A wind power converter open-circuit fault diagnosis method based on digital twin technology and current characteristic analysis

Abstract

The application discloses a wind power converter open-circuit fault diagnosis method based on digital twinning technology and current characteristic analysis, which comprises the following steps: a digital twinning model based on a deep neural network is established, wind speed, rotating speed, DC bus voltage and pitch angle are taken as model input variables, and a reference value of a three-phase current space vector amplitude is output; three-phase current signals are collected in real time, a measured value of the current space vector amplitude is calculated, a fault severity coefficient is calculated by using the measured value and the reference value, a double-IGBT open-circuit fault is determined when the fault severity coefficient exceeds a threshold value, and otherwise, a single-IGBT open-circuit fault is determined; effective current interval proportion analysis is performed on the three-phase current signals; a fault phase is located by combining a sampling point proportion and the fault type; and a fault bridge arm is judged by analyzing a fault phase current normalized current average value. The application realizes rapid and accurate diagnosis of the wind power converter open-circuit fault by combining the digital twinning technology and the current characteristic analysis, and has the characteristics of high diagnosis precision and good real-time performance.

Description

Technical Field

[0001] This invention relates to wind power system fault diagnosis and maintenance technology, specifically to a method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis. Background Technology

[0002] As wind power accounts for an increasing proportion of renewable energy, the reliability of wind power converters, as the core component connecting generators and the power grid, directly affects the operating efficiency and stability of the entire wind power system. Insulated-gate bipolar transistors (IGBTs) in the converter are prone to open-circuit faults during long-term operation. If not diagnosed and addressed promptly, this can lead to equipment damage or even system shutdown, resulting in significant economic losses.

[0003] Existing IGBT open-circuit fault diagnosis methods can be mainly divided into three categories: signal analysis-based, model-based, and data-driven. Traditional signal analysis methods (such as current waveform analysis and spectrum analysis) are simple and easy to implement, but they are sensitive to noise and have difficulty distinguishing between single and dual transistor faults. Model-based methods rely on accurate system modeling and have poor adaptability under varying operating conditions. While data-driven methods (such as machine learning) have strong generalization capabilities, they usually require a large number of fault samples and have high computational complexity.

[0004] In addition, existing methods still have shortcomings in terms of fault location accuracy and real-time performance: on the one hand, most diagnostic schemes are unable to effectively distinguish between single IGBT and dual IGBT open circuit faults; on the other hand, the location of faulty phases and faulty bridge arms often relies on multi-step threshold judgments, which are easily affected by current fluctuations and may lead to misjudgments. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis, which can adapt to varying operating conditions and accurately identify the fault type and locate the fault location.

[0006] Technical solution: The present invention provides a method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis, comprising:

[0007] Obtain the wind speed v of the wind power converter w Rotational speed ω m DC bus voltage U dc Pitch angle β p Establish a digital twin model based on deep neural networks, using wind speed v w Rotational speed ω m DC bus voltage U dc Pitch angle β pThe input variables for the deep neural network-based digital twin model are I, which outputs a reference value for the magnitude of the three-phase current space vector. svr ;

[0008] Real-time acquisition of three-phase current signals; calculation of the measured value I of the current space vector amplitude based on the real-time acquisition of three-phase current signals. svm Using the measured value I svm Compared with reference value I svr Calculate the severity coefficient α of the fault s When the fault severity coefficient α s Exceeding the preset threshold α of the fault severity coefficient th If the fault occurs, it is determined to be a dual IGBT open circuit fault; otherwise, it is determined to be a single IGBT open circuit fault.

[0009] The percentage D of sampling points in the three-phase current signal that are above a preset threshold I0 is calculated. ph This is used to perform effective current range proportion analysis on three-phase current signals; combined with the sampling point proportion D ph The fault phase is located based on the determined fault type;

[0010] The normalized average current value of the identified faulty phase current is analyzed, and the faulty bridge arm is determined by the positive or negative polarity of the normalized average current value.

