Stage-sensitive based wind power system stability risk assessment method and system, and medium

Abstract

The present application relates to the technical field of wind power system modeling and small signal stability analysis, and discloses a wind power system stability risk assessment method and system based on stage sensitivity and a medium. The external response of normal grid connection disturbance and recovery process is obtained, the response stage is divided and the stage contribution is checked, the bidirectional stage correlation feature is extracted, the stage weight is distributed through the stage prior of the control ring group and the effective stage mask, and the control parameter is identified synchronously; the parameter checked through physical back substitution is substituted into the wind power system physical model, linearization is carried out under the specified operating point and grid parameter, the complete closed-loop characteristic value, spectral abscissa and oscillation mode damping ratio are calculated, the small signal stability risk result is output, and the connection of the stage sensitive parameter identification result to the grid connection stability analysis is realized.

Description

Technical Field

[0001] This invention relates to the field of wind power system modeling and small disturbance stability analysis technology, and in particular to a stage-sensitive wind power system stability risk assessment method, system and medium. Background Technology

[0002] With the large-scale grid connection of wind turbines, the accuracy of wind power converter control parameters directly affects the dynamic analysis and simulation verification of the grid-connected system. However, actual control parameters are usually difficult to obtain directly and need to be identified based on externally measurable responses during disturbances. Existing technologies CN116227320A identify doubly-fed induction generator (DFIG) wind turbine control parameters by screening response features and using recurrent neural networks and optimization algorithms, while CN118199031A obtains phase-locked loop (PLL) and converter control parameters through time-domain fitting and algorithmic fitting, respectively.

[0003] However, existing technologies typically use complete responses or relevant features directly for identification, failing to adequately characterize the non-uniform distribution of control parameter sensitive information during disturbance and recovery phases. This results in effective response information under normal grid-connected weak disturbance conditions being easily weakened by low-sensitivity data, making it difficult to accurately and synchronously identify multiple control loop parameters of the wind power converter.

[0004] When the control parameter identification results are used for grid-connected stability analysis, they also need to be combined with the physical equations of the wind turbine and the power grid, and a closed-loop dynamic model needs to be obtained at an operating point consistent with the evaluation conditions. The parameter back-substitution response satisfies the verification conditions, indicating that the parameter set can reproduce the corresponding response under the verification conditions. However, it is still necessary to determine the model's damping margin and oscillation damping through closed-loop stability analysis to avoid judging the small-disturbance stability risk of the grid-connected system solely based on the response fitting results.

[0005] The information disclosed in this background section is intended only to enhance the understanding of the general background of this disclosure and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] This invention provides a method, system, and medium for assessing the stability risk of wind power systems based on stage sensitivity, which can effectively solve the problems in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A stage-sensitive wind power system stability risk assessment method, the method comprising: Acquire external response data of the wind power converter during the disturbance and recovery process while maintaining normal grid-connected control mode; The external response data is divided into multiple response stages, and response subsequences for each response stage are constructed. A multi-stage response sequence is then formed based on the stage statistical characteristics of each response stage. Calculate the stage contribution information of each response stage, verify the effectiveness of each response stage, and construct a stage-sensitive response sequence based on the stage statistical characteristics of the effective response stages in the original evolution order. Along the perturbation evolution direction and the recovery constraint direction, the positive stage correlation features and the reverse stage correlation features of the stage sensitive response sequence are extracted respectively. According to the phase-locked loop control group, grid-side converter control group, and rotor-side converter control group to which the control parameter to be identified belongs, the stage contribution information is converted into stage priors and effective stage masks for each control loop group, and the stage priors and effective stage masks are used to guide the stage weight redistribution of each control loop group to obtain the target stage features corresponding to each control loop group. Based on the target stage characteristics corresponding to each control loop group, the control parameter identification results of multiple control loops are obtained synchronously through the dedicated output branches of each control loop group. The control parameter identification results are substituted into the wind power system physical model for response back-substitution verification. A closed-loop small disturbance model under the specified operating point and grid parameters is established using the verified control parameters. The complete characteristic values ​​of the closed-loop small disturbance model are calculated, and the stability risk index is determined based on the complete characteristic values. The small disturbance stability risk assessment results under the specified operating point and grid parameters are then output.

[0008] Furthermore, the disturbance effect is a weak voltage disturbance that does not trigger the low voltage ride-through, current limiting, or protection logic of the wind power converter.

[0009] Furthermore, the plurality of response stages include a first response stage as a pre-disturbance steady-state stage, a second response stage as a disturbance initial response stage, a third response stage as a disturbance holding stage, a fourth response stage as a recovery attenuation stage, and a fifth response stage as a post-disturbance steady-state stage.

[0010] Furthermore, the wind power converter is a doubly fed wind turbine converter, and the external response data includes three-phase stator voltage, three-phase stator current, stator-side active power, stator-side reactive power, grid connection point voltage RMS value and DC bus voltage. The stage statistical characteristics are the statistical characteristics corresponding to the external response data in each response stage.

[0011] Furthermore, the determination of the stage contribution information includes: A parameter perturbation is applied to each of the control parameters to be identified, and the response sensitivity is calculated based on the change in external response before and after the parameter perturbation. The response sensitivity is normalized according to the control parameter scale and the response data scale, and the stage contribution information is determined according to the energy proportion of the normalized response sensitivity in each response stage.

[0012] Furthermore, based on the stage contribution information corresponding to each of the control parameters to be identified, the stage prior corresponding to each control loop group is determined according to the control loop group to which the control parameter to be identified belongs, and attention bias information in the stage weight redistribution process is generated based on the stage prior; an effective stage mask is generated based on the validity verification result of each response stage to mask invalid response stages in the stage weight redistribution process.

[0013] Furthermore, the forward stage association features and the reverse stage association features are extracted through a forward gated loop unit and a reverse gated loop unit, respectively. The forward stage association features and the reverse stage association features corresponding to the same response stage are concatenated to form stage association features. The stage association features are input into a multi-head attention mechanism, and the attention bias information and the effective stage mask are added to the attention score of the multi-head attention mechanism to mask invalid response stages and redistribute stage weights to each response stage to obtain the target stage features.

[0014] Furthermore, the plurality of control loops include a phase-locked loop, a grid-side converter DC voltage outer loop, a grid-side converter current inner loop, a rotor-side converter speed outer loop, and a rotor-side converter current inner loop. The control parameter identification results include the proportional control parameters and integral control parameters corresponding to each of the control loops.

[0015] Furthermore, when the wind power converter is a doubly-fed induction generator (DFIG) wind turbine converter, the small disturbance stability risk assessment includes: writing the control parameter identification results, which have been verified by response back substitution and have valid indicators, into a physical model containing the DFIG induction generator, rotor-side converter, grid-side converter, DC bus, phase-locked loop, and power grid; obtaining the equilibrium state that satisfies dynamic equilibrium and network constraints based on the specified operating point and power grid parameters; linearizing the physical model in this equilibrium state; and eliminating the algebraic variables when the Jacobian matrix of the algebraic constraints is invertible with respect to the algebraic variables, thereby obtaining a closed loop that retains all independent dynamic states of the physical model. A small-loop perturbation model is used; all eigenvalues ​​of the closed-loop small-loop perturbation model are calculated, and the maximum value of the real part of the eigenvalues ​​is used as the spectral abscissa. When oscillation modes exist, the damping ratio of each oscillation mode and its minimum value are calculated. Based on the spectral abscissa, the minimum damping ratio and the pre-set risk assessment threshold, the stability risk results under the specified operating point and grid parameters are output. When there are no oscillation modes, the damping ratio index is marked as inapplicable, and a judgment is made based on the spectral abscissa. When the parameter back-substitution check fails, the equilibrium state is obtained, the Jacobian matrix is ​​not invertible, or the result is at the judgment boundary, a verification flag is output.

[0016] A wind power system stability risk assessment system based on phase-sensitive response, the system comprising: The external response acquisition module acquires external response data of the wind power converter during the disturbance and recovery process while maintaining normal grid-connected control mode. The phased response construction module divides the external response data into multiple response phases, constructs response subsequences for each response phase, and forms a multi-phase response sequence based on the phase statistical characteristics of each response phase. The phase contribution verification module calculates the phase contribution information of each response phase, verifies the effectiveness of the response phase, and constructs a phase-sensitive response sequence based on the phase statistical characteristics of the effective response phases in the original evolution order. The bidirectional correlation extraction module extracts the positive and negative stage correlation features of the stage-sensitive response sequence along the perturbation evolution direction and the recovery constraint direction, respectively. The stage weight allocation module converts the stage contribution information into stage priors and effective stage masks for each control loop group according to the phase-locked loop control group, grid-side converter control group, and rotor-side converter control group to which the control parameter to be identified belongs. The stage priors and effective stage masks are then used to guide the stage weight reallocation of each control loop group to obtain the target stage features corresponding to each control loop group. The control parameter identification module obtains the control parameter identification results of multiple control loops simultaneously through the dedicated output branches of each control loop group, based on the target stage characteristics corresponding to each control loop group. The modeling and stability risk assessment module substitutes the control parameter identification results output by the control parameter identification module into the wind power system physical model for response back-substitution verification. Using the verified control parameters, a closed-loop small disturbance model is established under specified operating points and grid parameters. The complete closed-loop characteristic values ​​and stability risk indicators are calculated, and the small disturbance stability risk assessment results under the specified operating points and grid parameters are output.

