Wind power generation abnormal data analysis method and system

Abstract

The invention discloses a wind power generation abnormal data analysis method and system, and relates to the technical field of data analysis, and the method comprises the following steps: constructing a wind direction change rate enhanced perception model, analyzing the potential omen of wind direction abrupt change based on an ultra-short time scale wind direction change trend curve and a wind speed fluctuation coupling index collected at multiple measurement points, and determining the wind direction abrupt change. Generating a risk early warning label of wind wheel pointing deviation; and based on the risk early warning label, executing a high-frequency yaw disturbance prediction mechanism, and predicting an inflow angle continuous offset window caused by yaw response lag by using a nonlinear time sequence evolution trend and a short-term wind direction reversal probability curve. The wind direction sudden change early warning is realized through multi-measuring-point wind direction enhanced perception and wind speed coupling analysis, the response precision is improved and the energy consumption is reduced in combination with predictive yaw compensation and torque balance, the yaw parameters are dynamically optimized by using adaptive closed loop and reinforcement learning, the inflow angle is kept stable for a long time, and the power generation efficiency and the structural safety are improved.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, specifically to a method and system for analyzing abnormal data from wind power generation. Background Technology

[0002] Anomaly data analysis in wind power generation refers to the statistical analysis, pattern recognition, and intelligent diagnosis of collected operational data (such as wind speed, rotational speed, power generation, current, voltage, bearing temperature, vibration signals, etc.) during wind power generation. This analysis identifies data points or trends deviating from normal operating conditions, thereby determining whether equipment failures, control system anomalies, or environmental interference exist. Its main function is to proactively detect potential safety hazards and performance degradation, improve the operational reliability and power generation efficiency of wind power systems, reduce downtime and economic losses, and provide data support and decision-making basis for intelligent operation and maintenance, predictive maintenance, and remote monitoring of wind farms. It is a key technological link in realizing the full lifecycle management and intelligent upgrading of wind power.

[0003] The existing technology has the following shortcomings: In existing technologies, under conditions of drastic wind direction fluctuations, the yaw response process often has a slight delay. If this delay accumulates continuously over multiple consecutive wind direction switching cycles, it can easily lead to a persistent deviation between the actual direction of the wind turbine and the inflow wind direction. This causes the blade inflow angle to deviate from the optimal aerodynamic conditions for a long period of time, resulting in uneven stress on the wind turbine. Under high variable load conditions, this can induce fatigue crack propagation or even structural fracture at the hub, seriously affecting the overall operational safety and structural reliability of the turbine.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for analyzing abnormal data in wind power generation, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for analyzing abnormal wind power generation data, comprising the following steps: An enhanced perception model for wind direction change rate is constructed. Based on the wind direction change trend curve and wind speed fluctuation coupling index collected from multiple measurement points on an ultra-short time scale, potential precursors of sudden wind direction changes are analyzed, and risk warning labels for wind turbine pointing deviation are generated. Based on risk warning labels, a high-frequency yaw disturbance prediction mechanism is implemented. By utilizing the nonlinear time series evolution trend and the short-term wind direction reversal probability curve, the continuous inflow angle offset window caused by yaw response lag is predicted, and the corresponding predictive yaw delay level is output. The predictive yaw delay level is injected into the dynamic target yaw angle generation process. Combined with the current wind turbine azimuth information, blade attitude parameters and inertial response boundary conditions, the feedforward yaw angle adjustment including the hysteresis compensation factor is calculated so that the control command can be corrected in advance to follow the wind direction change trend. Based on the feedforward yaw angle adjustment, the bearing transient torque response equalization algorithm is triggered. By adjusting the start-up inertia release curve of the servo drive, the rotation start-up angular acceleration of the yaw system is matched with the direction and intensity of the predicted wind direction change, while optimizing drive energy consumption. The nonlinear difference between the changing trend of the wind turbine's inertial torque and the actual execution error of the yaw angle is continuously monitored. Based on this nonlinear difference, an adaptive closed-loop correction path is established to dynamically adjust the response frequency and angular velocity of the servo drive and suppress the cumulative amplification effect of the positioning deviation. In the adaptive closed-loop correction path, wind direction perception results, yaw control parameters and execution error trajectories are integrated to construct a dynamic control and scheduling engine with short-term memory and error reinforcement learning capabilities. This enables the self-updating of yaw response parameters in each control cycle, thereby maintaining the long-term stability of the blade inflow angle and optimal aerodynamic conditions.

[0007] Preferably, constructing an enhanced wind direction change rate sensing model includes the following steps: Three wind direction measurement points were set up at the top of the wind turbine tower, the root of the blade, and the front of the nacelle, respectively. Wind direction data at an ultra-short time scale were collected at a sampling interval of 200 milliseconds to construct a three-dimensional wind direction data matrix with timestamp alignment. The wind direction change rate and acceleration are obtained by performing second-order central difference operation on the data of each measuring point in the three-dimensional wind direction data matrix. After smoothing, the wind direction change trend function of each measuring point is extracted, and a representative wind direction change trend curve is generated by weighted averaging according to spatial weights. Correlation and covariance analysis were performed based on the wind direction change trend curve and the instantaneous wind speed change rate sequence to calculate the wind speed fluctuation coupling index, and risk assessment was initiated when the threshold conditions were continuously met. The input feature vector is constructed based on the maximum wind direction change rate, change duration, peak coupling index, and instantaneous offset angle. The risk level is then output by the Gaussian fuzzy evaluation model, and a corresponding risk warning label is generated.

[0008] Preferably, the risk warning labels are divided into three categories: no warning, moderate warning, and severe warning, and serve as the trigger signal and weight adjustment factor for the subsequent predictive yaw angle adjustment logic.

[0009] Preferably, performing high-frequency yaw disturbance prediction based on risk warning labels includes the following steps: Based on the wind turbine pointing offset risk label, wind direction trend curve, wind speed disturbance curve and predicted yaw angle sequence from the past 30 seconds to the current moment are collected to construct a wind direction disturbance time series set and perform wavelet transform to extract the disturbance principal component sequence. Based on the perturbation principal component sequence, Markov chain is used to statistically analyze the short-term wind direction reversal probability and generate a time series curve of wind direction reversal probability. The perturbation principal component sequence is coupled with the wind direction reversal probability curve. Combined with the nacelle azimuth, predicted wind direction and blade aerodynamic response delay, the predicted inflow angle is calculated and a continuous inflow angle offset window is constructed. Based on the time length of the offset window, the maximum offset angle, the probability of occurrence and the rate of change, the predictive yaw delay level is output.

