A parameter identification method for asymmetric fault ride-through of wind turbine generators

Abstract

A parameter identification method for asymmetric fault ride-through of wind turbines is proposed. This method extracts positive and negative sequence voltage components and phase jump angles from three-phase discrete waveform data. Then, a three-segment chain power characteristic function with coupled unbalance correction terms is constructed to fit the power response characteristics of the fault suppression segment, the fault clearing instantaneous point, and the recovery process segment, respectively. Finally, an asymmetric sensitivity-weighted integral operator is used to extract integral characteristic quantities with different physical meanings, constructing a system of simultaneous equations for closed-form solution, identifying nine control parameters with clear physical meanings in one step. This method utilizes an integral operator to automatically cancel 100Hz harmonic pulsations and accurately characterizes the coupling relationship between active power load reduction and negative sequence compensation through unbalance correction. It can stably reconstruct the centerline trajectory reflecting the actual control logic from manufacturer black-box waveform data, enabling high-fidelity parameter identification and measured data fitting for asymmetric fault ride-through characteristics of wind turbines.

Description

Technical Field

[0001] This invention belongs to the field of power system transient stability analysis technology, specifically relating to a parameter identification method for asymmetric fault ride-through of wind turbine units. Background Technology

[0002] As the scale of wind turbine grid connection continues to expand, the transient response characteristics of wind turbines during grid faults have an increasingly significant impact on the safe and stable operation of the power system. For direct-drive wind turbines using permanent magnet synchronous generators, negative sequence components appear in the three-phase voltages during asymmetrical faults such as single-phase grounding and two-phase short circuits. The positive and negative sequence components act together on the grid-connected converter, causing significant second-harmonic fluctuations in the active and reactive power outputs of the wind turbine. The waveform data provided by manufacturers typically only includes discrete quantities such as three-phase voltage and power, making it difficult to directly obtain internal control parameters. Therefore, it is necessary to identify the fault ride-through characteristics based on measured data. Existing fitting methods often use point-by-point error for parameter identification, which is easily affected by power pulsation peaks and valleys and instantaneous fluctuations during fault clearing, making it difficult to accurately extract the power trajectory centerline. Furthermore, some models only consider positive sequence voltage drops, failing to independently characterize the effects of negative sequence voltage and voltage imbalance, making it difficult to distinguish between the positive sequence support effect and the negative sequence compensation effect. In addition, asymmetrical fault clearing may be accompanied by positive sequence voltage phase jumps, and the power recovery process after a fault also exhibits nonlinear characteristics. Existing methods struggle to simultaneously characterize power suppression during a fault, power jumps at the moment of cutoff, and subsequent recovery trajectories, making it difficult to achieve accurate parameter identification and fitting for the entire asymmetric fault traversal process. Summary of the Invention

[0003] To address the aforementioned issues, this invention aims to solve the problems of 100Hz pulsation in power waveform data under asymmetric faults, difficulty in separating positive and negative sequence control effects, and difficulty in accurately fitting power jumps and recovery trajectories after fault clearance. It provides a parameter identification method for asymmetric fault ride-through of wind turbine units.

[0004] A parameter identification method for asymmetric fault ride-through of wind turbine generators, the method specifically comprising:

[0005] Step 1: Read the discrete waveform data of the symmetrical fault ride-through process of the wind turbine. The discrete waveform data includes at least the time series t, the three-phase voltage series U (A, B, C), the active power series P, and the reactive power series Q. Perform power frequency cycle moving average processing on the active power series and reactive power series: using a 20ms sliding window and a 10ms step size, obtain the average active power series P. data (i) and the average reactive power sequence Q data(i) This preprocessing step aims to reduce the interference of the 100Hz harmonic power pulsation unique to asymmetric faults on subsequent feature extraction, so that the ASWI operator can focus on the identification of the control axis.

[0006] Step 2: Based on the positive-sequence voltage amplitude sequence and negative-sequence voltage amplitude sequence obtained from the ABC three-phase voltage sequence, determine the fault start time t0 and the fault clearing time t according to the amplitude and rate of change of the positive-sequence voltage amplitude sequence. clear And based on the average active power sequence relative to the pre-fault steady-state active power P pre The convergence state determines the recovery end time t end The recording process is divided into a fault phase, a recovery starting point after fault clearing, and a fault recovery phase.

[0007] Step 3: Calculate the real-time voltage imbalance u(i) and the average voltage imbalance u during the fault phase based on the positive-sequence voltage amplitude sequence and the negative-sequence voltage amplitude sequence. avg Simultaneously, the phase angle sequence of the positive sequence voltage is extracted using a software phase-locked loop operator, and the voltage phase jump angle is calculated based on the phase angle difference before and after the fault clearing time. The reason for using positive sequence voltage is that: under asymmetrical faults, the voltage drop depth of each phase is uneven, and single-phase voltage may misjudge the start and end time of the fault; positive sequence voltage comprehensively reflects the amplitude of the positive sequence component of the three-phase voltage, and can stably characterize the severity of the power grid fault.

[0008] Step 4: Construct a three-segment chain power characteristic model for imbalance correction. The model includes active power characteristic functions and reactive power characteristic functions during the fault, active power characteristic functions and reactive power characteristic functions at the recovery starting point after fault clearing, and active power characteristic functions and reactive power recovery process characteristic functions during the fault recovery stage.

[0009] The three-segment chain power characteristic model is specifically as follows:

[0010] The active power characteristic function P1(t) during the fault stage is:

[0011] ;

[0012] Where P1(t) is the fitted calculated value of the active power at time t during the asymmetric fault; P pre The measured average active power during the steady-state operation phase before the fault occurred; U + (t) represents the measured per-unit value of the positive sequence voltage at time t extracted from the three-phase voltage by discrete symmetrical component transformation; U n The system's rated voltage per unit value; u(t) is the real-time voltage imbalance; k p It is the positive sequence voltage suppression index; This is the imbalance sensitivity coefficient.

