Data-driven fault identification method and equipment based on fault characteristics of wind turbine generator

By building a controlled current source model and a CNN-GRU fault identification model, the problem of inverse power failure identification is solved, high-precision fault type and position identification is achieved, and the power grid fault diagnosis capability is improved.

CN118940638AActive Publication Date: 2024-11-12NORTH CHINA ELECTRIC POWER UNIV
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411096291.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-12
Publication Date
2024-11-12
Estimated Expiration
2044-08-12

AI Technical Summary

Technical Problem

The existing technology is difficult to accurately identify and locate inverse power failures, especially in permanent magnet direct drive wind turbines. The fault characteristics rely too much on control strategies and parameters, resulting in difficulty in calculating short-circuit current in actual projects, affecting the power grid relay. Protective performance.

Method used

Build a controlled current source model, combine it with the CNN-GRU fault identification model, simulate the fault condition through the simulation model, obtain electrical quantity data, and build a fault sample set to achieve high-precision identification of fault types and locations.

Benefits of technology

It improves the accuracy and positioning accuracy of fault recognition, can be recognized at 100% fault type and fault locations of faults of more than 94%, which is better than traditional models and enhances the grid fault diagnosis capability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940638A_ABST
    Figure CN118940638A_ABST
Patent Text Reader

Abstract

The invention discloses a data-driven fault identification method and equipment based on fault characteristics of a wind turbine generator set, and the method comprises the steps: taking a permanent magnet direct drive type wind driven generator (PMSG) in an inverter power supply as a research object, and taking the fault characteristics of the permanent magnet direct drive type wind driven generator into consideration based on a control strategy; a simulation model of a multi-unit gathering and sending-out line is constructed through a simplified modeling method, an electrical quantity change rule of an output line under the condition of a power grid fault steady state is analyzed on the basis of the simulation model, and a CNN-GRU data driving type power grid fault identification network model suitable for fault electrical quantity information of the multi-unit gathering and sending-out line is constructed according to the change rule. And the accuracy of line fault diagnosis and positioning is further improved through deep learning. According to the data-driven fault identification method based on the fault characteristics of the wind turbine generator, modeling processing is simplified, the fault identification accuracy is high, and positioning is accurate.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine fault analysis, and in particular to a data-driven fault identification method and device based on wind turbine fault characteristics. Background Art

[0002] The scale of full-power inverter power supplies represented by photovoltaic power supplies (PV) and permanent magnet direct-drive wind turbines (PMSG) in my country has expanded rapidly. The biggest feature of inverter power supplies is that they are interconnected with the power grid through converters. When the external hardware parameters and fault conditions are certain, their fault characteristics depend to a large extent on the control strategy and control parameters of the converter. High-proportion new energy sources such as wind power and photovoltaics are connected to the power grid in the form of centralized power sources or distributed power sources. Their fault current characteristics are quite different from those of traditional synchronous generators. The adaptability of traditional AC system relay protection principles in new energy grid-connected systems has become a hot topic and key research topic in today's power system. Accurate and reliable fault identification methods are crucial to improving relay protection performance and ensuring the safe and stable operation of new power systems.

[0003] After the inverter power supply is connected to the power system on a large scale, the topology and fault characteristics of the fault network are changed. The complex short-circuit current fed by the inverter power supply will bring many adverse effects to the relay protection of the power grid, thereby threatening the safe and stable operation of the power system. At the same time, in the literature "Gao Jingyi, Weng Hanli, Lin Xiangning, et al. Analysis of the performance degradation mechanism of the power frequency variable distance protection on the full-power inverter power supply side [J / OL]. Electric Power Automation Equipment, 1-13.", considering that the analytical research on the fault current fed by the inverter power supply is too dependent on the control strategy and parameters, and the "black box" of the parameters in the actual project makes the analytical expression difficult to use for the short-circuit current calculation in the actual project, the in-depth analysis of the fault current characteristics of the inverter power supply is of great practical significance for accurately evaluating its impact on the relay protection of the power grid and improving and perfecting the protection performance. In the document "Kuang Xiaoyun, Fang Yu, Guan Hongbing, et al. Full-time domain short-circuit current calculation method for networks containing new energy inverter power supplies [J]. Electric Power Automation Equipment, 2020, 40(05): 113-122.", starting from the dynamic response characteristics of the inverter controller, a calculation model for the short-circuit current of the inverter power supply under symmetrical and asymmetrical faults is established, and the fault current characteristics of the inverter power supply are deeply analyzed; in the document "Jia Ke, Yang Zhe, Zhao Qijuan, et al. High-frequency sudden change distance protection for the transmission line of the new energy station [J]. Power System Technology, 2019, 43(09): 3271-3280.", a full-time domain short-circuit current calculation method based on the steady-state and transient time scales of the inverter power supply fault is proposed. The fault steady-state calculation method regards the inverter power supply as a voltage-controlled current source, which solves the problem that the traditional steady-state short-circuit current calculation method cannot fully consider the control strategy and nonlinear characteristics of the inverter power supply.

[0004] Inverter-type power supply electric field tie lines usually use current differential protection as their main protection, and staged distance protection and zero-sequence current protection as their backup protection. Among them, distance protection is greatly affected by the complex fault characteristics of inverter-type power supply, and the potential safety hazards are prominent, which is also the difficulty and focus of research. Taking into account the weak feedback characteristics of the inverter power supply and the instability of the equivalent impedance angle after grid connection, which may lead to false operation or refusal of distance protection, in the literature "Kong Xiangping, Zhang Zhe, Yin Xianggen, et al. Research on fault current characteristics and fault analysis methods of power grids containing inverter distributed power sources [J]. Proceedings of the CSEE, 2013, 33(34): 65-74+13.", researchers introduced the high-frequency sudden change distance protection principle suitable for the access of new energy sources, and in the literature "Fang Y, Jia K, Yang Z, et al. Impact of Inverter-Interfaced Renewable Energy Generators on Distance Protection and an Improved Scheme [J]. IEEE Transactions on Industrial Electronics, 2018, PP(99): 1-1.", they provided effective suggestions for the restoration of the fault response capability of the power frequency variable distance protection when the inverter power supply is connected to the grid by using measures such as delay elements on the source side of the inverter power supply and zero-sequence distance protection relay. The document "Chang Nana, Song Guobing, Chang Zhongxue, et al. Adaptive coordination scheme for single-ended backup protection based on transient characteristic distribution [J]. Automation of Electric Power Systems, 2023, 47(24): 143-155." proposes that the traditional single-ended backup protection adopts step time coordination, but its setting task is heavy and it is difficult to adapt to the random access scenario of inverter-type power sources. The document "Shen S, Lin D, Wang H, et al. An Adaptive Protection Scheme for Distribution Systems With DGs Based on Optimized Thevenin Equivalent Parameters Estimation [J]. IEEE Transactions on Power Delivery, 2017, 32(99): 411-419." proposes an adaptive multi-level time-limited overcurrent protection scheme suitable for renewable energy power sources. The coordination scheme of the main and backup protection can meet the requirements of selectivity and sensitivity at the same time. However, when the main protection of the staged current protection refuses to operate, the backup protection close to the power supply side takes a long time to operate, and the backup protection operation performance is poor.