[0010] Preferably, the predictive yaw delay level is divided into four levels: Level 1 indicates a low risk of yaw lag, Level 2 indicates a certain probability of continuous inflow angle deviation, Level 3 indicates a high probability of deviation caused by lag, and Level 4 indicates a high risk of delay that could lead to yaw control disorder and abnormal stress on the entire machine. The level is used as a quantitative control indicator for the subsequent precision adjustment of target yaw angle correction and the active scheduling of servo drive response frequency.

[0011] Preferably, injecting the predictive yaw delay level into the dynamic target yaw angle generation process includes the following steps: The predictive yaw delay level is received and used as a weighting factor in the calculation of the target yaw angle, and the nacelle azimuth, rotor rotation center azimuth and blade attitude parameters are obtained in real time. Based on the nacelle azimuth angle and the predicted wind direction angle, combined with the blade attitude and inertial response boundary conditions, the expected aerodynamic inflow angle change trend curve within the prediction time window is calculated. The compensation coefficient corresponding to the delay level is nonlinearly superimposed with the trend curve to generate a target yaw angle sequence containing hysteresis compensation factor. Over-correction is avoided by smoothing filtering and lead limit. The difference between the target yaw angle sequence and the current position of the cabin is calculated to obtain the instantaneous yaw angle adjustment. After energy consumption optimization, a control command is generated and output to the yaw drive actuator.

[0012] Preferably, the bearing transient torque response equalization algorithm triggered by the feedforward yaw angle adjustment includes the following steps: By receiving the feedforward yaw angle adjustment and combining it with the wind turbine main shaft speed, wind speed change rate and predicted wind direction change intensity and direction, a starting torque demand matrix is ​​established to determine the starting inertia release strategy. The servo drive start-up inertia release curve is calculated based on the start-up inertia release strategy. The curve includes an acceleration up section, a stable section and a gradual descent section, and is matched with the predicted wind direction change characteristics. During the execution process, the nacelle angular acceleration and bearing reaction torque are collected in real time, and the difference between them and the release curve is calculated and the drive input is adjusted in a closed loop to make the actual angular acceleration aligned with the target curve. Before completing the target yaw angle adjustment, energy consumption optimization is performed based on the fit between the release curve and the predicted wind direction change trend, and the high torque output time is controlled to reduce energy consumption and mechanical shock.

[0013] Preferably, the establishment of an adaptive closed-loop correction path based on the trend of inertial moment variation and the actual yaw angle error includes the following steps: During the yaw operation, the inertial moment data of the wind turbine main shaft and the real-time yaw angle of the nacelle are collected simultaneously to calculate the trend curve of the inertial moment change and the actual execution error. By mapping the trend curve of the change in inertial moment to the execution error input nonlinear mapping function, a nonlinear difference reflecting the coupling characteristics of inertial change and error is obtained; Based on the nonlinear difference, an adaptive closed-loop correction path is dynamically generated to determine the servo drive response frequency and target angular velocity of each correction node and refresh them in each yaw execution cycle. During the execution of the closed-loop path, when the difference converges to the tolerance range, the response frequency and angular velocity are reduced to enter a steady-state maintenance mode, and the drive energy consumption is reduced through energy optimization.

[0014] Preferably, the self-updating of yaw response parameters by fusing multi-source data in the adaptive closed-loop correction path and based on short-term memory and error reinforcement learning includes the following steps: Collect and fuse wind direction change rate, wind direction change probability, inflow angle prediction curve, target yaw angle, servo drive response frequency, angular velocity setpoint and execution error curve to construct a synchronous input data matrix; The input data and yaw execution results of the most recent control cycles are stored in a sliding window dataset to form a time series feature set. An error reinforcement learning decision model is built based on the time series feature set and the current real-time data to output the optimal control parameter adjustment scheme. In each control cycle, the feedforward yaw angle compensation value, servo drive response frequency and angular velocity limit range are updated, and the execution results are fed back to the short-term memory window to achieve continuous optimization.

[0015] The wind power generation anomaly data analysis system includes a wind direction change enhanced perception module, a high-frequency yaw disturbance prediction module, a dynamic target yaw angle generation module, a bearing transient torque response equalization module, an adaptive closed-loop correction module, and a dynamic control and scheduling module. The wind direction change enhancement perception module constructs a wind direction change rate enhancement perception model. Based on the ultra-short timescale wind direction change trend curve and wind speed fluctuation coupling index collected from multiple measurement points, it analyzes potential precursors of sudden wind direction changes and generates risk warning labels for wind turbine pointing deviation. The high-frequency yaw disturbance prediction module, based on risk warning labels, executes a high-frequency yaw disturbance prediction mechanism. It uses nonlinear time series evolution trends and short-term wind direction reversal probability curves to predict the continuous inflow angle offset window caused by yaw response lag and outputs the corresponding predictive yaw delay level. The dynamic target yaw angle generation module injects the predictive yaw delay level into the dynamic target yaw angle generation process. Combining the current wind turbine azimuth information, blade attitude parameters and inertial response boundary conditions, it calculates the feedforward yaw angle adjustment amount including the hysteresis compensation factor, so that the control command can be corrected in advance to follow the wind direction change trend. The bearing transient torque response equalization module triggers the bearing transient torque response equalization algorithm based on the feedforward yaw angle adjustment. By adjusting the starting inertia release curve of the servo drive, it makes the rotational starting angle acceleration of the yaw system match the direction and intensity of the predicted wind direction change, while optimizing drive energy consumption. The adaptive closed-loop correction module continuously monitors the nonlinear difference between the changing trend of the wind turbine's inertial torque and the actual execution error of the yaw angle. Based on this nonlinear difference, an adaptive closed-loop correction path is established to dynamically adjust the response frequency and angular velocity of the servo drive and suppress the cumulative amplification effect of the positioning deviation. The dynamic control and scheduling module integrates wind direction perception results, yaw control parameters, and execution error trajectories in the adaptive closed-loop correction path to construct a dynamic control and scheduling engine with short-term memory and error reinforcement learning capabilities. This enables the self-updating of yaw response parameters in each control cycle, thereby maintaining the long-term stability of the blade inflow angle and optimal aerodynamic conditions.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention constructs an enhanced sensing model of wind direction change rate and combines it with multi-point ultra-short timescale wind direction trend and wind speed fluctuation coupling analysis to achieve early identification and risk classification warning of sudden wind direction changes. Compared with the traditional method that relies on single-point sensing and fixed threshold triggering, it significantly improves the sensitivity and accuracy of wind direction change detection, enabling yaw control to intervene in advance before significant wind direction changes, reducing the probability of continuous yaw error from the source, and ensuring that the blades maintain a near-optimal inflow angle for more operating time.