[0013] The reactive power characteristic function Q1(t) during the fault stage is:

[0014] Q1(t) = Q pre +K pos ·(U n -U + (t))-K neg ·U - (t));

[0015] Where Q1(t) is the fitted calculated value of reactive power at time t during the asymmetric fault; Q pre K represents the measured average reactive power during the steady-state operation phase before the fault occurred. pos For positive-sequence reactive power support gain; K neg For negative sequence reactive power compensation gain; U - (t) represents the measured per-unit value of the negative sequence voltage at time t extracted from the three-phase voltage by discrete symmetric component transformation;

[0016] The active power characteristic function P2(t) at the recovery starting point after fault clearance is:

[0017] P2(t) = P1(t) clear )+γ asym ·(P pre -P1(t clear ))·u avg ;

[0018] Where P2(t) is the fitted calculated value of the instantaneous active power during asymmetric fault clearing; P1(t) clear The characteristic function of the fault segment is calculated at t. clear Final value of the active power centerline at time γ; asym For asymmetric jump correction gain; P pre -P1(t clear ) represents the dynamic space where active power needs to be restored; u avg The average voltage imbalance is calculated from discrete measured data during the fault period.

[0019] The reactive power characteristic function Q2(t) at the recovery starting point after fault clearance is:

[0020] ;

[0021] Where Q2(t) is the fitted calculated value of the reactive power during fault clearing; It represents the average value of the discrete reactive power within a specific window before fault clearance; This is the phase jump sensitivity coefficient;

[0022] The active power characteristic function P3(t) during the fault recovery phase is:

[0023] ;

[0024] Where P3(t) is the fitted calculated value of active power at time t during the fault recovery phase; tt clear T is the relative time offset calculated from the time the fault was cleared. rec_asym P is the asymmetric active power recovery time constant; n is the trajectory shape exponent; P pre This represents the average active power during steady-state operation before the fault.

[0025] The active power characteristic function Q3(t) during the fault recovery phase is:

[0026] Q3(t) = Q pre +(Q2(t)-Q pre )·[exp(-(tt clear ) / τ asym )];

[0027] Where Q3(t) is the calculated reactive power value fitted at time t during the fault recovery phase; Q pre The measured reactive power value during the steady-state operation phase before the fault; τ asym The reactive power recovery time constant;

[0028] Step 5: Construct the asymmetric sensitivity-weighted integral operator:

[0029] ;

[0030] Where stage refers to the corresponding identification stage. To correspond to the observation bias in the identification stage, σ represents the order of asymmetric sensitivity. By introducing this new dimension of imbalance sensitivity order σ, the "suppression caused by voltage drop" and "suppression caused by imbalance" are extracted separately from the same data segment, solving the industry problem of parameter interference in asymmetric fault fitting. ζ represents the time-domain order, t start T corresponds to the start time of the identification phase. stage This corresponds to the duration of the identification phase. This represents the average sliding window step size.

[0031] By adjusting the asymmetric sensitivity order σ and the time domain order ζ of the operator, integral feature quantities with different physical meanings are extracted from the measured data of the fault stage and the fault recovery stage and the three-segment chain power characteristic model, including total area features, unbalance weighted area features and energy centroid distribution features.

[0032] Step 6: Based on the integral feature quantity extracted in Step 5, construct a system of simultaneous equations for the model parameters. By solving the system of equations, identify all undetermined parameters in the three-segment chain feature model at once. Then, use the identified parameters to reconstruct a simulation trajectory that matches the axis characteristics in the measured data, thus completing the parameter identification of the asymmetric fault ride-through characteristics of the wind turbine. The parameters to be identified include: the imbalance sensitivity coefficient of the parameters to be identified during the fault stage. Positive sequence voltage suppression index k p Negative sequence reactive power compensation gain K neg Positive sequence reactive power support gain K pos The recovery starting point parameter to be identified after fault clearance is the asymmetric jump correction gain. and phase jump sensitivity coefficient The parameter to be identified during the fault recovery phase is the asymmetric active power recovery time constant T. rec_asym Trajectory shape exponent n and asymmetric reactive power return time constant .

[0033] Preferably, the sliding window length is W = 20ms (the number of sampling points for one power frequency cycle), which is suitable for sampling frequencies f. s Recording data ≥1kHz. The confirmation window is 5ms (approximately 1 / 4 of the power frequency cycle), and the observation window is 5ms. Determine the fault start time t0 and the fault clearing time t. clear and recovery end time t end The specific criteria are as follows:

[0034] Fault start time t0: When the absolute value of the voltage change rate is greater than the first threshold. If the voltage amplitude is lower than the drop threshold (0.9pu) and remains in this state within the first observation window, the sampling point that first meets the condition is determined as the fault start time t0;

[0035] Fault clearing time t clear When the rate of change of voltage is greater than the second threshold If the voltage amplitude exceeds the recovery threshold (0.85 pu) and remains in this state within the second observation window, the starting sampling point of the second observation window is determined as the fault clearing time t. clear ;

[0036] Restore end time t end When the active power enters the steady-state convergence band centered on the active power before the fault. If the system does not exit the preset stabilization time (50ms), the starting sampling point of the stabilization interval will be determined as the recovery end time.

[0037] Both the fault stage and the fault recovery stage include corresponding boundary sampling points.

[0038] Preferably, the calculation methods for the positive-sequence voltage amplitude and the negative-sequence voltage amplitude sequences are as follows:

[0039] ;

[0040] ;

[0041] Where α=e j·2π / 3 ;U a U b and U c These are the three-phase voltages;

[0042] The real-time voltage imbalance u(i) is:

[0043] ;

[0044] Average voltage imbalance u during the fault phase avg for:

[0045] ;

[0046] Where, N f This represents the total number of moving average windows included within the fault phase.

[0047] Preferably, the N f for:

[0048] .