[0005] When the new energy station is connected to the grid, the permanent magnet direct-drive wind turbine adopts the PQ (Power-Quality) control strategy to achieve active current and reactive current tracking reference value changes, and then output reference power. After the transmission line fails, in order to avoid the problem of system power imbalance at the moment of grid failure, the state stipulates that dynamic reactive current must be injected into the power system when the wind farm station fails, and the wind turbine must have low voltage ride through (LVRT) characteristics. Considering the overcurrent limiting conditions of the full inverter power supply, when an asymmetric fault occurs in the power grid, in order to ensure the symmetry of the three-phase current and alleviate the problems of poor performance of the wind turbine and overheating of the device, the permanent magnet direct-drive wind turbine usually simultaneously puts into the negative sequence current suppression strategy. Considering that the fault characteristics of permanent magnet direct-drive wind turbines are too dependent on control strategies and parameters, and the "black box" of parameters in actual projects makes it difficult to use the fault characteristics for short-circuit current calculations in actual projects, currently only researchers have derived analytical expressions for the fault current of permanent magnet direct-drive wind turbines based on control strategies and operating characteristics, but there is no study that has constructed a fault identification model for multi-unit transmission lines under fault steady-state conditions that takes into account the fault characteristics of permanent magnet direct-drive wind turbines through simplified wind turbine modeling methods. Summary of the invention

[0006] In order to solve the above problems in the prior art, the present invention provides a data-driven fault identification method based on wind turbine fault characteristics, which simplifies modeling and processing, has high fault identification accuracy and accurate positioning.

[0007] In order to achieve the above-mentioned object of the invention, the present invention provides a data-driven fault identification method based on the fault characteristics of a wind turbine generator set, comprising the following steps: Based on the low voltage ride-through strategy and the negative sequence current suppression strategy, a controlled current source model of a permanent magnet direct-drive wind turbine is constructed; a multi-unit transmission line includes the permanent magnet direct-drive wind turbine; the controlled current source model is: ; ; In the formula, is the short-circuit current amplitude, is the rated current, is the per unit voltage of the fan side, is the impedance phase angle on the PMSG side, pu is the per-unit value; Constructing a simulation model of the multi-unit collection and transmission line according to the controlled current source model and the multi-unit collection and transmission line; Acquire a fault input data set and input it into the simulation model, and run the simulation model so that a simulated fault occurs in the simulation model; Collecting a sample set of electrical quantity data of the output line in the simulation model within a preset time window after the simulated fault occurs; According to the electrical quantity data sample set, obtaining the fault output data set; Obtaining a fault sample set of the simulation model according to the fault input number set and the fault output number set; Constructing a CNN-GRU fault identification model based on the simulation model according to the fault sample set; Fault identification is performed on the multi-unit transmission line according to the CNN-GRU fault identification model.

[0008] Further, the electrical quantity data sample set is a set of electrical quantity data samples of the output line; the electrical quantity data samples include three-phase voltage effective values ​​and three-phase current effective values; the step of obtaining the fault output number set is: processing the electrical quantity data sample set according to a normalization formula to obtain the fault output number set, the fault output number set is a set of electrical quantity data final values; the normalization formula is: ; In the formula, is the maximum value of the electrical quantity data sample, is the minimum value of the electrical quantity data sample, The electrical quantity data samples currently processed by the normalization formula are: It is the final value of electrical quantity data.

[0009] Furthermore, the three-phase voltage effective value is: ; In the formula, The first Sampling points Phase voltage instantaneous value, M is the number of sampling points in one power frequency cycle, for Phase voltage effective value, is the order of sampling points within a power frequency cycle, is the phase in three-phase electricity, Phase in turn takes the three-phase power a Mutually, b Mutually, c Mutually.

[0010] Furthermore, the three-phase current effective value is: ; In the formula, The first Sampling points Phase current instantaneous value,M is the number of sampling points in one power frequency cycle, for Phase current effective value, is the order of sampling points within a power frequency cycle, is the phase in three-phase electricity, Phase in turn takes the three-phase power a Mutually, b Mutually, c Mutually.

[0011] Furthermore, the preset time window is 0.45s to 0.5s after the fault occurs.

[0012] Furthermore, the fault sample set includes a training set and a test set; The training set is used to construct the CNN-GRU fault identification model; The test set is used to test the CNN-GRU fault identification model constructed according to the training set.

[0013] Furthermore, the CNN-GRU fault identification model includes a fault type identification unit and a fault location identification unit; the fault type identification unit is used to identify the fault type of the multi-unit transmission line, and the fault location identification unit is used to identify the fault location of the multi-unit transmission line; Based on the fault type, the fault sample set includes a training set 1 and a test set 1, and the fault type identification unit is obtained according to the training set 1 and the test set 1; Based on the fault location, the fault sample set includes a training set 2 and a test set 2, and the fault location identification unit is obtained according to the training set 2 and the test set 2.

[0014] Further, the simulation model includes a fault simulation module and a program editing module; The program editing module is used to convert the fault input data set into a recognizable signal and transmit it to the fault simulation module; The fault simulation module is used to generate the simulated fault according to the identifiable signal converted from the fault input data set, and obtain the fault output data set according to the simulated fault condition. Further, the data format of the fault type of the controlled current source model is in binary form.

[0015] An electronic device, used to implement the data-driven fault identification method based on the fault characteristics of a wind turbine generator set, comprising a program editing module, a fault simulation module, a data monitoring module, a data analysis module, a data storage module and a fault identification module: The program editing module is used to convert the fault input data set into a recognizable signal and transmit it to the fault simulation module; The fault simulation module is used to receive the identifiable signal of the fault input data set of the program editing module and run a simulated fault; The data monitoring module is used to monitor the fault simulation module and output the electrical quantity data sample during the fault simulation process; The data parsing module is used to process the electrical quantity data samples and obtain a fault output number set based on the fault input number set, and to generate the fault sample set according to the fault input number set and the fault output number set; and to construct the CNN-GRU fault identification model according to the fault sample set; The data storage module is used to store the CNN-GRU fault identification model; The fault identification module is used to monitor the multi-unit transmission line and perform fault identification on the multi-unit transmission line according to the CNN-GRU fault identification model stored in the data storage module.