[0017] This invention generates a feedforward yaw angle adjustment amount containing a hysteresis compensation factor under the drive of a predictive yaw delay level, and combines it with a bearing transient torque response equalization algorithm to achieve precise matching between yaw action and the predicted wind direction change direction and intensity. This not only improves the timeliness and following accuracy of yaw response, but also dynamically optimizes the starting inertia release curve during the drive process, effectively reducing the energy consumption and mechanical shock of the servo drive, and extending the service life of bearings and transmission components, thereby significantly improving the reliability and economy of the whole machine under complex wind conditions.

[0018] This invention constructs an adaptive closed-loop correction path by continuously monitoring the nonlinear difference between the trend of inertial torque change and the execution error. Based on this, it integrates short-term memory and error reinforcement learning mechanisms to achieve adaptive updates of yaw response parameters in each control cycle. This enables the control strategy to be continuously optimized as the operating environment changes, effectively suppressing the cumulative amplification effect of positioning deviation. This ensures the stability of the blade inflow angle and optimal aerodynamic conditions during long-term operation of the unit, further improving the power generation efficiency and structural safety of the wind turbine throughout its entire life cycle. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0020] Figure 1 This is a flowchart of the wind power generation anomaly data analysis method of the present invention.

[0021] Figure 2 This is a schematic diagram of the modules of the wind power generation anomaly data analysis system of the present invention. Detailed Implementation

[0022] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0023] This invention provides, for example Figure 1 The wind power generation anomaly data analysis method shown includes the following steps: An enhanced perception model for wind direction change rate is constructed. Based on the wind direction change trend curve and wind speed fluctuation coupling index collected from multiple measurement points on an ultra-short time scale, potential precursors of sudden wind direction changes are analyzed, and risk warning labels for wind turbine pointing deviation are generated. This step proposes a method for constructing an enhanced wind direction change rate sensing model to improve the sensitivity to sudden wind direction changes and the accuracy of risk identification during wind power generation, thereby providing accurate and timely predictive support for subsequent yaw angle adjustments. The method includes the following steps: Three wind direction measurement points were set up at the top of the wind turbine tower, the blade root, and the front of the nacelle in the wind farm. A three-axis ultrasonic anemometer was used to collect wind direction data at each measurement point in real time, with a sampling interval of 200 milliseconds, forming a raw wind direction angle sequence on an ultra-short timescale. Using 25 sets of wind direction angle data collected from each measurement point within 5 seconds as the basic input unit, the three data points were aligned using timestamps to construct a three-dimensional wind direction data matrix D[i, j, t], where i represents the measurement point number, j represents the sampling number, and t represents the unified timestamp. This approach not only covers the spatial distribution of wind direction disturbances around the nacelle but also ensures temporal synchronization, providing high-quality input for subsequent trend fitting and prediction.

[0024] Second-order central difference calculations are performed on the data at each measuring point in the aforementioned wind direction data matrix to obtain the wind direction change rate V1[i,t] and acceleration V2[i,t] for each point within a continuous time interval. Local regression smoothing is applied to the V1 and V2 curves generated for each measuring point to extract the wind direction change trend function T[i,t] for the current time period at each measuring point. Furthermore, the three T[i,t] functions are weighted and averaged using the measuring point spatial weight coefficient W[i] (set based on the spatial distance and height difference from the wind turbine center) to obtain the representative wind direction change trend curve T_avg(t) around the current engine compartment. This trend curve reflects the overall trend and dominant direction of the wind direction change rate in real time and is the core basis for judging the risk of sudden wind direction changes.

[0025] Based on the obtained T_avg(t), the instantaneous wind speed S(t) recorded by the wind speed sensor installed on the top of the cabin during the corresponding time period is calculated, and the wind speed change rate sequence S_diff(t) is formed by the difference between adjacent samples. Covariance analysis and Pearson correlation calculation are performed on T_avg(t) and S_diff(t) to obtain the coupling strength index C(t) between the two, which is used to measure whether rapid changes in wind direction are accompanied by wind speed disturbances. When C(t) exceeds a preset threshold (e.g., 0.7) and lasts for more than 1 second, it is considered that there is a strong coupling precursor to a sudden change in wind direction. In order to eliminate false alarms caused by single-point anomalies, in practical applications, it is set that at least two consecutive cycles of sudden changes in wind direction trends meet the coupling condition before proceeding to the next step of risk label generation.

[0026] Based on four parameters—the maximum derivative of T_avg(t) (i.e., the peak value of the wind direction change rate), the duration of change Δt, the peak value of the coupling index C(t), and the instantaneous offset angle Δθ—an input feature vector F = [dT / dt_max, Δt, C_max, Δθ] is constructed. This vector is then fed into a pre-trained risk assessment model based on Gaussian fuzzy membership functions, outputting the wind turbine pointing offset risk level L corresponding to the current time slice. The risk levels are divided into three levels: L1 indicates low offset risk, L2 indicates a significant trend of sudden wind direction change, and L3 indicates a high-risk offset precursor approaching the wind turbine's aerodynamic instability zone. The corresponding wind turbine pointing offset risk warning labels are: no warning, moderate warning, and severe warning. This label will serve as the trigger signal and weight adjustment factor for the predictive yaw angle adjustment logic in subsequent steps, directly participating in target yaw angle generation, inertia release decision-making, and control frequency correction strategy formulation.

