A fan stability test data analysis method and device

Abstract

The application discloses a fan stability test data analysis method and device, and the method comprises the following steps: collecting three-domain core channel time sequence and performing time delay alignment, calculating effective test reference indexes, and obtaining effective time sequence data sets; calculating working condition coupling disturbance causal coefficients of high-fidelity data segments in the effective time sequence data sets, determining high-risk working condition segment data sets and core risk frequency bands; separating and determining dominant cause excitation sources for extreme high-risk data segments and their core risk frequency bands, calculating instability positioning coefficients to determine key disturbance amplification and conduction paths; constructing a correction parameter data set according to the derivative correlation parameters of the dominant cause excitation sources, and obtaining a fan stability margin; based on the effective test reference indexes, the working condition coupling disturbance causal coefficients, the instability positioning coefficients and the fan stability margin, calculating a final score of the comprehensive stability of the fan, and outputting a risk level and a rectification list. The application improves the instability risk identification and abnormal root cause positioning capability of the wind turbine generator.

Description

Technical Field

[0001] This invention relates to the field of wind power equipment operation and maintenance and fault diagnosis, and in particular to a method and apparatus for analyzing wind turbine stability test data. Background Technology

[0002] Wind turbine generators are exposed to complex outdoor environments for a long time and are significantly affected by natural conditions such as wind shear, turbulence fluctuations, and extreme wind conditions. At the same time, the mechanical structure of the unit operates repeatedly for a long time, the load on the transmission components changes dynamically, and electrical factors such as grid voltage fluctuations, harmonic interference, and grid-connected operation control disturbances are superimposed. The whole machine is in a complex operating state with multiple sources of disturbance superimposed.

[0003] Wind turbine systems are complex nonlinear systems with deep coupling of aerodynamics, mechanics, and electrical systems. There are objective physical relationships between the subsystems that interact and influence each other. Small disturbances in a single part or dimension can be transmitted and amplified step by step through the system relationships, leading to various hidden dangers such as excessive vibration of the unit, abnormal load, decreased operational stability, and grid connection fluctuations. In the long run, this can easily cause fatigue damage to components and frequent failures, seriously threatening the operational safety and service reliability of wind turbine units.

[0004] Currently, the wind power industry still faces many common technical shortcomings in monitoring and assessing potential turbine anomalies and instability, as well as in tracing fault sources: Existing monitoring and analysis generally adopts a mode of independent detection in different domains and judgment based on a single indicator. This severs the coupling and correlation characteristics between multiple systems of the unit, ignores the problems induced by complex disturbances across domains and links, can only identify single-point obvious faults, and is difficult to discover the hidden instability risks caused by the coupling of multiple factors, resulting in insufficient coverage of hidden danger identification. The actual operating signals of the unit have typical characteristics of non-stationarity, strong time-varying, and multiple disturbance superposition. Disturbances from different sources and of different types are intertwined and mixed together. Traditional signal analysis methods are difficult to effectively decouple and separate multi-source disturbances, and cannot clearly distinguish the boundaries between external environmental disturbances, equipment abnormalities, and grid-side interference, which greatly hinders the root cause investigation. Conventional state analysis is mostly based on data correlation, which can only characterize the degree of correlation between parameters. It cannot determine the directionality and driving force of disturbance, and it is difficult to distinguish between effective correlation caused by real physical action and false correlation caused by random fluctuations and noise interference. The problems of misjudgment and omission are prominent, and the analysis results are not scientific enough. Disturbances exhibit an objective law of progressive transmission and evolution within the unit. Different transmission links have significantly different effects on suppressing or amplifying disturbances. Existing technologies generally lack a systematic analysis of disturbance propagation paths and transmission evolution mechanisms, making it impossible to pinpoint key weak links in disturbance diffusion and difficult to explain common industry problems where the risk level of similar anomalies varies greatly under different operating conditions. At present, the assessment of instability and anomalies is mostly qualitative and empirical, lacking a unified quantitative evaluation system. It is impossible to quantitatively characterize the strength of disturbances, risk levels and the degree of contribution of causes, and it is difficult to accurately distinguish between core dominant causes and secondary related factors. Ultimately, the granularity of the source tracing results is coarse, which is difficult to directly guide precise on-site maintenance, optimization of operating parameters and adjustment of stability control strategies.

[0005] Based on this, the present invention proposes a method and apparatus for analyzing wind turbine stability test data. Summary of the Invention

[0006] This invention provides a method for analyzing wind turbine stability test data, characterized by comprising: S1. Collect the time series of the three core channels, perform time delay alignment on the time series, calculate the effective test benchmark index, and obtain the effective time series dataset; S2. Calculate the causality coefficient of the coupling disturbance of the operating condition in the high-fidelity data segment of the effective time series dataset, and determine the high-risk operating condition segment dataset and the core risk frequency band. S3. For the high-risk operating condition data segment and its core risk frequency band, separate and determine the dominant inducing source of excitation, calculate the instability coefficient and determine the key disturbance amplification and propagation path. S4. Construct a modified parameter dataset based on the derived related operating condition parameters of the dominant excitation source, calculate the critical decay margin and disturbance rejection redundancy of the parameters in the dataset, and obtain the wind turbine stability margin. S5. Based on effective test benchmark indicators, causal coefficient of coupled disturbance under operating conditions, instability coefficient, and wind turbine stability margin, calculate the final score of wind turbine comprehensive stability and output a list of risk levels and rectification priorities.

[0007] The wind turbine stability test data analysis method described above involves collecting time series sequences from three core channels, aligning these time series sequences with time delays, calculating effective test benchmark indicators, and obtaining an effective time series dataset. This process is specifically divided into the following sub-steps: The time series of the three core channels in the wind turbine stability test is collected. The time series delay is aligned based on the time series causal delay estimation algorithm based on the transfer entropy. The three-domain segmented data set is generated by the sliding window. Traverse the three-domain segmented data set, calculate the causal stability and causal fidelity of each data segment in the set, and obtain the multi-domain causal fidelity of each data segment; Based on the causal stability of operating conditions and the causal fidelity of multiple domains, the effective test benchmark index is calculated and invalid data segments are removed, and the effective time series dataset is output.

[0008] The wind turbine stability test data analysis method described above includes calculating the causality coefficient of coupled disturbances under operating conditions for high-fidelity data segments in the effective time-series dataset, and determining the high-risk operating condition data segments and core risk frequency bands. This is specifically divided into the following sub-steps: Calculate the set of coupled sensitive frequency bands for each high-fidelity data segment in the effective time-series dataset; Based on the set of coupling-sensitive frequency bands, the three-domain cross-frequency causal coupling degree and disturbance energy dissipation rate of the high-fidelity data segment in the effective time series dataset are calculated. Based on the three-domain cross-frequency causal coupling degree and disturbance energy dissipation rate, the operating condition coupling disturbance causal coefficient is generated, and the high-risk operating condition segment dataset and core risk frequency band are obtained.

[0009] The wind turbine stability test data analysis method described above, wherein calculating the set of coupled sensitive frequency bands for each high-fidelity data segment in the effective time-series dataset, specifically comprises the following sub-steps: By using ensemble empirical mode decomposition, the time series data of the aerodynamic, mechanical, and electrical domains in each high-fidelity data segment of the effective time series dataset are adaptively decoupled across the entire frequency band to obtain the three-domain Hilbert marginal spectrum of a single data segment. Based on the wind turbine system design parameters and measured frequency domain characteristics, a continuous frequency band range containing the core characteristic frequencies is defined as the sensitive candidate frequency band pool. For the frequency bands to be determined in the sensitive candidate frequency band pool, the energy proportions of the aerodynamic, mechanical, and electrical domains in the frequency bands to be determined are calculated based on the three-domain Hilbert marginal spectrum. The effective three-domain cross-coupled sensitive frequency bands are determined according to the overlap of the three domains, and the set of coupled sensitive frequency bands of the current high-fidelity data segment is obtained. Repeat the above process to complete the calculation of the set of coupled sensitive frequency bands for all high-fidelity data segments in the effective time series dataset.