[0016] The beneficial effects of the present invention are: The present invention intervenes from the time scale of traditional single-ended backup protection, takes the permanent magnet direct-drive wind generator (PMSG) in the inverter power supply as the research object, and considers that the analytical research on the fault current of the permanent magnet direct-drive wind generator is too dependent on the control strategy and parameters, and the "black box" of the parameters in the actual engineering makes it difficult to use the analytical expression for the short-circuit current calculation of the permanent magnet direct-drive wind generator in the actual engineering. A controlled current source model taking the control strategy into account is constructed, and a simulation model of the multi-unit transmission line is constructed based on the controlled current source model, so as to realize the simplified modeling processing of the high-order mathematical model system.

[0017] Based on the simulation model, the present invention deeply analyzes the changing rules of the electrical quantity data of the grid connection point under different fault conditions in the steady-state state of the power grid, builds a CNN-GRU fault identification model suitable for the fault sample set of the multi-unit transmission line, and obtains the processed values ​​of the three-phase voltage effective value and the three-phase current effective value of various fault conditions under different positions and different transition resistances of the output line by batch simulation means as the fault identification sample set, and further improves the accuracy of fault diagnosis and positioning of the multi-unit transmission line through artificial intelligence deep learning algorithm.

[0018] The CNN-GRU fault identification model provided by the present invention can better realize fault type identification, and the identification accuracy of the fault type can reach 100%; the CNN-GRU fault identification model has a strong spatiotemporal feature extraction capability, and the fault identification accuracy is higher than that of the traditional fault identification model, and can effectively realize fault location identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1It is a schematic diagram of the process of the present invention; Figure 2 It is a schematic diagram of the overall structure of the controlled current source model in the present invention; Figure 3 It is a schematic diagram of the internal structure of the simplified fan module of the present invention; Figure 4 Schematic diagram of the steady-state output results of the controlled current source model: (a) is the output voltage waveform, (b) is the output current waveform, (c) is the output current sequence component, and (d) is the output power; Figure 5 It is a schematic diagram of the simulation model in the present invention; Figure 6 The waveform diagram of the short-circuit fault result of the simulation model in the present invention: (a) is the voltage waveform of the three-phase short-circuit grid-connected point, (b) is the current waveform of the three-phase short-circuit grid-connected point, (c) is the voltage waveform of the A-phase grounding grid-connected point, and (d) is the current waveform of the A-phase grounding grid-connected point; Figure 7 It is a schematic diagram of the improved three-phase fault generator module in the present invention; Figure 8 It is a structural diagram of the CNN-GRU fault identification model in the present invention; Fig. 9 This is a schematic diagram of the fault identification process of the present invention; Fig.10 This is the training process of the CNN-GRU fault identification model in the present invention for the fault type; Fig.11 It is the identification result of the CNN-GRU fault identification model for the fault type in the present invention; Fig.12 This is the training process of the CNN-GRU fault identification model in the present invention for the fault location; Fig.13 It is the identification result of the fault location by the CNN-GRU fault identification model in the present invention; Fig.14 Comparison of fault location identification results of different fault identification models in the prior art: (a) SVM fault identification model, (b) GRU fault identification model, and (c) CNN fault identification model.

[0020] Among them, the figure markings are: 1. Wind turbine simplification module; 2. Inverter port electrical quantity measurement module; 3. Wind turbine outlet feeder; 4. AC power grid; 5. Wind turbine control link; 6. Controlled current source; 7. RL line; 8. Controlled current source output measurement module; 9. Electrical quantity data sample calculation module; 10. Delay link. DETAILED DESCRIPTION

[0021] The complexity and variability of the fault characteristics of inverter power supply after grid connection require that the power grid fault identification technology can quickly and accurately perform fault diagnosis and location. Traditional power grid fault identification methods mostly use switch quantity as the fault feature set and complete fault identification through expert system, Petri net, Bayesian network and other algorithms. See the literature "Zhou Ziguan, Bai Xiaomin, Li Zaihua, et al. Intelligent transmission network fault diagnosis method using knowledge grid technology [J]. Proceedings of the CSEE, 2010, 30(04): 8-15", "Xie Hongtao, Tong Xiaoyang. Comprehensive power grid fault diagnosis method based on hierarchical fuzzy Petri net [J]. Power System Technology, 2012, 36(01): 246-252.", "He Xiaofei, Tong Xiaoyang, Sun Mingwei. Distributed power grid fault diagnosis based on Bayesian network and DS evidence theory [J]. Automation of Electric Power Systems, 2011, 35(10): 42-47.". With the development of power system data acquisition systems, the grid dispatching end can obtain various types of data information. Electrical quantity information such as voltage and current is superior to switch quantity information in terms of completeness and accuracy. Deep learning has a strong ability to extract data features and can autonomously extract features from fault samples. Therefore, artificial intelligence deep learning algorithms based on power big data can be applied in the field of power grid fault identification. In the literature "Xu Shuwei, Qiu Caiming, Zhang Dongxia, et al. Transmission line fault type identification based on deep learning [J]. Proceedings of the CSEE, 2019, 39(01): 65-74+321.", a convolutional neural network model is constructed to diagnose the fault of the photovoltaic power generation system with the OU process. The accuracy of the diagnosis result is much higher than that of traditional machine learning algorithms such as support vector machine and random forest, and the accuracy is almost unaffected by the fault location and category; in the literature "Wei Dong, Gong Qingwu, Lai Wenqing, et al. Research on the method of fault judgment and fault phase selection of transmission lines based on convolutional neural network [J]. Proceedings of the CSEE, 2016, 36(S1): 21-28.", a convolutional neural network fault diagnosis model with dual classification channels is established with the fault current signal of the two-terminal AC system as the fault feature set, which solves the two non-independent classification problems of fault judgment and fault phase selection; in the literature "Sun Hao. Research on the method of AC and DC power grid fault diagnosis based on deep learning [D]. Beijing Jiaotong University, 2020.” In this paper, the full bus electrical quantity information of the AC / DC hybrid power grid model is collected as the fault feature set, and a deep belief network fault identification model integrating the dimensionality reduction function of the automatic encoder is constructed to realize the application of deep learning in the fault identification of AC / DC hybrid power grid.

[0022] Based on the above research status, from the time scale intervention of the traditional single-end backup protection, the permanent magnet direct-driven wind turbine in the inverter power supply is taken as the research object, and the steady-state equivalent model of the wind farm transmission line fault considering the control strategy and fault current characteristics is constructed to realize the simplified modeling processing of the high-order mathematical model system. Based on the above model, the variation law of the machine-side electrical quantity under the condition of power grid fault is deeply analyzed, and the CNN-GRU (Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU)) data-driven power grid fault identification model suitable for fault electrical quantity information is constructed. The accuracy of wind farm transmission line fault diagnosis and positioning is further improved through artificial intelligence deep learning algorithm, which provides a reasonable reference for the subsequent research of new backup protection coordination schemes.