[0027] This embodiment discloses a specific and repeatable method for enhancing the perception of wind direction change rate. This method achieves high-confidence, low-false-report dynamic perception of potential wind turbine deviation by employing multi-point high-frequency sampling, trend fitting modeling, wind speed disturbance coupling index calculation, and fuzzy logic classification label generation. This significantly outperforms traditional methods relying on single-point measurements, static threshold judgments, or moving average strategies. It exhibits stronger robustness and practicality in handling high-load wind conditions and rapid wind direction disturbances. This method provides strong front-end data support for subsequent yaw control response optimization.

[0028] Based on risk warning labels, a high-frequency yaw disturbance prediction mechanism is implemented. By utilizing the nonlinear time series evolution trend and the short-term wind direction reversal probability curve, the continuous inflow angle offset window caused by yaw response lag is predicted, and the corresponding predictive yaw delay level is output. This step proposes a method for predicting high-frequency yaw disturbances based on risk warning labels, aiming to address the persistent inflow angle shift caused by yaw response lag in wind power generation devices under conditions of severe wind direction fluctuations. Unlike existing technologies that statically set yaw initiation thresholds and calculate wind direction averages over fixed periods, this method uses real-time generated rotor pointing deviation risk labels as trigger conditions. Combining nonlinear time series analysis and short-term probability modeling, it dynamically calculates the wind energy inflow angle shift range caused by yaw response delay, and based on this, determines whether yaw control may lag in the future and the degree of lag, providing more forward-looking and quantitative predictive support. The method includes the following steps: Based on the wind turbine pointing offset risk label obtained in the previous step, the high-frequency yaw disturbance prediction process is activated. Specifically, the wind direction trend curve, wind speed disturbance curve, and predicted yaw angle sequence from the past 30 seconds to the current moment are used as input to construct a wind direction disturbance time series set S(t). Wavelet transform is performed on the set S(t) to decompose the wind direction change patterns at different scales, and the energy change trends of the main disturbance frequency bands within 3 seconds before and after the current moment are extracted to form the disturbance principal component sequence P(t). This sequence is used to characterize the evolution trend of wind direction disturbance at different time resolutions, providing a multi-dimensional change basis for the next step of probabilistic modeling. Compared with existing schemes that use moving average or Fourier transform for periodic analysis, this processing method can more sensitively capture the non-stationary, highly abrupt wind direction disturbance patterns.

[0029] Based on the perturbation principal component sequence P(t), a short-term wind direction reversal probability curve is constructed. This step employs a Markov chain nested sequence approach to statistically analyze the historical probability distribution of wind direction reversal within a short time window (e.g., 3 seconds). Specifically, taking the current wind direction trend as the ground state, the probability of it transforming into the opposite direction (change greater than 90°) in historical wind conditions is analyzed, and a "wind direction reversal probability time series curve" Q(t) is constructed. This curve expresses the probability of a significant reversal of the current wind direction within a short period at each moment. Combining the rate of change of the perturbation principal component, the risk level of the current wind turbine being at a "perturbation inflection point" can be further quantified, thereby inferring the wind direction change challenges that yaw control will face.

[0030] The generated perturbation principal component sequence P(t) and the wind direction reversal probability curve Q(t) are dynamically coupled for analysis to construct a continuous inflow angle offset window. To construct this window, the predicted inflow angle θ(t) is first calculated based on the angle between the current nacelle azimuth and the predicted wind direction, combined with the rotor rotation direction and blade aerodynamic response delay. Within the prediction period of 3 to 5 seconds, an LSTM (Long Short-Term Memory) model is used to perform multi-step predictions of θ(t). Combining the dynamic changes of P(t) and Q(t), the time interval T[θ_dev] where continuous deviations from the aerodynamic design angle may occur is identified; this is the continuous inflow angle offset window. The existence of this window indicates that if the yaw angle is not adjusted in time, the rotor will continuously be in a suboptimal aerodynamic inflow state, resulting in periodic asymmetric forces.

[0031] Based on four key indicators—the duration of the continuous inflow angle offset window T[θ_dev], the maximum offset angle, the probability of occurrence, and the rate of change—a predictive yaw delay level assessment model is constructed. This model classifies yaw delay levels according to the severity of the risk: Level 1 indicates a low risk of yaw lag; Level 2 indicates a certain probability of continuous inflow angle offset; Level 3 indicates an almost certain lag leading to offset; and Level 4 indicates a high-risk delay that could cause yaw control malfunctions and abnormal turbine stress. The predictive yaw delay level, as a quantitative indicator, not only provides a basis for adjusting the accuracy of subsequent target yaw angle corrections but also allows for proactive scheduling of the response frequency of the yaw control drive mechanism, improving the wind turbine's adaptability to complex wind conditions.

[0032] The high-frequency yaw disturbance prediction method based on risk warning labels proposed in this embodiment is significantly different from the existing technology that uses fixed sampling period and wind direction difference threshold control. It not only enhances the ability to identify sudden wind direction trends through nonlinear time series multi-scale analysis, but also realizes the quantitative prediction of future yaw lag risk through short-term reversal probability modeling. Finally, it outputs a predictive yaw delay level that can be used by active control strategies, and constructs a complete closed-loop mechanism from "trend identification" to "risk prediction" to "quantitative output". It provides key data support and technical path for realizing intelligent wind power generation control, and has high engineering practical value and innovation.

[0033] The predictive yaw delay level is injected into the dynamic target yaw angle generation process. Combined with the current wind turbine azimuth information, blade attitude parameters and inertial response boundary conditions, the feedforward yaw angle adjustment including the hysteresis compensation factor is calculated so that the control command can be corrected in advance to follow the wind direction change trend. This step proposes a method to inject predictive yaw delay levels into the dynamic target yaw angle generation process. This method is used to proactively compensate for yaw response lag during wind power generation and to pre-correct the rotor orientation, keeping the rotor inflow angle within the optimal aerodynamic range. Unlike existing technologies that rely solely on the difference between the current wind direction and the nacelle azimuth to set the target yaw angle, this method introduces predictive yaw delay levels as a dynamic control variable. It calculates this by combining rotor azimuth, blade attitude, and inertial response boundary conditions, generating a feedforward yaw angle adjustment that includes a hysteresis compensation factor. This allows the yaw control command to adapt to future wind direction changes in advance, thereby significantly improving yaw accuracy and response efficiency. The method includes the following steps: The system receives the predictive yaw delay level output from the previous prediction stage and incorporates it as a weighting factor into the dynamic target yaw angle calculation process. Specifically, the predictive delay level is denoted as L, and the delay compensation coefficient is denoted as k(L). This coefficient is set according to the delay level classification; for example, level 1 corresponds to a lower compensation coefficient to reduce unnecessary adjustments, while level 3 or 4 corresponds to a higher compensation coefficient to achieve a more aggressive yaw lead. Simultaneously, the system acquires real-time attitude parameters such as the current nacelle azimuth angle α_current, the rotor rotation center azimuth, and the blade pitch and angles, providing complete initial state input for subsequent corrections. The key feature of this step is that it directly embeds the quantified delay prediction results into the target calculation logic, achieving deep coupling with wind condition changes, rather than post-correction as in traditional methods.