[0011] Furthermore, a digital twin model based on deep neural networks is established, using wind speed v w Rotational speed ω m DC bus voltage U dc Pitch angle β p The input variables for the deep neural network-based digital twin model are I, which outputs a reference value for the magnitude of the three-phase current space vector. svr ;

[0012] Design a deep neural network-based digital twin model that includes multi-channel one-dimensional convolutional layers and fully connected layers;

[0013] Select wind speed v w Rotational speed ω m DC bus voltage U dc and propeller pitch angle β p As input variables for digital twin models based on deep neural networks;

[0014] The input variables first pass through a 32-channel one-dimensional convolutional layer for activation, batch normalization, and pooling operations. They then undergo the same processing in 64-channel and 128-channel one-dimensional convolutional layers. Finally, three fully connected layers output a reference value I for the magnitude of the three-phase current space vector. svr ;

[0015] The training dataset is obtained through simulation or field measurement. The training set and validation set are divided proportionally. The Adam optimizer is used to train the digital twin model based on the deep neural network. The batch size and number of iterations are set. If the model error meets the preset standard, the model is saved; otherwise, the hyperparameters are adjusted and the model is retrained.

[0016] Furthermore, the fault severity coefficient α s The calculation formula is as follows:

[0017]

[0018] Furthermore, the fault severity coefficient is preset to a threshold α. th It falls between the average severity of a single IGBT failure and the average severity of a dual IGBT failure.

[0019] Furthermore, the percentage D of the sampling points of the three-phase current signal that are above a preset threshold I0 ph The calculation formula is as follows:

[0020]

[0021] Where, n eff,ph This indicates that the single-phase current satisfies |i| within a unit period. ph [k]|>I0 is the number of sampling points; N is the total number of samples in the calculation period; ph∈[a,b,c], k∈[0,N-1] represents the specific sampling point; I0 ​​is the set current threshold.

[0022] Furthermore, the formula for calculating the set current threshold I0 is as follows:

[0023]

[0024] Where, β d This is the adjustment coefficient.

[0025] Furthermore, combined with the proportion of sampling points D ph The fault phase is located based on the determined fault type, including:

[0026] For a single IGBT fault, select the sampling point ratio D. ph The phase with the smallest value is designated as the faulty phase.

[0027] For dual IGBT faults, check whether there is a single-phase sampling point percentage D. ph Less than threshold D th If a faulty phase exists, it is directly determined to be a dual IGBT faulty phase, and the diagnostic process ends; if no phase meets the conditions, the proportion of sampling points D is determined. ph The two phases with the smallest values ​​are the faulty phases.

[0028] Furthermore, the threshold D th Set the threshold D to meet the following conditions. th Less than D under all heterogeneous dual IGBT faults ph Value, and threshold D th It is a constant greater than 0.

[0029] Furthermore, the normalized current average value analysis is performed on the identified fault phase current. The polarity of the normalized current average value is used to determine the faulty bridge arm, including:

[0030] Calculate the normalized mean current of the fault phase. The calculation formula is as follows:

[0031]

[0032] Furthermore, the normalized current average value analysis is performed on the identified fault phase current. The polarity of the normalized current average value is used to determine the faulty bridge arm, including:

[0033] like If the fault is found, the fault is determined to be a lower IGBT; otherwise, the fault is determined to be an upper IGBT.

