PMSG system converter open circuit fault diagnosis method and device

Abstract

The present invention discloses a PMSG system converter open circuit fault diagnosis method and device, the method steps include: S1. Select the fault feature that causes the PMSG system state to change and add it to the spare feature pool; S2. Perform three-dimensional projection on each fault feature in the spare feature pool, judge the sensitivity of each fault feature to the fault position and wind speed, and screen out the fault features that are sensitive to the fault position and insensitive to the wind speed; S3. Calculate the intra-class distance of each screened fault feature under different wind speed conditions and fault conditions; S4. According to the sensitivity state and intra-class distance of the fault feature, screen the fault feature to obtain the final target fault feature; S5. Extract sample data of each target fault feature, train the nonlinear network to obtain the fault diagnosis model, so as to realize the converter open circuit fault diagnosis. The present invention has the advantages of simple implementation method, high diagnostic efficiency and accuracy, strong flexibility, etc.

Description

Technical Field

[0001] The present invention relates to the technical field of PMSG (Permanent Magnet Synchronous Generator) systems, and in particular to a PMSG system converter open circuit fault diagnosis method and device. Background Art

[0002] PMSG is one of the most common generator types in wind power generation systems, and the converter in its drive system is a key component in PMSG. Whether the PMSG wind power system can be safely connected to the grid is based on the normal operation of the key component converter. However, the main components of the converter are power electronic devices, which are highly sensitive to the environment. In actual operating conditions, the failure rate of the converter has always been the first among the electrical failures of the wind power system, among which the switch open circuit failure is one of the most common failure forms. After the switch open circuit failure occurs, it will not cause overcurrent and overvoltage. Considering the economic efficiency of grid-connected operation, the PMSG system will still be connected to the grid after the failure occurs, that is, the system maintains abnormal grid-connected operation. However, if the fault is latent for a long time, the operating pressure of other electrical equipment will gradually increase, and the risk of secondary failure will increase significantly, threatening the safety of the power grid, and it is very likely to further evolve into a more serious fault problem. Therefore, it is of great significance to realize the detection of switch open circuit failure in the machine-side converter of the PMSG drive system for the safe and stable operation of the PMSG system.

[0003] The working reliability of the freewheeling diode in the machine-side converter of the PMSG drive system is higher than that of the power switch, but it may still cause an open-circuit fault of a single power switch. In the case of an open-circuit fault of a single power switch, the freewheeling diode can form a channel with other devices for the bidirectional flow of the stator current of the faulty phase. That is, when the cause of the open-circuit fault is a single power tube, the fault current can be conducted through the freewheeling diode to form a bidirectional flow. However, due to the small current deviation and unclear fault characteristics, the open-circuit fault of a single power switch is highly hidden and difficult to detect.

[0004] At present, there are two main diagnostic methods for switch open circuit faults: model analysis method and signal analysis method. Model analysis method realizes fault diagnosis and analysis by constructing an analysis model, but it has high requirements on model accuracy and is suitable for systems with simple input-output relationships. Signal analysis method is a common way to realize fault diagnosis. Among them, current fault diagnosis method is widely used because it does not need to add sensors and has low cost. However, traditional fault diagnosis methods based on current characteristics are usually for whole tube open circuit faults. The fault current of whole tube open circuit faults is unipolar, and the fault characteristics are more obvious than single power tube open circuit faults. When a single power switch open circuit fault occurs, the fault current has nonlinear distortion, and the magnitude and direction of the instantaneous distortion are irregular. In addition, the current signal load characteristics are obvious, which will hinder the capture of fault characteristics or limit the ability to characterize characteristics. Therefore, in order to ensure the diagnostic effect, traditional fault diagnosis methods based on current characteristics must rely heavily on the ability to capture fault characteristics. With regard to the selection of fault features, the existing technology usually relies on experience or expert knowledge for feature selection, which is not only inefficient, but also the prominence and feature characterization capabilities of the selected fault features are still limited, resulting in low accuracy of fault diagnosis.

[0005] Some practitioners have proposed that in the feature extraction process, the required features should be selected by considering the characteristics of different fault locations to improve the effectiveness of fault diagnosis. However, this type of method usually only considers the fault location characteristics, that is, the corresponding features are selected specifically for the fault location characteristics, and the sensitivity to other factors such as wind speed is not considered. The features selected based on the single fault location characteristics are still unable to fully and accurately characterize the open-circuit fault state of the PMSG system converter, and therefore the fault cannot be accurately diagnosed. Summary of the invention

[0006] The technical problem to be solved by the present invention is: in response to the technical problems existing in the prior art, the present invention provides a PMSG system converter open-circuit fault diagnosis method and device with simple implementation method, high diagnostic efficiency and accuracy, which does not need to rely on experience to select features, and can effectively extract fault features with prominent features and strong characterization capabilities to construct a diagnostic model, thereby realizing efficient and accurate PMSG system converter open-circuit fault diagnosis.