[0034] Based on the angle Δθ between the nacelle azimuth and the current predicted wind direction, and combined with blade attitude parameters and rotor radius, the expected aerodynamic inflow angle variation trend curve I(t) within the prediction time window T is calculated. During the calculation, two supplementary parameters are introduced: the rate of change of the projected area of ​​the blade's windward surface and the rate of change of the local angle of attack. This ensures that the trend curve reflects the dynamic impact of wind direction changes on the blade forces. Simultaneously, by incorporating inertial response boundary conditions, the minimum and maximum available acceleration required for the nacelle to shift from the current azimuth to the target azimuth are calculated to determine the feasibility of achieving target correction within the time window T. In this way, the target yaw angle is set not only based on the spatial azimuth difference but also integrates aerodynamic response and structural inertial constraints, representing a significant innovation compared to existing methods.

[0035] The compensation coefficient k(L) corresponding to the predicted yaw delay level is nonlinearly superimposed with the expected inflow angle change trend curve I(t) to generate a target yaw angle sequence β(t) including a hysteresis compensation factor. To avoid control oscillations caused by overcorrection, a quadratic smoothing filter and a lead limiter are introduced in the calculation, allowing the compensation lead angle to dynamically change within the maximum lead angle β_max. For example, when the forecast shows that the future wind direction will rapidly veer to the right and the delay level is 4, the target yaw angle will be corrected to the right by a large margin in advance; if the delay level is only 1, the lead is very small or even zero. This process ensures that the compensation strategy is proactive without increasing energy consumption or causing mechanical shocks due to over-adjustment.

[0036] The difference between the target yaw angle sequence β(t), which includes a hysteresis compensation factor, and the current position of the nacelle is calculated to obtain the instantaneous yaw angle adjustment Δβ_adj, and a control command signal is generated and output to the yaw drive actuator. Before output, an energy consumption optimization calculation is performed, that is, the amplitude and acceleration of the control signal are dynamically adjusted according to the current wind speed, yaw drive start-up power, and expected execution time to ensure that the drive energy consumption and mechanical wear are reduced while meeting the advance correction accuracy. The output timing of this control command is synchronized with the wind direction prediction cycle, so that the yaw action is initiated in advance before the wind direction change occurs, and the wind turbine is close to the optimal inflow angle position when the actual wind direction change arrives, thereby maximizing aerodynamic efficiency and reducing structural fatigue.

[0037] This step directly incorporates the predictive yaw delay level into the dynamic target yaw angle generation. By combining the rotor's current azimuth, blade attitude, and inertial constraints to calculate the feedforward yaw angle adjustment including a hysteresis compensation factor, it achieves a shift from passive, delayed yaw control to proactive, advance yaw control. This method not only improves the timeliness and accuracy of yaw maneuvers but also effectively reduces the risk of persistent inflow angle deviation. It significantly outperforms existing post-response and single-angle difference-driven control strategies, demonstrating outstanding technological advancement and engineering application value.

[0038] Based on the feedforward yaw angle adjustment, the bearing transient torque response equalization algorithm is triggered. By adjusting the starting inertia release curve of the servo drive, the rotational starting angle acceleration of the yaw system is matched with the direction and intensity of the predicted wind direction change, while optimizing drive energy consumption. This step proposes a method for triggering the transient torque response balancing of the bearing based on the feedforward yaw angle adjustment. This method aims to rationally allocate the release of starting inertia of the yaw drive when yaw control is initiated early, ensuring that the angular acceleration of the nacelle rotation startup is highly matched with the direction and intensity of the predicted wind direction change, while simultaneously reducing drive energy consumption and mitigating mechanical shock. This method differs from existing technologies that use fixed yaw drive acceleration or rely solely on torque thresholds for startup. Instead, it achieves a flexible, efficient, and controllable startup process by dynamically sensing and intelligently adjusting the mechanical response to deeply couple the mechanical response with the predicted wind conditions. The method includes the following steps: The system receives the feedforward yaw angle adjustment Δβ_adj output from the previous calculation stage and uses it as one of the core input variables for torque balance calculation. To achieve dynamic control, it is also necessary to simultaneously acquire quantitative indicators of the current turbine shaft speed, wind speed change rate, and predicted wind direction change intensity and direction. Based on these inputs, a "starting torque demand matrix" is established. This matrix corresponds to different starting inertia release strategies under different wind direction changes (clockwise or counterclockwise) and different wind direction change intensities. For example, when the predicted wind direction change is large and fast, the matrix outputs a higher starting torque target; when the predicted change is gradual, it outputs a lower starting torque target to reduce ineffective energy consumption. The core innovation of this step lies in binding the feedforward yaw angle adjustment to the wind direction change trend in real time, rather than using a fixed setpoint.

[0039] Based on the output of the starting torque demand matrix, the required servo drive starting inertia release curve is calculated. This curve, with time on the horizontal axis and drive torque on the vertical axis, is divided into three stages: an acceleration phase, a stabilization phase, and a descent phase. The shape of the acceleration phase is determined by the predicted wind direction change acceleration and the feedforward adjustment, ensuring that the angular acceleration at the moment of nacelle startup can quickly overcome static friction and synchronously follow the wind direction change rhythm. The stabilization phase maintains a constant torque level that just meets the requirements for predicted wind direction correction, avoiding excessive acceleration that could cause inertial overshoot. The descent phase gradually releases torque as the engine approaches the target yaw angle, reducing mechanical shock. This curve generation method is significantly superior to the fixed linear acceleration mode in existing technologies because it can dynamically match the characteristics of wind condition changes.