[0023] In order to clearly illustrate the technical features of this solution, this solution is described below through a specific implementation method.

[0024] See also Figure 1 The embodiment of the present invention provides a data-driven fault identification method based on wind turbine fault characteristics, comprising the following steps: Based on the low voltage ride-through strategy and negative sequence current suppression strategy, a controlled current source model of a permanent magnet direct-drive wind turbine generator is constructed; the multi-unit transmission line includes a permanent magnet direct-drive wind turbine generator; the controlled current source model is: ; ; In the formula, is the short-circuit current amplitude, is the rated current, is the per unit voltage of the fan side, is the impedance phase angle on the PMSG side, pu is the per-unit value; The controlled current source model can effectively simulate the fault characteristics of the permanent magnet direct-drive wind turbine generator in the steady-state condition of fault, and can also simulate the operating characteristics of the permanent magnet direct-drive wind turbine generator in the steady-state condition of non-fault condition. A simulation model of the multi-unit convergence transmission line is constructed based on the controlled current source model and the multi-unit convergence transmission line. The simulation model is used to analyze the variation law of the output line electrical quantity of the multi-unit convergence transmission line in the steady-state condition of fault, and can also analyze the variation law of the output line electrical quantity of the multi-unit convergence transmission line in the steady-state condition of non-fault condition; Obtaining a fault input data set and inputting it into a simulation model, and running the simulation model so that a simulated fault occurs in the simulation model; Collecting an electrical quantity data sample set in the simulation model within a preset time window after the simulated fault occurs, the electrical quantity data sample set is a collection of electrical quantity data samples of the output line; the electrical quantity data samples include three-phase voltage effective values ​​and three-phase current effective values; According to the electrical quantity data sample set, a fault output number set is obtained; according to the fault input number set and the fault output number set, a fault sample set of the simulation model is obtained; According to the fault sample set, a CNN-GRU fault identification model based on the simulation model is constructed, namely, a CNN-GRU data-driven power grid fault identification network model; Fault identification is performed on multi-unit transmission lines based on the CNN-GRU fault identification model.

[0025] Taking the permanent magnet direct-drive wind turbine in the inverter power supply as the research object, the control strategy adopted by the permanent magnet direct-drive wind turbine is analyzed. For a fixed wind turbine, its control strategy is fixed, and it will respond to a specific current for the voltage disturbance brought by the outside world. Therefore, the fault current and fault voltage have a certain mapping relationship and can be equivalent to a controlled current source model. In this embodiment, the control strategy is a low voltage ride-through control strategy and a negative sequence current suppression control strategy. Based on the control strategy, the feed short-circuit current expression of the permanent magnet direct-drive wind turbine is derived, that is, the controlled current source model of the permanent magnet direct-drive wind turbine is constructed; the steps of constructing the controlled current source model are: First, the negative sequence current control strategy of permanent magnet direct-drive wind turbine is analyzed: When the power grid is operating normally, the inverter power supply adopts the PQ control strategy. When a symmetrical or asymmetrical short circuit occurs in the power grid, the power outer loop is locked and the inverter , The reference value is given. To derive the dq axis current expression, firstly, the three-phase voltage and current of the inverter port are transformed into dq, see equations (1)-(3): (1) (2) (3) In the formula, , , is the three-phase voltage at the inverter port, is the d-axis component of the inverter port voltage, is the q-axis component of the inverter port voltage, T is the dq transformation matrix, , , Output three-phase current for the inverter, is the d-axis component of the inverter output current, is the q-axis component of the inverter output current, is the phase-locked loop phase angle; Considering the three-phase abc coordinate system, the three-phase voltage equation at the inverter port is (4): (4) In the formula, , , is the three-phase voltage at the inverter grid connection point, is the inverter filter inductor, is the parasitic resistance of the inverter filter inductor; According to equations (1) and (3), the three-phase voltage of the inverter grid connection point is transformed into dq, and then the following equation can be obtained: , , in the dq coordinate system, , , , Substituting into equation (4), we can get the inverter port voltage equation as equation (5): (5) In the formula, is the d-axis component of the inverter port voltage, is the q-axis component of the inverter port voltage, is the d-axis component of the inverter grid-connected point voltage, is the q-axis component of the inverter grid-connected point voltage, is the inverter output voltage angular frequency, is the d-axis component of the inverter output current, is the q-axis component of the inverter output current, which is obtained by dq transformation of the inverter port voltage and current; (6) In the formula, is the current inner loop proportional adjustment coefficient, is the current inner loop integral adjustment coefficient, is the reference value of the d-axis component of the output voltage, is the reference value of the q-axis component of the output voltage, is the reference value of the current d-axis component, is the reference value of the current q-axis component, For the integration phase, is the voltage amplitude of the inverter port; Based on this, the inverter port voltage equation (Equation (5)) and the current inner loop control equation (Equation (6)) are combined to obtain the inverter d and q axis current differential equations during the fault period: (7) In the formula, is the reference value of the d-axis component of the positive sequence current, is the reference value of the q-axis component of the positive sequence current; (8) In the formula, , , , is the coefficient to be determined.

[0026] By analyzing the current expression, it can be obtained that the inverter d and q axis currents after the fault are composed of two groups of decaying exponential components and one group of constant constant components. Since the time constant of the exponential decay component is short, the d and q axis current components decay rapidly to their reference values. It can be considered that the inverter power supply output short-circuit current can quickly follow its reference value, which is: (9) According to the national standard, distributed power sources must have a certain low voltage ride-through capability, and when the positive sequence voltage at the wind turbine grid connection point drops to 20%, the wind turbine needs to remain connected to the grid and run continuously for 625ms. Under the low voltage ride-through control strategy, the wind turbine-side inverter immediately disconnects the PQ outer loop. In response to the reactive current support strategy, according to the degree of grid connection point voltage drop after the fault, the reference value of the positive sequence inner loop should satisfy equation (10).

[0027] (10) Where: To control the positive sequence inner loop reactive current reference value; To control the positive sequence inner loop active current reference value; is the per unit voltage at the fan side; It is the reference active power per unit value during normal operation; When <0.2pu, the fan is disconnected from the grid.

[0028] Secondly, short-circuit current analysis of permanent magnet direct-drive wind turbine fault steady state: When a fault occurs, the terminal voltage of the permanent magnet direct-drive wind turbine generator drops, and the q-axis current value representing the reactive power gradually increases according to the voltage drop degree, while the d-axis current value representing the active power first needs to maintain the active power during normal operation, but after the terminal current reaches the maximum 2 times the rated current specified by the inverter, the d-axis current gradually decreases. At this time, the short-circuit current fed by the permanent magnet direct-drive wind turbine generator satisfies formula (11).