[0049] Preferably, the input to the software phase-locked loop (PLL) operator is a three-phase voltage sampling sequence. The PLL operator tracks the positive-sequence voltage phase through Clark / Park transform and a proportional-integral regulator. The bandwidth of the software PLL operator is 20Hz to 50Hz. The average positive-sequence voltage phase θ within a 5ms window before the fault clearing time is calculated. ) and the average phase value θ of the positive sequence voltage within a 5ms window after the fault clearing time. ), and calculate the phase jump angle. :

[0050] ;

[0051] When the absolute value of the phase difference is greater than π, the phase difference is subjected to 2π foldback processing, so that... Through the above method, this invention transforms the original three-phase discrete waveform data into a scalar sequence that reflects the asymmetric transient characteristics, providing key data support for the subsequent construction of the power characteristic function of the coupling imbalance correction term.

[0052] Preferred active power parameters during the fault phase and k p The identification process is as follows:

[0053] make ;

[0054] Construct a system of two nonlinear equations:

[0055] ;

[0056] ;

[0057] The Levenberg-Marquardt nonlinear least squares method is used to solve the two-variable nonlinear equation system to obtain the active power parameters during the fault stage. and k p Identification.

[0058] Preferably, the reactive power parameter K during the fault phase pos and K neg The identification process is as follows:

[0059] make ;

[0060] Construct a system of two nonlinear equations:

[0061] make If ζ=0, then we have:

[0062] ;

[0063] make If ζ=0, then we have:

[0064] ;

[0065] The parameter K is obtained by simultaneously solving the two sets of linear equations. pos and K neg Identification.

[0066] Preferred recovery start parameters after fault clearance and The identification process is as follows:

[0067] Calculate the average active power observation value P within the preset narrow window after the fault is cleared. obs and average observed reactive power Q obs ;

[0068] Substituting the positive sequence voltage and real-time voltage imbalance in the fault-end window into the active power characteristic function of the fault stage, we obtain the final value P1(t) of the active power centerline during the fault stage. clearThe fault end window is backed down by three sliding window steps relative to the fault clearing time.

[0069] Constructing equations for parameter identification

[0070]

[0071] .

[0072] Preferably, the fault recovery phase parameter T rec_asym The identification process for n is as follows:

[0073] make ;

[0074] make Let ζ=0 and ζ=1 respectively, and construct the zeroth-order area moment and first-order time moment for the fault recovery phase:

[0075]

[0076] ;

[0077]

[0078] ;

[0079] Among them, T stage =t end -t clear ; Calculate the normalized time center G= When G < 0.38, the initial value of n is set to 1; when G > 0.62, the initial value of n is set to 3; otherwise, the initial value of n is set to 2; n∈{1,2,3,4} is enumerated sequentially, and the Levenberg-Marquardt method is used to optimize T for each value of n. rec_asym Select the T with the smallest residual norm. rec_asym Combined with n, the parameter T is completed. rec_asym Identification of n.

[0080] Preferably, the fault recovery phase parameters The identification process is as follows:

[0081] Select the option that satisfies |Q during the fault recovery phase. data (t)-Q pre |>10 -5 , and t i >t clear Using an effective sampling window, perform linear regression on the formula Q3(t):

[0082] ;

[0083] Calculation parameters :

[0084] ;

[0085] When the number of effective sampling windows is less than 5, Set it to 0.1s.

[0086] This invention introduces an imbalance sensitivity factor and negative sequence compensation gain, enabling the power characteristic function to sense the degree of imbalance and reflect the game relationship between the manufacturer's active power load reduction and negative sequence compensation, thus ensuring physical consistency in parameter identification. The summation characteristics of the integral operator are used to automatically cancel 100Hz harmonic pulsations in the recorded waveform data, focusing the identification results on the true control centerline and avoiding interference from oscillation peaks. By introducing a phase jump compensation term at the recovery starting point and assigning a natural zero derivative characteristic to the active power recovery function, accurate characterization of the instantaneous step cut and the "soft start" in the initial recovery stage is achieved, ensuring fitting accuracy at segmented junctions and throughout the entire recovery process. The entire method is based on feature mapping and simultaneous equations, avoiding the uncertainty and multiple solutions problems of traditional heuristic optimization algorithms, and achieving rapid, repeatable, and unique identification of the wind turbine manufacturer's black-box control parameters. Attached Figure Description

[0087] Appendix Figure 1 This is a flowchart of a parameter identification method for asymmetric fault ride-through of wind turbine units disclosed in an embodiment of the present invention. Detailed Implementation

[0088] To make the technical solution of the present invention easier to understand, a parameter identification method for asymmetric fault ride-through of wind turbine units disclosed in the present invention will now be clearly and completely described in conjunction with embodiments and accompanying drawings.

[0089] Step 100: Read the discrete waveform data of the symmetrical fault ride-through process of the wind turbine. The discrete waveform data includes at least the time series t, the three-phase voltage series U (A, B, C), the active power series P, and the reactive power series Q. Perform power frequency cycle moving average processing on the active power series and the reactive power series: using a 20ms sliding window and a 10ms step size, obtain the average active power series P. data (i) and the average reactive power sequence Q data (i) This preprocessing step aims to reduce the interference of the 100Hz harmonic power pulsation unique to asymmetric faults on subsequent feature extraction, so that the ASWI operator can focus on the identification of the control axis.