[0017] A computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, can implement the phase-sensitive wind power system stability risk assessment method described above.

[0018] The technical solution of this invention can achieve the following technical effects: By acquiring external response data during the normal grid connection disturbance and recovery process of wind power converters, a phased response sequence is constructed and the effectiveness of the phase is verified by combining the response sensitivity. Then, the bidirectional phase correlation features of the disturbance and recovery directions are extracted, and the phase weights are redistributed by combining the phase importance represented by the phase sensitivity contribution information. In this way, multiple control loop parameters can be identified simultaneously, which effectively solves the problem that it is difficult to accurately identify multiple control loop parameters simultaneously under normal grid connection weak disturbance conditions of wind power converters due to the uneven distribution of control parameter sensitivity information in different response phases.

[0019] This invention also connects the control parameters obtained from the stage-sensitive response with the physical model of the wind power system, so that the identification results can be used for closed-loop small disturbance analysis under specified operating conditions. By simultaneously checking the real parts of all eigenvalues ​​and the damping of the oscillation mode, it distinguishes between instability, insufficient damping margin, and insufficient oscillation damping. Furthermore, by evaluating the scope of application of the results through parameter validity and model solution constraints, it provides a basis for verifying stability risks under corresponding operating conditions.

[0020] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1This is a flowchart illustrating a phase-sensitive wind power system stability risk assessment method. Figure 2 This is a topology diagram of a typical vector control structure doubly fed wind turbine converter grid-connected model. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0025] Example 1;

[0026] like Figure 1 As shown, this application provides a stage-sensitive wind power system stability risk assessment method, the method including: S10: Acquire external response data of the wind power converter during the disturbance and recovery process while maintaining normal grid-connected control mode; S20: Divide the external response data into multiple response stages, construct response subsequences for each response stage, and form a multi-stage response sequence based on the stage statistical characteristics of each response stage. S30: Calculate the stage contribution information of each response stage, verify the effectiveness of the response stage, and construct the stage sensitive response sequence according to the original evolution order based on the stage statistical characteristics of the effective response stages. S40: Extract the positive stage correlation features and the reverse stage correlation features of the stage-sensitive response sequence along the perturbation evolution direction and the recovery constraint direction, respectively; S50: According to the phase-locked loop control group, grid-side converter control group and rotor-side converter control group to which the control parameter to be identified belongs, the stage contribution information is converted into the stage prior and effective stage mask of each control loop group respectively, and the stage prior and effective stage mask are used to guide the stage weight redistribution of each control loop group to obtain the target stage characteristics corresponding to each control loop group. S60: Based on the target stage characteristics of each control loop group, the control parameter identification results of multiple control loops are obtained synchronously through the dedicated output branches of each control loop group. S70: Substitute the control parameter identification results into the wind power system physical model for response back-substitution verification, use the verified control parameters to establish a closed-loop small disturbance model under specified operating point and grid parameters, calculate the complete closed-loop characteristic value and stability risk index, and output the corresponding small disturbance stability risk assessment results.

[0027] Specifically, this embodiment takes the converter of a doubly fed wind turbine as the object to be identified. The doubly fed wind turbine includes a phase-locked loop, a grid-side converter, a rotor-side converter, a DC bus, and a grid connection point measurement device. The control parameters to be identified include the proportional control parameters and integral control parameters corresponding to the phase-locked loop, the outer loop of the DC voltage of the grid-side converter, the inner loop of the current of the grid-side converter, the outer loop of the speed of the rotor-side converter, and the inner loop of the current of the rotor-side converter. The wind power converter is controlled in normal grid-connected control mode, and a weak voltage disturbance is applied at the grid connection point that can stimulate control dynamics but does not trigger undervoltage ride-through, current limiting, or protection logic. Normal grid-connected control mode means that the wind power converter continues to operate according to the phase-locked loop, power, DC voltage, and current control logic used before the disturbance, without switching to fault ride-through or protection control logic. The weak voltage disturbance can be generated by small step jumps, pseudo-random small disturbances, natural voltage fluctuations in the field, or voltage drops on the test platform. The data collection begins before the disturbance occurs, including three-phase stator voltage, three-phase stator current, stator-side active power, stator-side reactive power, grid connection point voltage RMS value, and DC bus voltage, and continues until the wind power converter re-enters a stable state after the disturbance is resolved. External response data refers to a multivariate time series formed by arranging the above response quantities in ascending order of sampling time. Stator-side active power and stator-side reactive power can be directly obtained by the measuring device or calculated based on the stator voltage and stator current components in a synchronous rotating coordinate system. Based on the disturbance start time, disturbance recovery start time, pre-disturbance observation time, post-disturbance observation time, initial disturbance response window length, and recovery attenuation window length, the external response data is divided into the pre-disturbance steady-state stage, the initial disturbance response stage, the disturbance holding stage, the recovery attenuation stage, and the post-disturbance steady-state stage. The response stage refers to the continuous time interval with the same or similar control dynamic meaning during the disturbance effect and recovery process. The pre-disturbance steady-state stage is used to provide the response baseline, the initial disturbance response stage is used to characterize the rapid dynamics of the phase-locked loop and the inner current loop, the disturbance holding stage is used to characterize the outer loop regulation, power control, and DC bus voltage changes, the recovery attenuation stage is used to characterize the controller integral state release, power callback, and dynamic attenuation, and the post-disturbance steady-state stage is used to characterize the steady-state level of the system after recovery. Statistical characteristics of each stage are calculated for the external response data within each response stage, and the statistical characteristics of each stage are arranged according to the order of disturbance evolution to form a multi-stage response sequence. Response characteristics refer to statistical quantities used to describe the change state of each response quantity within a response stage. In this embodiment, the stage mean is used as the response characteristic. The multi-stage response sequence is a feature matrix with response stages as rows and external response variables as columns. The order of stages in the matrix is ​​consistent with the time order of disturbance and recovery process. The stage statistical characteristics can also be root mean square value, maximum value, minimum value, peak-to-peak value, slope, area, recovery time, or damping ratio estimate. During the offline training phase, different parameter combinations are set for each control parameter to be identified. The corresponding external response data is obtained through simulation models or experimental platforms, and multi-stage response sequences are constructed in the same way. Each multi-stage response sequence and the corresponding real control parameter vector form a training sample, and all training samples form a training sample set. The mean of input features, the standard deviation of input features, the mean of control parameters, and the standard deviation of control parameters are all calculated only based on the training sample set and used for subsequent input standardization and output inverse standardization. For each control parameter to be identified, positive and negative parameter perturbations are performed based on the current parameter value to obtain the corresponding external response under the two parameter values. The difference between the two external responses is divided by twice the parameter perturbation amount to obtain the response sensitivity of the control parameter to the external response data. The response sensitivity is used to represent the degree of change of each external response quantity when the control parameter changes by a unit. Based on the standard deviation of the control parameter to be identified in the training sample set and the standard deviation of the responses of each external response quantity in the training sample set, the response sensitivity is normalized to eliminate the influence caused by the different dimensions of the control parameter and the response quantity. For each control parameter to be identified, the L2 norm energy of the normalized response sensitivity in each response stage is calculated, and the sensitivity energy in a response stage is divided by the sum of the sensitivity energies of all response stages to obtain the stage sensitivity contribution of the control parameter to be identified in that response stage. The stage sensitivity contribution refers to the relative contribution of a response stage to the identification information of the corresponding control parameter. The stage sensitivity contribution is used on the one hand to check whether each response stage contains effective parameter identification information, and on the other hand to characterize the relative importance of each response stage to the control parameter to be identified, and serves as the physical prior information for the subsequent stage weight redistribution. For each response stage, the maximum stage sensitivity contribution of all control parameters to be identified within that response stage is selected, and this maximum stage sensitivity contribution is compared with the average contribution level when the sensitivity is uniformly distributed across all response stages. If the maximum stage sensitivity contribution is not lower than the average contribution level, the corresponding response stage is determined as a valid response stage. Validity verification involves using the stage sensitivity contribution to confirm whether the response stage contains sufficient valid dynamic information to support the identification of at least one control parameter to be identified. Validity verification is performed on each response stage in the multi-stage response sequence, maintaining the original arrangement order and corresponding response characteristics of each response stage. An valid stage mask is generated based on the validity verification results of each response stage, letting the th... The validity of each response phase is indicated by When valid When invalid Before the state update of the bidirectional gated loop unit, the validity flag is used to mask each response stage so that the invalid response stage does not update and does not change the hidden state of the adjacent valid stage; the valid stage mask is further used for the attention score calculation of the subsequent multi-head attention mechanism so that the invalid response stage does not participate in feature propagation in both the bidirectional stage association extraction and stage weight redistribution stages. The stage features of each response stage are standardized using the mean and standard deviation of the input features from the training sample set. Let the standardized stage feature be... The input features of each stage are The state update of the bidirectional gated loop unit is performed according to the validity flag; the forward processing direction refers to the time sequence from the pre-disturbance steady state stage through the disturbance initial response stage, the disturbance holding stage and the recovery decay stage to the post-disturbance steady state stage, and the reverse processing direction is opposite to the forward processing direction. The forward-gated loop unit processes each response stage sequentially along the forward processing direction to obtain forward stage correlation features; the reverse-gated loop unit processes each response stage sequentially along the reverse processing direction to obtain reverse stage correlation features; the forward stage correlation features are used to characterize the dynamic influence of the preceding perturbation stage on the subsequent stage, and the reverse stage correlation features are used to characterize the reverse correlation between the recovery state and the preceding perturbation stage; the forward stage correlation features and the reverse stage correlation features corresponding to the same response stage are concatenated to form a stage correlation feature matrix; The stage association feature matrix is ​​input into the multi-head attention mechanism, and the weight allocation of each response stage is guided according to the aforementioned stage contribution information. Each attention head performs query mapping, key mapping, and value mapping on the stage association features. Based on the correlation between query features and key features, and combined with the stage contribution information, the stage weights between different response stages are determined, and then the value features are weighted using stage correlation weights. The weighted features output by each attention head are concatenated and the output mapping is completed to obtain the target stage features. The target stage features refer to the features formed after bidirectional stage association extraction and stage weight redistribution, which are used to highlight the response stages that contribute more to the identification of control parameters and reduce the impact of low-sensitivity response stages on parameter mapping. Effective stage pooling is performed on the target stage features of each control loop group, and the corresponding dedicated output branch of the control loop group is input. The normalized control parameter estimates are denormalized using the control parameter mean and standard deviation in the training sample set to obtain the physical quantity estimates of each control parameter. The proportional control parameters and integral control parameters corresponding to the phase-locked loop, grid-side converter DC voltage outer loop, grid-side converter current inner loop, rotor-side converter speed outer loop, and rotor-side converter current inner loop are output simultaneously. After the parameter identification model completes offline training, for a new wind power converter to be identified, the external response data is obtained in the same way, the response stages are divided, a multi-stage response sequence is constructed, the validity is verified, the bidirectional stage correlation features are extracted, and the stage weights are redistributed. The control parameter identification results of multiple control loops can be obtained directly without repeatedly calling the simulation model for iterative optimization for each set of data to be identified.