[0029] (11) In the formula is the short-circuit current amplitude, is the side impedance phase angle of permanent magnet direct-drive wind turbine generator, is the rated current.

[0030] The short-circuit current fed by the permanent magnet direct-drive wind turbine generator can be calculated as: (12) (13) In the formula, is the short-circuit current amplitude, is the rated current, is the per unit voltage of the fan side, is the impedance phase angle on the PMSG side, pu is the per-unit value; Combining equations (12) and (13), it can be seen that the amplitude and phase of the short-circuit current fed by the permanent magnet direct-drive wind turbine generator mainly depend on the voltage drop degree and the low voltage ride-through strategy selected. When the current does not reach the limit value during the fault, the permanent magnet direct-drive wind turbine generator can be regarded as a voltage-controlled current source model with both amplitude and phase controlled. When the current reaches the limit value, it can be regarded as a voltage-controlled current source model with only phase controlled in the short-circuit calculation. That is, equations (12) and (13) can be regarded as a controlled current source model of a permanent magnet direct-drive wind turbine generator constructed based on the low voltage ride-through strategy and the negative sequence current suppression strategy.

[0031] In this embodiment, a voltage-controlled current source model of a permanent magnet direct-drive wind turbine is constructed in MATLAB / Simulink, such as Figure 2-4 shown. Figure 2 The overall structure of the controlled current source model of a permanent magnet direct-drive wind turbine generator includes a simplified wind turbine module 1, an inverter port electrical quantity measurement module 2, a wind turbine outlet feeder 3, and an AC power grid 4. The simplified wind turbine module 1 consists of Figure 3 As shown in the figure, the inverter port electrical quantity measurement module 2 is used to collect the three-phase voltage and the three-phase current of the inverter port. The fan outlet feeder 3 is used to connect the fan simplified module 1 and the AC power grid 4. Considering that the outlet voltage of the permanent magnet direct-drive wind turbine generator is generally 690V, the controlled current source model of the permanent magnet direct-drive wind turbine generator is connected to the 690V, 50Hz AC power supply through a transmission line, wherein the transmission line parameters R=0.12Ω / km, L=1e-4 H / km, and the line length is 10km, so as to realize the complete circuit construction of the controlled current source model. Figure 3 The internal structure of the simplified wind turbine module 1 is shown in FIG. 5 , wherein the internal modularization of the wind turbine control link 5 is constructed according to formulas (1) to (11) and the overall packaging is the wind turbine control link 5. In the present invention, the permanent magnet direct-drive wind turbine generator is simplified to a controlled current source model, so three controlled current sources are selected to simulate a , b , cThree-phase current, considering that the controlled current source needs to run in parallel with the resistor, a resistor is connected in parallel to each controlled current source to ensure its normal operation. Figure 3 In the controlled current source 6, in order to ensure that the environment of the controlled current source 6 is a basic circuit (the controlled current source 6 cannot operate independently), an RL connection line needs to be placed between the controlled current source 6 and the subsequent controlled current source output measurement module 8. Since this controlled current source 6 is a voltage-controlled current source, it is necessary to collect the output voltage of the controlled current source 6 as its own input variable. Therefore, a controlled current source output measurement module 8 is set here to collect its output voltage. At the same time, the right side of the controlled current source output measurement module 8 a , b , c The port is used to transmit the output current of the controlled current source 6 (because Figure 2 The simplified module 1 of the medium-sized wind turbine is obtained by encapsulation of this module, so the right side of this measurement module a , b , c Ports and Figure 2 On the right side of the simplified module 1 of the blower a , b , c port is the same port); Figure 3 The controlled current source output measurement module 8 and the electrical quantity data sample calculation module 9 are used to calculate or verify the instantaneous values ​​of the three-phase voltage and the three-phase current, and then calculate the effective values ​​of the three-phase voltage and the three-phase current.

[0032] In addition, in order to avoid the occurrence of algebraic loops during the fault simulation process, a delay link 10 is added to ensure the correctness of the fault simulation.

[0033] The steady-state output of the controlled current source model is as follows: Figure 4 As shown by Figure 4 It can be seen that the controlled current source model can effectively output three-phase symmetrical voltage and current under steady-state conditions, and effectively output active power without outputting reactive power. That is, the controlled current source model can better simulate the steady-state operation process of permanent magnet direct-drive wind turbines.

[0034] Build in MATLAB / Simulink Figure 5The simulation model of the multi-unit transmission line shown in the figure. Among them, PMSG1, PMSG2, and PMSG3 are all controlled current source models of permanent magnet direct-drive wind turbines constructed by considering the low voltage ride-through strategy and the negative sequence current suppression strategy. After passing through the 0.69 / 35kv box transformer, each controlled current source model is respectively collected to the 35kv busbar with 1km, 10km, and 25km long lines, and then connected to the system 35 / 220kv main transformer, and then connected to the 220kV, 50Hz AC system through a total length of 100km overhead line. Among them, the wind farm transmission line is a full-line overhead line with a voltage level of 220kV and a total length of 100km. The positive sequence parameters r1= 0.036Ω / km, l1= 1.62e-3 H / km, c1= 9.08e-9 F / km, and the zero sequence parameter is 3 times the positive sequence parameter.

[0035] In order to verify the effectiveness and reliability of the simulation model of the multi-unit transmission line built in this paper under the condition of fault steady state, the three-phase short circuit fault (10km away from the grid connection point, transition resistance 30Ω) and A-phase ground short circuit fault (20km away from the grid connection point, transition resistance 20Ω) at different locations of the 220kV AC transmission line were simulated respectively. On this basis, the voltage at the PMSG grid connection point ( Figure 5 middle , that is, the instantaneous value of three-phase voltage), current ( Figure 5 middle The three-phase current instantaneous value) information and the voltage drop information of each PMSG outlet are analyzed. The results are as follows Figure 6 shown.

[0036] Depend on Figure 6 It can be seen that when a three-phase short circuit with a transition resistance of 20Ω occurs 10km away from the grid-connected point of the controlled current source model, the three-phase voltage at the grid-connected point drops rapidly and enters the fault steady-state stage after about 0.022s; when a phase A ground short circuit fault with a transition resistance of 30Ω occurs 20km away from the wind turbine grid-connected point, the grid-connected point voltage drops asymmetrically, but due to the use of a negative-sequence current suppression control strategy, the grid-connected point current can still maintain three-phase symmetry after the fault occurs, the output negative-sequence current in the fault steady-state can be approximated to 0, and the fault transient time is relatively short.