[0090] Step 110: Based on the positive-sequence voltage amplitude sequence and negative-sequence voltage amplitude sequence obtained from the ABC three-phase voltage sequence, determine the fault start time t0 and the fault clearing time t according to the amplitude and rate of change of the positive-sequence voltage amplitude sequence. clear And based on the average active power sequence relative to the pre-fault steady-state active power P pre The convergence state determines the recovery termination time t end The recording process is divided into a fault phase, a recovery starting point after fault clearing, and a fault recovery phase, specifically:

[0091] The sliding window length is set to W = 20ms (the number of sampling points for one power frequency cycle), which is suitable for sampling frequencies f. s Recording data ≥1kHz. The confirmation window is 5ms (approximately 1 / 4 of the power frequency cycle), and the observation window is 5ms. Determine the fault start time t0 and the fault clearing time t. clear and recovery end time t end The specific criteria are as follows:

[0092] Fault start time t0: When the absolute value of the voltage change rate is greater than the first threshold. If the voltage amplitude is lower than the drop threshold (0.9pu) and remains in this state within the first observation window, the sampling point that first meets the condition is determined as the fault start time t0;

[0093] Fault clearing time t clear When the rate of change of voltage is greater than the second threshold If the voltage amplitude exceeds the recovery threshold (0.85 pu) and remains in this state within the second observation window, the starting sampling point of the second observation window is determined as the fault clearing time t. clear ;

[0094] Restore end time t end When the active power enters the steady-state convergence band centered on the active power before the fault. If the system does not exit the preset stabilization time (50ms), the starting sampling point of the stabilization interval will be determined as the recovery end time.

[0095] Step 120: Calculate the real-time voltage imbalance u(i) and the average voltage imbalance u during the fault phase based on the positive-sequence voltage amplitude sequence and the negative-sequence voltage amplitude sequence. avg Specifically, this embodiment employs a discrete symmetric component transformation algorithm, starting from the three-phase instantaneous voltage sampling point U. a (i), U b (i), U c (i) Calculate the positive sequence voltage magnitude sequence U + (i) and the negative sequence voltage magnitude sequence U - (i):

[0096] ;

[0097] ;

[0098] Where α=e j·2π / 3 U a U b and U c These are the three-phase voltages;

[0099] Based on this definition, real-time voltage imbalance is:

[0100] ;

[0101] Average voltage imbalance u during the fault phase avg for:

[0102] ;

[0103] Where, N f This represents the total number of moving average windows included within the fault phase. .

[0104] Meanwhile, to fit the impact of grid phase sequence reconfiguration during asymmetric fault clearance, this invention uses a software phase-locked loop (SPLL) operator to track the positive-sequence voltage vector in real time. phase angle The input to the software phase-locked loop (PLL) operator is a three-phase voltage sampling sequence. The PLL operator tracks the positive-sequence voltage phase using Clark / Park transform and a proportional-integral regulator. The bandwidth of the PLL operator is 20Hz to 50Hz. The average positive-sequence voltage phase θ within a 5ms window before the fault clearing time is calculated. ) and the average phase value θ of the positive sequence voltage within a 5ms window after the fault clearing time. ), and calculate the phase jump angle. :

[0105] ;

[0106] When the absolute value of the phase difference is greater than π, the phase difference is subjected to 2π foldback processing, so that... Through the above method, this invention transforms the original three-phase discrete waveform data into a scalar sequence that reflects the characteristics of asymmetric transients, providing crucial data support for the subsequent construction of the power characteristic function of the coupling imbalance correction term. The rationale for using positive-sequence voltage is that: under asymmetric faults, the voltage drop depth of each phase is uneven, and single-phase voltage may misjudge the fault start and end times; positive-sequence voltage comprehensively reflects the amplitude of the positive-sequence components of the three-phase voltages, and can stably characterize the severity of power grid faults.

[0107] Step 130: Under asymmetric fault conditions, the transient characteristics of a direct-drive wind turbine are modulated by both positive and negative sequence voltage components, resulting in complex frequency doubling pulsations and nonlinear evolution characteristics in its output power. This invention constructs a three-segment chain characteristic model for imbalance correction, aiming to extract the average power centerline affected by imbalance from the pulsating background of the manufacturer's recorded waveform data using mathematical methods, and to accurately fit the entire process logic of fault suppression, instantaneous connection, and trajectory recovery.

[0108] The model not only considers the power load reduction caused by voltage amplitude drops, but also specifically introduces an imbalance sensitivity factor and a phase jump compensation operator. Through this piecewise feature mapping, the present invention can transform complex converter imbalance control strategies into physically meaningful analytical functions, thereby achieving high-fidelity reproduction of asymmetric control intentions from different manufacturers.

[0109] The three-segment chain power characteristic model is specifically as follows:

[0110] In stage [t0,t clear Within this embodiment, considering the physical behavior of the manufacturer's logic in actively adjusting the converter current capacity to suppress 100Hz torque ripple, active and reactive characteristic functions of the coupling imbalance correction term are constructed:

[0111] The core logic of the active power characteristic function P1(t) during the fault stage lies in fitting how the manufacturer can sacrifice active power output to free up current space for negative sequence compensation under unbalanced conditions:

[0112] ;

[0113] Where P1(t) is the fitted calculated value of the active power at time t during the asymmetric fault; P pre The measured average active power during the steady-state operation phase before the fault occurred; U + (t) represents the measured per-unit value of the positive sequence voltage at time t extracted from the three-phase voltage by discrete symmetrical component transformation; U n The system's rated voltage per unit value; u(t) is the real-time voltage imbalance; k p k is the positive sequence voltage suppression exponent. p >1 indicates superlinear decay, k p =1 indicates linear following. The unbalance sensitivity coefficient has a feasible region of [0, 5.0]. This parameter characterizes the control weight by which the manufacturer uses active power load shedding to obtain negative sequence compensation capacity. The larger the value, the higher the manufacturer's priority in suppressing imbalance, and the more obvious the downward shift of the active power axis.

[0114] when When (approaching symmetrical fault), Automatically sets to 0, the formula degenerates into The pure positive sequence voltage power law form.

[0115] The reactive power characteristic function Q1(t) during the fault stage is used to fit the manufacturer's differentiated support strategies for positive and negative sequence voltage components, and its mathematical expression is defined as:

[0116] Q1(t) = Q pre +K pos ·(U n -U + (t))-K neg ·U - (t));

[0117] Where Q1(t) is the fitted calculated value of reactive power at time t during the asymmetric fault; Q pre K represents the measured average reactive power during the steady-state operation phase before the fault occurred. pos For positive-sequence reactive power support gain, feasible region

[0118] [-5, 10], reflecting the reactive power support stiffness of the manufacturer in boosting the system voltage amplitude; K neg U represents the negative-sequence reactive power compensation gain, with a feasible region of [0, 8.0], used to fit the control strength by which the manufacturer injects negative-sequence reactive current to counteract the effects of grid imbalance; - (t) is the measured per-unit value of the negative sequence voltage at time t extracted from the three-phase voltage by discrete symmetric component transformation.