[0010] The wind turbine stability test data analysis method described above involves separating and identifying the dominant excitation sources for the extreme high-risk operating condition data segment and its core risk frequency band within the high-risk operating condition data set, and calculating the instability coefficient to determine the key disturbance amplification and propagation path. This process is specifically divided into the following sub-steps: For the extreme high-risk operating condition data segment and its core risk frequency band in the high-risk operating condition segment dataset, three types of excitation sources are separated and the excitation sources are paired with the instability propagation path to obtain the excitation-response path pairing set; Based on the incentive-response path pairing set, the causal contribution rate of each incentive source is calculated within the core risk frequency band, and the dominant incentive source and its path transmission loss rate are determined. The path transmission loss rate is made into an index of the same direction. The instability coefficient is calculated by combining the causal contribution rate of the corresponding dominant incentive source. The key disturbance amplification transmission path and the secondary cooperative instability path are divided according to the size of the coefficient.

[0011] The wind turbine stability test data analysis method described above involves constructing a modified parameter dataset based on the derived associated operating condition parameters of the dominant excitation source, calculating the critical attenuation margin and disturbance rejection redundancy of the parameters in the dataset, and obtaining the wind turbine stability margin. This method is specifically divided into the following sub-steps: Based on the dominant cause and excitation source, corresponding related operating condition parameters are derived. After data segmentation and instability coefficient correction, a corrected parameter dataset is obtained, and a nonlinear mapping function is generated for each related operating condition parameter. Based on the modified parameter dataset, the critical value of parameter instability is obtained by combining the nonlinear mapping function. The critical attenuation margin and disturbance rejection redundancy of the parameters are calculated based on the critical value of parameter instability. The stability margin of the wind turbine is generated by allocating weights through the instability coefficient.

[0012] The wind turbine stability test data analysis method described above, based on effective test benchmark indicators, operating condition coupled disturbance causality coefficient, instability displacement coefficient, and wind turbine stability margin, calculates the final comprehensive stability score of the wind turbine and outputs a risk level and rectification priority list. Specifically, it consists of the following sub-steps: Based on the importance classification of performance indicators according to IEC standards, weights are assigned to effective test benchmark indicators and wind turbine stability margin, and weighted fusion is performed to obtain the wind turbine stability performance compliance rate. The causality coefficient of the coupled disturbance under operating conditions and the instability coefficient are assigned weights by dynamic weighting, and the risk acceptability of the wind turbine is calculated by weighting. Based on the dual-dimensional evaluation rules of the IEC wind power stability test standard, the wind turbine stability performance compliance rate and wind turbine risk acceptability are weighted and the two are weighted to output the final score of the wind turbine's comprehensive stability. Based on the final comprehensive stability score of the wind turbine, a list of risk levels and rectification priorities is determined.

[0013] The present invention also provides a wind turbine stability test data analysis device, comprising: Acquisition module: used to acquire the time series of the three core channels, perform time delay alignment on the time series, calculate the effective test benchmark indicators, and obtain the effective time series dataset; High-risk operating condition segment module: used to calculate the causal coefficient of operating condition coupling disturbance in the high-fidelity data segment of the effective time series dataset, and to determine the high-risk operating condition segment dataset and the core risk frequency band; The dominant inducing factor module is used to separate and determine the dominant inducing factor excitation source for the extreme high-risk operating condition data segment and its core risk frequency band in the high-risk operating condition data segment, and to calculate the instability coefficient to determine the key disturbance amplification and propagation path. Stability margin assessment module: used to construct a modified parameter dataset based on the derived related operating condition parameters of the dominant excitation source, calculate the critical decay margin and disturbance rejection redundancy of the parameters in the dataset, and obtain the wind turbine stability margin; Risk Level and Rectification List Module: Based on effective test benchmark indicators, operating condition coupling disturbance causality coefficient, instability coefficient, and wind turbine stability margin, this module calculates the final comprehensive stability score of the wind turbine and outputs a risk level and rectification priority list.

[0014] The beneficial effects achieved by this invention are as follows: This invention abandons the limitations of traditional single-domain independent analysis, fully takes into account the coupling and linkage characteristics of multiple systems such as pneumatic, mechanical and electrical systems, avoids severing the objective physical relationship of the unit, and can effectively identify complex and hidden instability hazards induced by multiple factors, greatly improving the comprehensiveness and completeness of abnormal risk identification. It adapts to the non-stationary and multi-disturbance aliasing operation characteristics of wind turbine timing signals, effectively distinguishes and decouples disturbances from different sources and of different types, clearly delineates the boundaries of environmental disturbances, equipment malfunctions, and grid-side interference, and reduces analysis bias caused by signal aliasing. Breaking through the technical limitations of simple correlation analysis, it can accurately identify the direction of disturbance and causal driving relationship, effectively eliminate false correlation interference caused by random fluctuations and environmental noise, eliminate false coupling misjudgment, and make the results of disturbance law analysis more consistent with the real physical mechanism. Following the step-by-step transmission law of actual unit operation, the entire process of disturbance transmission, evolution, amplification or dissipation is fully restored, the transmission differences of each link and the weak points of the system are clarified, and the problem of different risk levels of the same anomaly under different operating conditions is reasonably explained. Establishing a standardized and quantitative evaluation mechanism can objectively quantify the contribution of various disturbances, accurately classify the core dominant causes and secondary related factors, achieve precise matching and positioning of physical anomaly types, solve the problems of traditional qualitative analysis being crude and having poor source tracing direction, and the results can directly support on-site maintenance and fault management. Attached Figure Description

[0015] 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.

[0016] Figure 1 This is a flowchart of a wind turbine stability test data analysis method provided in Embodiment 1 of this application.

[0017] Figure 2 This is a schematic diagram of a wind turbine stability test data analysis device provided in Embodiment 2 of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 like Figure 1 As shown, Embodiment 1 of this application provides a method for analyzing wind turbine stability test data, including: S1. Collect the time series of the three core channels, perform time delay alignment on the time series, calculate the effective test benchmark index, and obtain the effective time series dataset.

[0020] S11. Collect the time series of the three core channels in the wind turbine stability test, complete the time series delay alignment based on the time series causal delay estimation algorithm based on the transfer entropy, and generate the three-domain segmented data set through the sliding window.

[0021] The three-domain core channel time series refers to the original discrete time series data of the aerodynamic, mechanical, and electrical domains that are collected synchronously during the stability test of the wind turbine under all operating conditions. It includes eight types of original time series data. The aerodynamic domain includes the time series data of impeller inflow wind speed and impeller inflow wind direction angle. The mechanical domain includes the time series data of tower forward and backward vibration, tower left and right vibration, tower vertical vibration, main shaft real-time torque, and blade pitch angle. The electrical domain includes the time series data of grid voltage at the unit's grid connection point.

[0022] To address the common issues in wind turbine full-condition stability test data, such as mixed operating conditions, mismatched acquisition delays across multiple channels, and lack of clear physical causal relationships in data segments, the following process was used to complete the causal segmentation of the three core channel time series and the multi-channel causal delay alignment: First, the basic time scale of the sliding window is determined based on the inherent physical link of energy transmission in the wind turbine: strictly following the real physical causal transmission link between the upstream and downstream of the wind turbine: airflow wind energy capture → impeller aerodynamic load generation → main shaft mechanical transmission → generator electrical energy conversion → grid-connected power output. Combined with the wind turbine's factory-designed rated speed parameters, the duration of a single impeller rotation cycle is calculated. The three complete impeller rotation cycles are set as the minimum time width of the sliding window for data analysis, ensuring that each sliding window can completely cover the entire three-domain transmission process from wind energy input to grid power output. At the same time, the sliding step size is determined to be a single impeller rotation cycle, thereby achieving appropriate overlap between adjacent data segments and continuous, complete traversal of the full time-series data.