[0037] The steps of obtaining the fault output data set are as follows: the electrical quantity data sample set is processed according to the normalization formula to obtain the fault output data set, which is the set of final values ​​of the electrical quantity data; considering that the voltage and current data in the electrical quantity data sample set have different dimensions, in order to reduce the impact of different electrical quantity data on the training differences of the simulation model, improve the generalization performance of the network and accelerate the convergence to the optimal solution, the busbar electrical quantity data samples collected within the time window need to be normalized; the normalization formula is: ; In the formula, is the maximum value of the electrical quantity data sample, is the minimum value of the electrical quantity data sample, The electrical quantity data samples currently processed by the normalization formula are: It is the final value of electrical quantity data.

[0038] Furthermore, the effective value of the three-phase voltage is: ; In the formula, The first Sampling points Phase voltage instantaneous value, M is the number of sampling points in one power frequency cycle, for Phase voltage effective value, is the order of sampling points within a power frequency cycle, is the phase in three-phase electricity, Phase in turn takes the three-phase power a Mutually, b Mutually, c Mutually.

[0039] Furthermore, the three-phase current effective value is: ; In the formula, The first Sampling points Phase current instantaneous value, M is the number of sampling points in one power frequency cycle, for Phase current effective value, is the order of sampling points within a power frequency cycle, is the phase in three-phase electricity, Phase in turn takes the three-phase power a Mutually, b Mutually, c Mutually.

[0040] In some preferred embodiments, the preset time window is 0.45s to 0.5s after the fault occurs. Taking into account the actual requirements of fault diagnosis time and network generalization, the model in this preferred embodiment uses a sampling frequency of 10kHz and needs to collect data within a preset time window of 0.3s after the fault occurs. The preferred preset time window is 0.45s to 0.5s after the fault occurs.

[0041] In some preferred embodiments, the fault input data includes the fault type of the controlled current source model and the fault location of the simulation model. In order to ensure the applicability of the simulation model to power grid fault diagnosis, multiple groups of different transition resistance values ​​are set as interference to expand the dimension of the fault sample set. In some preferred embodiments, the simulation model uses the randperm function to randomly generate 8 fault distances in each section. When 7 fault resistances are known as interference items, batch simulation faults are performed for 10 fault types and three-phase current effective values ​​and three-phase voltage effective values ​​are calculated. A total of 1680 groups of data are obtained in the fault sample set, and then the fault sample set is classified and constructed according to the fault type and fault location. The number of samples of each type is shown in Tables 1 and 2.

[0042] Table 1 Classification statistics of fault sample sets based on fault types Table 2 Classification statistics of fault sample sets based on fault location In some preferred embodiments, the fault sample set includes a training set and a test set; The training set is used to build the CNN-GRU fault identification model; The test set is used to test the CNN-GRU fault identification model built based on the training set.

[0043] Taking Tables 1 and 2 as examples, the fault sample set obtained through the above steps includes 1680 groups of data. In order to realize the subsequent construction and testing of the CNN-GRU fault identification model, the fault sample set is divided into a training set and a test set according to a certain ratio, wherein the training set is used to realize the construction of a high-precision CNN-GRU fault identification model; the test set is used to test the above CNN-GRU fault identification model, and the error comparison between the test classification result and the actual classification result is carried out to realize the rational evaluation of the constructed CNN-GRU fault identification model.

[0044] The CNN-GRU fault identification model constructed with the fault sample sets in Tables 1 and 2 as the research object is as follows: Figure 8 As shown: (1) Input link: The input link includes the input layer and the data folding layer. The function of the data folding layer is to convert the sequence data, i.e. the fault sample set, into a tensor data test data set through data preprocessing for the subsequent input and processing of the convolutional neural network. The test data set obtained after data preprocessing is used as the input of the model and enters the input layer in the form of a cell array. After data preprocessing by the data folding layer, it is converted into an array form recognizable by the convolutional neural network and then enters the subsequent network training link.

[0045] (2) CNN layer: The CNN layer in the model consists of a two-dimensional convolutional neural network layer with 128 hidden layers and a step size of 2*1*1, a batch normalization layer, an elu activation layer, and an average pooling layer with a step size of 1*1. Considering that the collected data are the effective values ​​of three-phase voltage and three-phase current under fault steady-state conditions, and their own disturbance data are relatively small, a single convolutional neural network structure is sufficient to achieve accurate and fast feature extraction, while reducing the network complexity and the number of hyperparameters, thereby improving the network training efficiency.

[0046] (3) GRU layer: The GRU layer is preceded by a data expansion layer and a flattening layer. The data expansion layer corresponds to the preceding data folding layer and is intended to restore the data converted into tensors to their original data form. The flattening layer converts the data dimension after feature extraction by the convolutional neural network, converting its input spatial dimension into a channel dimension. The GRU layer consists of a GRU feature extraction layer, a GRU data output layer, and a discard layer. The construction of a two-layer GRU structure can effectively ensure the full extraction of training set data features. The GRU feature learning layer contains 128 hidden layers, and the data output type is set to "sequence output". The GRU data output layer contains 128 hidden layers, and the data output type is set to "last bit output". The setting of the discard layer can effectively alleviate the overfitting phenomenon that may occur due to too many network parameters, and achieve a regularization effect to a certain extent. (4) Output layer: The output layer includes a fully connected layer, a Softmax layer, and a classification output layer. The output of the fully connected layer is the number of fault identification network labels. By setting the "Classification" classification output layer, the function of the overall deep learning network can be emphasized as "classification training."

[0047] Preferably, the CNN-GRU fault identification model includes a fault type identification unit and a fault location identification unit, such as Fig. 9 The fault type identification unit is used to identify the fault type of the multi-unit transmission line, and the fault location identification unit is used to identify the fault location of the multi-unit transmission line; Based on the fault type, the fault sample set includes a training set 1 and a test set 1, that is, the fault sample set is first classified based on the fault type as a classification method, and then divided into a training set 1 and a test set 1 after classification, and a fault type identification unit is obtained according to the training set 1 and the test set 1; Based on the fault location, the fault sample set includes training set 2 and test set 2, that is, the fault sample set is first classified based on the fault location as the classification method, and then divided into training set 2 and test set 2 after classification, and the fault location identification unit is obtained according to training set 2 and test set 2.

[0048] In order to ensure the applicability of the CNN-GRU fault identification model for fault identification, multiple groups of different transition resistance values ​​are set as interference to expand the dimension of the training data set. The model uses the randperm function to randomly generate 8 fault distances in each segment. When 7 types of fault resistance are known as interference items, batch simulation is performed for 10 types of faults. A total of 1680 groups of samples are obtained. The fault sample set is classified by fault type as the classification method as shown in Table 1.