[0028] The technical solution of this invention obtains external response data during the normal grid connection disturbance and recovery process of a wind power converter, constructs a stage response sequence, verifies the stage effectiveness by combining response sensitivity, extracts bidirectional stage correlation features in the disturbance and recovery directions, and redistributes stage weights by combining the stage importance represented by stage sensitivity contribution information, thereby simultaneously identifying multiple control loop parameters. This effectively solves the problem that it is difficult to accurately identify multiple control loop parameters simultaneously under normal grid connection weak disturbance conditions of wind power converters due to the uneven distribution of control parameter sensitivity information in different response stages.

[0029] Furthermore, the disturbance effect is a weak voltage disturbance that does not trigger the low-voltage ride-through, current limiting, or protection logic of the wind power converter; As a preferred embodiment of the above, the disturbance effect adopts a weak voltage disturbance at the grid connection point; a weak voltage disturbance refers to a voltage change process that can stimulate the control dynamics of the phase-locked loop, grid-side converter and rotor-side converter, while preventing the wind power converter from entering the low voltage ride-through, current limiting or protection control state; during the disturbance, the wind power converter continues to operate using the normal grid connection control logic before the disturbance, and the control mode does not switch. In practice, after the wind power converter is stably connected to the grid, a small step change is applied to the grid connection point voltage, causing the grid connection point voltage to drop by 5% to 10% from the level before the disturbance. After maintaining this for a predetermined time, the disturbance is removed, allowing the grid connection point voltage to return to the level before the disturbance. External response data are collected before the disturbance occurs, during the disturbance holding period, and after the disturbance is resolved. The disturbance amplitude is set according to the low voltage ride-through threshold, current limiting threshold, and protection threshold of the wind power converter to ensure that the disturbance process can generate a measurable dynamic response without triggering the corresponding control logic. In other implementations, weak voltage disturbances can also be pseudo-random small disturbances, natural voltage fluctuations in the field, or voltage drops generated by the experimental platform. When using natural voltage fluctuations in the field, voltage change segments that do not cause low voltage ride-through, current limiting, or protection actions are selected as disturbances and recovery processes. When using voltage drops from the experimental platform, the grid connection point voltage is adjusted through a grid simulation device, and the wind power converter is kept in normal grid connection control mode.

[0030] Furthermore, the multiple response stages include a first response stage as the pre-disturbance steady-state stage, a second response stage as the initial response stage of the disturbance, a third response stage as the disturbance sustaining stage, a fourth response stage as the recovery and decay stage, and a fifth response stage as the post-disturbance steady-state stage.

[0031] As a preferred embodiment of the above, the start time of the disturbance is determined. Disturbance resection or recovery start time Observation time before disturbance Observation time after disturbance Length of the initial response window to the disturbance and recovery attenuation window length ,in This indicates the moment when the grid connection point voltage begins to deviate from the pre-disturbance level. This indicates the moment when the grid connection point voltage begins to recover. and These represent the time lengths used to observe the steady-state level before and after the disturbance, respectively. This indicates the length of the window used to capture rapid dynamics after a disturbance occurs. This indicates the window length used to capture decay dynamics after the disturbance is recovered; First, the nominal division of the five response phases is represented by the following time windows: ; ; ; ; ; in to These represent the steady-state region before disturbance, the initial response region after disturbance, the sustained response region after disturbance, the recovery and decay region, and the steady-state region after disturbance, respectively. Will Used to provide a response baseline, Used to characterize the fast dynamics of phase-locked loops and the inner current loop, Used to characterize outer loop regulation, power control, and DC bus voltage changes. Used to characterize the controller's integral state release, power callback, and dynamic decay processes. Used to characterize the steady-state level of the system after recovery; The boundary of the actual response phase is determined by the grid connection point voltage event, the normalized response rate of change, and the recovery residual. The above time window is used to describe the phase composition. The phase characteristics are calculated using the following detected and verified boundary. If the detection fails, the corresponding rollback is performed. Let the sampling time be ,in To analyze the starting point of the record, For sampling sequence number, Let be the sampling period; let the external response vector be . ,in The number of response channels; take no less than [number] before the disturbance. The baseline set consists of consecutive valid sampling points. ,in , The minimum baseline duration specified for the model verification task. This represents the floor function; the first... The baseline mean, standard deviation, and measurement resolution of each response channel are denoted as follows: , and And construct the scaling matrix: ; in, The value is given by the resolution of the measuring device or the minimum quantization step size of the data acquisition card, therefore There is no division by zero term; define the normalized rate of change of response and the recovered residual: ; ; in, Let the effective value of the grid connection point voltage be... Its baseline mean and measurement resolution are respectively and Define the relative voltage deviation: ; The rate of change threshold, recovery threshold, and voltage event threshold are determined by the baseline empirical quantiles: ; ; ; in, Indicating sample experience Quantiles , and These are the allowable false trigger rates for the rate of change, recovery residual, and voltage event detection, respectively, and are registered in the model verification task book or data quality rules; when the baseline samples are insufficient to calculate the required quantiles, adaptive boundary conditions are not enabled, and a baseline data insufficiency flag is output; assuming the nominal frequency of the power grid is... The number of consecutive confirmed points is This ensures that the duration of continuous confirmation is not less than one fundamental frequency cycle; Disturbance application time Pick Continuously established The first point; the time of perturbation removal Pick after Continuously established The first point; the peak moment of the rate of change Take interval Internal satisfaction and right The first local maximum value that holds true for all conditions; the end time of a fast response. Pick after Continuously established The first point; restore the end time. Pick after and Simultaneously established The first point; the endpoint of the analysis The endpoint of this data recording is given, and must be later than [date / time]. At least one fundamental frequency period; The stage boundary must satisfy: ; Formed when satisfied , , , and ;in, The key is to retain the PLL and the fast dynamics of the inner current loop. The key features to be retained are the DC bus voltage outer loop, speed outer loop, and power regulation. The focus is on preserving the integral state release and recovery decay; the above correspondence only represents the main sources of sensitivity. If the disturbance command time stamp can be obtained, it will be used for verification. and ;like Not in If detected earlier, then and Merge into a single disturbed dynamic phase, retaining only the degradation event record without entering parameter identification; if Not in Pre-detection, sampling interval exceeds If a new event is triggered again before the previous event resumes or if any boundary is out of order, the window is marked as a stage division failure; when there is no instruction time stamp, for each detected complete group and Events are handled separately, only those events that have been resolved are handled separately. The next event is initiated only after the previous one has occurred; incomplete or overlapping events are not included in parameter identification. Define phase boundary state variables : Take when complete detection and satisfying ordered relation ;only Not in When the event is detected before the perturbation is applied, removed, and restored, the rapid response phase and the perturbation holding phase are combined and taken. When there are missing recovery boundaries, sampling interruptions, overlapping events, or out-of-order boundaries, take... ;only The five-stage complete window then proceeds to subsequent parameter identification. The window is only used for event archiving and boundary algorithm verification.

[0032] Furthermore, the wind power converter is a doubly fed wind turbine converter. The external response data includes three-phase stator voltage, three-phase stator current, stator-side active power, stator-side reactive power, grid connection point voltage RMS value and DC bus voltage. The stage statistical characteristics are the statistical characteristics corresponding to the external response data in each response stage.