[0119] when When (approaching symmetrical fault), K neg Automatically sets to 0, the formula degenerates into .

[0120] Because the measured data exhibits severe 100Hz oscillations under asymmetric fault conditions, the target of this formula fitting is the centerline of the trajectory in the manufacturer's data. This is achieved by adjusting... and K neg This invention can accurately reproduce the dynamic game logic of how manufacturers balance "maintaining active power output" and "suppressing frequency doubling ripple" under limited converter capacity. This allows the fitting model to not only fit the measured points, but also to quantify the control performance of the unit under unbalanced power grids through parameter back-calculation.

[0121] At the time t of the asymmetric fault termination clearWhen the system voltage abruptly changes from an unbalanced state to a balanced state, the manufacturer's control logic typically "resets" or "resets" the power command based on the severity of the imbalance during the fault and the voltage phase shift at the moment of cutoff. This invention constructs a starting point characteristic function containing negative-sequence weighting terms and phase correction terms to fit the unique jump characteristics generated by the manufacturer's waveform recording data at this instant.

[0122] The active power characteristic function P2(t) at the recovery starting point after fault clearance is used to fit the initial height during the transition from the asymmetric suppressed state to the symmetric recovered state, and its mathematical expression is defined as:

[0123] P2(t) = P1(t) clear )+γ asym ·(P pre -P1(t clear ))·u avg ;

[0124] Where P2(t) is the fitted calculated value of the instantaneous active power during asymmetric fault clearing; P1(t) clear The characteristic function of the fault segment is calculated at t. clear The final value of the active power centerline at time t, during calculation and The value of the window at the end of the fault should be taken (backwards by 3 window steps ≈ 30ms), not t. clear The value at the time of determination is used to avoid overestimation caused by crossing into the recovery segment data. This item reflects the chain-like connection characteristic of the asymmetric fault traversal process; γ asym For asymmetric jump correction gain, the feasible region is [0, 30]. This parameter characterizes the manufacturer's logic control strength over the power reset amplitude at the moment of asymmetric fault termination, reflecting the correction weight of the imbalance on the recovery start position; P pre -P1(t clear ) represents the dynamic space where active power needs to be restored; u avg The average voltage imbalance is calculated from discrete measured data during the fault period.

[0125] The reactive power characteristic function Q2(t) at the recovery starting point after fault clearance is used to fit the reactive power step characteristics caused by the phase jump, and its mathematical expression is defined as:

[0126] ;

[0127] Where Q2(t) is the fitted calculated value of the reactive power during fault clearing; It is the average value of the discrete reactive power within a specific window before fault clearance, used to eliminate the interference of 100Hz pulsation on the identification of the starting point; This is the phase jump sensitivity coefficient, with a feasible region of [0, 20]. This parameter reflects how the manufacturer's logic instantaneously adjusts the reactive power output value based on the severity of the grid phase jump.

[0128] The data points fluctuate wildly and irregularly at the moment of asymmetric fault clearance. This invention addresses this by introducing... and Two key features enable the model to perceive the pre-fault state and shear off the instantaneous impact. When fitting the CSV waveform, γ is adjusted... asym and The model can accurately simulate manufacturer data at t clear The step logic at the point: If the reactive power drops significantly during a sudden phase jump in the waveform recorded by the manufacturer, the algorithm will automatically identify the larger drop. This compensation mechanism based on physical offsets ensures seamless fitting of the chain model during asymmetric transitions.

[0129] like <0.001rad (no obvious phase jump), , ;like , (Degradation to steady-state reactive power); if Cutting

[0130] In stage [t] clear ,t end Within this timeframe, the active power needs to robustly return from the starting point P2(t) after the jump to the steady-state operating value P. pre In view of the characteristics of large residual stress in the shaft system and high suppression requirements after asymmetric failure, this embodiment abandons the conventional linear or tangent model and adopts a cosine operator with variable curvature control capability for trajectory fitting to ensure that the fitted curve has excellent smoothness at the initial connection point.

[0131] The active power characteristic function P3(t) during the fault recovery phase is:

[0132] ;

[0133] Where P3(t) is the fitted calculated value of active power at time t during the fault recovery phase; tt clear T is the relative time offset calculated from the time the fault was cleared. rec_asym For the asymmetric active power recovery time constant, the feasible region is [0.05, 5.0]. When hour, , That is, the power is exactly at Complete the transition from P2(t) to P within the specified time. preThe full regression. This parameter reflects the total power regression time set by the manufacturer after an asymmetric fault; n is the trajectory shape exponent, which takes a positive integer value. n=1 represents the standard cosine recovery, and n≥2 represents the higher-order cosine recovery. This parameter is used to adjust the "softening" of the recovery trajectory in the initial stage: when the n value is large, the curve at t clear The slope in the vicinity is close to zero, fitting the extremely gentle starting logic of the manufacturer; when the value of n is small, the curve recovers faster. By identifying n, this embodiment can accurately fit the trajectory shape of different manufacturers after balancing "rapid support" and "shaft protection" in the early stage of asymmetric recovery.

[0134] Quantitative definition of softening degree: "Softening degree" is defined as... At t=t clear The characteristic that all derivatives up to the (2n-1)th order are zero. When n=1, only the second derivative is non-zero (lightest softening); when n=4, the first seven derivatives are all zero (heaviest softening). The specific value of n is identified by the ASWI centroid moment, realizing the objective quantification of the degree of softening. Judgment rule: if the centroid is forward (recovery energy is concentrated in the front), the initial value of n is 1 or 2; if the centroid is backward (recovery energy is concentrated in the rear), the initial value of n is 3 or 4.

