostic result is that the switching device at position A has failed can be detected to obtain three-phase current data.
[0115] S706: Fix the specified convolutional layer parameters of the candidate fault diagnosis model and train the candidate fault diagnosis model based on historical data to obtain the target fault diagnosis model.
[0116] In this embodiment, the first, second, and third convolutional layers of the candidate fault diagnosis model can be frozen, and then the candidate fault diagnosis model can be trained using measured historical data pairs to obtain the target fault diagnosis model after transfer learning.
[0117] This application can ensure high classification accuracy even with limited measured sample data, improve the accuracy of the fault diagnosis model, enhance the flexibility of inverter fault diagnosis, and improve fault diagnosis and location.
[0118] Figure 8 This is a flowchart of a fault diagnosis method for an inverter according to an embodiment of this application, as shown below. Figure 8 As shown, the method includes the following steps:
[0119] S801, acquire the target three-phase current data of the inverter under test.
[0120] To diagnose a fault in an inverter under test, it is necessary to obtain the output signal of the inverter under test, that is, the target three-phase current data.
[0121] S802 inputs the target three-phase current data into the target fault diagnosis model to obtain the target diagnosis results of the inverter under test.
[0122] The target three-phase current data is input into the target fault diagnosis model to obtain the target diagnosis results corresponding to the inverter. The target fault diagnosis model is obtained according to the training method for fault diagnosis models applicable to inverters as described above.
[0123] This application embodiment inputs the target three-phase current data into the target fault diagnosis model to obtain the diagnostic results of the inverter under test. This application embodiment has high classification accuracy, improves the accuracy of the fault diagnosis model, enhances the flexibility and efficiency of inverter fault diagnosis, and improves fault location effectiveness.
[0124] Figure 9 This is a structural diagram of a training device for a fault diagnosis model applicable to an inverter, according to one embodiment of this application. Figure 9 As shown, the training device 900 for the fault diagnosis model of the inverter includes:
[0125] The first acquisition module 910 is used to acquire multiple sample data pairs, each sample data pair including the three-phase current data of the inverter and the corresponding reference diagnostic results;
[0126] The second acquisition module 920 is used to input three-phase current data into the initial fault diagnosis model in order to obtain the predictive diagnosis results of the inverter.
[0127] Training module 930 is used to train the initial fault diagnosis model based on the predicted diagnosis results and the reference diagnosis results to obtain candidate fault diagnosis models;
[0128] The transfer learning module 940 is used to perform transfer learning on the candidate fault diagnosis model to obtain the target fault diagnosis model.
[0129] In some implementations, the first acquisition module 910 is also used for:
[0130] Obtain the main circuit topology and modulation type of the inverter;
[0131] Based on the main circuit topology and modulation type, the simulation system is invoked to construct an inverter simulation model;
[0132] Fault simulation information is obtained based on reference diagnostic results;
[0133] The fault simulation information is sent to the simulation system. The fault simulation information is used to instruct the simulation system to generate a simulated fault diagnosis model of the inverter based on the fault simulation information. The three-phase current data is acquired under the simulated fault diagnosis.
[0134] In some implementations, the second acquisition module 920 is also used for:
[0135] Standardize the current data of any channel in the three-phase current data to obtain the standardized current data of any channel;
[0136] The normalized current data of any channel is noise-added using Gaussian random noise to obtain the noise-added three-phase current data.
[0137] In some implementations, training module 930 is also used for:
[0138] Based on the predicted diagnostic results and the reference diagnostic results, the loss function of the initial fault diagnosis model is determined;
[0139] The initial fault diagnosis model is adjusted in reverse according to the loss function, and the adjusted initial fault diagnosis model is returned for the next training until the training ends and a candidate fault diagnosis model is generated.
[0140] In some implementations, the initial fault diagnosis model includes convolutional layers, pooling layers, and fully connected layers. The second acquisition module 920 is also used for:
[0141] Based on the first convolutional layer, feature extraction is performed on the three-phase current data to obtain the first feature representation;
[0142] The first feature is pooled based on the first pooling layer to obtain the first high-frequency feature representation and the first low-frequency feature representation;
[0143] A fused feature representation is obtained based on the first high-frequency feature representation and the first low-frequency feature representation;
[0144] The dimensionality of the fused feature representation is reduced based on the second pooling method to obtain the target feature representation.