[0033] As a preferred embodiment of the above, the externally measurable response vector at each sampling time is acquired or calculated: ; in Indicates the first Each sampling time, , , These represent the three-phase stator voltages, , , These represent the three-phase stator currents, Indicates the active power on the stator side. Indicates the reactive power on the stator side. Indicates the effective value of the grid connection point voltage. Indicates the DC bus voltage. Indicates the first The externally measurable response vector at each sampling time; For the first Statistical characteristics of the externally measurable response vector calculation stage within each response stage: ; in Indicates the first Within the first response phase Each sampling time, Indicates the first The number of sampling points within each response phase Indicates the first Statistical characteristics of each response phase; Arrange the stage statistical characteristics of each response stage according to the order of perturbation evolution to obtain the stage-sensitive response sequence: ; in Indicates a phase-sensitive response sequence. Indicates the number of response stages. This indicates the feature dimension of a single stage.

[0034] Furthermore, the determination of stage contribution information includes: Apply parameter perturbation to each control parameter to be identified, and calculate the response sensitivity based on the change in external response before and after parameter perturbation; The response sensitivity is normalized based on the control parameter scale and the response data scale, and the stage contribution information is determined based on the energy proportion of the normalized response sensitivity in each response stage. Furthermore, based on the stage contribution information corresponding to each control parameter to be identified, the stage prior corresponding to each control loop group is determined according to the control loop group to which the control parameter to be identified belongs, and attention bias information is generated in the stage weight redistribution process based on the stage prior; an effective stage mask is generated based on the validity verification results of each response stage to mask the invalid response stage in the stage weight redistribution process.

[0035] As a preferred embodiment of the above, for the first Calculate the response sensitivity of each control parameter to an externally measurable response: ; in Indicates the first One control parameter to be identified This represents the perturbation parameter used in finite difference finite differences. Indicates the first The control parameter to be identified is in the first... The response sensitivity corresponding to each sampling time; Indicates the first The externally measurable response vector at each sampling time; The response sensitivity is normalized using the standard deviation of the parameters and the standard deviation of the response quantity. ; in Indicates the normalized response sensitivity; Indicates the first The standard deviation of each control parameter in the training samples; This represents a diagonal matrix composed of the standard deviations of the various response quantities. Indicates the first The standard deviation of each response quantity in the training samples; Calculate the normalized response sensitivity at the th Energy percentage within each response phase: ; in Indicates the first The control parameter in the first... The relative sensitivity contribution in each response phase Represents the L2 norm, Indicates the total number of response stages. This indicates the stage index used when summing all response stages. Indicates the first The set of sampling times included in each response phase. Indicates the first The set of sampling times contained in each response phase; according to and Verify the effectiveness of the response phase, including Indicates the first The maximum relative contribution of each response stage to all control parameters to be identified. Represents the set of effective response phases. Indicates sensitivity at The average contribution level when the response stages are uniformly distributed; As a further optimization of the above-mentioned methods for determining and verifying the effectiveness of stage contribution information, the response stage can also be screened from the perspective of the joint identifiability of multiple control parameters to be identified. This avoids the impact of the response correlation between multiple control parameters on the synchronous identification results when judging solely based on the stage sensitivity contribution of a single control parameter. Specifically: A multi-parameter joint sensitivity matrix is ​​constructed in a unified dimensionless coordinate system, and an effective stage is selected based on numerical rank, dynamic stage coverage, condition number, and minimum eigenvalue. Let the parameter to be identified be ,in For the number of parameters; the first The mean and standard deviation of each parameter in the training set are: and Its standardized value is The normalized response value is ;No. The dimensionless sensitivity of each parameter is listed as follows: ; When the analytical derivative cannot be obtained, the central difference method is used. ,in For the first Unit vectors in each coordinate direction, This refers to the smallest parameter perturbation that can cause a change in response exceeding the measurement resolution during model verification. set up For baseline set The dimensionless covariance is defined in the original text, which is a dimensional response noise covariance. ; make To achieve maximum normalized measurement resolution, To calculate the machine precision of the floating-point format used, the regularization is: The whitening sensitivity is: ; in, for An identity matrix of order 1. The largest eigenvalue is represented; the inverse square root is calculated by performing eigenvalue decomposition on the above-mentioned translation-regularized matrix; [The remaining text appears to be incomplete and requires further context.] Composed of whitening sensitivity columns To ensure consistency across different sample dimensions, each stage is interpolated to a fixed time using normalized stage time. Each node; for the candidate stage set , No. The stacking matrix of the training samples is: ; in, This indicates vertical stacking in the row direction. , This represents the number of training samples. The summary information matrix of the training set is as follows: ; make The numerical rank threshold is The candidate set must simultaneously satisfy the following conditions: and and at least cover The two dynamic stages; sequentially sorting all qualified sets according to the stage number. Ascending order, condition number Ascending order, minimum eigenvalue Sort by descending order and lexicographical order of stage numbers, and the first set is determined as follows: This eliminates the uncertainty of choice caused by "smaller" or "larger" options; express ,otherwise . , and After summarizing and fixing the training set on the same calibration operating domain, when the identified sample exceeds the operating domain, switch to the corresponding calibration domain or output an out-of-domain flag; if there is no qualified set, output all parameters or the corresponding parameter group cannot be jointly identified, and do not force the identification result to be given. Using the aforementioned joint sensitivity information matrix, under low voltage ride-through, current limiting, and protection constraints, the smallest weak disturbance that meets the identification requirements is selected from a finite candidate set. To be compatible with the five-stage partitioning of application, hold, and removal, candidate perturbations are limited to single weak voltage steps or equivalent rectangular pulses; let the perturbation variable be... ,in This represents the change in voltage amplitude. To maintain the duration, the perturbation waveform is as follows: right It is true at all times, and 0 at all other times; let the voltage command resolution be... The minimum and maximum number of continuous sampling points are and Then the finite candidate set is: ; in, , To ensure that the maximum candidate amplitude does not exceed the equipment protection boundary, , The duration of disturbance retention shall not exceed the maximum allowable duration for a single verification. Indicates the floor operation; total number of candidates This refers to the maximum number of searches. All of the above quantities are read and recorded from the controller configuration, sampling settings, and verification task sheet. Let the low voltage ride-through threshold, converter current limit value, and protection status variables be respectively... , and Voltage and current prediction errors on the independent model validation set The quantiles are respectively and The voltage and current measurement resolutions are respectively and Define safety margin and The candidate perturbation must satisfy the following conditions throughout the entire process: ; in, The converter current vector, The probability of exceeding the safety prediction error limit is registered in the task specification. This is calculated using a simulation model, digital twin model, or validated sensitivity model of the current calibrated operating domain. For non-zero nominal PI parameters The recognition accuracy constraint is: ; in, For the first The standard deviation of the training set for each parameter. The relative standard uncertainty of the target registered by the applicant in the model verification task book or acceptance rules; only when... Calculate its inverse when the aforementioned full-rank condition is met. If the nominal value of the parameter to be identified is zero, then the relative constraint is replaced with... , The absolute standard uncertainty is registered in the task specification. The disturbance energy is... Optimal perturbation according to Ascending order Ascending sum The ascending order is determined from candidates that satisfy safety, full rank, and precision constraints; if or If not configured, only security and identifiability filtering will be performed, and a precision requirement flag will be output; traversal If none of the candidate solutions are feasible, terminate the active perturbation design and wait for a single natural voltage event that meets the same safety conditions. If the conditions are still not met, output an insufficient data flag. In a unified dimensionless coordinate system, a joint sensitivity matrix corresponding to multiple control parameters to be identified is constructed, and the set of effective response stages is determined based on the numerical rank, dynamic stage coverage, condition number, and minimum eigenvalue of the information matrix corresponding to the joint sensitivity matrix.

[0036] Furthermore, the positive stage association features and the negative stage association features are extracted through the positive gated recurrent unit and the negative gated recurrent unit, respectively. The positive stage association features and the negative stage association features corresponding to the same response stage are concatenated to form the stage association features. The stage association features are input into the multi-head attention mechanism, and the attention bias information and the effective stage mask are added to the attention score of the multi-head attention mechanism to mask the invalid response stages and redistribute the stage weights of each response stage to obtain the target stage features.

[0037] Furthermore, the multiple control loops include a phase-locked loop, an outer loop for DC voltage of the grid-side converter, an inner loop for current of the grid-side converter, an outer loop for speed of the rotor-side converter, and an inner loop for current of the rotor-side converter. The control parameter identification results include the proportional control parameters and integral control parameters corresponding to each control loop.