[0023] Next, using the time-series data of impeller inflow wind speed and impeller inflow wind direction angle in the aerodynamic domain as the baseline causal reference source, a time-series causal delay estimation algorithm based on transfer entropy is used to sequentially traverse the remaining tower forward-backward vibration, tower left-right vibration, tower vertical vibration, main shaft real-time torque, blade pitch angle, and grid voltage time-series data at the unit's grid connection point. The time delay offset of these six types of time-series data relative to the impeller inflow wind speed and impeller inflow wind direction angle time-series data is calculated one by one. The non-physical inherent time difference of these six types of time-series data caused by the sampling delay of the acquisition sensor, the signal transmission link delay, and the hardware acquisition buffer deviation is quantified. Based on these six types of time delay offsets, the corresponding time-series data sequences are corrected by forward / backward millisecond-level time-series shift, completing the global unified time-series alignment of the three core channel time-series sequences.

[0024] Finally, based on the sliding window with a determined window width and sliding step size, the three core channel time series after being traversed and equally segmented and aligned is processed. At the same time, a unique time series label is attached to each segmented data segment. The label records two items: first, the start and end time intervals corresponding to the data segment; second, the basic mark of the validity of each group of time series data, and the status is marked according to whether there are original data quality problems such as missing, distortion, and jump in each group of data within the data segment.

[0025] All data segments form a three-domain segmented data set that is continuously divided, time-aligned, and carries time-series labels.

[0026] S12. Traverse the three-domain segmented data set and calculate the causal stability of each data segment and the multi-domain causal fidelity.

[0027] First, traverse all sliding window data segments in the three-domain segmented data set, extract the time series data of impeller inflow wind speed and impeller inflow wind direction angle in a single data segment, calculate the coefficient of variation of the two sets of time series data respectively, and perform equal weight fusion to obtain the fused coefficient of variation. After normalizing the range of the fused coefficient of variation in all data segments, the aerodynamic input steady-state coefficient with a value range of [0,1] is obtained. Based on the transfer entropy algorithm, the aerodynamic-mechanical causal transfer intensity of the real-time torque of the main shaft, the forward / backward / left / right / vertical vibration of the tower, and the blade pitch angle time series data relative to the impeller inflow wind speed and impeller inflow wind direction angle time series data in a single data segment is calculated. The aerodynamic-mechanical causal transfer intensity is normalized by the maximum value of the aerodynamic-mechanical causal transfer intensity in all data segments to obtain the aerodynamic-mechanical causal matching coefficient with a value range of [0,1]. Based on the transfer entropy algorithm, the mechanical-electrical causal transfer strength of the grid voltage time series data of the unit connection point relative to the real-time torque time series data of the main shaft within a single data segment is calculated. The mechanical-electrical causal transfer strength is normalized by the maximum value of the mechanical-electrical causal transfer strength in all data segments to obtain the mechanical-electrical causal matching coefficient with a value range of [0,1]. The pneumatic input steady-state coefficient, pneumatic-mechanical causal matching coefficient, and mechanical-electrical causal matching coefficient are multiplied by weight to obtain the working condition causal steady-state coefficient corresponding to a single data segment, with a value range of [0,1].

[0028] Secondly, for the same sliding window data segment, the transfer entropy algorithm is used to calculate the transfer entropy values ​​of the time series data between the aerodynamic domain and the mechanical domain, and between the mechanical domain and the electrical domain in the data segment in turn. The maximum value of all transfer entropy values ​​in a single data segment is used as the normalization basis to normalize each transfer entropy value, so as to obtain the causal correlation degree of a single set of time series data in the data segment. The closer the value is to 1, the stronger the causal transmission strength and the higher the correlation between the upstream and downstream of the time series. The number of time series data groups with causal correlation below a preset threshold within the data segment is counted and the proportion of abnormal time series data groups is calculated. Based on the proportion of abnormal time series data groups and the penalty intensity coefficient, a fidelity correction term is constructed through a one-sided exponential decay function. In this embodiment, the penalty intensity coefficient is set to 3. The larger the penalty intensity coefficient is, the stronger the penalty for abnormal data. When the proportion of abnormal data increases, the fidelity correction term decays faster. The average causal correlation of all normal time-series data within the data segment is multiplied by the fidelity correction term to obtain the multi-domain causal fidelity of the data segment, with a value range of [0,1].

[0029] Finally, the causal stability of the operating conditions and the causal fidelity of the multi-domain causal data segments that have been traversed are summarized in turn, and then stored in a unified and organized manner.

[0030] S13. Based on the causal stability of the operating conditions and the causal fidelity of the multi-domain, calculate the effective test benchmark index and remove invalid data segments, and output the effective time series dataset.

[0031] Based on the requirements for test data validity in the IEC standards for wind turbine mechanical load testing and wind power grid connection characteristic testing, this embodiment sets weights of 0.6 and 0.4 for the operating condition causal stability and multi-domain causal fidelity of each data segment, respectively, to complete the weighted fusion of the two indicators and obtain the effective test benchmark index for a single data segment. The closer the index value is to 1, the higher the validity and causal matching degree of the data segment. Secondly, based on industry testing specifications, a validity threshold of 0.85 is set to retain high-fidelity data segments with effective test benchmark indices higher than the validity threshold.

[0032] The retained high-fidelity data segments are spliced ​​together in time series to form an effective time series dataset.

[0033] S2. Calculate the causal coefficient of the coupling disturbance of the operating conditions in the high-fidelity data segment of the effective time series dataset, and determine the high-risk operating condition segment dataset and the core risk frequency band.

[0034] S21. Calculate the set of coupled sensitive frequency bands for each high-fidelity data segment in the effective time-series dataset.

[0035] First, Hilbert Huang Transform (HHT) is used to adaptively decouple the high-fidelity data segment in the effective time series dataset across the entire frequency band. First, the time series data in the aerodynamic, mechanical, and electrical domains of the high-fidelity data segment are decomposed into multiple Intrinsic Mode Function (IMF) components and low-frequency trend terms by ensemble empirical mode decomposition (EEMD). After removing high-frequency noise IMF components and low-frequency trend terms, Hilbert Transform is then applied to each retained IMF component to obtain the instantaneous frequency and instantaneous amplitude of each component. The Hilbert spectrum is then calculated, and the Hilbert marginal spectrum of each domain is fitted to form the three-domain Hilbert marginal spectrum of a single data segment. Next, based on the wind turbine system design parameters and measured frequency domain characteristics, a continuous frequency band interval containing the following core characteristic frequencies is defined as the sensitive candidate frequency band pool, specifically including: impeller rotation characteristic frequency, first / second-order bending natural frequency of the tower, natural frequency of the main shaft / gearbox transmission system, grid-connected fundamental frequency, and integer harmonics. For the frequency band to be judged within the sensitive candidate frequency band pool, the energy proportions of the aerodynamic, mechanical, and electrical domains within the frequency band to be judged are calculated based on the three-domain Hilbert marginal spectrum. The energy overlap of the three domains is defined as the minimum value of the energy proportions of the aerodynamic, mechanical, and electrical frequency bands. If and only if the overlap of the three domains is ≥30%, the frequency band is determined to be an effective three-domain cross-coupling sensitive frequency band and is included in the current set of coupled sensitive frequency bands of the high-fidelity data segment.

[0036] Repeat the above process to complete the calculation of the set of coupled sensitive frequency bands for all high-fidelity data segments in the effective time series dataset.

[0037] S22. Based on the set of coupled sensitive frequency bands, calculate the three-domain cross-frequency causal coupling degree and disturbance energy dissipation rate of the high-fidelity data segment in the effective time series dataset.