[0007] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0008] A PMSG system converter open circuit fault diagnosis method, the steps comprising:

[0009] S1. According to the operating state of the PMSG system at different fault locations when a machine-side converter fault occurs during the operation of the PMSG system, all fault features that cause the PMSG system state to change are selected and added to the backup feature pool;

[0010] S2. Perform three-dimensional projection on each fault feature in the backup feature pool, cluster each fault feature after projection, determine the sensitivity of each fault feature to the fault location and wind speed according to the number of clusters formed by clustering, and select fault features that are sensitive to the fault location and insensitive to the wind speed;

[0011] S3. Calculate the intra-class distance of each fault feature selected in step S2 under different wind speed conditions and fault conditions;

[0012] S4. According to the sensitivity of the fault feature to the fault location type and the corresponding intra-class distance, the final target fault feature is again selected from the fault features selected in step S2;

[0013] S5. Extract sample data of each target fault feature, train a nonlinear network to obtain a fault diagnosis model, so as to realize open circuit fault diagnosis of the PMSG system converter.

[0014] Furthermore, the step S1 includes:

[0015] Construct the model expression of the three-phase current of the PMSG drive system:

[0016]

[0017] Among them, i s represents the three-phase current vector, s represents the auxiliary logic vector, R represents the generator resistance vector, L represents the generator reactance vector, e represents the generator potential vector, c represents the preset coefficient value determined according to the initial state, u dc Indicates the DC side voltage value;

[0018] When a machine-side converter fault occurs during the operation of the PMSG system, the fault feature corresponding to the data change in the model expression of the three-phase current of the PMSG drive system is selected and added to the spare feature pool.

[0019] Furthermore, the fault features in the backup feature pool include three-phase features of A, B, and C, and the three-phase features of A, B, and C of each fault feature are described in the form of a 1*3 vector, wherein for the DQ current vector, a 1*3 form is formed by combining the amplitude and the vector angle. In step S2, the three-dimensional projection is completed by projecting the ABC three-phase features of each fault feature to the xyz three-dimensional orthogonal coordinates respectively.

[0020] Further, the step S2 comprises:

[0021] S201. Perform three-dimensional projection on the fault characteristics of a single fault type under the same wind speed, wherein the fault types include a single switch fault, a non-same-phase double switch fault, and a same-phase double switch fault, and determine whether the number of clusters formed by clustering after the three-dimensional projection is equal to the number of converter switches of the PMSG system. If so, it is determined that the current fault characteristics are sensitive to the fault location, and the process proceeds to step S202; otherwise, it is determined that the current fault characteristics are not sensitive to the fault location, and the process proceeds to step S203;

[0022] S202. Perform three-dimensional projection on the fault features of a single fault type at two different wind speeds, and determine whether the number of clusters formed by clustering after the three-dimensional projection is equal to the number of converter switches of the PMSG system. If so, it is determined that the current fault feature is not sensitive to the wind speed; otherwise, it is determined that the current fault feature is sensitive to the wind speed and the process proceeds to step S203;

[0023] S203. Delete the current fault feature from the backup feature pool.

[0024] Furthermore, in step S3, the intra-class distance of the jth class is calculated according to the following formula:

[0025]

[0026] Among them, m j represents the feature mean vector of the jth class, n j is the number of characteristic samples contained in the jth class, and The feature i vector representing the jth class;

[0027] The expression for calculating the total intra-class distance is:

[0028]

[0029] Among them, c represents the total number of feature categories, and N represents the total number of feature samples contained in the jth category.

[0030] Furthermore, in step S4, the compression degree corresponding to the fault feature is determined according to the intra-class distance, and the fault feature that is sensitive to the fault location under the required fault diagnosis type and whose compression degree meets the preset requirements is selected as the final target fault feature.

[0031] Furthermore, step S5 includes: using the sample data corresponding to the screened target fault characteristics to train the BP neural network to obtain a fault diagnosis model, and determining whether the diagnosis rate of the obtained fault diagnosis model is greater than a preset threshold value, and if so, outputting the final fault diagnosis model, otherwise adjusting the number of samples or adjusting the nonlinear optimization function of the intermediate layer of the neural network training, and retraining the nonlinear network.