[0040] During servo drive execution, the current angular acceleration of the engine compartment and the bearing reaction torque are acquired in real time, and the difference between these and the target inertia release curve is calculated. When the difference exceeds the set tolerance range, the drive input current or hydraulic output pressure is immediately adjusted to increase or decrease the instantaneous torque, so that the actual angular acceleration is re-aligned with the target curve. This closed-loop adjustment ensures uniform force distribution on the bearing during the startup phase, avoiding excessive load on one side of the bearing for a long time, and significantly reducing the risk of bearing fatigue damage and premature wear. Unlike traditional methods that rely solely on a single torque setting during startup, this method corrects the torque in real time throughout the entire startup process, resulting in a smoother and more reliable response.

[0041] Before adjusting the target yaw angle, an energy consumption optimization calculation is performed based on the fit between the current torque release curve and the predicted wind direction change trend. This optimization process calculates the energy consumed per unit angle correction and, combined with the predicted wind direction change trend within the next 2 seconds, determines whether to prematurely end the high torque output or extend the low torque stable section, aiming to achieve the goal of "achieving necessary corrections with minimal energy consumption." The optimized control commands directly act on the drive actuator, enabling it to accurately complete the yaw action while reducing energy consumption to an optimal level and significantly reducing gear meshing impact and bearing end-face load fluctuations, thereby extending the overall service life of the drive mechanism and bearings.

[0042] This step dynamically generates and corrects the servo drive start-up inertia release curve in real time under the drive of the feedforward yaw angle adjustment, so that the yaw start-up process can be highly adapted to the predicted wind direction and intensity. At the same time, it realizes bearing force balance and drive energy consumption optimization, which overcomes the problems of single start-up response, coarse inertial control and excessive energy consumption in the existing technology. It has significant technical progress and engineering application value in terms of yaw control stability, response speed and structural durability.

[0043] The nonlinear difference between the changing trend of the wind turbine's inertial torque and the actual execution error of the yaw angle is continuously monitored. Based on this nonlinear difference, an adaptive closed-loop correction path is established to dynamically adjust the response frequency and angular velocity of the servo drive and suppress the cumulative amplification effect of the positioning deviation. This step proposes an adaptive closed-loop correction method based on the trend of inertial torque variation and the actual yaw angle execution error, used to continuously suppress the accumulation and amplification of yaw positioning deviation during wind turbine operation. Unlike existing methods that rely on fixed frequency and fixed angular velocity for yaw adjustment, this method continuously monitors the trend of inertial torque variation of the wind turbine during yaw execution and calculates the nonlinear difference between this trend and the actual yaw angle execution error in real time. An adaptive closed-loop correction path is dynamically constructed using this difference, and the response frequency and angular velocity of the servo drive are adjusted accordingly to achieve long-term stable and accurate positioning. The method includes the following steps: During yaw execution, a high-precision torque sensor and rotary encoder synchronously acquire the inertial torque data sequence M(t) on the wind turbine main shaft and the real-time yaw angle α_actual(t) of the nacelle. By performing second-order difference analysis on M(t), the rate of change of inertial torque dM / dt is obtained, and combined with wind turbine speed and wind direction change data to generate the inertial torque change trend curve T_M(t). Simultaneously, the target yaw angle α_target(t) is acquired, and the actual execution error E(t) = α_target(t) − α_actual(t) is calculated, providing the necessary input for subsequent nonlinear difference calculations. Unlike existing technologies that only evaluate the error after adjustment, this method tracks the dynamic relationship between inertial torque and angle deviation in real time during the yaw maneuver.

[0044] The inertial moment variation trend curve T_M(t) and the execution error E(t) are input into the nonlinear mapping function F(·), which is constructed through polynomial fitting and weight optimization to capture the coupling characteristics between inertial variation and execution error. Specifically, the nonlinear difference ΔNL(t) = F[T_M(t), E(t)], where F includes the inertial sensitivity coefficient, error sensitivity coefficient, and directional adjustment factor, used to describe the degree of influence of execution error on cabin response when the inertial moment variation is large. This nonlinear difference reflects the difficulty of yaw adjustment and the inertial compensation requirement under the current state, and is the core basis for establishing an adaptive closed-loop correction path.

[0045] An adaptive closed-loop correction path is dynamically generated based on the nonlinear difference ΔNL(t). This path consists of multiple correction nodes, each defined by the servo drive response frequency f_adj(t) and the target angular velocity ω_adj(t) at a specific time point, and forms a smooth adjustment trajectory through Bézier curve interpolation. The specific adjustment rules are as follows: when ΔNL(t) is positively large, the response frequency is increased and the angular velocity is appropriately increased to quickly eliminate the accumulated deviation; when ΔNL(t) is negatively large, the angular velocity is reduced while maintaining a high response frequency to avoid reverse overshoot and mechanical shock. This closed-loop path is dynamically refreshed within each yaw execution cycle to ensure that the adjustment strategy matches the current wind conditions and structural response state, thereby preventing the positioning error from gradually amplifying during multi-cycle yaw adjustments.

[0046] During the execution of the closed-loop correction path, new inertial torque trends and execution error data are continuously collected and compared with the current ΔNL(t). When the difference gradually converges to the set tolerance range, the response frequency and angular velocity of the servo drive are gradually reduced, entering a steady-state maintenance mode to maintain the long-term stability of the target yaw angle. During this process, energy optimization calculations are performed to ensure that the overall energy consumption of the servo drive is reduced without affecting accuracy, thus minimizing fatigue accumulation in bearings and transmission mechanisms. Unlike traditional single proportional control, this method introduces real-time feedback of the dynamic relationship between inertial trend and error during the control process, making the adjustment action predictable and adaptive, significantly improving the long-term operational reliability and positioning accuracy of the wind turbine generator under complex wind conditions.

[0047] This step constructs an adaptive closed-loop correction path by continuously monitoring the nonlinear coupling relationship between the changing trend of inertial torque and the execution error. During execution, the response frequency and angular velocity of the servo drive are dynamically adjusted. This not only suppresses the cumulative amplification effect of positioning deviation, but also optimizes energy consumption and mechanical force distribution. It overcomes the shortcomings of existing technologies in yaw control, such as slow response, easy error accumulation, and lack of dynamic correction mechanism. It has significant technological advancements and engineering application value.