[0038] First, for a single effective three-domain cross-coupled sensitive frequency band in the set of coupled sensitive frequency bands of the high-fidelity data segment, the effective intrinsic mode function components corresponding to the coupled sensitive frequency band are extracted from the effective intrinsic mode function component sets of the aerodynamic, mechanical, and electrical domains, respectively. This yields the three-domain frequency domain component time series, i.e., a single cross-frequency calculation sequence pair. Zero-mean and unit-variance standardization is then applied to all frequency domain component sequences within all single cross-frequency calculation sequence pairs in the current high-fidelity data segment to eliminate computational interference caused by differences in physical dimensions and amplitude magnitudes. Finally, all standardized frequency domain component sequences in the set of coupled sensitive frequency bands are summarized to establish a one-to-one mapping relationship between the unique identifier of the current high-fidelity data segment, the set of coupled sensitive frequency bands, and the three-domain frequency domain component time series, forming the standard frequency domain component sequence set of the current high-fidelity data segment. Next, for a single high-fidelity data segment and a single set of coupled sensitive frequency bands, the corresponding single-set cross-frequency calculation sequence pair is retrieved from the set of standardized frequency domain component sequences. Using the aerodynamic domain standardized frequency domain component time series as the source signal and the mechanical domain standardized frequency domain component time series as the target signal, the directional frequency domain transfer entropy value of the aerodynamic domain to the mechanical domain path is calculated. Similarly, using the mechanical domain standardized frequency domain component time series as the source signal and the electrical domain standardized frequency domain component time series as the target signal, the directional frequency domain transfer entropy value of the mechanical domain to the electrical domain path is calculated. Using the aerodynamic domain standardized frequency domain component time series as the source signal and the electrical domain standardized frequency domain component time series as the target signal, the directional frequency domain transfer entropy value of the aerodynamic domain to the electrical domain path is calculated. Through the above process, the three-path directional frequency domain transfer entropy of this single set of coupled sensitive frequency bands is obtained. The three-path directional frequency domain transfer entropy of all single-set frequency bands within the set of coupled sensitive frequency bands of the current high-fidelity data segment is calculated iteratively.

[0039] In the current high-fidelity data segment and single-group coupling sensitive frequency band, the phase of each source signal sequence is randomly shuffled to generate 100 causally uncorrelated replacement sequences. The directional frequency domain transfer entropy value from the replacement sequence to the target signal is calculated to form a causal null hypothesis distribution. The 95th percentile of the null hypothesis distribution is taken as the causal significance threshold. Only the directional frequency domain transfer entropy greater than the causal significance threshold is retained as the effective causal coupling strength. The directional frequency domain transfer entropy below the threshold is directly assigned to 0, thus completely eliminating pseudo-coupling interference caused by random fluctuations. The effective directional frequency domain transfer entropy values ​​of each path within the current high-fidelity data segment are summed to obtain the full-link effective coupling strength of the three paths in the high-fidelity data segment. The effective coupling strengths of the three paths in the current data segment are then summed with equal weights to obtain the total coupling strength of the data segment. The total coupling strength of all high-fidelity data segments is then iterated over, and the global maximum value is taken to normalize the total coupling strength of the current high-fidelity data segment, resulting in a three-domain cross-frequency causal coupling degree with a value in the range [0,1]. The closer this coupling degree is to 1, the stronger the cross-frequency causal coupling correlation between the three domains of the current high-fidelity data segment, and the higher the risk of instability and linkage.

[0040] Secondly, for each set of independent coupled sensitive frequency bands corresponding to the current high-fidelity data segment, for the frequency range of the coupled sensitive frequency band, integral operations are performed on the three-domain Hilbert marginal spectrum of the current high-fidelity data segment to accumulate the total energy of each domain within the frequency range, thus obtaining the single-domain frequency band energy; subsequently, all single-domain frequency band energies in the current data segment are summarized to obtain the total full-link energy of the aerodynamic domain, mechanical domain, and electrical domain under the data segment; Next, in this data segment, the ratio of the total energy of the mechanical domain to the total energy of the aerodynamic domain represents the energy conversion efficiency from the aerodynamic domain input to the mechanical domain; the ratio of the total energy of the electrical domain to the total energy of the mechanical domain represents the energy conversion efficiency from the mechanical domain to the electrical domain output. By multiplying the energy conversion efficiencies of the two links, the total energy transfer coefficient, i.e., the disturbance energy dissipation rate, representing the total energy transfer from the initial aerodynamic domain input to the final electrical domain output is obtained. This coefficient can intuitively reflect whether the overall trend of disturbance energy change during the entire transfer process is to accumulate and amplify or to gradually dissipate.

[0041] When the disturbance energy dissipation rate is greater than 0 and less than 1, it means that the disturbance energy is amplified after being transmitted through the entire three-domain link. The smaller the dissipation rate, the stronger the energy amplification effect and the higher the risk of system instability. When the dissipation rate is equal to 1, it means that the disturbance energy is neither dissipated nor amplified and is rigidly transmitted through the entire link. When the dissipation rate is greater than 1, it means that the disturbance energy is dissipated after being transmitted through the entire link. The larger the dissipation rate, the stronger the dissipation effect and the better the system stability.

[0042] Finally, the calculation of the three-domain cross-frequency causal coupling degree and disturbance energy dissipation rate of all high-fidelity data segments was completed.

[0043] S23. Based on the three-domain cross-frequency causal coupling degree and disturbance energy dissipation rate, generate the operating condition coupling disturbance causal coefficient to obtain the high-risk operating condition segment dataset and core risk frequency band.

[0044] The disturbance energy dissipation rate is reverse-directed and converted into a disturbance energy amplification coefficient in the [0,1] interval to ensure that both the three-domain cross-frequency causal coupling degree and the disturbance energy dissipation rate have the same causal characteristic that the larger the value, the higher the risk. The three-domain cross-frequency causal coupling degree and the disturbance energy amplification coefficient are fused by equal weight normalization to obtain the operating condition coupling disturbance causal coefficient in the [0,1] interval. The closer the coefficient value is to 1, the stronger the three-domain coupling disturbance, the greater the energy amplification potential, and the higher the instability risk under the operating condition of the high-fidelity data segment. The closer the coefficient value is to 0, the weaker the three-domain coupling effect, the less significant the disturbance energy amplification, and the more stable the system operation. High-risk operating condition segment identification and core risk frequency band locking are completed based on the causal coefficient of operating condition coupling disturbance: Combining the overall distribution pattern of the causal coefficient of operating condition coupling disturbance of all high-fidelity data segments, and taking into account the sensitivity and false judgment rate of risk assessment, 0.6 is set as the risk judgment threshold in this embodiment; The causal coefficient of operating condition coupling disturbance of each high-fidelity data segment is judged one by one, and the data segment with the coefficient value higher than 0.6 is judged as a high-risk operating condition data segment and included in the high-risk operating condition segment dataset for unified management.

[0045] Simultaneously, for each high-risk operating condition data segment, the effective value of the three-path directional frequency domain transfer entropy, the energy of the three-domain sub-bands, and the total energy of the three-domain full link of the data segment are retrieved from the set of independent coupled sensitive frequency bands for that data segment. The directional frequency domain transfer entropy values ​​of the three paths of the coupled sensitive frequency band are summed to obtain the comprehensive value of the cross-domain coupling strength of the independent coupled sensitive frequency band. The proportion of the energy of the three-domain sub-bands of the coupled sensitive frequency band to the total energy of the corresponding domain full link is calculated, and the arithmetic mean of the three proportions is taken to obtain the comprehensive value of the three-domain energy concentration of the frequency band. The comprehensive value of the cross-domain coupling strength and the comprehensive value of the three-domain energy concentration are normalized and fused with equal weights to obtain the risk priority coefficient of each independent coupled sensitive frequency band.