[0034] Beneficial effects: Compared with the prior art, the significant technical effects of the present invention are as follows: (1) The present invention innovatively combines digital twin technology with current feature analysis, and realizes intelligent diagnosis of open circuit faults of converters by constructing a digital twin model based on deep neural networks; (2) It establishes a mapping relationship between non-electrical parameters such as wind speed and rotational speed and current space vector, and realizes accurate identification of fault types through dynamically generated reference values; (3) It proposes an effective current range ratio analysis method, and effectively overcomes the misjudgment problem of traditional methods under noise interference by statistically analyzing the proportion of sampling points of current signals above a set threshold; (4) It can quickly locate the faulty bridge arm by only calculating the normalized mean of the faulty phase current, which greatly improves the diagnostic efficiency; (5) Compared with traditional methods, the present invention not only realizes accurate differentiation between single-tube and double-tube open circuit faults, but also has the advantages of small calculation volume, strong real-time performance, and good engineering applicability, providing effective technical support for improving the operational reliability of wind power converters. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the process of the present invention;

[0036] Figure 2 This is a structural diagram of a permanent magnet direct-drive wind power system;

[0037] Figure 3 A structural diagram of the digital twin model;

[0038] Figure 4A rendering of a digital twin model;

[0039] Figure 5 Fault severity estimation curves for four typical open-circuit faults;

[0040] Figure 6 The normalized mean current and effective current ratio curves are shown for phase A bridge open circuit.

[0041] Figure 7 The normalized mean current and effective current ratio curves are shown for phase A with the upper and lower bridges open circuit.

[0042] Figure 8 The normalized mean current and effective current ratio curves for phase A and phase B bridge open circuit conditions;

[0043] Figure 9 The graph shows the normalized mean current and the percentage of effective current when the upper bridge of phase A and the lower bridge of phase C are open. Detailed Implementation

[0044] The technical solution of the present invention will now be described in detail with reference to specific embodiments and accompanying drawings.

[0045] like Figure 1 As shown, the wind power converter open-circuit fault diagnosis method based on digital twin technology and current characteristic analysis of the present invention includes the following steps:

[0046] S1. Obtain the wind speed v of the wind power converter. w Rotational speed ω m DC bus voltage U dc Pitch angle β p Establish a digital twin model based on deep neural networks, using wind speed v w Rotational speed ω m DC bus voltage U dc Pitch angle β p The input variables for the deep neural network-based digital twin model are I, which outputs a reference value for the magnitude of the three-phase current space vector. svr .

[0047] The specific implementation process of step S1 is as follows:

[0048] S1.1, such as Figure 2 As shown, taking a permanent magnet direct-drive wind power system as an example, the wind speed v of the wind power converter is obtained. w Rotational speed ω m DC bus voltage U dc Pitch angle β p With wind speed v w Rotational speed ω mThe main input variable is the DC bus voltage U. dc Pitch angle β p These are auxiliary input variables.

[0049] S1.2 Design a digital twin model based on a deep neural network, including multi-channel one-dimensional convolutional layers and fully connected layers, such as... Figure 3 As shown, the multi-channel one-dimensional convolutional layer includes a 32-channel one-dimensional convolutional layer, a 64-channel one-dimensional convolutional layer, and a 128-channel one-dimensional convolutional layer; the number of fully connected layers is three, namely a 512-channel fully connected layer, a 128-channel fully connected layer, and a 32-channel fully connected layer.

[0050] S1.3, Select wind speed v w Rotational speed ω m DC bus voltage U dc and propeller pitch angle β p As an input variable for a deep neural network-based digital twin model, the wind speed v can be established using this model. w Rotational speed ω m The mapping relationship between non-electrical parameters and the three-phase current space vector.

[0051] S1.4, the input variables first pass through a 32-channel one-dimensional convolutional layer for activation, batch normalization, and pooling operations. Then, they are processed in the same way in 64-channel and 128-channel one-dimensional convolutional layers. Finally, the reference value I of the three-phase current space vector amplitude is output through three fully connected layers. svr .

[0052] S1.5. Obtain the training dataset using simulation or field measurement, divide the training set and validation set proportionally, use the Adam optimizer to train the digital twin model based on deep neural network, set the batch size and number of iterations, save the model if the model error meets the preset standard, otherwise adjust the hyperparameters and retrain.