[0032] A PMSG system converter open circuit fault diagnosis device, comprising:

[0033] A feature selection unit is used to select all fault features that cause the PMSG system state to change according to the operating state of the PMSG system at different fault locations when a machine-side converter fault occurs during the operation of the PMSG system, and add them to a backup feature pool;

[0034] A feature projection unit is used to perform three-dimensional projection on each fault feature in the backup feature pool, cluster each projected fault feature, judge the sensitivity of each fault feature to the fault location and wind speed according to the number of clusters formed by clustering, and screen out fault features that are sensitive to the fault location and insensitive to the wind speed;

[0035] A distance calculation unit is used to calculate the intra-class distance of each fault feature screened by the feature projection unit under different wind speed conditions and fault conditions;

[0036] A feature screening unit, configured to screen out a final target fault feature from each fault feature screened by the first feature screening unit according to a sensitivity state of the fault feature to the fault location type and the corresponding intra-class distance;

[0037] The model training unit is used to extract sample data of each target fault feature and train a nonlinear network to obtain a fault diagnosis model to realize open circuit fault diagnosis of the PMSG system converter.

[0038] A computer device comprises a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0039] A computer-readable storage medium storing a computer program, wherein the computer program implements the above method when executed.

[0040] Compared with the prior art, the advantages of the present invention are as follows: the present invention realizes feature selection by combining feature projection and distance analysis clustering method, realizes qualitative analysis of feature sensitivity state by feature projection, and realizes quantitative analysis of distance within feature selection class by distance analysis, which can effectively improve the highlighting and feature characterization capabilities of the selected fault features, solve the problem that feature selection in traditional fault diagnosis methods needs to rely on expert knowledge or experience, not only improves the flexibility of feature selection, but also can effectively improve the utilization rate of features; at the same time, the selected fault features are used to train the fault diagnosis model using a nonlinear network, and the powerful data processing ability of the nonlinear network for samples can be used to achieve rapid optimization, thereby realizing efficient and accurate PMSG system converter open-circuit fault diagnosis; at the same time, by comprehensively considering the sensitivity of the fault location and wind speed for feature extraction, it can ensure that the features that can completely and accurately characterize the open-circuit fault state of the PMSG system converter are effectively screened out, thereby further improving the effectiveness of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of the implementation flow of the PMSG system converter open circuit fault diagnosis method of this embodiment.

[0042] Figure 2 It is a detailed flow chart of realizing open circuit fault diagnosis of PMSG system converter in a specific application embodiment of the present invention.

[0043] Figure 3 It is a schematic diagram of the PMSG topology system structure in a specific application embodiment.

[0044] Figure 4 It is a schematic diagram of the three-phase current output waveform under different switch open circuit fault conditions of the machine-side converter of the PMSG system obtained at the first wind speed (7m / s) in a specific application embodiment.

[0045] Figure 5 It is a schematic diagram of the three-phase current output waveform under different switch open circuit fault conditions of the machine-side converter of the PMSG system obtained at the second wind speed (10.5m / s) in the specific application embodiment.

[0046] Figure 6 It is a schematic diagram of the DQ current output waveform under different switch open circuit fault conditions of the machine-side converter of the PMSG system at the first wind speed (7m / s) in a specific application embodiment.

[0047] Figure 7 It is a schematic diagram of the DQ current output waveform under different switch open circuit fault conditions of the machine-side converter of the PMSG system at the second wind speed (10.5m / s) in a specific application embodiment.

[0048] Figure 8It is a schematic diagram of the result after three-phase projection of the three-phase current obtained in a specific application embodiment.

[0049] Fig. 9 It is a schematic diagram of the statistical results of the system sensitivity obtained after projection analysis of various features obtained in a specific application embodiment.

[0050] Fig.10 Schematic diagram of the inter-class distance curve of each feature obtained in a specific application embodiment. DETAILED DESCRIPTION

[0051] The present invention is further described below in conjunction with the accompanying drawings and specific preferred embodiments, but the protection scope of the present invention is not limited thereby.

[0052] like Figure 1 As shown, the steps of the PMSG system converter open circuit fault diagnosis method of this embodiment include:

[0053] S1. Feature selection: According to the operating state of the PMSG system at different fault locations when a machine-side converter fault occurs during the operation of the PMSG system, all fault features that cause the PMSG system state to change are selected and added to the backup feature pool;

[0054] S2. Feature projection: Perform three-dimensional projection on each fault feature in the backup feature pool, cluster each fault feature after projection, and judge the sensitivity of each fault feature to the fault location and wind speed according to the number of clusters formed by clustering, and select the fault features that are sensitive to the fault location but insensitive to the wind speed;

[0055] S3. Distance calculation: Calculate the distance within the class of each fault feature selected in step S2 under different wind speed conditions and fault conditions;

[0056] S4. Feature screening: According to the sensitivity of the fault feature to the fault location type and the corresponding intra-class distance, the final target fault feature is screened out from the fault features screened out in step S2;

[0057] S5. Fault diagnosis: Extract sample data of each target fault feature, train the nonlinear network to obtain a fault diagnosis model, and realize open circuit fault diagnosis of the PMSG system converter.