[0038] As a preferred embodiment of the above, the stage-sensitive response sequence is standardized using training set statistics: ; in Represents a stage-sensitive response sequence The Middle The first response phase One characteristic, and They represent the first The mean and standard deviation of each feature in the training set. This represents the standardized stage characteristics; The standardized first Each stage feature vector By inputting the forward-gated loop unit and the reverse-gated loop unit respectively, we obtain: ; ; in, For the first The validity indicators of each response phase; when At that time, the state update of the corresponding response phase participating in the gating loop unit is performed. When the previous valid state remains unchanged, the invalid response phase is prevented from affecting subsequent valid response phases through the hidden state; the reverse gated loop unit uses the same valid phase mask for processing. express The OK; This is a positive stage correlation feature, representing the positive hidden state in the direction of perturbation evolution; This is a reverse stage association feature, representing the reverse hidden state in the direction of recovery constraint; By concatenating the forward and backward hidden states, we obtain the stage-related features. And form a stage-related feature matrix. ,in Indicates the first The bidirectional correlation characteristics of each response stage A stage-related feature matrix representing all response stages; The following first explains the basic operations of a single attention branch. The actual control loop group branches are expressed using weights incorporating stage priors and effective stage masks, as described later. The general query, key, and value matrices correspond to the same-name matrices in each group during group calculations. For the... Individual attention head calculation: ; ; ; ; ; And spliced ​​together ; in , and They represent the first The query matrix, key matrix, and value matrix of each attention head. , , and This represents a trainable weight matrix. Indicates the first Each attention head has a feature dimension. This represents the stage-related weight matrix. Indicates the first The stage-weighted features of the attention head output, Indicates the number of attention heads. This indicates the stage characteristics after the stage weights have been redistributed; The stage sensitivity contribution is converted into stage priors, attention biases and effective stage masks for each control loop group of PLL, GSC and RSC. No. The total energy of whitening sensitivity of each parameter in all training samples and the first parameter The stage energies are as follows: ; The energy set is obtained by using empty perturbation calibration samples with only content measurement resolution perturbation without changing parameters. ,by As the first The lower bound of information for each parameter, among which This is the false alarm rate for uncontrolled disturbances specified in the task description. Only when... Contribution during the time-computation phase: ; like Then the first The parameter is marked as insufficient information, and the uniform prior is not used to mask this state. Assume a control loop group... , For group The set of parameter subscripts; the prior for the group stage is: ; When the denominator is zero or the information of any of the required parameters in the group is insufficient, output a flag indicating that the parameter group is unrecognizable; let For trainable scalars, Defined only as the lower limit for the calculation of dimensionless logarithms and and construct ,in for One-dimensional column vector; effective stage mask The a-th and b-th elements are in When it is 0, in When the time is negative infinity, only invalid key stages are masked; Shared BiGRU stage correlation features, and set up dedicated multi-head attention and output heads for different control loops, while using joint sensitivity weighted error to constrain the direction of easy confusion of multiple parameters; Input the normalized stage sensitive response matrix into the shared BiGRU to obtain ,in To share the dimension of hidden features. For the ... Group 1 One's attention, ,definition: ; in, For the query, key, and value projection matrix, therefore , This represents the dimension of a single attention head. The attention matrix is: ; The softmax function is calculated along the bond-phase dimension. The multi-head output is: ; in, Output projection matrices for each group. The dimension of the group features; express No. Line, group context vector is If group Include If there are 1 parameter, the group output header will first provide the standardized parameters: ; in, and These are the output weights and biases, respectively; the results from each group are concatenated. Then, press Denormalization yields physical control parameters. ; make Fix the matrix for the training set The Diagonal elements; only when all required parameters satisfy... Only then can it be constructed. and ; No. The standardized parameter error of each training sample is The batch sample size is The error of the joint sensitivity weighted parameter is: ; Common parameter error Constrain the overall deviation of each parameter. use The direction of the joint error in the external response of the additional constraint of the cross term is easily confused; This indicates the correlation of response sensitivity and is not evidence of physical topological coupling within the controller; any If the information limit is not exceeded, instead of continuing training by adding an arbitrarily small constant, the corresponding parameter is output as an indistinguishable flag. The pooling and fully connected expressions without control loop group subscripts are used to illustrate basic operations. This embodiment uses the effective stage pooling and dedicated output branches of the aforementioned control loop groups, and concatenates the outputs of each group into a standardized parameter vector. The basic pooling operation averages the stage features after stage weight redistribution along the stage dimension to obtain the context vector: ; in express The OK, Indicates the number of response stages. A context vector representing the integration of information from each stage; Inputting the context vector into the fully connected layer yields standardized parameter estimates: ; in and This represents the weight matrix of the fully connected layer. and This represents the bias vector. Represents a non-linear activation function. This represents the standardized parameter estimate; By inverting the standardized parameter estimates using the mean and standard deviation of the training set parameters, we obtain: ; in Indicates the first The physical quantity estimate of each control parameter. Indicates the first Standardized estimates of each control parameter and They represent the first The mean and standard deviation of each control parameter in the training set; When the wind power converter is identified as a doubly-fed wind turbine converter, the output is: ; in This represents the proportional-integral control parameter estimation vector. and This indicates the proportional and integral parameters of the phase-locked loop. and This indicates the proportional and integral parameters of the outer loop DC voltage of the grid-side converter. and This indicates the proportional and integral parameters of the grid-side converter's inner current loop. and This indicates the outer loop proportional and integral parameters of the rotor-side converter speed. and This indicates the proportional and integral parameters of the rotor-side converter current inner loop; Let the first The perturbation of each sample is described as follows The running condition vector is It includes at least wind speed, active power, reactive power, grid voltage, network equivalence conditions, and initial state before disturbance; if all samples are fixed at the same operating point, the fixed conditions are specified in the model's applicable scope; a differentiable forward model is trained using samples generated by the same equipment model as the identification task, but independent of the identification test set. After outputting a complete response, it follows the same boundary as the measured data. Construction phase response matrix When training the recognition network, its parameters are fixed. The back-substitution response is: ; in, This indicates that stage features are calculated based on the same stage boundary; the first... Stage-standardized residual covariance Stage residuals estimated from independently calibrated or validated samples; if Then As the lower bound of the eigenvalues, we obtain ,in For the first Maximum normalized measurement resolution of stage features; Obtained from the phase priors of the three control loop groups The loss for weighted response reproduction in the high-sensitivity phase is: ; in, and These are the responses at the reproduction and target stages, respectively. For the available loss set... The total loss is To prevent a certain weight from approaching zero, let ,in For unconstrained training variables, , , Due to losses in the first round of training, This represents the maximum relative weight change factor registered in the training configuration. If there is no independently validated differentiable forward model, then... Delete Do not enable and Only the trained physical model is back-substituted; After training is completed, Substitute the simulation model or digital twin model of the wind power converter into the same... , and Calculate the back-substitution error The acceptance threshold is: ; in, To determine the number of independent validation samples, The allowed false rejection rate for model validation; take This corresponds to the 95th percentile of experience and is only considered a preferred configuration. Output identification parameters and valid flags when they are met; output verification flags when they are not met. If historical qualified values ​​exist, retain the parameters and timestamp of the most recent successful verification; if the first identification fails, do not output new valid identification values, but only return the nominal parameters with sources in the controller configuration as reference, along with invalid, low confidence, and write-to-controller flags.

[0039] In step S70, after obtaining the identification parameters with valid identifiers, a physical model of the wind power system for stability analysis is established. This physical model uses parameter substitution to verify the equipment structure, control logic, coordinate transformation, and other available equipment parameters. The identified proportional and integral parameters are written into the corresponding phase-locked loop, grid-side converter, and rotor-side converter control equations, and then connected to the network equations of the power grid being evaluated.

[0040] Taking a doubly-fed induction generator (DFIG) wind power system as an example, the physical model includes the electromagnetic dynamics of the DFIG induction generator, the mechanical speed dynamics, the control dynamics of the rotor-side and grid-side converters, the DC bus energy storage dynamics, and the grid connection circuit. It also retains the states of filters, lines, and other dynamic components in the adopted model. For the electromagnetic state in synchronous rotating coordinates, corresponding independent flux linkages or current states can be used; the proportional-integral controller retains its integral state, the phase-locked loop retains its angle and integral states, and the DC bus retains its voltage state. Ten identification parameters are used to determine the coefficients in the corresponding control equations. A unified phase reference is used to handle coordinate redundancy, retaining all independent dynamic states and their closed-loop coupling relationships.