[0145] The target feature representation is mapped using a fully connected layer to obtain the prediction and diagnosis results.
[0146] In some implementations, the second acquisition module 920 is also used for:
[0147] A one-dimensional discrete wavelet transform is performed on the first high-frequency feature representation and the first low-frequency feature representation to obtain the second low-frequency feature representation and multiple second high-frequency feature representations.
[0148] Based on the second convolutional layer, feature extraction is performed on multiple second high-frequency feature representations to obtain the third feature representation. Based on the third convolutional layer, feature extraction is performed on the second low-frequency feature representations to obtain the fourth feature representation.
[0149] The third and fourth feature representations are channel-connected to obtain the fused feature representation.
[0150] In some implementations, the transfer learning module 940 is also used for:
[0151] Obtain historical data pairs of the inverter. Each historical data pair includes the measured three-phase current data of the inverter and the measured diagnostic results.
[0152] The specified convolutional layer parameters of the candidate fault diagnosis model are fixed, and the candidate fault diagnosis model is trained based on historical data to obtain the target fault diagnosis model.
[0153] This application adopts a combination of deep learning and transfer learning to obtain multiple sample data pairs, input the sample data pairs into the fault diagnosis model, and then train the transferred model with a small amount of measured data samples through training and transfer learning. This can ensure high classification accuracy even with a small amount of measured sample data, improve the accuracy of the fault diagnosis model, and thus promote the stable and reliable operation of the new energy power generation system.
[0154] Figure 10 This is a structural diagram of a fault diagnosis device for an inverter according to an embodiment of this application, as shown below. Figure 10 As shown, the inverter fault diagnosis device 1000 includes
[0155] The first acquisition module 1010 is used to acquire the target three-phase current data of the inverter under test;
[0156] The second acquisition module 1020 is used to input the target three-phase current data into the target fault diagnosis model in order to obtain the target diagnosis result of the inverter under test. The target fault diagnosis model is obtained according to the training device for the fault diagnosis model applicable to the inverter as described above.
[0157] This application embodiment inputs the target three-phase current data into the target fault diagnosis model to obtain the diagnostic results of the inverter under test. This application embodiment has high classification accuracy, improves the accuracy of the fault diagnosis model, enhances the flexibility and efficiency of inverter fault diagnosis, and improves fault location effectiveness.
[0158] To implement the above embodiments, this application also proposes an electronic device 1100, such as... Figure 11 As shown, the electronic device 1100 includes: a processor 1110 and a memory 1120 communicatively connected to the processor. The memory 1120 stores instructions that can be executed by at least one processor. The instructions are executed by at least one processor 1110 to implement the training method for a fault diagnosis model applicable to an inverter or the fault diagnosis method for an inverter as described in this application.
[0159] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0160] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0161] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0162] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0163] Based on the same concept, embodiments of this application also provide a computer-readable storage medium storing computer instructions thereon, wherein the computer instructions are used to cause a computer to execute the training method for a fault diagnosis model applicable to an inverter or the fault diagnosis method for an inverter as described in the above embodiments.
[0164] Based on the same concept, embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, provides the training method for a fault diagnosis model applicable to an inverter or the fault diagnosis method for an inverter as described in the above embodiments.
[0165] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. This application can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.
[0166] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.
[0167] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.
[0168] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of the invention. Therefore, if these modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.