[0048] In the adaptive closed-loop correction path, the wind direction perception results, yaw control parameters and execution error trajectory are integrated to build a dynamic control and scheduling engine with short-term memory and error reinforcement learning capabilities. This enables the self-updating of yaw response parameters in each control cycle, thereby maintaining the long-term stability of the blade inflow angle and the optimal aerodynamic conditions. This step proposes a method to self-update yaw response parameters by fusing multi-source operational data in an adaptive closed-loop correction path and utilizing short-term memory and error reinforcement learning. This method aims to maintain the stability of the blade inflow angle and optimal aerodynamic conditions during long-term operation of wind turbine generators. Unlike existing technologies that only perform single parameter corrections based on the current error within a fixed period, this method introduces a short-term memory mechanism and a reinforcement learning decision model to deeply fuse wind direction perception results, yaw control parameters, and execution error trajectories. Within each control cycle, the yaw response parameters are dynamically updated based on historical trends and real-time changes, achieving continuous optimization of the control strategy. The method includes the following steps: Multi-source input data is collected and fused, including the wind direction change rate, wind direction change probability, and inflow angle prediction curve from the wind direction sensing process, as well as the target yaw angle, servo drive response frequency, and angular velocity setpoint used in the current yaw control. Simultaneously, the error curve E(t) between the actual yaw angle and the target yaw angle within the current execution cycle is obtained, and the error change rate and direction are recorded. To ensure the timeliness and consistency of the data, the above multi-source data are synchronously processed according to a unified timestamp, and an input data matrix D_input[n, k] is constructed, where n is the sampling time and k is the data dimension. This step differs from existing methods that only use a single wind direction difference or a single error value; by synchronously fusing multiple variables, it provides rich and complete feature information for subsequent learning and memorization.

[0049] Based on the adaptive closed-loop correction path, a short-term memory mechanism is introduced. The input data matrix and corresponding yaw response results from the most recent control cycles (e.g., the past 5-10 cycles) are stored in a sliding window dataset, forming a traceable temporal feature set. This short-term memory not only saves the target and actual execution results for each cycle but also retains the wind conditions, control parameter settings, and error change trajectory at that time. This allows for the selection of better initial control parameters in new control cycles by referencing recent adjustment effects and environmental feature similarity. This time-based memory method significantly differs from the short-sighted strategy of "directly correcting the current error" in existing technologies, giving control decisions historical relevance and environmental adaptability.

[0050] Based on short-term memory datasets and current real-time data, an error reinforcement learning decision model is constructed. This model uses the multi-source inputs of the current control cycle as the state S_t, and the actions A_t are adjusting yaw control parameters (such as target yaw angle lead, servo drive acceleration gain, and response frequency correction coefficient). The reward function R_t is the weighted sum of the reduction in error amplitude and energy consumption in the next cycle. By iteratively updating the policy function π(S_t), the model can adaptively select the optimal control parameter adjustment scheme under different wind conditions and operating states, thereby achieving dynamic evolution of the yaw response strategy. Compared with existing methods that manually set fixed parameters, this reinforcement learning process can continuously optimize control performance in complex and variable operating environments.

[0051] Within each control cycle, based on the optimal parameter adjustment scheme output by the error reinforcement learning model, the key control parameters in the current adaptive closed-loop correction path are updated, including the feedforward yaw angle compensation value, servo drive response frequency, and angular velocity limit range, and applied in real time to the next yaw execution. Simultaneously, the execution results of this cycle (including error change trends, energy consumption statistics, and blade stress state changes) are fed back to the short-term memory window for decision training in the next cycle, achieving continuous self-learning and self-optimization. In this way, this method can continuously maintain the blade inflow angle close to the design optimum during long-term operation, effectively suppressing the accumulation of inflow angle deviation even under extreme conditions such as rapid wind direction fluctuations and unstable wind speeds, thus maintaining overall aerodynamic efficiency and structural safety.

[0052] This step deeply integrates wind direction sensing results, yaw control parameters, and execution error trajectories, and combines short-term memory and error reinforcement learning to achieve adaptive updates of yaw response parameters in each control cycle. This method overcomes the shortcomings of existing technologies that rely on fixed parameters, lack historical correlation, and lack self-learning capabilities. It transforms the yaw control of wind turbines from static rule-driven to dynamic, intelligent decision-driven, resulting in significant improvements in long-term operational stability, aerodynamic efficiency, and structural lifespan.

[0053] This invention constructs an enhanced sensing model of wind direction change rate and combines it with multi-point ultra-short timescale wind direction trend and wind speed fluctuation coupling analysis to achieve early identification and risk classification warning of sudden wind direction changes. Compared with the traditional method that relies on single-point sensing and fixed threshold triggering, it significantly improves the sensitivity and accuracy of wind direction change detection, enabling yaw control to intervene in advance before significant wind direction changes, reducing the probability of continuous yaw error from the source, and ensuring that the blades maintain a near-optimal inflow angle for more operating time.

[0054] This invention generates a feedforward yaw angle adjustment amount containing a hysteresis compensation factor under the drive of a predictive yaw delay level, and combines it with a bearing transient torque response equalization algorithm to achieve precise matching between yaw action and the predicted wind direction change direction and intensity. This not only improves the timeliness and following accuracy of yaw response, but also dynamically optimizes the starting inertia release curve during the drive process, effectively reducing the energy consumption and mechanical shock of the servo drive, and extending the service life of bearings and transmission components, thereby significantly improving the reliability and economy of the whole machine under complex wind conditions.

[0055] This invention constructs an adaptive closed-loop correction path by continuously monitoring the nonlinear difference between the trend of inertial torque change and the execution error. Based on this, it integrates short-term memory and error reinforcement learning mechanisms to achieve adaptive updates of yaw response parameters in each control cycle. This enables the control strategy to be continuously optimized as the operating environment changes, effectively suppressing the cumulative amplification effect of positioning deviation. This ensures the stability of the blade inflow angle and optimal aerodynamic conditions during long-term operation of the unit, further improving the power generation efficiency and structural safety of the wind turbine throughout its entire life cycle.