[0041] The physical model used for stability analysis is written as: ; ; in, It is a column vector composed of all independent physical dynamic states. For algebraic variable column vector, Let be the vector on the right-hand side of the dynamic equation. For algebraic constraint vectors, The control parameter vector obtained from the aforementioned identification and verified by physical back-substitution; The fixed operating conditions and grid parameter vectors for this assessment include at least active power, reactive power or their corresponding given values, wind speed, grid-connected voltage, network topology, and equivalent parameters of lines or the grid. Interrelated quantities must satisfy power flow and equipment balance relationships and are not treated as independent arbitrary inputs. Equipment ratings and unidentified parameters are obtained from the original physical model or equipment configuration and are written in as known coefficients. and ; According to the specified and Seeking and The equilibrium state where both are zero is denoted as... and This state should be within the calibration operating range of the identification model and the normal grid-connected control mode, with each segment control element operating within the same allowable linearization segment. Taking the partial derivatives with respect to both dynamic and algebraic variables in this equilibrium state yields the following results. , , and The four of them are as follows: right , right , right and right The Jacobian matrix. When it is reversible, we get: ; ; in, The dynamic state increment is the state of relative equilibrium. To obtain the closed-loop small perturbation state matrix that retains all independent physical dynamic states after eliminating algebraic variables. , , and All calculations were performed under the same equilibrium state, the same identification parameters, and the same operating conditions. All closed-loop eigenvalues ​​were calculated. ; ; in, for The order of for An identity matrix of order 1, where λ is an eigenvalue variable. For the r-th eigenvalue, r traverses all eigenvalues ​​by algebraic multiplicity. There are eigenvalues, where Re represents the real part. Let be the horizontal axis of the spectrum, with units of per second. For eigenvalues ​​with non-zero imaginary parts, calculate the oscillation mode damping ratio: ; ; Where Im represents the imaginary part, | | represents the modulus of the complex eigenvalue. Let r be the dimensionless damping ratio of the r-th oscillation mode. The minimum damping ratio for all oscillation modes is given, and conjugate eigenvalues ​​correspond to the same damping ratio. If all eigenvalues ​​are real numbers, the minimum value is not calculated, the oscillation mode damping ratio index is marked as inapplicable, and the determination is continued based on the spectral abscissa. Prior to the assessment, register the spectral abscissa limits according to actual operational requirements. and the lower limit of the oscillation mode damping ratio and its sources, among which Less than zero, Greater than zero and less than one; When the output is greater than zero, there is a risk of instability due to small disturbances. Less than zero but greater than Risk of insufficient output attenuation margin; existence of oscillation mode and Less than At the same time, there is a risk of insufficient output oscillation damping. Only... Strictly less than And when oscillation modes exist Strictly greater than Only when the specified operating point and grid parameters are met will the output result satisfy the set stability margin requirement; when there is no oscillating mode, only the spectral abscissa condition applies. When the value equals zero or the corresponding indicator equals the set limit, it is marked as a judgment boundary and a verification flag is output. No result that meets the stability margin requirement is output. Parameters failed back-substitution verification, evaluation conditions exceeded the calibrated operating range, equilibrium state could not be obtained, and the control operating section did not meet the required linearization conditions. If the process is irreversible or all characteristic values ​​cannot be reliably obtained, record the corresponding reasons and output a verification flag; the evaluation results correspond to the specified model, operating point, and power grid parameters.

[0042] Example 2;

[0043] Based on the same inventive concept as Embodiment 1, the present invention also provides a wind power system risk assessment system based on stage-sensitive response, the system comprising: The external response acquisition module acquires external response data of the wind power converter during the disturbance and recovery process while maintaining normal grid-connected control mode. The phased response construction module divides the external response data into multiple response phases, constructs response subsequences for each response phase, and forms a multi-phase response sequence based on the phase statistical characteristics of each response phase. The phase contribution verification module calculates the phase contribution information of each response phase, verifies the effectiveness of the response phase, and constructs a phase-sensitive response sequence based on the phase statistical characteristics of the effective response phases in the original evolution order. The bidirectional correlation extraction module extracts the positive and negative stage correlation features of the stage-sensitive response sequence along the perturbation evolution direction and the recovery constraint direction, respectively. The stage weight allocation module converts the stage contribution information into stage priors and effective stage masks for each control loop group according to the phase-locked loop control group, grid-side converter control group, and rotor-side converter control group to which the control parameter to be identified belongs. The stage priors and effective stage masks are then used to guide the stage weight reallocation of each control loop group to obtain the target stage features corresponding to each control loop group. The control parameter identification module obtains the control parameter identification results of multiple control loops simultaneously through the dedicated output branches of each control loop group, based on the target stage characteristics corresponding to each control loop group. The modeling and stability risk assessment module substitutes the control parameter identification results output by the control parameter identification module into the wind power system physical model for response back-substitution verification. Using the verified control parameters, a closed-loop small disturbance model is established under specified operating points and grid parameters. The complete closed-loop characteristic values ​​and stability risk indicators are calculated, and the small disturbance stability risk assessment results under the specified operating points and grid parameters are output.

[0044] The system described above in this invention can effectively implement a stage-sensitive wind power system stability risk assessment method, and the technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0045] Example 3;

[0046] Based on the same inventive concept as Embodiment 1, the present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, can implement a phase-sensitive wind power system stability risk assessment method.

[0047] Example 4;

[0048] like Figure 2 As shown, the complete implementation process of this invention is illustrated using a grid-connected model of a doubly-fed induction generator (DFIG) converter with a typical vector control structure. The DFIG includes a PLL phase-locked loop, a GSC DC voltage outer loop, a GSC current inner loop, an RSC speed outer loop, and an RSC current inner loop. The parameters to be identified are: .

[0049] Step 1: Establish or connect the grid-connected object of the doubly-fed induction generator converter to be identified. The object includes the wind turbine, doubly-fed induction generator, rotor-side converter, grid-side converter, DC bus, phase-locked loop, and grid connection point measurement device; record the sampling period. ,in Let the time interval between two adjacent samples be denoted as ; let the be . Each sampling time is ,in Indicates the start time of data recording. Indicates the sampling sequence number; determines the start time of the disturbance. Disturbance recovery start time Observation time before disturbance Length of the initial response window to the disturbance , restore attenuation window length and recovery observation time after disturbance The variables mentioned above are used to define the data acquisition interval and response stage boundary in the embodiments, respectively. The parameter vector to be identified is denoted as: ; in Indicates the proportional parameter. Indicates integral parameters, PLL indicates phase-locked loop, GSC indicates grid-side converter, RSC indicates rotor-side converter, and subscripts are used. Indicates the outer loop of DC voltage, subscript Indicates the outer ring of rotational speed, subscript Indicates the inner current loop; The training sample statistics and sensitivity information used in this step are pre-calibrated offline, and the identification network is trained offline before being applied to the object to be identified. A finite-safe weak perturbation is selected; the low-voltage ride-through threshold is read. Traffic limiting value Protection status Voltage command resolution and allowed duration range ,form ;in, and For the minimum and maximum number of continuous sampling points, To account for the maximum candidate amplitude that does not exceed the protection boundary even after taking into account the safety margin, , , Therefore, the maximum number of checks is [number]. A single weak step or rectangular pulse; let and These are the independent verification of the lumped voltage and current prediction errors, respectively. Quantiles and The resolutions for voltage and current measurements are calculated respectively. and ,in Register the probability of exceeding the safety error limit in the task description; check each one. , and , The converter current vector is used; the joint sensitivity information is pre-calibrated during the offline training phase according to the rules described in step seven, and the information matrix of candidate disturbances is predicted accordingly. This matrix is ​​the training set average of the inner product of the whitening sensitivities of all parameters under candidate perturbations; for those satisfying the safety conditions and Candidates with full rank values, check ,in The standard deviation of the training set parameters. For non-zero nominal values ​​in the controller configuration, Register the target relative standard uncertainty for the model verification task; select the smallest candidate in the order of disturbance energy, duration and amplitude; output the corresponding flag when the accuracy requirement is not configured; stop active disturbance and wait for a natural voltage event that meets the same safety conditions when all candidates are unqualified; if still unqualified, output insufficient data.

[0050] Step 2: Apply a weak voltage disturbance at the grid connection point, setting the grid connection point voltage setpoint to... satisfy when , when , when ,in This indicates the rated voltage at the grid connection point before the disturbance. Indicates the magnitude of voltage sag; defines the voltage sag ratio. In this embodiment, the preferred embodiment is... That is, the grid connection point voltage is from Reduce to or The rationale is that this range can stimulate measurable dynamics in PLL, GSC, and RSC, while typically not triggering low-voltage ride-through, current limiting, or protection logic; in field applications, natural voltage fluctuations can be used instead of active disturbances, but these natural voltage fluctuations should meet certain requirements. And the protection logic was not triggered, among which and These represent the effective values ​​of the grid connection point voltage before and after the disturbance.

[0051] Step 3, at the measurement point from to Three-phase stator voltage was collected during the period. , , Three-phase stator current , , Effective value of grid connection point voltage and DC bus voltage subscript , , Indicates the differentiation of the three phases. Indicates the effective value of the grid connection point voltage. Indicates DC bus voltage; if active power and reactive power If the values ​​are not directly provided by the measuring device, they are calculated based on the voltage and current in a synchronously rotating coordinate system. ; ; in and Indicates stator voltage shaft and Axial components, and Represents stator current shaft and Axial components, Indicates the active power on the stator side. This indicates the reactive power on the stator side.

[0052] Step four, organize the sampling data obtained in step three into an externally measurable response vector: ; in Indicates the first The externally measurable response vector at each sampling time has a dimension of , Indicates the number of response variables; [to include all] Stored as response sequences in ascending order of sampling time. ,in This indicates the total number of sampling points.

[0053] Establish baseline scale and detection threshold. Take the duration before the perturbation. The continuous valid data within is ,according to structure ,in For the first Channel baseline standard deviation Measure the resolution for this channel. Calculate separately. , and ,in The baseline response mean. , and These are the grid connection point voltage, its baseline mean, and the grid connection point voltage measurement resolution, respectively. Indicating sample experience Quantiles; the permissible false trigger rate registered in the task description. , and calculate , and ;make This serves as the number of consecutive confirmation points. If the baseline sample is insufficient, this identification process terminates at this step and outputs a baseline data insufficiency flag.

[0054] Step 5: First, give the nominal time windows for the five response stages based on the disturbance time. , , , and ,in This indicates the steady-state region before the disturbance. This indicates the initial response region to the disturbance. Indicates the disturbance sustaining response zone. Indicates the recovery attenuation region. Indicates the steady-state region after the disturbance; counts the number of sampling points in each stage. ,in Indicates the first The number of sampling points within each response phase This indicates the number of elements in the set. The actual stage boundary and the number of sampling points are updated based on the following event detection results.