Claims
1. A training method for a fault diagnosis model suitable for inverters, characterized in that, include: Acquire multiple sample data pairs, each of which includes the three-phase current data of the inverter and the corresponding reference diagnostic results; The three-phase current data is input into the initial fault diagnosis model to obtain the predictive diagnosis results of the inverter. The initial fault diagnosis model is trained based on the predicted diagnosis results and the reference diagnosis results to obtain candidate fault diagnosis models; The candidate fault diagnosis model is transferred to obtain the target fault diagnosis model. Also includes: Obtain the main circuit topology and modulation type of the inverter; Based on the main circuit topology and the modulation type, the simulation system is invoked to construct an inverter simulation model; Fault simulation information is obtained based on the reference diagnostic results; The fault simulation information is sent to the simulation system, and the fault simulation information is used to instruct the simulation system to generate a simulated fault diagnosis of the inverter simulation model based on the fault simulation information. The three-phase current data is obtained under the simulated fault diagnosis. Before inputting the three-phase current data into the initial fault diagnosis model, the method further includes: The current data of any channel in the three-phase current data is standardized to obtain the standardized current data of any channel. The normalized current data of any one of the channels is noise-added according to Gaussian random noise to obtain the noise-added three-phase current data; The initial fault diagnosis model includes convolutional layers, pooling layers, and fully connected layers. The step of inputting the three-phase current data into the initial fault diagnosis model to obtain the predicted diagnosis results includes: Based on the first convolutional layer, feature extraction is performed on the three-phase current data to obtain a first feature representation; The first feature is pooled based on the first pooling layer to obtain a first high-frequency feature representation and a first low-frequency feature representation; A fused feature representation is obtained based on the first high-frequency feature representation and the first low-frequency feature representation; The dimensionality of the fused feature representation is reduced based on the second pooling method to obtain the target feature representation. The target feature representation is mapped based on the fully connected layer to obtain the prediction and diagnosis results; The step of obtaining the fused feature representation based on the first high-frequency feature representation and the first low-frequency feature representation includes: Perform a one-dimensional discrete wavelet transform on the first high-frequency feature representation and the first low-frequency feature representation to obtain a second low-frequency feature representation and multiple second high-frequency feature representations; Based on the second convolutional layer, feature extraction is performed on the multiple second high-frequency feature representations to obtain a third feature representation; based on the third convolutional layer, feature extraction is performed on the second low-frequency feature representations to obtain a fourth feature representation. The third feature representation and the fourth feature representation are connected via channels to obtain the fused feature representation.
2. The method according to claim 1, characterized in that, The step of training the initial fault diagnosis model based on the predicted diagnosis results and the reference diagnosis results to obtain candidate fault diagnosis models includes: Based on the predicted diagnostic results and the reference diagnostic results, the loss function of the initial fault diagnosis model is determined; The initial fault diagnosis model is adjusted in reverse according to the loss function, and the adjusted initial fault diagnosis model is returned for the next training until the training ends and the candidate fault diagnosis model is generated.
3. The method according to claim 1, characterized in that, The step of performing transfer learning on the candidate fault diagnosis model to obtain the target fault diagnosis model includes: Obtain historical data pairs of the inverter, wherein any historical data pair includes the measured three-phase current data of the inverter and the measured diagnostic results; The specified convolutional layer parameters of the candidate fault diagnosis model are fixed, and the candidate fault diagnosis model is trained based on the historical data to obtain the target fault diagnosis model.
4. A fault diagnosis method for an inverter, characterized in that, include: Acquire the target three-phase current data of the inverter under test; The target three-phase current data is input into the target fault diagnosis model to obtain the target diagnosis result of the inverter under test. The target fault diagnosis model is obtained according to the training method described in any one of claims 1-3.
5. A training device for a fault diagnosis model suitable for inverters, characterized in that, The training device implements the training method as described in claim 1, including: The first acquisition module is used to acquire multiple sample data pairs, each of which includes the three-phase current data of the inverter and the corresponding reference diagnostic results. The second acquisition module is used to input the three-phase current data into the initial fault diagnosis model in order to obtain the predictive diagnosis results of the inverter. The training module is used to train the initial fault diagnosis model based on the predicted diagnosis results and the reference diagnosis results to obtain candidate fault diagnosis models. The transfer learning module is used to perform transfer learning on the candidate fault diagnosis model to obtain the target fault diagnosis model.
6. A fault diagnosis device for an inverter, characterized in that, include: The first acquisition module is used to acquire the target three-phase current data of the inverter under test; The second acquisition module is used to input the target three-phase current data into the target fault diagnosis model to obtain the target diagnosis result of the inverter under test, wherein the target fault diagnosis model is obtained according to the training device as described in claim 5.