[0135] P pre This represents the average active power during steady-state operation before the fault; when hour, The above domain limitation guarantees that The cosine independent variable is always in Within the interval, the function value is non-negative and will not have a complex value.

[0136] The active power characteristic function Q3(t) during the fault recovery phase is used to fit the exponential decay characteristic of reactive power returning to the steady-state reactive power value from the phase sequence reorganization state Q2(t):

[0137] Q3(t) = Q pre +(Q2(t)-Q pre )·[exp(-(tt clear ) / τ asym )];

[0138] Where Q3(t) is the calculated reactive power value fitted at time t during the fault recovery phase; Q pre The measured reactive power value during the steady-state operation phase before the fault; τ asym The reactive power recovery time constant reflects the speed at which the manufacturer's controller withdraws the reactive power support to the normal operating command after phase sequence reconfiguration.

[0139] Because measured waveform data from asymmetric faults are typically accompanied by decaying second-harmonic reverberations in the early stages of recovery, traditional linear fitting is prone to resulting in large initial residuals due to neglecting the "start-up buffer." The P3(t) formula used in this invention addresses this issue at t=t clear The derivative at this point is naturally zero, a mathematical property that aligns with the cautious recovery strategy employed by the manufacturer's controller after experiencing asymmetric voltage fluctuations. This is achieved through the relationship between the shape exponent n and the time constant T. rec_asym With dual identification, this invention can fit a smooth, robust active centerline trajectory that conforms to the manufacturer's true physical control intent from data affected by residual vibration interference.

[0140] Step 140: This invention proposes a feature extraction operator with automatic elimination capability of frequency doubling pulsation. By adjusting the observation weights for "voltage imbalance" and "time evolution," the discrete waveform lattice is mapped into solvable algebraic characteristic quantities. An asymmetric sensitivity-weighted integral operator is constructed.

[0141] ;

[0142] Where stage refers to the corresponding identification stage. To correspond to the observation bias in the identification stage, σ represents the order of asymmetric sensitivity. By introducing this new dimension of imbalance sensitivity order σ, the "suppression caused by voltage drop" and "suppression caused by imbalance" are extracted separately from the same data segment, solving the industry problem of parameter interference in asymmetric fault fitting. ζ represents the time-domain order, t start T corresponds to the start time of the identification phase. stage This corresponds to the duration of the identification phase. This represents the average sliding window step size.

[0143] By adjusting the asymmetric sensitivity order σ and the time domain order ζ of the operator, integral feature quantities with different physical meanings are extracted from the measured data of the fault stage and the fault recovery stage and the three-segment chain power characteristic model, including total area features, unbalance weighted area features and energy centroid distribution features.

[0144] Step 150: Based on the integral feature quantity extracted in Step 140, construct a system of simultaneous equations for the model parameters. By solving the system of equations, identify all undetermined parameters in the three-segment chain feature model at once. Then, use the identified parameters to reconstruct a simulation trajectory that matches the axis characteristics in the measured data, thus completing the parameter identification of the asymmetric fault ride-through characteristics of the wind turbine. The parameters to be identified include: the imbalance sensitivity coefficient of the parameters to be identified during the fault stage. Positive sequence voltage suppression index k p Negative sequence reactive power compensation gain K negPositive sequence reactive power support gain K pos The recovery starting point parameter to be identified after fault clearance is the asymmetric jump correction gain. and phase jump sensitivity coefficient The parameter to be identified during the fault recovery phase is the asymmetric active power recovery time constant T. rec_asym Trajectory shape exponent n and asymmetric reactive power return time constant .

[0145] Specifically:

[0146] Active power parameters during fault phase and k p The identification process is as follows:

[0147] make ;

[0148] Construct a system of two nonlinear equations:

[0149] ;

[0150] ;

[0151] The Levenberg-Marquardt nonlinear least squares method is used to solve the two-variable nonlinear equation system to obtain the active power parameters during the fault stage. and k p Identification. A global search is initiated with 5 sets of random initial values ​​for Latin Hypercube. Convergence criterion: and Uniqueness verification: If the parameter deviation of the 5 convergence points is > 1%, select the solution with the smallest residual.

[0152] Failure handling: If convergence is not achieved after 10 iterations, revert to the single-parameter simplified model – fixed. (Degenerate to a pure positive sequence voltage model) ), only with Solve .

[0153] reactive power parameter K during fault phase pos and K neg The identification process is as follows:

[0154] make ;

[0155] Construct a system of two nonlinear equations:

[0156] make If ζ=0, then we have:

[0157] ;

[0158] make If ζ=0, then we have:

[0159] ;

[0160] This system of equations applies to K. pos K neg For a system of linear equations, the analytical solution can be obtained directly by matrix inversion:

[0161] ;

[0162] Feasible domain: , If the solution exceeds the feasible region, prune to the nearest boundary.

[0163] Recovery start parameters after fault clearance and The identification process is as follows:

[0164] Calculate the average active power observation value P within the preset narrow window after the fault is cleared. obs and average observed reactive power Q obs ;

[0165] Substituting the positive sequence voltage and real-time voltage imbalance in the fault-end window into the active power characteristic function of the fault stage, we obtain the final value P1(t) of the active power centerline during the fault stage. clear The fault end window is backed down by three sliding window steps relative to the fault clearing time.

[0166] Constructing equations for parameter identification

[0167]

[0168] .

[0169] Feasible domain: , If the solution exceeds the feasible region, prune to the nearest boundary.

[0170] Boundary handling:

[0171] like → ;

[0172] like → ;

[0173] like rad → ;

[0174] like → Take ;

[0175] like → Cut .