[0049] In order to verify the classification accuracy of the constructed CNN-GRU fault identification model, fault type identification is performed in a massive range by dividing each segment. First, based on the set sampling frequency, the three-phase voltage effective value and the three-phase current effective value in the preset time window are intercepted according to the steady-state process after the fault occurs to obtain the fault input data set; after about 2 hours of batch simulation of faults, a fault output data set based on the fault input data can be obtained, which contains a total of 1680 groups of data, and is used as the fault sample set. After the collected fault sample set is converted into data type and reclassified according to the fault type, it is divided into training set 1 and test set 1 to obtain the fault type identification unit, and the classification label required for subsequent fault type identification is matched for the CNN-GRU fault identification model; then, the randomly shuffled data samples and labels are converted into cell data types and dimensions to generate data types that can be recognized by the deep learning network, and the fault type samples of 1680 groups of data are divided into training set 1 of 1200 groups of data and test set 1 of 480 groups of data through proportional division. In the fault identification training process of constructing the CNN-GRU fault identification model from the training set 1, the Adam gradient descent algorithm is used, and the algorithm settings are shown in Table 3 below: Table 3. Parameter settings of fault identification training algorithm based on fault type The training process of the fault type identification unit of the CNN-GRU fault identification model is as follows: Fig.10 As shown in Figure 2, when the number of iterations exceeds 1000, the training accuracy of the fault type identification unit of the CNN-GRU fault identification model can reach 100%, and the loss rate is close to 0; the identification results of the fault type identification unit for the fault type are as follows: Fig.11 As shown in the figure, the result is displayed in the form of a confusion matrix. The diagonal of the confusion matrix represents the number of samples corresponding to various fault types in the test set 1, and the "1-10" displayed on the horizontal axis is the label corresponding to the 10 fault types. From the confusion matrix results, it can be seen that the fault type identification unit of the CNN-GRU fault identification model can well realize fault type identification, and the identification accuracy of the fault type can reach 100%.

[0050] In order to realize the fault section division, the simulation model divides the 220kv transmission line into three sections, in which each section of the transmission line is divided into two, and the length of each section of the AC transmission line is set as an adjustable variable in the working area. The specific location of the fault, that is, the fault location, is simulated by changing the line length. Among them, the total length of the L1 line is 20km, and the total length of the L2 and L3 lines is 40km. In order to ensure that the topological structure and line impedance of the multi-unit transmission line remain unchanged during the fault, the total length of a single line must be kept constant when the line length is variable. The fault sample set is classified by the fault location as the classification method as shown in Table 2.

[0051] There are 1680 groups of samples collected by the signal acquisition module, and the dimension of each group of samples is 3000. Directly using a large amount of data in the fault sample set to train the CNN-GRU fault identification model will cause the CNN-GRU fault identification model to fall into the local optimum during the training process, resulting in extremely low fault diagnosis accuracy. Considering that the effective values ​​of voltage and current are approximately constant under the steady-state fault condition, high-dimensional samples can be reduced in dimension. The data with reduced dimension can represent the original data, thereby reducing the redundant information of the data and reducing the computational complexity. The fault sample set after dimensionality reduction is divided into training set and test set as described above. After the collected fault sample set is converted into data type and reclassified according to the fault location, it is divided into training set 2 and test set 2 and the fault location identification unit is obtained to match the classification label required for subsequent fault location identification for the CNN-GRU fault identification model; then, the randomly disrupted data samples and labels are converted into cell data type and dimensionally transformed to generate data types that can be recognized by the CNN-GRU fault identification model. The fault sample set of 1680 groups of data is divided into training set 2 of 1380 groups of data and test set 2 of 300 groups of data through proportional division. The training process uses the Adam gradient descent algorithm, and the algorithm settings are shown in Table 4: Table 4 Parameter settings of fault identification training algorithm based on fault location The training process of the fault location identification unit of the CNN-GRU fault identification model and the identification results of the fault location are shown in the figure. Fig.12 and 13 As shown in the figure, the fault location identification unit of the CNN-GRU fault identification model has obvious errors in the classification of three-section transmission line faults. The accuracy of the identification results of L1, L2, and L3 are 99.0%, 94.1%, and 95.9% respectively. Among them, one fault on the L1 line was misjudged as occurring on the L2 line, 6 faults on the L2 line were misjudged as occurring on the L3 line, and 4 faults on the L3 line were misjudged as occurring on the L2 line. Through the fault inversion data retrieval, it can be seen that the several misjudged faults all occurred at the exit or end of their respective lines.

[0052] In order to verify the superiority of the CNN-GRU fault identification model for fault identification, the traditional SVM fault identification model, CNN fault identification model and GRU fault identification model were selected for fault identification training under the same fault sample set. The parameters of the CNN fault identification model and the GRU fault identification model are the same as those of the CNN-GRU fault identification model constructed in this paper. Fig.14 As shown in Table 5.

[0053] From the comparison results, it can be seen that the training effect of traditional single networks as fault identification models is poor and it is impossible to effectively identify the fault location. The CNN-GRU fault identification model has a strong spatiotemporal feature extraction capability, and the fault identification accuracy on the three lines is above 94%, which can effectively identify the fault location.

[0054] Table 5 Comparison of fault location identification results Further, the simulation model includes a fault simulation module and a program editing module; The program editing module is used to convert the fault input data set into a recognizable signal and transmit it to the fault simulation module; using the model simulation and program programming in the MATLAB / Simulink platform, the model is nested in the editor using loop statements to realize the interaction between the editor and the model. If only the successive simulation method is used, it takes a long time to build a high-dimensional massive fault sample for fault identification training, and the simulated data set is not conducive to classification and sorting. Therefore, it is particularly important to realize batch simulation of the model on the MATLAB / Simulink platform. Before the batch simulation starts, it is necessary to use the "sim" function in the m file to run the Simulink model, and pass the model parameters to the simulation model through the "sim" function. Batch simulation can pass different model parameters to the simulation model by simply changing the programming statement through the cooperation between the editor and the model, and complete batch simulation under different fault conditions.

[0055] The fault simulation module is used to simulate a fault according to an identifiable signal converted from a fault input data set, and obtain a fault output data set according to the simulated fault condition.

[0056] Furthermore, the data format of the fault type of the controlled current source model is in binary form. The traditional three-phase fault generator module usually uses the "Three-Phase Fault" module in Simulink. Although this module can set different fault moments, it cannot set the fault type and the transition resistance corresponding to the fault at the same time. Therefore, the traditional three-phase fault generator module is used as the basis for construction. Figure 7The improved three-phase fault generator module shown in FIG. Among them, the control module can realize the fault type conversion as shown in Table 6, and the fault type control module in the batch simulation fault generation module is set to a binary array to realize the successive simulation of 10 fault types, that is, simulated faults of 10 fault types occur respectively.