[0053] In this embodiment, the training set and validation set are divided in an 8:2 ratio. The Adam optimizer is used for training, the batch size is 64, and the number of iterations is 50. If the model error meets the preset standard, the model is saved; otherwise, the hyperparameters are adjusted and the model is retrained. Figure 4 Figure (a) shows the mean squared error loss curves of the model on the training and test sets. Figure 4 Figure (b) in the figure shows a comparison curve of the normalized values ​​of the model. From Figure 4As shown in Figure a, the loss curves of the model on both the training and test sets steadily decrease with each training epoch, eventually converging, and the final values ​​of the two sets are quite close, indicating that the model has good generalization ability and does not exhibit overfitting or underfitting. Further analysis of the comparison between predicted and measured values ​​(Figure b) reveals that the measured values ​​show a scatter distribution with a clear trend, while the predicted values ​​exhibit a smoother, less noisy fitting curve, remaining stable across different numerical ranges, indicating high prediction accuracy, consistent with the conclusions of the loss curves. These two figures corroborate each other, demonstrating that the model performs excellently in both training efficiency and prediction performance, and can be reliably applied to practical tasks.

[0054] S2. Real-time acquisition of three-phase current signals, and calculation of the measured value I of the current space vector amplitude based on the real-time acquired three-phase current signals. svm The calculation formula is as follows:

[0055]

[0056] Using measured value I svm Compared with reference value I svr Calculate the severity coefficient α of the fault s When the fault severity coefficient α s Exceeding the preset threshold α of the fault severity coefficient th If the fault occurs, it is determined to be a dual IGBT open circuit fault; otherwise, it is determined to be a single IGBT open circuit fault.

[0057] In this embodiment, the fault severity coefficient α s The calculation formula is as follows:

[0058]

[0059] Determining whether a single IGBT or dual IGBTs are open-circuit faults: If α s >α th If the fault is found to be open in both IGBTs, it is determined to be a dual IGBT open-circuit fault; otherwise, it is a single IGBT open-circuit fault. The fault severity coefficient has a preset threshold α. th It falls between the average severity of a single IGBT failure and the average severity of a dual IGBT failure. For example... Figure 5 As shown, Figure 5 The diagram shows the fault severity estimation curves under four typical open-circuit faults: open circuit on phase A upper bridge, open circuit on both phase A upper and lower bridges, open circuit on both phase A upper bridge and phase B upper bridges, and open circuit on both phase A upper bridge and phase C lower bridges. It can be seen that α is the fault severity estimation curve under the dual IGBT open-circuit condition. s It should be significantly greater than the single IGBT open-circuit state, i.e., α s This can effectively distinguish between the two fault states, as shown in the figure α. th Set to 27%.

[0060] S3. Calculate the percentage D of sampling points in the three-phase current signal that are above the preset threshold I0. ph This is used to perform effective current range proportion analysis on three-phase current signals; combined with the sampling point proportion D ph The fault phase is located based on the determined fault type.

[0061] The specific implementation process of step S3 is as follows:

[0062] S3.1 Calculate the percentage D of sampling points of the three-phase current signal that are above the preset threshold I0. ph This allows for the analysis of the effective current range proportion of three-phase current signals, and the calculation formula is as follows:

[0063]

[0064] Where, n eff,ph This indicates that the single-phase current satisfies |i| within a unit period. ph [k]|> The number of sampling points for I0; N is the total number of samples in the calculation period, and the calculation period is greater than the fundamental current period; ph∈[a,b,c] represents a specific phase, and k∈[0,N-1] represents a specific sampling point; I0 ​​is the set current threshold, and its calculation formula is as follows:

[0065]

[0066] Where, β d This is the adjustment coefficient, and its value is a small positive number.

[0067] S3.2, Combined with the sampling point proportion D of each phase calculated in step S3.1 ph The faulty phase is located based on the fault type (single IGBT or dual IGBT fault) determined in step S2. The specific determination rules are as follows:

[0068] For a single IGBT fault, select the sampling point ratio D. ph The phase with the smallest value is designated as the faulty phase.