[0058] This embodiment realizes feature selection by combining feature projection and distance analysis clustering, realizes qualitative analysis of feature sensitivity state by feature projection, and realizes quantitative analysis of distance within feature selection class by distance analysis, which can effectively improve the highlighting and feature characterization capabilities of the selected fault features, solve the problem that feature selection in traditional fault diagnosis methods needs to rely on expert knowledge or experience, not only improves the flexibility of feature selection, but also effectively improves the utilization rate of features; at the same time, the selected fault features are trained with BP neural network fault diagnosis model, and the powerful nonlinear processing ability of nonlinear network for samples can be used to achieve rapid optimization, thereby realizing efficient and accurate PMSG system converter open circuit fault diagnosis. At the same time, this embodiment extracts features by comprehensively considering the sensitivity of fault location and wind speed, which can ensure the effective screening of features that can fully and accurately characterize the PMSG system converter open circuit fault state, thereby further improving the effectiveness of fault diagnosis.

[0059] In step S1 of this embodiment, when a machine-side converter fault occurs during the operation of the PMSG system, the system operation state is analyzed to identify the characteristics of the system state change after the fault, and if there is a change, the fault characteristics are incorporated into the fault diagnosis backup feature pool. Specifically, according to the offline simulation calculation and analysis of the system operation state, the distortion characteristics that may be caused by the fault are determined, and the fault characteristics corresponding to the data changes in the model expression of the three-phase current of the PMSG drive system caused by the machine-side converter fault during the operation of the PMSG system are selected and added to the backup feature pool, so that all possible distortion characteristics are stored in the backup feature pool as candidate characteristics.

[0060] The detailed process of determining the fault characteristics that cause the system state to change after the fault in this embodiment is as follows:

[0061] First, construct a vector representation of the mixed logic dynamic system (MLD) model based on the PMSG system:

[0062] i′=Ai+Be+Cs (1)

[0063] Among them, i=[i a i b i c ] is the state vector of the PMSG system, and the three elements i a ,i b and i c is the three-phase current; s is the auxiliary logic variable vector, which is generally composed of the trigger pulse signal and other introduced auxiliary variables, representing the discrete input vector of the system; e = [e as e bs e cs ] Tis the generator potential vector of the PMSG system, which can be regarded as the continuous input vector of the system. The three elements e as , e bs and e cs is the three-phase potential of the generator; A, B and C are the coefficient matrices.

[0064] The control behavior of the above model can be simply described as follows: the generator potential (continuous input vector) and the auxiliary logic variable (discrete input vector) work together to affect the output state of the PMSG system.

[0065] Define the time scale at the second level (S level), equation (1) can be written in vector form as equation (2), that is:

[0066]

[0067] in, represents the generator reactance vector, represents the generator resistance vector, represents the generator potential vector, Represents the three-phase current vector. The logic vector s is composed of logic values, which have different values ​​depending on the system operation status.

[0068] Adding superscript * to the operating state under fault operation, the MLD model under fault operation can be expressed as:

[0069]

[0070] Substituting equation (3) from equation (2), we obtain equation (4):

[0071]

[0072] in, represents the current residual vector, u dc Indicates the DC side voltage value.

[0073] In formula (4), the influence of e is eliminated, that is, the influence from the external environment is eliminated, which can play a decoupling role.

[0074] By further rearranging equation (4), we can obtain the logical vector expression under fault operation:

[0075]

[0076] According to the mathematical expression of formula (2), using s * Can be obtained:

[0077]

[0078] Wherein, c is a preset coefficient determined according to the initial state.

[0079] Formula (6) is the general expression of the three-phase current of the PMSG drive system. This expression can be used to determine whether the fault feature is a feature that causes the system state to change when the system fails. Specifically, the backup feature is determined based on whether the data on the right side of Formula (6) changes after the fault. That is, if the fault feature can cause the data in Formula (6) to change after the fault occurs, the fault feature is included in the backup feature pool.