[0055] Detect events and form an orderly phase. According to continuous Click OK ,according to back continuous Click OK ;exist The first one in the country Determination of local maximum values ​​for points that are continuously non-increasing , and then back continuous Click OK ;exist Later and Simultaneously continuous Click OK .check ,in To record the endpoint. Completely detected and with ordered boundaries, the time-boundary boundary state variables of the timing stage. ;only Missing events, but when applied, excised, and restored, the events are complete, merging the rapid and hold phases, making... It only archives events; it restores events with missing boundaries, interrupted sampling, out-of-order boundaries, or overlapping events. And terminate the window recognition. Only for Window formation to And calculate the stage mean of each channel to form Stage response matrix.

[0056] Step 6: Calculate the stage mean characteristics for each response stage: ; in Indicates the first The statistical characteristics of each response phase; to Arranged in order of perturbation evolution, we obtain Stage-sensitive response sequence ,in This represents the model input matrix in the embodiment, where rows correspond to response stages and columns correspond to externally measurable response variables.

[0057] Step 7: During the offline training phase, a training sample set is formed using simulation samples or experimental samples with different combinations of PI parameters. ; in Indicates the sample number. This represents the total number of training samples. Indicates the first Phase-sensitive response sequences of individual samples, Indicates the first The true control parameter vector corresponding to each sample; the mean of the input features is calculated only from the training set. Input feature standard deviation Parameter mean and parameter standard deviation ,in Indicates the input feature number. Indicates the output parameter number; according to and Complete standardization, among which Represents standardized input features. This represents standardized parameters.

[0058] Calculate the joint sensitivity in a unified coordinate system. Use the mean of the training set parameters. and standard deviation definition and calculate ,in For the first A dimensionless sensitivity series of parameters. Derived from the baseline original dimensionless covariance. get ;make To achieve maximum normalized measurement resolution, For the machine precision of the floating-point format used, according to Determine the lower bound of the covariance eigenvalues, then calculate. ,in It is a 10th order identity matrix. The time for each stage is normalized and interpolated to a fixed value. Point, for candidate set Construct the first Matrix of training samples ,in Then by Calculate training samples , This represents the number of training samples. If the analytical derivative cannot be obtained, then... Calculation, where For the first A unit vector in each parameter direction. It is the smallest parameter disturbance that can cause a change in response exceeding the measurement resolution.

[0059] The determination of effective stages is used to generate effective stage masks in subsequent multi-head attention mechanisms, without changing the original arrangement and characteristics of each response stage in the stage-sensitive response sequence; the determination of effective stages and group priors retains those that simultaneously satisfy... and Furthermore, candidates covering at least two dynamic stages are selected, and the order is determined by ascending stage number, ascending condition number, descending minimum eigenvalue, and lexicographical order of stage number. ,in and These are the minimum and maximum eigenvalues, respectively. This is the numerical rank threshold. (Definition) and The two are respectively the first The training set average total sensitivity energy of the parameters and the first parameter Stage average energy. Energy of the sample calibrated by empty perturbation. Quantiles provide the lower bound of information. , The false positive rate for empty disturbances; when Time calculation Otherwise, mark the first Some parameters are unidentifiable. They are categorized by PLL parameter index set. GSC parameter subscript set and RSC parameter subscript set calculate ,in Indicates a control loop group. In the Phase belongs to If the value is 1, then the value is 0; otherwise, the value is 0. , and Within the same calibrated operating domain, the training set is aggregated and then fixed.

[0060] Step 8: Calculate the standardized stage-sensitive response sequence. The input is a bidirectional gated recurrent unit (BiGRU), where Indicates by The standardized input matrix is ​​composed of a forward gated recurrent unit (GRU) that evolves in the direction of the perturbation. arrive Calculation, reverse GRU from the direction of restoration constraint arrive Calculation; splicing to obtain and ,in This represents the stage-related feature matrix.

[0061] Step nine, The input is a multi-head attention mechanism (MHA). The basic attention operations are explained below; the actual control loop branching is expressed using the group priors and effective stage masks given later. The general query, key, and value matrices correspond to the same-named matrices in the group calculation, and the attention head indices in the two notations correspond accordingly. For the... Individual attention head calculation , , , and ,in Indicates the attention head number. , and These represent the query matrix, key matrix, and value matrix, respectively. , and This represents a trainable projection matrix. This indicates the dimension of the attention head feature. Represents the stage weight matrix. Indicates the first The outputs of each attention head are concatenated to obtain... ,in Indicates the number of attention heads. Indicates the output projection matrix. This represents the stage-weighted feature matrix.

[0062] Training sharing - dedicated recognition network. Standardization will be implemented. The stage response matrix is ​​fed into the shared BiGRU to obtain... ,in To share the hidden feature dimension, for each control loop group... Passing the exam One's attention, to , and form Query, key and value matrix, , and For the corresponding projection matrix, The dimension of a single attention head. (Will be determined by...) The resulting bias and invalid phase mask Attention scores are directly incorporated. Each group is obtained through multi-head concatenation, effective stage pooling, and independent output heads. , and The three parameters are standardized estimates of PLL, GSC, and RSC, which are concatenated and then... Ten physical control parameters are obtained. Let... For the first The trainable scalar of the group For machine precision, calculation and ,in A 5-dimensional all-one column vector; effective stage mask The value is 0 when the bond is valid and negative infinity when it is invalid. Individual attention points obtain group stage characteristics Then, press Pooling and Output, where For the number of attention heads, for No. OK, and These are the group output weights and biases, respectively.

[0063] Step 10: Obtain group parameters by pooling and dedicated output branches according to the effective stages of each control loop group, and concatenate them into a standardized parameter vector. The following pooling and fully connected expressions without control loop group subscripts are used to illustrate the basic operations, where... Pooling along the stage dimension yields: ; in express The OK, The context vector is represented; standardized parameter estimates are obtained through a fully connected layer. ,in and This represents the weight matrix of the fully connected layer. and This represents the bias vector. Represents a non-linear activation function. This represents the standardized parameter estimation vector.

[0064] Apply joint sensitivity-weighted error and optional response reproduction constraints. structure ,in It is a matrix composed of its diagonal elements. This is the joint sensitivity correlation matrix; if the information content of any diagonal does not exceed If the condition is met, training for the corresponding parameters will stop and an unrecognizable flag will be output. Indicates the first Standardized parameter error for each sample, This represents the number of training batch samples, calculated as follows: and These two are the ordinary parameter error and the joint sensitivity weighted error, respectively. If an independently verified differentiable forward model exists... Then the first Perturbation of each sample Operating conditions and estimated parameters Input the model and calculate the response reproduction loss at the same stage boundary. Otherwise, the loss will not be enabled. For each enabled loss... ,make ,in As trainable variables, and The weights are determined by the initial loss and the maximum relative weight change factor allowed by the training configuration, thus preventing any weight from degenerating to zero. When enabling response reproduction constraints, first... The response was obtained during the reproduction phase, among which This indicates that stage features are extracted according to the same stage boundary; let and calculate ,in The first value is estimated from independent calibration samples and processed by the lower bound of eigenvalues. Stage residual covariance. Weight boundaries are clearly defined. and , For the first round of losses, The maximum relative weight change factor registered for the training configuration.

[0065] Step eleven, according to Output the estimated physical parameters, where Indicates the first The identification results of each control parameter.

[0066] Perform physical backsubstitution and output the results. Substitute the 10 identified parameters into the wind power converter simulation model or digital twin model, under the same disturbances as measured. Operating conditions and stage boundaries Calculation ,in and The first Simulated and measured responses at each stage and The definition used is the same as the training phase. (By...) One independent validation sample and allowable false rejection rate get .when Time output Effective stage set, stage boundary state variables Output a valid flag; otherwise, output a verification flag. If historical qualified results exist, retain the most recent qualified parameters and timestamp; on the first failure, only return the nominal parameters with a source in the controller configuration, along with invalid, low confidence, and write-to-controller flags.

[0067] Step 12: Conduct a small-disturbance stability risk assessment using the ten control parameters verified through physical back-substitution. Figure 2 In the doubly-fed induction generator (DFIG) wind turbine grid-connected physical model shown, the proportional-integral (PI) parameters corresponding to the PLL, GSC DC voltage outer loop, GSC current inner loop, RSC speed outer loop, and RSC current inner loop are replaced respectively, while other equipment parameters retain their sourced configuration values. The wind speed, active and reactive power operation requirements, grid-connected voltage, network topology, and equivalent parameters of the lines or grid for this evaluation are selected to obtain a compatible equilibrium state and confirm that it is within the calibrated operating range and normal grid-connected control operating section. The calculations are performed according to the aforementioned physical model equations. , , and ,exist Reversible formation Retaining all independent dynamic states of the generator, machinery, DC bus, controller, and modeled grid-connected circuits, calculate all closed-loop eigenvalues. and when oscillation modes exist Compare with the registered limits, output the instability risk, insufficient damping margin risk, insufficient oscillation damping risk, whether the set stability margin requirements are met or the verification flag is checked, and record the parameters, model and operating condition used.

[0068] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are for illustrative purposes only and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. It is clear that those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application is intended to include such modifications and modifications.