[0069] For dual IGBT faults: a) First check if there is a sampling point percentage D for a certain phase. ph Less than threshold D th (D th The setting satisfies the following condition: threshold D th Less than D under all heterogeneous dual IGBT faults ph Value, and threshold D th (a) If a constant greater than 0 exists, the phase is directly determined to be a dual IGBT faulty phase, and the diagnostic process ends; b) If no phase meets the condition, the proportion of sampling points D is determined. ph The two phases with the smallest values ​​are the faulty phases.

[0070] S4. Perform normalized current average (NCA) analysis on the determined fault phase current, and determine the faulty bridge arm by the positive and negative polarities of the normalized current average (NCA).

[0071] The specific implementation process of step S4 is as follows:

[0072] S4.1 Calculate the normalized mean current of the fault phase. The calculation formula is as follows:

[0073]

[0074] S4.2. The faulty bridge arm is determined by the polarity of the normalized average current value, as follows: If the normalized average current value is positive, that is... This indicates a fault in the lower bridge arm; if the normalized average current is negative, that is... This indicates a fault in the upper bridge arm. Figure 6 Figure (a) shows the normalized mean current curve of phase A with the upper bridge open circuit. Figure 6 Figure (b) shows the percentage of effective current range under open-circuit conditions on phase A upper bridge. Figure 7 Figure (a) shows the normalized mean current curves for phase A with the upper and lower bridges open. Figure 7 Figure (b) shows the percentage of effective current range under open circuit conditions for phase A upper and lower bridges. Figure 8 Figure (a) shows the normalized mean current curves for the upper bridge of phase A and the upper bridge of phase B with the circuit open. Figure 8 Figure (b) shows the percentage of effective current range under open circuit conditions for phase A and phase B bridges. Figure 9 Figure (a) shows the normalized mean current curves for the upper bridge of phase A and the lower bridge of phase C with the circuit open. Figure 9 Figure (b) shows the percentage of effective current ranges when the upper bridge of phase A and the lower bridge of phase C are open. Figure 6 , Figure 7 , Figure 8 ,and Figure 9 The normalized mean current and effective current ratio curves are displayed for four typical open-circuit faults: A-phase upper bridge open circuit, A-phase upper and lower bridge open circuit, A-phase upper bridge and B-phase upper bridge open circuit, and A-phase upper bridge and C-phase lower bridge open circuit. These curves can effectively locate the faulty phase and the faulty bridge arm.

[0075] This invention can improve the operational reliability and fault diagnosis accuracy of wind power systems, and can cope with the impact of harsh wind power operating environments such as wind speed fluctuations and mechanical stress changes on converters, thus providing technical support for the safe and stable operation of wind power systems.

[0076] This invention combines digital twin technology and current characteristic analysis to achieve rapid and accurate diagnosis of open-circuit faults in wind power converters, featuring high diagnostic accuracy and good real-time performance. By constructing a high-fidelity virtual model, this invention provides a dynamic reference benchmark for fault diagnosis; while the analysis strategy combining effective current percentage and normalized current mean can extract more discriminative fault features from the current signal. Therefore, developing a diagnostic method that integrates digital twins and multi-dimensional current characteristic analysis is of great significance for improving the accuracy and robustness of open-circuit fault diagnosis in wind power converters.