[0080] Considering that the converter state is described by a three-phase expression, that is, the fault features in the standby feature pool include ABC three-phase features. This embodiment uses a 1*3 vector form to describe the ABC three-phase features of each fault feature, wherein for the DQ current vector, a 1*3 form is formed by combining the amplitude and the vector angle (for example, the D-axis current vector average value, the Q-axis current average value, and the vector angle value formed by the average values ​​of the two vectors). In step S2 of this embodiment, the ABC three-phase features of each fault feature are projected to the xyz three-dimensional orthogonal coordinates respectively to complete the three-dimensional projection, thereby ensuring that all features can be projected in three dimensions.

[0081] In this embodiment, step S2 includes:

[0082] S201. Perform three-dimensional projection on the fault characteristics of a single fault type under the same wind speed. The fault types include single switch fault, non-same-phase double switch fault and same-phase double switch fault. Determine whether the number of clusters formed by clustering after three-dimensional projection is equal to the number of converter switches of the PMSG system. If yes, determine that the current fault characteristics are sensitive to the fault location, and proceed to step S202; otherwise, determine that the current fault characteristics are not sensitive to the fault location, and proceed to step S203;

[0083] S202. Perform three-dimensional projection on the fault features of a single fault type at two different wind speeds, and determine whether the number of clusters formed by clustering after the three-dimensional projection is equal to the number of converter switches of the PMSG system. If so, it is determined that the current fault feature is not sensitive to the wind speed; otherwise, it is determined that the current fault feature is sensitive to the wind speed and the process proceeds to step S203;

[0084] S203. Delete the current fault feature from the backup feature pool.

[0085] In a specific application embodiment, it is assumed that {x a ,y b ,z c} is the three-phase current feature, and the three-dimensional projection is to project each component onto the xyz axis, and the projection is a point in the three-dimensional coordinate system. The open-circuit fault of the PMSG system converter is divided into three categories: single switch fault, non-same-phase double switch fault and same-phase double switch fault. First, a single fault sample under a single wind speed is selected for projection. If the projection shows a cluster equal to the number of converter switches, it is determined that the single fault sample under a single wind speed of this feature is sensitive to the fault location; then single fault samples under two different wind speeds are selected for projection. If the projection can be distinguished into clusters equal to the number of converter switches (that is, the number of clustered trees is equal to the number of converter switches), it is determined that the single fault sample of multiple wind speeds of this feature is also sensitive. If the distinguishable clusters of the features under different faults are different from the theoretical fault number, it means that the fault feature of this type is not sensitive, and finally the fault features that are sensitive to the fault location and insensitive to wind speed are screened out.

[0086] Taking a single switch fault as an example, the features occurring under a single switch fault are projected in three dimensions. If the number of projected clusters is equal to the number of switches, the feature is judged to be sensitive to the single switch fault. If they are not equal, it means it is not sensitive. The sensitivity analysis is performed on different fault types in turn to confirm the sensitivity of the feature to the fault type. Then, the features with the fault type sensitivity divided are analyzed for wind speed sensitivity. Assuming that feature A is sensitive to a single switch fault, a sample of another wind speed is added to the feature sample of the original wind speed condition. If the number of clusters obtained after the three-dimensional projection is equal to the number of original clusters, it means that the fault feature is not sensitive to wind speed and is initially screened as a diagnostic feature. If the number of clusters is not equal, it means that it is sensitive to wind speed and the feature is removed from the candidate feature pool.

[0087] In step S3 of this embodiment, the intra-class distance of the jth class is calculated specifically according to the following formula:

[0088]

[0089] Among them, m j represents the feature mean vector of the jth class, n j is the number of characteristic samples contained in the jth class, and The feature i vector representing the jth class;

[0090] The expression for calculating the total intra-class distance is:

[0091]

[0092] Among them, c represents the total number of feature categories, and N represents the total number of feature samples contained in the jth category.

[0093] S w The value of takes the smallest value, indicating that the corresponding compression degree is higher and the compression effect is better. wThe value of is used to filter out the final target fault features for model training.

[0094] In step S4 of this embodiment, the compression degree corresponding to the fault feature is determined according to the intra-class distance, and the fault feature that is sensitive to the fault location under the required fault diagnosis type and the compression degree meets the preset requirements is selected as the final target fault feature. In a specific application embodiment, according to the fault diagnosis requirements and the sensitivity of the feature to the fault location type, as well as the feature compression degree and distinguishability, a feature or feature combination with good pre-diagnosis efficiency is selected. For example, assuming that only a single-tube open circuit under multiple wind speeds needs to be diagnosed, and after feature analysis, it is obtained that the DC component is sensitive to the single-tube open circuit and the non-co-phase double-tube open circuit, is insensitive to the wind speed, and the clustering compression degree and distinguishability are good (meeting the preset requirements), and the covariance is only sensitive to the co-phase double-tube open circuit fault, then the DC component can be used as a screened feature, and the covariance can be deleted in the spare feature pool. Through the above steps, fault features that are sensitive to the fault location type, have a good feature compression degree, and can be distinguished can be screened out, so that the highlighting of the features and the feature characterization capabilities can be ensured to the greatest extent possible.