[0046] For each high-risk operating condition data segment, the risk priority coefficients of all single-group independent coupled sensitive frequency bands are sorted from high to low. The independent frequency band with the highest coefficient is selected as the core risk frequency band exclusive to that high-risk operating condition data segment. This completes the dual accurate positioning of the high-risk operating condition data segment and the core risk frequency band, providing a clear target for subsequent targeted instability prevention and anomaly tracing.

[0047] S3. For the high-risk operating condition data segment and its core risk frequency band, separate and determine the dominant inducing source of excitation, and calculate the instability coefficient to determine the key disturbance amplification and transmission path.

[0048] S31. For the extreme high-risk operating condition data segment and its core risk frequency band in the high-risk operating condition segment data set, separate three types of excitation sources and pair the excitation sources with the instability propagation path to obtain the excitation-response path pairing set.

[0049] Based on the high-risk operating condition dataset, the high-risk operating condition data segments in the dataset are prioritized according to the causality coefficient of the operating condition coupling disturbance from high to low. The extreme high-risk operating condition data segment with the highest coefficient is selected as the core verification object. The core risk frequency band corresponding to this data segment is used as the target analysis frequency band to complete the analysis of the excitation source and transmission path. The specific analysis process is as follows: First, for the high-risk operating condition data segment to be analyzed, its corresponding core risk frequency band is used as the sole analysis boundary. Combining the inherent physical characteristic frequency bands of the three types of excitation sources in the wind turbine system, frequency domain causal filtering is used to perform directional decoupling of the effective inherent mode function components of this data segment: For the aerodynamic characteristic frequency band in the core risk frequency band, feature components strongly correlated with impeller inflow wind speed fluctuations and inflow wind direction angle turbulence disturbances are extracted, separating independent components of the aerodynamic excitation source; for the inherent characteristic frequency band in the mechanical domain, feature components strongly correlated with tower triaxial vibration anomalies, main shaft drive torque fluctuations, and pitch actuator adjustment disturbances are extracted, separating independent components of the mechanical excitation source; for the characteristic frequency band in the electrical domain, feature components strongly correlated with grid voltage fluctuations and harmonic distortion at the unit's grid connection point are extracted, separating independent components of the electrical excitation source. During the separation process, based on the effective value of the directional frequency domain transfer entropy of the corresponding frequency band of the data segment, cross-coupling interference components between different excitation sources are eliminated, ensuring that the three types of excitation source components are independent of each other, without causal confusion, and fully preserving the core characteristics of each excitation source.

[0050] Secondly, based on the actual physical causal transmission chain of the wind turbine, the instability transmission path is decomposed into three levels: The primary transmission path corresponds to the aerodynamic domain → mechanical domain transmission system link. The input end is the independent component of the aerodynamic excitation source separated from the aerodynamic domain, and the output end is the components of the tower's front-to-back / left-to-right / vertical vibration and the real-time torque response of the main shaft in the mechanical domain. The secondary transmission path corresponds to the closed-loop link of pitch adjustment within the mechanical domain. The input end is the tower vibration and main shaft real-time torque response components output by the primary path, and the output end is the blade pitch angle adjustment response component of the mechanical domain. The three-level transmission path corresponds to the mechanical domain → electrical domain grid-side link. The input end is the real-time torque of the main shaft and the blade pitch angle response components in the mechanical domain, and the output end is the grid voltage response components at the unit connection point in the electrical domain.

[0051] Finally, the three independent excitation source components obtained from the separation are paired with the hierarchically decomposed path pairs for directional causal matching. The effective value of the directional frequency domain transfer entropy of the core risk frequency band corresponding to the current data segment is used as the criterion for determining causal correlation. Only when the effective value of the directional transfer entropy from the excitation input end to the response output end is greater than the significance threshold is the pairing confirmed as a valid causal link. Subsequently, a unique causal label is added to each effective excitation source and path response pairing. The label contains four core pieces of information: excitation source type, transmission path hierarchy, corresponding core risk frequency band, and effective value of directional transfer entropy. Finally, these are summarized to form a set of excitation-response path pairs with causal labels specific to this extreme high-risk working condition data segment.

[0052] S32. Based on the incentive-response path pairing set, calculate the causal contribution rate of each incentive source in the core risk frequency band, and determine the dominant incentive source and its path transmission loss rate.

[0053] For each set of excitation-response path pairs in the data segment of extreme high-risk working conditions, with its core risk frequency band as the only calculation boundary, in order to solve the shortcomings of simply using energy proportion to distinguish between causal relationships and false couplings, and simply using transfer entropy to quantify energy-driven weights, the causal contribution rate of excitation sources is calculated by the following formula, which accurately quantifies the true driving contribution of the three types of excitation sources to the target instability response. The specific formula is as follows: , in, Let be the causal contribution rate of the m-th excitation source. The closer the contribution rate value is to 1, the stronger the driving effect of the excitation source on the instability response, and the more likely it is to be the core instability cause. To integrate and normalize the baseline total, For the m-th type of excitation source, the effective instability-inducing driving energy within the core risk frequency band, after causal coupling and conduction delay correction, plus the total effective contribution after adding the excitation source frequency band energy proportion correction weight term, is defined as follows: m is the excitation source number, which is fixed at 3 types in this embodiment: m=1 for aerodynamic excitation, m=2 for mechanical excitation, and m=3 for electrical excitation; r is the target instability response, corresponding to the path output response component of the excitation-response path pair paired with the current excitation source in the hierarchical decomposition of step S31. These are the lower and upper limits of the core risk frequency band for the current data segment, respectively. The effective value of the directional frequency domain transfer entropy at frequency f is given for the m-th excitation source to the target instability response r. Let f be the Hilbert marginal spectral energy density of the m-th excitation source at frequency f, corresponding to the frequency domain energy distribution of the independent components of the excitation source obtained in step S31. The frequency domain propagation delay parameter from the m-th excitation source to the target instability response r at frequency f is calculated from the phase information of the effective value of the directional frequency domain transfer entropy in sub-step S22, and characterizes the actual propagation time of the disturbance in the physical link. As a causal time delay correction factor, it corrects the energy phase deviation caused by the perturbation conduction time delay, ensuring that the energy calculation fully conforms to the actual physical conduction process, and the energy terms of non-causal time series will be significantly attenuated. To adjust the weighting term for the energy proportion of the excitation source frequency band, Let be the energy proportion of the m-th excitation source within the core risk frequency band, representing the degree of energy focusing of the excitation source within the core risk analysis frequency band, with a value of [0,1]. This is a nonlinear focusing correction coefficient with a value range of [0.5, 2]. It is used to correct the spurious contribution deviation caused by simple energy integration and ensure that the causal contribution rate calculation closely matches the actual perturbation driving capability of the excitation source. Let f be the Hilbert marginal spectrum energy density of the target instability response r at frequency f, which characterizes the energy intensity of the actual instability response of the system. for The maximum effective value of the directional frequency domain transfer entropy from all excitation sources to the unstable response of the target within the frequency band is used as a global normalization constraint. To normalize the weights, ensure It strictly falls within the [0,1] interval.

[0054] According to the above formula, the causal contribution rates of the three excitation sources (aerodynamic, mechanical, and electrical) for the extreme high-risk operating condition data segment are calculated sequentially. Using the causal contribution rate as the core criterion, the excitation source with the highest contribution rate is identified as the core instability-dominant cause of this high-risk operating condition data segment. Simultaneously, the frequency integral energy, frequency energy percentage, and peak energy density data of the independent component of this excitation source within the core risk frequency band are matched with the inherent characteristic frequency band library of typical physical anomalies in the corresponding wind turbine domain to initially identify candidate physical anomaly types. Then, combining the excitation-response path and the effective value of the directional frequency domain transfer entropy corresponding to this excitation source, the causal significance of the candidate anomaly types is verified, eliminating misjudgments due to similar frequency domain characteristics, and finally determining the specific physical anomaly type. The remaining two types of excitation sources are identified as secondary causes and secondary related factors, respectively, in descending order of contribution rate.