Claims

1. A wind power converter open-circuit fault diagnosis method based on digital twin technology and current feature analysis, characterized in that: include: Obtain the wind speed of the wind power converter Rotation speed DC bus voltage Pitch angle Establish a digital twin model based on deep neural networks, using wind speed... Rotation speed DC bus voltage Pitch angle The input variables for the deep neural network-based digital twin model are the reference values ​​for the three-phase current space vector amplitudes output by the deep neural network-based digital twin model. ; Real-time acquisition of three-phase current signals; calculation of the measured value of the current space vector amplitude based on the real-time acquisition of three-phase current signals. Using measured values Compared with reference value Calculate the severity coefficient of the fault When the severity coefficient of the fault Exceeding the preset threshold of the fault severity coefficient If the fault occurs, it is determined to be a dual IGBT open circuit fault; otherwise, it is determined to be a single IGBT open circuit fault. By calculating the three-phase current signal at a preset threshold The above sampling point percentage This is used to perform effective current range proportion analysis on three-phase current signals; combined with sampling point proportion. The fault phase is located based on the determined fault type; The normalized average current value of the identified faulty phase current is analyzed, and the faulty bridge arm is determined by the positive or negative polarity of the normalized average current value.

2. The method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis according to claim 1, characterized in that: Establish a digital twin model based on deep neural networks, using wind speed Rotation speed DC bus voltage Pitch angle The input parameters for the deep neural network-based digital twin model are used to output reference values ​​for the three-phase current space vector amplitudes. ; Design a deep neural network-based digital twin model that includes multi-channel one-dimensional convolutional layers and fully connected layers; Select wind speed Rotation speed DC bus voltage and propeller pitch angle As input parameters for a deep neural network-based digital twin model; The input variables first pass through a 32-channel one-dimensional convolutional layer for activation, batch normalization, and pooling operations, then enter a 64-channel one-dimensional convolutional layer for the same processing. Finally, the input variables are output as reference values ​​for the three-phase current space vector amplitudes through three fully connected layers. ; The training dataset is obtained through simulation or field measurement. The training set and validation set are divided proportionally. The Adam optimizer is used to train the digital twin model based on the deep neural network. The batch size and number of iterations are set. If the model error meets the preset standard, the model is saved; otherwise, the hyperparameters are adjusted and the model is retrained.

3. The method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis according to claim 1, characterized in that, The severity coefficient of the fault The calculation formula is as follows: 。 4. The method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis according to claim 1, characterized in that, The fault severity coefficient has a preset threshold. It falls between the average severity of a single IGBT failure and the average severity of a dual IGBT failure.

5. The method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis according to claim 1, characterized in that, The percentage of sampling points of the three-phase current signal above the preset threshold I0 The calculation formula is as follows: ； in, This indicates that the single-phase current satisfies the following within a unit period. The number of sampling points; This represents the total number of samples in the calculation period; , Indicates a specific sampling point; This is the set current threshold.

6. The method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis according to claim 5, characterized in that, The formula for calculating the set current threshold I0 is as follows: ； in, This is the adjustment coefficient.

7. The method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis according to claim 1, characterized in that, Combined with the proportion of sampling points The fault phase is located based on the determined fault type, including: For single IGBT faults, select the proportion of sampling points. The phase with the smallest value is designated as the faulty phase. For dual IGBT faults, check whether there is a single-phase sampling point ratio. Less than the threshold If a phase exists, it is directly determined to be a dual IGBT faulty phase, and the diagnostic process ends; if no phase meets the conditions, the proportion of sampling points is determined. The two phases with the smallest values ​​are the faulty phases.

8. The method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis according to claim 7, characterized in that, The threshold Set a threshold that meets the following conditions. Less than all out-of-phase dual IGBT faults Value, and threshold It is a constant greater than 0.

9. The method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis according to claim 1, characterized in that, Normalized current average analysis is performed on the identified fault phase current. The polarity of the normalized current average value is used to determine the faulty bridge arm, including: Calculate the normalized mean current of the fault phase. The calculation formula is as follows: ； in, This represents the total number of samples in the calculation period.

10. The method for diagnosing open-circuit faults in wind power converters based on digital twin technology and current characteristic analysis according to claim 9, characterized in that, Normalized current average analysis is performed on the identified fault phase current. The polarity of the normalized current average value is used to determine the faulty bridge arm, including: like If the fault is found, the lower IGBT is identified as faulty; otherwise, the upper IGBT is identified as faulty.

Images