[0095] In this embodiment, the nonlinear network specifically adopts a BP neural network, and step S5 specifically includes: using the sample data corresponding to the selected target fault feature to train the BP neural network to obtain a fault diagnosis model, judging whether the diagnosis rate of the fault diagnosis model obtained is greater than a preset threshold, if so, outputting the final fault diagnosis model, otherwise adjusting the number of samples and retraining the BP neural network. The selected final target fault feature can be trained using an offline BP neural network training method to obtain a fault diagnosis model, and then the trained fault diagnosis model is used to perform fault diagnosis on the open circuit fault of the converter in the PMSG system. During the training process, the model parameters can be continuously adjusted, and the model parameter adjustment includes changing the number of training samples (appropriately increasing or decreasing the number of samples) and adjusting the nonlinear optimization function of the intermediate layer of the neural network training until the diagnosis rate reaches the preset condition, indicating that the most suitable fault feature has been found for fault diagnosis, and the accuracy and reliability of fault diagnosis can be ensured. It can be understood that other nonlinear networks other than the BP neural network can also be used to train the fault diagnosis model according to actual needs.

[0096] like Figure 2As shown, this embodiment is aimed at PMSG system converter open circuit fault diagnosis. First, based on the fault current characteristics, features that reflect the system state and may be distorted due to the fault are selected and added to the spare feature pool as spare features. After the spare features are three-dimensionally projected, it is first determined whether they are sensitive to the fault location. If they are determined to be sensitive, it is further determined whether they are sensitive to the wind speed. Fault features that are sensitive to the fault location and insensitive to the wind speed are selected and then cluster distance analysis is performed. The intra-class distance data is calculated to analyze the compression degree and distinguishability of the features, so as to combine the projection analysis and distance analysis results to screen out the final fault features, and input the final screened fault features into the BP neural network for model training until the training obtains a model that meets the preset accuracy conditions (accuracy reaches 99%) as a fault diagnosis model, which can achieve efficient and accurate PMSG system converter open circuit fault diagnosis.

[0097] To verify the effectiveness of the present invention, the above method of the present invention is used in a specific application embodiment to perform a diagnosis test on an open circuit fault of a PMSG system converter, wherein the PMSG topology system is as follows: Figure 3 As shown, the converter is an AC-DC-AC full-power structure, and the grid-side filter inductor L ks (k=a,b,c) is saturated and linear, the three-phase potential of the generator (e as 、e bs and e cs ) is symmetrical and is a three-phase stable pure sine wave potential. and is the three-phase stator current.

[0098] like Figure 4 The following are the three-phase current output waveform results under different switch open circuit fault conditions of the PMSG system side converter at a wind speed of 7m / s. The specific PMSG system parameters are: generator rated power 1.58MVA, rated voltage 0.69KV, rated frequency 50Hz, voltage coefficient 1500V, resistance 0.017pu, inductance 0.064pu, moment of inertia 2pu, DC side capacitance 200mH, grid equivalent resistance 0.2mH. Figure 5 The following are the three-phase current output waveform results under different switch open circuit fault conditions of the machine-side converter of the PMSG system at a wind speed of 10.5m / s. Figure 6 The following are the DQ current output waveform results under different switch open circuit fault conditions of the machine-side converter of the PMSG system at a wind speed of 7m / s. Figure 7 The following are the DQ current output waveform results under different switch open circuit fault conditions of the machine-side converter of the PMSG system at a wind speed of 10.5m / s. Figure 8 The results of projecting the sampled A, B, and C three-phase current DC components onto the orthogonal coordinates X, Y, and Z axes respectively.