[0057] Table 6 Fault type settings An electronic device for realizing a data-driven fault identification method based on fault characteristics of a wind turbine generator system, comprising a program editing module, a fault simulation module, a data monitoring module, a data analysis module, a data storage module and a fault identification module: A program editing module, used to convert a fault input data set into a recognizable signal and transmit it to a fault simulation module; A fault simulation module, used for receiving a recognizable signal of a fault input data set of the program editing module and running a simulated fault; A data monitoring module, used to monitor the fault simulation module during the fault simulation process and output electrical quantity data samples; A data analysis module is used to process electrical quantity data samples and obtain a fault output number set based on a fault input number set, to generate a fault sample set based on the fault input number set and the fault output number set; and to construct a CNN-GRU fault identification model based on the fault sample set; Data storage module, used to store CNN-GRU fault identification model; The fault identification module is used to monitor the multi-unit transmission line and perform fault identification on the multi-unit transmission line according to the CNN-GRU fault identification model stored in the data storage module.

[0058] The system embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0059] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0060] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A data-driven fault identification method based on wind turbine fault characteristics, characterized in that: The following steps are involved: Based on the low voltage ride-through strategy and the negative sequence current suppression strategy, a controlled current source model of a permanent magnet direct-drive wind turbine is constructed; a multi-unit transmission line includes the permanent magnet direct-drive wind turbine; the controlled current source model is: ; ; In the formula, is the short-circuit current amplitude, is the rated current, is the per unit voltage of the fan side, is the impedance phase angle on the PMSG side, pu is the per-unit value; Constructing a simulation model of the multi-unit transmission line according to the controlled current source model and the multi-unit transmission line; acquiring a fault input data set and inputting it into the simulation model, and running the simulation model so that a simulated fault occurs in the simulation model; Collecting a sample set of electrical quantity data of the output line in the simulation model within a preset time window after the simulated fault occurs; obtaining a fault output data set based on the sample set of electrical quantity data; obtaining a fault sample set of the simulation model based on the fault input data set and the fault output data set; A CNN-GRU fault identification model based on the simulation model is constructed according to the fault sample set.

2. The data-driven fault identification method based on wind turbine fault characteristics according to claim 1 is characterized in that: The electrical quantity data sample set is a set of electrical quantity data samples of the output line, and the electrical quantity data samples include three-phase voltage effective values ​​and three-phase current effective values; the step of obtaining the fault output number set is: processing the electrical quantity data sample set according to a normalization formula to obtain the fault output number set, and the fault output number set is a set of electrical quantity data final values; the normalization formula is: ; In the formula, is the maximum value of the electrical quantity data sample, is the minimum value of the electrical quantity data sample, The electrical quantity data samples currently processed by the normalization formula are: It is the final value of electrical quantity data.

3. The data-driven fault identification method based on wind turbine fault characteristics according to claim 2 is characterized in that: The three-phase voltage effective value is: ; In the formula, The first Sampling points Phase voltage instantaneous value, M is the number of sampling points in one power frequency cycle, for Phase voltage effective value, is the order of sampling points within a power frequency cycle, is the phase in three-phase electricity, The phases are phase a, phase b, and phase c of the three-phase electricity in sequence.

4. The data-driven fault identification method based on wind turbine fault characteristics according to claim 2 is characterized in that: The three-phase current effective value is: ; In the formula, The first Sampling points Instantaneous value of phase current, M is the number of sampling points in one power frequency cycle, for Phase current effective value, is the order of sampling points within a power frequency cycle, is the phase in three-phase electricity, The phases are phase a, phase b, and phase c of the three-phase electricity in sequence.

5. The data-driven fault identification method based on wind turbine fault characteristics according to claim 1 is characterized in that: The preset time window is 0.45s to 0.5s after the fault occurs.

6. The data-driven fault identification method based on wind turbine fault characteristics according to claim 1, characterized in that: The fault sample set includes a training set and a test set; The training set is used to construct the CNN-GRU fault identification model; The test set is used to test the CNN-GRU fault identification model constructed according to the training set.

7. The data-driven fault identification method based on wind turbine fault characteristics according to claim 6 is characterized in that: The CNN-GRU fault identification model includes a fault type identification unit and a fault location identification unit; The fault type identification unit is used to identify the fault type of the multi-unit transmission line, and the fault location identification unit is used to identify the fault location of the multi-unit transmission line; Based on the fault type, the fault sample set includes a training set 1 and a test set 1, and the fault type identification unit is obtained according to the training set 1 and the test set 1; Based on the fault location, the fault sample set includes a training set 2 and a test set 2, and the fault location identification unit is obtained according to the training set 2 and the test set 2.

8. The data-driven fault identification method based on wind turbine fault characteristics according to claim 1, characterized in that: The simulation model includes a fault simulation module and a program editing module; The program editing module is used to convert the fault input data set into a recognizable signal and transmit it to the fault simulation module; The fault simulation module is used to generate the simulated fault according to the identifiable signal converted from the fault input number set, and obtain the fault output number set according to the simulated fault condition.

9. The data-driven fault identification method based on wind turbine fault characteristics according to claim 1, characterized in that: The data format of the fault type of the controlled current source model is in binary form.

10. An electronic device, used to implement the data-driven fault identification method based on wind turbine fault characteristics according to any one of claims 1 to 9, characterized in that: It includes program editing module, fault simulation module, data monitoring module, data analysis module, data storage module and fault identification module: The program editing module is used to convert the fault input data set into a recognizable signal and transmit it to the fault simulation module; The fault simulation module is used to receive the identifiable signal of the fault input data set of the program editing module and run a simulated fault; The data monitoring module is used to monitor the fault simulation module and output the electrical quantity data sample during the fault simulation process; The data parsing module is used to process the electrical quantity data samples and obtain a fault output number set based on the fault input number set, and to generate the fault sample set according to the fault input number set and the fault output number set; and to construct the CNN-GRU fault identification model according to the fault sample set; The data storage module is used to store the CNN-GRU fault identification model; The fault identification module is used to monitor the multi-unit transmission line and perform fault identification on the multi-unit transmission line according to the CNN-GRU fault identification model stored in the data storage module.

Citation Information

Patent Citations

  • Permanent magnetic direct-drive wind power generation system integrated fault diagnosis method

    CN104459388A

  • ANFIS-based electric vehicle permanent magnet synchronous motor fault classification method

    CN106291354A

  • Method for intelligently diagnosing wind turbine unit faults based on fault knowledge base

    CN106682814A

  • Open-circuit fault diagnosis method for direct-drive permanent-magnet wind power generator system current transformer

    CN107728063A

  • Modeling simplification and parameter identification method based on PSASP direct drive wind generation set

    CN108595861A