[0055] Secondly, for the current data segment under extremely high-risk operating conditions, targeting the excitation source that has been identified as the core instability-dominant cause, in order to quantify the disturbance propagation characteristics of each level of the corresponding excitation-response path pairing set, the path propagation loss rate is calculated, and the amplification / dissipation characteristics of the disturbance by the propagation path from the excitation source to the target instability response r are quantified. The specific formula is as follows: , in, Let be the conduction loss rate of the p-th hierarchical conduction path corresponding to the causal source in the set, with a value range of (0, +∞). A value less than 1 indicates that the path amplifies the disturbance; the smaller the value, the stronger the amplification effect, and it is a critical path for amplifying the disturbance. A value greater than 1 indicates that the path exhibits a dissipative attenuation effect on the disturbance; the larger the value, the stronger the attenuation effect. p represents the hierarchical transmission path number corresponding to the inducing excitation source in the set. These are the lower and upper limits of the core risk frequency band for the current data segment, respectively. This is the ratio of the average energy of the path excitation input after time delay correction to the average energy of the path response output. This represents the total number of frequency sampling points within the core risk frequency band. Let f be the Hilbert marginal spectral energy amplitude at frequency f at the input of the p-th path. Let f be the Hilbert spectrum energy amplitude at frequency f at the output of the p-th path. Let f be the frequency domain propagation delay parameter of the p-th path at frequency f, which characterizes the phase delay and physical propagation time of the unstable disturbance from the excitation input to the response output. It is calculated from the phase information of the directional frequency domain propagation entropy of this path. is the time delay loss correction factor for the p-th path, which corrects the energy attenuation caused by time delay during disturbance propagation. The higher the frequency and the greater the time delay, the stronger the loss correction effect. This is a correction term for path wave transmission characteristics. Within a time-domain window that strictly matches the core risk frequency band of the current high-risk operating condition data segment, the standard deviation of the time-domain component of the excitation input of the p-th transmission path is calculated. This standard deviation is used to quantify the severity of the inherent disturbance fluctuations of the excitation source of this path during the period of highest instability risk. Within a time-domain window that strictly matches the core risk frequency band of the current high-risk operating condition data segment, the standard deviation of the time-domain component of the response output of the p-th propagation path is calculated. This standard deviation is used to quantify the severity of fluctuations in the unstable response of this path during the period of highest risk. The absolute difference between the Pearson correlation coefficients of the path input and output ends and the original discrete time-series data sequences in the three domains represents the magnitude of the change in the degree of correlation of the original signal characteristics before and after the disturbance propagation. Let Pearson correlation coefficient be the time-domain component of the excitation input of p propagation paths and the original discrete-time data sequence in three domains. This coefficient characterizes the degree of causal correlation between the path input excitation components and the original measured reference signal. Let Pearson correlation coefficient be the time-domain component of the output response of p propagation paths and the original discrete-time data sequence in three domains. This coefficient characterizes the degree of causal correlation between the path output response components and the original measured reference signal. The larger of the two Pearson correlation coefficients at the input and output ends represents the highest original baseline correlation strength throughout the entire p-th transmission path. This is a normalization correction term for conduction distortion, used to characterize the degree of causal characteristic distortion before and after conduction along the entire path. The minimum value correction term is fixed at 1. This is used to avoid calculation errors where the denominator is 0.

[0056] S33. The path transmission loss rate is made into an index of the same direction. The instability coefficient is calculated by combining the causal contribution rate of the corresponding dominant incentive source. The key disturbance amplification transmission path and the secondary cooperative instability path are divided according to the size of the coefficient.

[0057] For the response path pairs corresponding to the dominant causal source in the stimulus-response path pairing set, their path transmission loss rate is reversed and converted into a path perturbation amplification coefficient in the range of [0,1], ensuring that the causal contribution rate of the stimulus source is in the same direction as the larger the value, the higher the risk.

[0058] The causal contribution rate of the dominant inducing factor is weighted and fused with the disturbance amplification coefficient of the corresponding path to obtain the instability position coefficient. The closer the coefficient value is to 1, the stronger the combined instability-driving capability of the inducing factor-path. The instability position coefficients of all response path pairs corresponding to the dominant inducing factors are sorted by size, and the path with the highest coefficient is selected as the key disturbance amplification and propagation path. The response component at the end of the path is the weak response component of the system. At the same time, secondary cooperative instability paths are identified based on the secondary coefficients to complete the causal localization of instability across the entire link.

[0059] S4. Construct a modified parameter dataset based on the derived related operating condition parameters of the dominant inducing source, calculate the critical decay margin and disturbance rejection redundancy of the parameters in the dataset, and obtain the wind turbine stability margin.

[0060] S41. Based on the dominant induced excitation source, the corresponding associated operating condition parameters are derived. After data segmentation and instability coefficient correction, a corrected parameter dataset is obtained, and a nonlinear mapping function is generated for each associated operating condition parameter.

[0061] First, taking the core instability-inducing source as the target, based on the physical domain to which the source belongs, the corresponding underlying raw direct sampling parameters are selected from 8 types of raw time series data. Combined with the inherent structure of the wind turbine, mechanical transmission and electrical operation fixed parameters, all potential related operating condition analysis parameters in this domain are derived and solved. Redundant parameters from other irrelevant domains are eliminated to construct the target operating condition parameter set specific to this analysis.

[0062] Secondly, for each parameter in the target operating condition parameter set, outliers are removed and the upper and lower boundaries of the valid parameter data are statistically analyzed to determine the full range of parameter values. For each operating condition parameter in the parameter set, the full range of values ​​is traversed and segmented using equally spaced intervals. Within each parameter interval, the instability response characteristic values ​​of the instability response monitoring nodes corresponding to the key disturbance amplification and transmission paths are extracted. Instability response monitoring nodes refer to tower vibration and main shaft torque monitoring nodes for mechanical domain paths, and grid connection point voltage monitoring nodes for electrical domain paths. The instability response characteristic value extraction indicators are clearly defined as follows: peak value and effective value for vibration parameters; fluctuation amplitude and mean for torque / voltage parameters; and rate of change and steady-state deviation for pitch / wind speed parameters. This ensures that the extracted response characteristic values ​​can accurately characterize the instability response state of the system within the specified interval.

[0063] Next, the unstable response characteristic value is corrected by multiplying the unstable position coefficient to eliminate the interference of low causality data. After the correction is completed, the parameter range, interval mean and corrected unstable response characteristic value of each parameter interval are integrated to construct the corrected parameter dataset. Finally, using the interval mean of each target operating condition parameter in the corrected parameter dataset as the independent variable and the weighted corrected instability response characteristic value within the corresponding interval as the dependent variable, a fitting sample pair is constructed. For each sample pair, a nonlinear least squares fitting algorithm is used, and an appropriate fitting function type is selected based on the wind turbine instability mechanism. The optimal parameters of the fitting function are solved through iterative calculation to ensure that the fitting function can accurately represent the nonlinear correlation between the operating condition parameters and the system response. During the fitting process, the goodness of fit is calculated to verify the fitting effect. If the goodness of fit is less than 0.9, the fitting function type is adjusted or the data is re-screened until the fitting accuracy requirements are met. After completing the coefficient solution and accuracy verification, all undetermined function coefficients are determined, and a continuous nonlinear mapping function is generated for each parameter sample pair. At the same time, a parameter-response change trend curve is plotted with the interval mean of the operating condition parameters on the horizontal axis and the weighted corrected instability response characteristic value on the vertical axis, clearly showing the influence of changes in each operating condition parameter on the system instability response.

[0064] S42. Based on the modified parameter dataset, the critical value of parameter instability is obtained by combining the nonlinear mapping function. The critical attenuation margin and disturbance rejection redundancy of the parameters are calculated based on the critical value of parameter instability. The stability margin of the wind turbine is generated by allocating weights through the instability coefficient.