[0099] like Fig. 9 For sampling I k , iR k , C, F FC , F msf , F vf , PSDC, (S, Φ, θ), (I c ,P DQ ), F ks The system sensitivity summary result diagram obtained after projection analysis of 11 features, the parameters correspond to the average value, effective value, autocorrelation function, cross-correlation function, frequency centroid, power spectrum density covariance, mean square frequency, probability density correlation coefficient, DQ combination feature 1 (area, vector angle, distribution angle of QD vector), DQ combination feature 2 (QD current vector centroid, distribution angle distribution range), DQ combination feature 3 (D, Q current vector probability density distribution range, DQ current vector angle probability density distribution range). For the convenience of description, the numbers are I k , iR k , C, F FC , F msf , F vf , PSDC, (S, Φ, θ), (I c ,P DQ ), F ks They correspond to 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, and 11 respectively. Fig. 9 The results of the sensitivity of the fault and wind speed are shown in Figure 2. By removing the features that do not meet the fault and wind speed sensitivity conditions, we can obtain the following: Fig.10 The intra-class distance curve is shown in Figure 2. Fig. 9 and Fig.10 It can be seen from the data that the feature samples for simultaneous diagnosis of single switch open circuit fault, same-phase double switch open circuit fault, non-same-phase double switch open circuit fault, etc. under multiple wind speed conditions need to be realized by combining the DC component with other features. PSDC and F msf The average value of the distance within the class is large, the compression degree is low, and the diagnostic effect is affected; the distance within class C is small, the compression degree is small, which is conducive to fault diagnosis. This shows that the present invention can effectively improve the fault diagnosis accuracy by combining the sensitive situation analysis of feature projection and distance analysis.

[0100] This embodiment further conducts experimental simulation on a platform with the following PMSG system parameters, wherein the generator rated power is 1.58MVA, rated voltage is 0.69KV, rated frequency is 50Hz, voltage coefficient is 1500V, resistance is 0.017pu, inductance is 0.064pu, moment of inertia is 2pu, DC side capacitance is 200mH, and grid equivalent resistance is 0.2mH. 40 groups of samples of the feature combination of DC component plus covariance are collected, and the nonlinear optimization function selects "logsig" and "triangdx". After BP neural network training, the obtained fault diagnosis result can reach 99.98%. This shows that the diagnostic method of the present invention has a very high fault diagnosis accuracy and can effectively improve the fault diagnosis accuracy.

[0101] The PMSG system converter open circuit fault diagnosis device of this embodiment includes:

[0102] A feature selection unit is used to select all fault features that cause the PMSG system state to change according to the operating state of the PMSG system at different fault locations when a machine-side converter fault occurs during the operation of the PMSG system, and add them to a backup feature pool;

[0103] A feature projection unit is used to perform three-dimensional projection on each fault feature in the backup feature pool, cluster each projected fault feature, judge the sensitivity of each fault feature to the fault location and wind speed according to the number of clusters formed by clustering, and screen out fault features that are sensitive to the fault location and insensitive to the wind speed;

[0104] A distance calculation unit is used to calculate the intra-class distance of each fault feature screened by the feature projection unit under different wind speed conditions and fault conditions;

[0105] A feature screening unit, configured to screen out a final target fault feature from each fault feature screened by the first feature screening unit according to a sensitivity state of the fault feature to the fault location type and the corresponding intra-class distance;

[0106] The model training unit is used to extract sample data of each target fault feature and train a nonlinear network to obtain a fault diagnosis model to realize open circuit fault diagnosis of the PMSG system converter.

[0107] The PMSG system converter open-circuit fault diagnosis device of this embodiment corresponds to the PMSG system converter open-circuit fault diagnosis method described above, and will not be described in detail again.

[0108] This embodiment also includes a computer device, including a processor and a memory, the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0109] This embodiment also includes a computer-readable storage medium storing a computer program, and when the computer program is executed, the method described above is implemented.

[0110] The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment, it is not intended to limit the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the scope of protection of the technical solution of the present invention.

Claims

1. A PMSG system converter open circuit fault diagnosis method, characterized in that the steps include: S1. According to the operating state of the PMSG system at different fault locations when a machine-side converter fault occurs during the operation of the PMSG system, all fault features that cause the PMSG system state to change are selected and added to the backup feature pool; S2. Perform three-dimensional projection on each fault feature in the backup feature pool, cluster each fault feature after projection, determine the sensitivity of each fault feature to the fault location and wind speed according to the number of clusters formed by clustering, and select fault features that are sensitive to the fault location and insensitive to the wind speed; S3. Calculate the intra-class distance of each fault feature selected in step S2 under different wind speed conditions and fault conditions; S4. According to the sensitivity of the fault feature to the fault location type and the corresponding intra-class distance, the final target fault feature is again selected from the fault features selected in step S2; S5. Extract sample data of each target fault feature, train a nonlinear network to obtain a fault diagnosis model, so as to realize open circuit fault diagnosis of the PMSG system converter.