Claims

1. A stage-sensitive wind power system stability risk assessment method, characterized in that, The method includes: Acquire external response data of the wind power converter during the disturbance and recovery process while maintaining normal grid-connected control mode; The external response data is divided into multiple response stages, and response subsequences for each response stage are constructed. A multi-stage response sequence is then formed based on the stage statistical characteristics of each response stage. Calculate the stage contribution information of each response stage, verify the effectiveness of each response stage, and construct a stage-sensitive response sequence based on the stage statistical characteristics of the effective response stages in the original evolution order. Along the perturbation evolution direction and the recovery constraint direction, the positive stage correlation features and the reverse stage correlation features of the stage sensitive response sequence are extracted respectively; According to the phase-locked loop control group, grid-side converter control group, and rotor-side converter control group to which the control parameter to be identified belongs, the stage contribution information is converted into stage priors and effective stage masks for each control loop group, and the stage priors and effective stage masks are used to guide the stage weight redistribution of each control loop group to obtain the target stage features corresponding to each control loop group. Based on the target stage characteristics corresponding to each control loop group, the control parameter identification results of multiple control loops are obtained synchronously through the dedicated output branches of each control loop group. The control parameter identification results are substituted into the wind power system physical model for response back-substitution verification. A closed-loop small disturbance model under the specified operating point and grid parameters is established using the verified control parameters. The complete characteristic values ​​of the closed-loop small disturbance model are calculated, and the stability risk index is determined based on the complete characteristic values. The small disturbance stability risk assessment results under the specified operating point and grid parameters are then output.

2. The method for assessing the stability risk of a wind power system based on stage sensitivity as described in claim 1, characterized in that, The disturbance is a weak voltage disturbance that does not trigger the low voltage ride-through, current limiting, or protection logic of the wind power converter.

3. The method for assessing the stability risk of a wind power system based on stage sensitivity as described in claim 1, characterized in that, The plurality of response stages include a first response stage as a pre-disturbance steady-state stage, a second response stage as a disturbance initial response stage, a third response stage as a disturbance holding stage, a fourth response stage as a recovery attenuation stage, and a fifth response stage as a post-disturbance steady-state stage.

4. The method for assessing the stability risk of a wind power system based on stage sensitivity as described in claim 1, characterized in that, The wind power converter is a doubly fed wind turbine converter. The external response data includes three-phase stator voltage, three-phase stator current, stator-side active power, stator-side reactive power, grid connection point voltage RMS value and DC bus voltage. The stage statistical characteristics are the statistical characteristics corresponding to the external response data in each response stage.

5. The method for assessing the stability risk of a wind power system based on stage sensitivity as described in claim 1, characterized in that, The determination of the contribution information at each stage and the validity verification of the response stage include: For the A parameter perturbation is applied to each control parameter to be identified. The response sensitivity is calculated based on the change in the external response before and after the parameter perturbation. The response sensitivity is then normalized according to the control parameter scale and the response data scale to obtain the normalized response sensitivity. ; Based on the energy percentage of the normalized response sensitivity in each response stage, calculate the first... The control parameter to be identified is in the first... The stage sensitivity contribution of each response stage: ； Calculate the first The maximum stage sensitivity contribution of each response stage to all control parameters to be identified: ； And determine the first according to the following formula Effectiveness indicators of each response phase : ； in, Index of the control parameter to be identified. For the response phase index, The total number of response phases, For the first The set of sampling times included in each response phase. Let be the set of sampling times contained in the i-th response phase. Represents the L2 norm; Indicates the first Each response phase is considered an effective response phase. Indicates the first The first response phase is an invalid response phase.

6. The method for assessing the stability risk of a wind power system based on stage sensitivity as described in claim 5, characterized in that, According to the control loop group to which the control parameter to be identified belongs, the stage sensitivity contribution is converted into stage prior, attention bias information and effective stage mask corresponding to each control loop group, including: Let the control loop group be , For belonging to the first The set of control parameter indices to be identified for each control loop group is used to determine the first control loop group based on the stage sensitivity contribution and effectiveness flag. The first control loop group is in the... Group-stage priors for each response phase: ； Based on the prior knowledge of the group stage, the first The first control loop group is in the... Attention bias in each response phase: ; in, ; Used to the first The trainable scalar corresponding to each control loop group The prior strength coefficient is mapped to a positive value. This is the positive lower limit for dimensionless logarithmic calculations; Construct bias vectors based on the attention biases corresponding to each response stage: ； And construct the first Attention bias matrices for each control loop group: ； in, for A one-dimensional column vector; Construct a valid stage mask matrix based on the validity flags of each response stage. , its first The elements are: ； in, For the query phase index, This is a key-stage index. This allows the invalid response phase to be masked in subsequent attention weight calculations when it is considered a key phase.

7. The method for assessing the stability risk of a stage-sensitive wind power system according to claim 6, characterized in that, The forward phase association features and the reverse phase association features are extracted by the forward gated loop unit and the reverse gated loop unit that update the state using the validity flag, respectively. make Indicates the first The input features of each response phase are used to update the hidden state of the forward-gated recurrent unit according to the following formula: ； The reverse gated loop unit uses the same effective stage mask processing, and its hidden state is updated according to the following formula: ； in, and They represent the first Each response phase corresponds to a forward hidden state and a reverse hidden state; when At this time, the corresponding gated loop unit keeps the adjacent hidden states unchanged so that the invalid response phase does not participate in the propagation of hidden states; The positive and negative hidden states corresponding to the same response stage are concatenated to form stage association features, and the stage association features are input into the multi-head attention mechanism corresponding to each control loop group. For the The first control loop group Each attention head has its stage weight matrix calculated according to the following formula: ； in, For the first The first control loop group The query matrix corresponding to each attention head. The corresponding key matrix, The feature dimension of a single attention head. For the first Attention bias matrices corresponding to each control loop group For the effective stage mask matrix, Calculation along the bond stage dimension; The corresponding value matrix is ​​weighted using the aforementioned stage weight matrix, and then multi-head concatenation is performed to obtain the first... The target stage characteristics corresponding to each control loop group.

8. The method for assessing the stability risk of a wind power system based on stage sensitivity according to claim 1, characterized in that, The multiple control loops include a phase-locked loop, a grid-side converter DC voltage outer loop, a grid-side converter current inner loop, a rotor-side converter speed outer loop, and a rotor-side converter current inner loop. The control parameter identification results include the proportional control parameters and integral control parameters corresponding to each control loop.

9. The method for assessing the stability risk of a stage-sensitive wind power system according to claim 4, characterized in that, The small disturbance stability risk assessment includes: The control parameter identification results obtained through response back-substitution verification will be written into the physical model including the doubly fed induction generator, rotor-side converter, grid-side converter, DC bus, phase-locked loop and power grid. The equilibrium state that satisfies dynamic equilibrium and network constraints is obtained based on the specified operating point and power grid parameters. The physical model is linearized based on the equilibrium state. When the Jacobian matrix of the algebraic constraints is invertible with respect to the algebraic variables, the algebraic variables are eliminated to obtain a closed-loop small perturbation model that retains all independent dynamic states of the physical model. Calculate all eigenvalues ​​of the closed-loop small perturbation model, take the maximum value of the real part of the eigenvalue as the abscissa of the spectrum, and calculate the damping ratio of each oscillation mode when there is an oscillation mode and obtain the minimum damping ratio; Based on the spectral abscissa, the minimum damping ratio, and the pre-set risk assessment threshold, output the stability risk results under the specified operating point and power grid parameters. When there is no oscillation mode, mark the damping ratio index as inapplicable and make a judgment based on the spectral abscissa. Output a verification flag when the parameter back-substitution verification fails, the equilibrium state is obtained without success, the Jacobian matrix is ​​invertible, or the result is at the judgment boundary.

10. A stage-sensitive wind power system stability risk assessment system, characterized in that, The system includes: The external response acquisition module acquires external response data of the wind power converter during the disturbance and recovery process while maintaining normal grid-connected control mode. The phased response construction module divides the external response data into multiple response phases, constructs response subsequences for each response phase, and forms a multi-phase response sequence based on the phase statistical characteristics of each response phase. The phase contribution verification module calculates the phase contribution information of each response phase, verifies the effectiveness of the response phase, and constructs a phase-sensitive response sequence based on the phase statistical characteristics of the effective response phases in the original evolution order. The bidirectional correlation extraction module extracts the positive and negative stage correlation features of the stage-sensitive response sequence along the perturbation evolution direction and the recovery constraint direction, respectively. The stage weight allocation module converts the stage contribution information into stage priors and effective stage masks for each control loop group according to the phase-locked loop control group, grid-side converter control group, and rotor-side converter control group to which the control parameter to be identified belongs. The stage priors and effective stage masks are then used to guide the stage weight reallocation of each control loop group to obtain the target stage features corresponding to each control loop group. The control parameter identification module obtains the control parameter identification results of multiple control loops simultaneously through the dedicated output branches of each control loop group, based on the target stage characteristics corresponding to each control loop group. The modeling and stability risk assessment module substitutes the control parameter identification results output by the control parameter identification module into the wind power system physical model for response back-substitution verification. Using the verified control parameters, a closed-loop small disturbance model is established under specified operating points and grid parameters. The complete closed-loop characteristic values ​​and stability risk indicators are calculated, and the small disturbance stability risk assessment results under the specified operating points and grid parameters are output.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which includes program instructions that, when executed by a processor, can implement the phase-sensitive wind power system stability risk assessment method as described in any one of claims 1-9.

Images