[0065] First, based on the IEC wind turbine design standards and the wind turbine factory safety limits, the limit threshold of the instability response characteristic value of each operating condition parameter in the correction parameter dataset is clearly defined. Based on the nonlinear mapping function, the instability critical value of the operating condition parameter when the instability response characteristic value reaches the limit threshold is obtained by reverse calculation. Secondly, based on the rated operating parameters of the wind turbine, the measured operating parameters, and the critical instability parameters, the absolute numerical distance between the measured operating parameters and the critical instability parameters under rated operating conditions is calculated. The distance is corrected by combining the nonlinear mutation characteristics when the parameters approach the instability threshold. Then, the global range of the corresponding parameters is used for normalization to complete the calculation of the critical attenuation margin of the current parameter. The closer the critical attenuation margin value is to 1, the greater the corresponding operating safety margin and the higher the stability margin. Next, the absolute difference between the numerical boundary of each parameter in the statistical correction parameter dataset under actual working conditions throughout the entire cycle and the critical parameter value for parameter instability is obtained to obtain the maximum disturbance fluctuation margin that each parameter can withstand. After the range normalization of the parameter, the independent disturbance resistance redundancy of each parameter is obtained one by one. The closer the value is to 1, the more sufficient the stability redundancy space of the parameter to resist external disturbances, and the better the stability under working condition fluctuations.

[0066] Finally, based on the instability coefficient, dynamic weights are set for the critical decay margin and disturbance immunity redundancy. The larger the instability coefficient, the more importance is placed on its own safety boundary margin. Therefore, the weight of the critical decay margin is increased, rather than being a fixed weight. Moreover, the weights of the critical decay margin and disturbance immunity redundancy are always 1. The critical decay margin and disturbance immunity redundancy are weighted and fused to obtain the wind turbine stability margin. Based on the wind turbine stability margin value, the stability margin is classified as follows: 0.8-1 is excellent, 0.6-0.8 is good, 0.4-0.6 is average, and below 0.4 is poor. This completes the quantitative rating of the wind turbine stability performance.

[0067] S5. Based on effective test benchmark indicators, causal coefficient of coupled disturbance under operating conditions, instability coefficient, and wind turbine stability margin, calculate the final score of wind turbine comprehensive stability and output a list of risk levels and rectification priorities.

[0068] A comprehensive analysis is conducted on the effective test benchmark indicators, operating condition coupling disturbance causality coefficient, instability coefficient, and wind turbine stability margin obtained sequentially to determine the overall stability of the wind turbine. First, the above four core indicators are processed to unify the data direction. After unification, all four indicators meet the unified rule that the larger the value, the better the system stability.

[0069] Based on the IEC wind power testing standard benchmark threshold, the four indicators are normalized twice to ensure that the values ​​of the four indicators fall within the [0,1] interval. Then, based on the importance classification of performance indicators according to the IEC standard, weights are assigned to the effective test benchmark indicators and the wind turbine stability margin, and weighted fusion is performed to obtain the wind turbine stability performance compliance rate. Next, weights are assigned to the operating condition coupling disturbance causality coefficient and the instability coefficient through dynamic weighting, and the wind turbine risk acceptability is calculated by weighting. Specifically, the basic weight is determined by the rule that the larger the instability coefficient, the higher the weight of the instability coefficient, and the corresponding decrease in the weight of the operating condition coupling disturbance causality coefficient. The weight of the instability coefficient is determined by the product of the basic weight and the fixed floating coefficient and the instability coefficient. The fixed floating coefficient is the difference between the maximum and minimum values ​​of the basic weight of the instability coefficient, and the weight of the operating condition coupling disturbance causality coefficient is 1 minus the weight of the instability coefficient.

[0070] Based on the two-dimensional evaluation rules of the IEC wind power stability test standard, the wind turbine stability performance compliance rate is set with a weight of 0.6 and the wind turbine risk acceptability is set with a weight of 0.4. The two are weighted and calculated to output the final score of the wind turbine's comprehensive stability.

[0071] Based on the final comprehensive stability score of the wind turbine and the requirements of the IEC standard, the stability level is determined as follows: a score greater than or equal to 0.8 and an instability coefficient less than 0.3 indicates low risk, excellent wind turbine stability, and compliance with the standard requirements; a score greater than or equal to 0.6 and less than 0.8 and an instability coefficient greater than or equal to 0.3 and less than 0.7 indicates medium risk, and the wind turbine stability is qualified; a score less than 0.6 or an instability coefficient greater than or equal to 0.7 indicates high risk, and the wind turbine stability does not meet the standard and rectification is required.

[0072] During rectification, based on the instability coefficients output in step S33 and the corresponding key disturbance amplification and propagation paths, the coefficients are sorted in descending order to form a clear rectification priority list.

[0073] Example 2 like Figure 2 As shown, Embodiment 2 of this application provides a wind turbine stability test data analysis device, comprising: Acquisition Module: Used to acquire time series sequences from the three core channels, perform time delay alignment on these sequences, calculate effective benchmark metrics, and obtain a valid time series dataset. Specifically, it consists of the following sub-modules: The three-domain segmentation submodule collects the time series of the three core channels in the wind turbine stability test, completes the time series delay alignment based on the time series causal delay estimation algorithm based on the transfer entropy, and generates a three-domain segmented data set through a sliding window.

[0074] Data fidelity submodule: Traverses the three-domain segmented data set, calculates the causal stability and causal fidelity of each data segment in the set, and obtains the multi-domain causal fidelity of each data segment.

[0075] Invalid data removal submodule: Based on the causal stability of operating conditions and the causal fidelity of multiple domains, it calculates the effective test benchmark index and removes invalid data segments, outputting the effective time series dataset.

[0076] High-Risk Operating Condition Segment Module: This module calculates the causality coefficient of operating condition coupling disturbances on high-fidelity data segments within the valid time-series dataset, and identifies the high-risk operating condition segment dataset and core risk frequency bands. It is specifically divided into the following sub-modules: Sensitive Frequency Band Submodule: Calculates the set of coupled sensitive frequency bands for each high-fidelity data segment in the effective time-series dataset.

[0077] The computational submodule calculates the three-domain cross-frequency causal coupling degree and disturbance energy dissipation rate of the high-fidelity data segment in the effective time-series dataset based on the set of coupling-sensitive frequency bands.

[0078] Core frequency band sub-module: Based on the three-domain cross-frequency causal coupling degree and disturbance energy dissipation rate, the operating condition coupling disturbance causal coefficient is generated to obtain the high-risk operating condition segment dataset and core risk frequency band.

[0079] The dominant triggering factor module is used to isolate and determine the dominant triggering factor excitation source for the extreme high-risk operating condition data segment and its core risk frequency band in the high-risk operating condition data segment, and to calculate the instability coefficient to determine the key disturbance amplification and propagation path. Specifically, it is divided into the following sub-modules: Response path submodule: For the extreme high-risk operating condition data segment and its core risk frequency band in the high-risk operating condition segment dataset, three types of excitation sources are separated and the excitation sources are paired with the instability propagation path to obtain the excitation-response path pairing set.

[0080] Dominant Inducing Factor Module: Based on the incentive-response path pairing set, calculate the causal contribution rate of each incentive source within the core risk frequency band, and determine the dominant inducing factor incentive source and its path transmission loss rate.

[0081] Critical Path Submodule: The path transmission loss rate is oriented in the same direction as the index. The instability coefficient is calculated by combining the causal contribution rate of the corresponding dominant incentive source. The critical disturbance amplification transmission path and the secondary cooperative instability path are divided according to the size of the coefficient.

[0082] Stability Margin Assessment Module: This module constructs a modified parameter dataset based on the derived associated operating condition parameters of the dominant excitation source, calculates the critical decay margin and disturbance rejection redundancy of the parameters in the dataset, and obtains the wind turbine stability margin. It is specifically divided into the following sub-modules: The associated parameter submodule derives the corresponding associated operating condition parameters based on the dominant inducing factors and excitation sources. After data segmentation and instability coefficient correction, a corrected parameter dataset is obtained, and a nonlinear mapping function is generated for each associated operating condition parameter.