2. The PMSG system converter open circuit fault diagnosis method according to claim 1, characterized in that: The step S1 comprises: Construct the model expression of the three-phase current of the PMSG drive system: Among them, i s represents the three-phase current vector, s represents the auxiliary logic vector, R represents the generator resistance vector, L represents the generator reactance vector, e represents the generator potential vector, c represents the preset coefficient value determined according to the initial state, u dc Indicates the DC side voltage value; When a machine-side converter fault occurs during the operation of the PMSG system, the fault feature corresponding to the data change in the model expression of the three-phase current of the PMSG drive system is selected and added to the spare feature pool.

3. The PMSG system converter open circuit fault diagnosis method according to claim 1, characterized in that: The fault features in the backup feature pool include three-phase features A, B, and C. The three-phase features A, B, and C of each fault feature are described in a 1*3 vector form. For the DQ current vector, a 1*3 form is formed by combining the amplitude and the vector angle. In step S2, the ABC three-phase features of each fault feature are projected to the xyz three-dimensional orthogonal coordinates respectively to complete the three-dimensional projection.

4. The PMSG system converter open circuit fault diagnosis method according to claim 1, characterized in that: The step S2 comprises: S201. Perform three-dimensional projection on the fault characteristics of a single fault type under the same wind speed, wherein the fault types include a single switch fault, a non-same-phase double switch fault, and a same-phase double switch fault, and determine whether the number of clusters formed by clustering after the three-dimensional projection is equal to the number of converter switches of the PMSG system. If so, it is determined that the current fault characteristics are sensitive to the fault location, and the process proceeds to step S202; otherwise, it is determined that the current fault characteristics are not sensitive to the fault location, and the process proceeds to step S203; S202. Perform three-dimensional projection on the fault features of a single fault type at two different wind speeds, and determine whether the number of clusters formed by clustering after the three-dimensional projection is equal to the number of converter switches of the PMSG system. If so, it is determined that the current fault feature is not sensitive to the wind speed; otherwise, it is determined that the current fault feature is sensitive to the wind speed and the process proceeds to step S203; S203. Delete the current fault feature from the backup feature pool.

5. The PMSG system converter open circuit fault diagnosis method according to claim 1, characterized in that: In step S3, the intra-class distance of the jth class is calculated according to the following formula: Among them, m j represents the feature mean vector of the jth class, n j is the number of characteristic samples contained in the jth class, and The feature i vector representing the jth class; The expression for calculating the total intra-class distance is: Among them, c represents the total number of feature categories, and N represents the total number of feature samples contained in the jth category.

6. The PMSG system converter open circuit fault diagnosis method according to any one of claims 1 to 5, characterized in that: In the step S4, the compression degree corresponding to the fault feature is determined according to the intra-class distance, and the fault feature that is sensitive to the fault location under the required fault diagnosis type and whose compression degree meets the preset requirements is selected as the final target fault feature.

7. The PMSG system converter open circuit fault diagnosis method according to any one of claims 1 to 5, characterized in that: The step S5 includes: using the sample data corresponding to the screened target fault characteristics to train the nonlinear network to obtain a fault diagnosis model, and judging whether the diagnosis rate of the obtained fault diagnosis model is greater than a preset threshold value, and if so, outputting the final fault diagnosis model; otherwise, adjusting the number of samples or adjusting the nonlinear optimization function of the intermediate layer of the nonlinear network training, and retraining the nonlinear network.

8. A PMSG system converter open circuit fault diagnosis device, characterized in that: include: A feature selection unit is used to select all fault features that cause the PMSG system state to change according to the operating state of the PMSG system at different fault locations when a machine-side converter fault occurs during the operation of the PMSG system, and add them to a backup feature pool; A feature projection unit is used to perform three-dimensional projection on each fault feature in the backup feature pool, cluster each projected fault feature, judge the sensitivity of each fault feature to the fault location and wind speed according to the number of clusters formed by clustering, and screen out fault features that are sensitive to the fault location and insensitive to the wind speed; A distance calculation unit is used to calculate the intra-class distance of each fault feature screened by the feature projection unit under different wind speed conditions and fault conditions; A feature screening unit, configured to screen out a final target fault feature from the fault features screened out by the feature projection unit according to a sensitivity state of the fault feature to the fault location type and the corresponding intra-class distance; The model training unit is used to extract sample data of each target fault feature and train a nonlinear network to obtain a fault diagnosis model to realize open circuit fault diagnosis of the PMSG system converter.

9. A computer device comprising a processor and a memory, wherein the memory is used to store a computer program, wherein: The processor is configured to execute the computer program to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed, the method according to any one of claims 1 to 7 is implemented.

Images