[0176] The fault recovery phase parameter T rec_asym The identification process for n is as follows:

[0177] make ;

[0178] make Let ζ=0 and ζ=1 respectively, and construct the zeroth-order area moment and first-order time moment for the fault recovery phase:

[0179]

[0180] ;

[0181]

[0182] ;

[0183] Among them, T stage =t end -t clear ; Calculate the normalized time center G= When G < 0.38, the initial value of n is set to 1; when G > 0.62, the initial value of n is set to 3; otherwise, the initial value of n is set to 2; n∈{1,2,3,4} is enumerated sequentially, and the Levenberg-Marquardt method is used to optimize T for each value of n. rec_asym Select the T with the smallest residual norm. rec_asym Combined with n, the parameter T is completed. rec_asym Identification of n.

[0184] The parameters of the fault recovery phase The identification process is as follows:

[0185] Select the option that satisfies |Q during the fault recovery phase. data (t)-Q pre |>10 -5 , and t i >t clear Using an effective sampling window, perform linear regression on the formula Q3(t):

[0186] ;

[0187] Calculation parameters :

[0188] ;

[0189] When the number of effective sampling windows is less than 5, Set it to 0.1s.

[0190] It should be noted that for those skilled in the art, several improvements, substitutions, modifications and refinements can be made without departing from the principles and spirit of this invention, and these improvements, substitutions, modifications and refinements should also be considered within the scope of protection of this invention.

Claims

1. A parameter identification method for asymmetric fault ride-through of wind turbine generators, characterized in that, The method is specifically as follows: Step 1: Read the discrete waveform data of the symmetrical fault ride-through process of the wind turbine. The discrete waveform data includes at least the time series t, the three-phase voltage series U (A, B, C), the active power series P, and the reactive power series Q. Perform power frequency cycle moving average processing on the active power series and the reactive power series to obtain the average active power series P. data (i) and the average reactive power sequence Q data (i); Step 2: Based on the positive-sequence voltage amplitude sequence and negative-sequence voltage amplitude sequence obtained from the ABC three-phase voltage sequence, determine the fault start time t0 and the fault clearing time t according to the amplitude and rate of change of the positive-sequence voltage amplitude sequence. clear And based on the average active power sequence relative to the pre-fault steady-state active power P pre The convergence state determines the recovery termination time t end The recording process is divided into a fault phase, a recovery starting point after fault clearing, and a fault recovery phase. Step 3: Calculate the real-time voltage imbalance u(i) and the average voltage imbalance u during the fault phase based on the positive-sequence voltage amplitude sequence and the negative-sequence voltage amplitude sequence. avg Simultaneously, the phase angle sequence of the positive sequence voltage is extracted using a software phase-locked loop operator, and the voltage phase jump angle is calculated based on the phase angle difference before and after the fault clearing time. ; Step 4: Construct a three-segment chain power characteristic model for imbalance correction. The model includes active power characteristic functions and reactive power characteristic functions during the fault, active power characteristic functions and reactive power characteristic functions at the recovery starting point after fault clearing, and active power characteristic functions and reactive power recovery process characteristic functions during the fault recovery stage. The three-segment chain power characteristic model is specifically as follows: The active power characteristic function P1(t) during the fault stage is: ； Where P1(t) is the fitted calculated value of the active power at time t during the asymmetric fault; P pre This represents the measured average active power during the steady-state operation phase before the fault occurred; U + (t) represents the measured per-unit value of the positive sequence voltage at time t extracted from the three-phase voltage by discrete symmetrical component transformation; U n The system's rated voltage per unit value; u(t) is the real-time voltage imbalance; k p It is the positive sequence voltage suppression index; This is the imbalance sensitivity coefficient; The reactive power characteristic function Q1(t) during the fault stage is: Q1(t)=Q pre +K pos ·(U n -U + (t))-K neg ·U - (t)); Where Q1(t) is the fitted calculated value of reactive power at time t during the asymmetric fault; Q pre K represents the measured average reactive power during the steady-state operation phase before the fault occurred. pos For positive-sequence reactive power support gain; K neg For negative sequence reactive power compensation gain; U - (t) represents the measured per-unit value of the negative sequence voltage at time t extracted from the three-phase voltage by discrete symmetric component transformation; The active power characteristic function P2(t) at the recovery starting point after fault clearance is: P2(t)=P1(t clear )+γ asym ·(P pre -P1(t clear ))·u avg ; Where P2(t) is the fitted calculated value of the instantaneous active power during asymmetric fault clearing; P1(t) clear The characteristic function of the fault segment is calculated at t. clear Final value of the active power centerline at time γ; asym For asymmetric jump correction gain; P pre -P1(t clear ) represents the dynamic space where active power needs to be restored; u avg The average voltage imbalance is calculated from discrete measured data during the fault period. The reactive power characteristic function Q2(t) at the recovery starting point after fault clearance is: ； Where Q2(t) is the fitted calculated value of the reactive power during fault clearing; It represents the average value of the discrete reactive power within a specific window before fault clearance; This is the phase jump sensitivity coefficient; The active power characteristic function P3(t) during the fault recovery phase is: ； Where P3(t) is the fitted calculated value of active power at time t during the fault recovery phase; tt clear T is the relative time offset calculated from the time the fault was cleared. rec_asym P is the asymmetric active power recovery time constant; n is the trajectory shape exponent; P pre This represents the average active power during steady-state operation before the fault. The active power characteristic function Q3(t) during the fault recovery phase is: Q3(t)=Q pre +(Q2(t)-Q pre )·[exp(-(t-t clear ) / τ asym )]; Where Q3(t) is the calculated reactive power value fitted at time t during the fault recovery phase; Q pre The measured reactive power value during the steady-state operation phase before the fault; τ asym The reactive power recovery time constant; Step 5: Construct the asymmetric sensitivity-weighted integral operator: ； Where stage refers to the corresponding identification stage. The observation bias corresponds to the identification stage, where σ is the order of asymmetric sensitivity, ζ is the time-domain order, and t is the value of t. start T corresponds to the start time of the identification phase. stage This corresponds to the duration of the identification phase. This represents the average sliding window step size. By adjusting the asymmetric sensitivity order σ and the time domain order ζ of the operator, integral feature quantities with different physical meanings are extracted from the measured data of the fault stage and the fault recovery stage and the three-segment chain power characteristic model, including total area features, unbalance weighted area features and energy centroid distribution features. Step 6: Based on the integral feature quantity extracted in Step 5, construct a system of simultaneous equations for the model parameters. By solving the system of equations, identify all undetermined parameters in the three-segment chain feature model at once. Then, use the identified parameters to reconstruct a simulation trajectory that matches the axis characteristics in the measured data, thus completing the parameter identification of the asymmetric fault ride-through characteristics of the wind turbine. The parameters to be identified include: the imbalance sensitivity coefficient of the parameters to be identified during the fault stage. Positive sequence voltage suppression index k p Negative sequence reactive power compensation gain K neg Positive sequence reactive power support gain K pos The recovery starting point parameter to be identified after fault clearance is the asymmetric jump correction gain. and phase jump sensitivity coefficient The parameter to be identified during the fault recovery phase is the asymmetric active power recovery time constant T. rec_asym Trajectory shape exponent n and asymmetric reactive power return time constant .