[0083] The margin calculation submodule obtains the parameter instability critical value based on the modified parameter dataset and the nonlinear mapping function. It calculates the critical attenuation margin and disturbance rejection redundancy based on the parameter instability critical value and generates the wind turbine stability margin by allocating weights through the instability coefficient.

[0084] Risk Level and Rectification List Module: Based on effective test benchmark indicators, operating condition coupling disturbance causality coefficient, instability coefficient, and wind turbine stability margin, this module calculates the final comprehensive stability score of the wind turbine and outputs a risk level and rectification priority list.

[0085] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for analyzing wind turbine stability test data, characterized in that, include: S1. Collect the three-domain core channel time sequence. The three-domain core channel time sequence refers to the original discrete time sequence of aerodynamic, mechanical and electrical domains that are synchronously collected in the full-condition stability test of the wind turbine. Perform time delay alignment on the time sequence, calculate the effective test benchmark index, and obtain the effective time sequence dataset. Specifically, it is divided into the following sub-steps: The time series of the three core channels in the wind turbine stability test is collected. The time series delay is aligned by the time series causal delay estimation algorithm based on the transfer entropy. The three-domain segmented data set is generated by the sliding window. Traverse the three-domain segmented data set, calculate the causal stability and causal fidelity of each data segment in the set, and obtain the multi-domain causal fidelity of each data segment; We perform weighted fusion based on the causal stability of operating conditions and the causal fidelity of multiple domains to obtain the effective test benchmark indicators for each data segment and remove invalid data segments, output the effective time series dataset, and retain the high-fidelity data segments whose effective test benchmark indicators are higher than the validity threshold. S2. Calculate the causality coefficient of coupling disturbances under operating conditions for the high-fidelity data segment in the effective time-series dataset, and determine the high-risk operating condition segment dataset and the core risk frequency band; this is specifically divided into the following sub-steps: The calculation of the set of coupled sensitive frequency bands for each high-fidelity data segment in the effective time-series dataset involves the following sub-steps: Using ensemble empirical mode decomposition, adaptive full-band decoupling is performed on the time-series data of the aerodynamic, mechanical, and electrical domains in each high-fidelity data segment of the effective time-series dataset to obtain the three-domain Hilbert marginal spectrum of a single data segment; based on the wind turbine system design parameters and measured frequency domain characteristics, a continuous frequency band interval containing the core characteristic frequency is delineated, i.e., the sensitive candidate frequency band pool; for the frequency bands to be determined within the sensitive candidate frequency band pool, the energy proportions of the aerodynamic, mechanical, and electrical domains in the frequency bands to be determined are calculated based on the three-domain Hilbert marginal spectrum; the effective three-domain cross-coupling sensitive frequency bands are determined according to the overlap of the three domains, resulting in the set of coupled sensitive frequency bands for the current high-fidelity data segment; the above process is repeated to complete the calculation of the set of coupled sensitive frequency bands for all high-fidelity data segments in the effective time-series dataset. Based on the set of coupling-sensitive frequency bands, the three-domain cross-frequency causal coupling degree and disturbance energy dissipation rate of the high-fidelity data segment in the effective time series dataset are calculated. Based on the three-domain cross-frequency causal coupling degree and disturbance energy dissipation rate, the operating condition coupling disturbance causal coefficient is generated to obtain the high-risk operating condition segment dataset and core risk frequency band. S3. For the high-risk operating condition data segment and its core risk frequency band, separate and determine the dominant inducing source of excitation, calculate the instability coefficient and determine the key disturbance amplification and propagation path. Specifically, it is divided into the following sub-steps: For the extreme high-risk operating condition data segment and its core risk frequency band in the high-risk operating condition segment dataset, three types of excitation sources are separated and the excitation sources are paired with the instability propagation path to obtain the excitation-response path pairing set; Based on the incentive-response path pairing set, the causal contribution rate of each incentive source is calculated within the core risk frequency band, and the dominant incentive source and its path transmission loss rate are determined. The path transmission loss rate is made into an index of the same direction. The instability coefficient is calculated by combining the causal contribution rate of the corresponding dominant incentive source. The key disturbance amplification transmission path and the secondary cooperative instability path are divided according to the size of the coefficient. S4. Construct a modified parameter dataset based on the derived related operating condition parameters of the dominant excitation source, calculate the critical decay margin and disturbance rejection redundancy of the parameters in the dataset, and obtain the wind turbine stability margin. S5. Based on effective test benchmark indicators, causal coefficient of coupled disturbance under operating conditions, instability coefficient, and wind turbine stability margin, calculate the final score of wind turbine comprehensive stability and output a list of risk levels and rectification priorities.

2. The method for analyzing wind turbine stability test data as described in claim 1, characterized in that, Based on the derived associated operating condition parameters of the dominant excitation source, a modified parameter dataset is constructed. The critical decay margin and disturbance rejection redundancy of the parameters in the dataset are calculated to obtain the wind turbine stability margin. This is specifically divided into the following sub-steps: Based on the dominant cause and excitation source, corresponding related operating condition parameters are derived. After data segmentation and instability coefficient correction, a corrected parameter dataset is obtained, and a nonlinear mapping function is generated for each related operating condition parameter. Based on the modified parameter dataset, the critical value of parameter instability is obtained by combining the nonlinear mapping function. The critical attenuation margin and disturbance rejection redundancy of the parameters are calculated based on the critical value of parameter instability. The stability margin of the wind turbine is generated by allocating weights through the instability coefficient.

3. The method for analyzing wind turbine stability test data as described in claim 1, characterized in that, Based on effective test benchmark indicators, causality coefficient of coupled disturbances under operating conditions, instability coefficient, and wind turbine stability margin, the final comprehensive stability score of the wind turbine is calculated, and a list of risk levels and rectification priorities is output. The specific steps are as follows: Based on the importance classification of performance indicators according to IEC standards, weights are assigned to effective test benchmark indicators and wind turbine stability margin, and weighted fusion is performed to obtain the wind turbine stability performance compliance rate. The causality coefficient of the coupled disturbance under operating conditions and the instability coefficient are assigned weights by dynamic weighting, and the risk acceptability of the wind turbine is calculated by weighting. Based on the dual-dimensional evaluation rules of the IEC wind power stability test standard, the wind turbine stability performance compliance rate and wind turbine risk acceptability are weighted and the two are weighted to output the final score of the wind turbine's comprehensive stability. Based on the final comprehensive stability score of the wind turbine, a list of risk levels and rectification priorities is determined.

4. A wind turbine stability test data analysis device, performing the method as described in any one of claims 1-3, characterized in that, The device includes: Acquisition module: used to acquire the time series of the three core channels, perform time delay alignment on the time series, calculate the effective test benchmark indicators, and obtain the effective time series dataset; High-risk operating condition segment module: used to calculate the causal coefficient of operating condition coupling disturbance in the high-fidelity data segment of the effective time series dataset, and to determine the high-risk operating condition segment dataset and the core risk frequency band; The dominant inducing factor module is used to separate and determine the dominant inducing factor excitation source for the extreme high-risk operating condition data segment and its core risk frequency band in the high-risk operating condition data segment, and to calculate the instability coefficient to determine the key disturbance amplification and propagation path. Stability margin assessment module: used to construct a modified parameter dataset based on the derived related operating condition parameters of the dominant excitation source, calculate the critical decay margin and disturbance rejection redundancy of the parameters in the dataset, and obtain the wind turbine stability margin; Risk Level and Rectification List Module: Based on effective test benchmark indicators, operating condition coupling disturbance causality coefficient, instability coefficient, and wind turbine stability margin, this module calculates the final comprehensive stability score of the wind turbine and outputs a risk level and rectification priority list.

Images