2. The parameter identification method for asymmetric fault ride-through of wind turbine units as described in claim 1, characterized in that, Determine the fault start time t0 and the fault clearing time t clear and recovery end time t end The specific criteria are as follows: Fault start time t0: When the absolute value of the voltage change rate is greater than the first threshold. If the voltage amplitude is lower than the drop threshold and remains in this state within the first observation window, the sampling point that first meets the condition is determined as the fault start time t0; Fault clearing time t clear When the rate of change of voltage is greater than the second threshold If the voltage amplitude exceeds the recovery threshold and remains in this state within the second observation window, the starting sampling point of the second observation window is determined as the fault clearing time t. clear ; Restore end time t end When the active power enters the steady-state convergence zone centered on the active power before the fault and remains stable for a preset time without jumping out, the starting sampling point of this stable interval is determined as the recovery end time. Both the fault stage and the fault recovery stage include corresponding boundary sampling points.

3. The parameter identification method for asymmetric fault ride-through of wind turbine units as described in claim 1, characterized in that, The calculation methods for the positive-sequence voltage amplitude and negative-sequence voltage amplitude sequences are as follows: ； ； Where α=e j·2π / 3 ;U a U b and U c These are the three-phase voltages; The real-time voltage imbalance u(i) is: ； Average voltage imbalance u during the fault phase avg for: ； Where, N f This represents the total number of moving average windows included within the fault phase.

4. The parameter identification method for asymmetric fault ride-through of wind turbine units as described in claim 3, characterized in that, The N f for: 。 5. The parameter identification method for asymmetric fault ride-through of wind turbine units as described in claim 1, characterized in that, The input to the software phase-locked loop (PLL) operator is a three-phase voltage sampling sequence. The PLL operator tracks the positive-sequence voltage phase using a Clark / Park transform and a proportional-integral regulator. The bandwidth of the software PLL operator is 20Hz to 50Hz. The average positive-sequence voltage phase θ within a 5ms window before the fault clearing time is calculated. ) and the average phase value θ of the positive sequence voltage within a 5ms window after the fault clearing time. ), and calculate the phase jump angle. : ； When the absolute value of the phase difference is greater than π, the phase difference is subjected to 2π foldback processing, so that... .

6. The parameter identification method for asymmetric fault ride-through of wind turbine units as described in claim 1, characterized in that, Active power parameters during fault phase and k p The identification process is as follows: make ; Construct a system of two nonlinear equations: ； ； The Levenberg-Marquardt nonlinear least squares method is used to solve the two-variable nonlinear equation system to obtain the active power parameters during the fault stage. and k p Identification.

7. The parameter identification method for asymmetric fault ride-through of wind turbine units as described in claim 1, characterized in that, reactive power parameter K during fault phase pos and K neg The identification process is as follows: make ; Construct a system of two nonlinear equations: make If ζ=0, then we have: ； make If ζ=0, then we have: ； The parameter K is obtained by simultaneously solving the two sets of linear equations. pos and K neg Identification.

8. The parameter identification method for asymmetric fault ride-through of wind turbine units as described in claim 1, characterized in that, Recovery start parameters after fault clearance and The identification process is as follows: Calculate the average active power observation value P within the preset narrow window after the fault is cleared. obs and average observed reactive power Q obs ; Substituting the positive sequence voltage and real-time voltage imbalance in the fault-end window into the active power characteristic function of the fault stage, we obtain the final value P1(t) of the active power centerline during the fault stage. clear The fault end window is backed down by three sliding window steps relative to the fault clearing time. Constructing equations for parameter identification ; 。 9. The parameter identification method for asymmetric fault ride-through of wind turbine units as described in claim 1, characterized in that, The fault recovery phase parameter T rec_asym The identification process for n is as follows: make ; make Let ζ=0 and ζ=1 respectively, and construct the zeroth-order area moment and first-order time moment for the fault recovery phase: ; ; ; ； Among them, T stage =t end -t clear ; Calculate the normalized time center G= When G < 0.38, the initial value of n is set to 1; when G > 0.62, the initial value of n is set to 3; otherwise, the initial value of n is set to 2; n∈{1,2,3,4} is enumerated sequentially, and the Levenberg-Marquardt method is used to optimize T for each value of n. rec_asym Select the T with the smallest residual norm. rec_asym Combined with n, the parameter T is completed. rec_asym Identification of n.

10. The parameter identification method for asymmetric fault ride-through of wind turbine units as described in claim 1, characterized in that, The parameters of the fault recovery phase The identification process is as follows: Select the option that satisfies |Q during the fault recovery phase. data (t)-Q pre |>10 -5 , and t i >t clear Using an effective sampling window, perform linear regression on the formula Q3(t): ； Calculation parameters : ； When the number of effective sampling windows is less than 5, Set it to 0.1s.

Images