[0056] This invention provides, for example Figure 2 The wind power generation anomaly data analysis system shown includes a wind direction change enhanced perception module, a high-frequency yaw disturbance prediction module, a dynamic target yaw angle generation module, a bearing transient torque response equalization module, an adaptive closed-loop correction module, and a dynamic control and scheduling module. The wind direction change enhancement perception module constructs a wind direction change rate enhancement perception model. Based on the ultra-short timescale wind direction change trend curve and wind speed fluctuation coupling index collected from multiple measurement points, it analyzes potential precursors of sudden wind direction changes and generates risk warning labels for wind turbine pointing deviation. The high-frequency yaw disturbance prediction module, based on risk warning labels, executes a high-frequency yaw disturbance prediction mechanism. It uses nonlinear time series evolution trends and short-term wind direction reversal probability curves to predict the continuous inflow angle offset window caused by yaw response lag and outputs the corresponding predictive yaw delay level. The dynamic target yaw angle generation module injects the predictive yaw delay level into the dynamic target yaw angle generation process. Combining the current wind turbine azimuth information, blade attitude parameters and inertial response boundary conditions, it calculates the feedforward yaw angle adjustment amount including the hysteresis compensation factor, so that the control command can be corrected in advance to follow the wind direction change trend. The bearing transient torque response equalization module triggers the bearing transient torque response equalization algorithm based on the feedforward yaw angle adjustment. By adjusting the starting inertia release curve of the servo drive, it makes the rotational starting angle acceleration of the yaw system match the direction and intensity of the predicted wind direction change, while optimizing drive energy consumption. The adaptive closed-loop correction module continuously monitors the nonlinear difference between the changing trend of the wind turbine's inertial torque and the actual execution error of the yaw angle. Based on this nonlinear difference, an adaptive closed-loop correction path is established to dynamically adjust the response frequency and angular velocity of the servo drive and suppress the cumulative amplification effect of the positioning deviation. The dynamic control and scheduling module integrates wind direction perception results, yaw control parameters, and execution error trajectories in the adaptive closed-loop correction path to construct a dynamic control and scheduling engine with short-term memory and error reinforcement learning capabilities. This enables the self-updating of yaw response parameters in each control cycle, thereby maintaining the long-term stability of the blade inflow angle and optimal aerodynamic conditions.

[0057] The wind power generation anomaly data analysis method provided in this embodiment of the invention is implemented through the above-mentioned wind power generation anomaly data analysis system. For details of the specific methods and processes of the wind power generation anomaly data analysis system, please refer to the embodiments of the above-mentioned wind power generation anomaly data analysis method, which will not be repeated here.

[0058] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for analyzing abnormal data in wind power generation, characterized in that, Includes the following steps: An enhanced perception model for wind direction change rate is constructed. Based on the wind direction change trend curve and wind speed fluctuation coupling index collected from multiple measurement points on an ultra-short time scale, potential precursors of sudden wind direction changes are analyzed, and risk warning labels for wind turbine pointing deviation are generated. Based on risk warning labels, a high-frequency yaw disturbance prediction mechanism is implemented. By utilizing the nonlinear time series evolution trend and the short-term wind direction reversal probability curve, the continuous inflow angle offset window caused by yaw response lag is predicted, and the corresponding predictive yaw delay level is output. The predictive yaw delay level is injected into the dynamic target yaw angle generation process. Combined with the current wind turbine azimuth information, blade attitude parameters and inertial response boundary conditions, the feedforward yaw angle adjustment including the hysteresis compensation factor is calculated so that the control command can be corrected in advance to follow the wind direction change trend. Based on the feedforward yaw angle adjustment, the bearing transient torque response equalization algorithm is triggered. By adjusting the starting inertia release curve of the servo drive, the rotational starting angle acceleration of the yaw system is matched with the direction and intensity of the predicted wind direction change, while optimizing drive energy consumption. The nonlinear difference between the changing trend of the wind turbine's inertial torque and the actual execution error of the yaw angle is continuously monitored. Based on this nonlinear difference, an adaptive closed-loop correction path is established to dynamically adjust the response frequency and angular velocity of the servo drive and suppress the cumulative amplification effect of the positioning deviation. In the adaptive closed-loop correction path, wind direction perception results, yaw control parameters and execution error trajectories are integrated to construct a dynamic control and scheduling engine with short-term memory and error reinforcement learning capabilities, so as to realize the self-updating of yaw response parameters in each control cycle.

2. The wind power generation anomaly data analysis method according to claim 1, characterized in that, The steps involved in constructing an enhanced perception model for the rate of wind direction change are as follows: Three wind direction measurement points were set up at the top of the wind turbine tower, the root of the blade, and the front of the nacelle, respectively. Wind direction data at an ultra-short time scale were collected at a sampling interval of 200 milliseconds to construct a three-dimensional wind direction data matrix with timestamp alignment. The wind direction change rate and acceleration are obtained by performing second-order central difference operation on the data of each measuring point in the three-dimensional wind direction data matrix. After smoothing, the wind direction change trend function of each measuring point is extracted, and a representative wind direction change trend curve is generated by weighted averaging according to spatial weights. Correlation and covariance analysis were performed based on the wind direction change trend curve and the instantaneous wind speed change rate sequence to calculate the wind speed fluctuation coupling index, and risk assessment was initiated when the threshold conditions were continuously met. The input feature vector is constructed based on the maximum wind direction change rate, change duration, peak coupling index, and instantaneous offset angle. The risk level is then output by the Gaussian fuzzy evaluation model, and a corresponding risk warning label is generated.

3. The wind power generation anomaly data analysis method according to claim 2, characterized in that, Risk warning labels are divided into three categories: no warning, moderate warning, and severe warning, and serve as trigger signals and weight adjustment factors for subsequent predictive yaw angle adjustment logic.

4. The wind power generation anomaly data analysis method according to claim 1, characterized in that, Performing high-frequency yaw disturbance prediction based on risk warning labels includes the following steps: Based on the wind turbine pointing offset risk label, wind direction trend curve, wind speed disturbance curve and predicted yaw angle sequence from the past 30 seconds to the current moment are collected to construct a wind direction disturbance time series set and perform wavelet transform to extract the disturbance principal component sequence. Based on the perturbation principal component sequence, Markov chain is used to statistically analyze the short-term wind direction reversal probability and generate a time series curve of wind direction reversal probability. The perturbation principal component sequence is coupled with the wind direction reversal probability curve. Combined with the nacelle azimuth, predicted wind direction and blade aerodynamic response delay, the predicted inflow angle is calculated and a continuous inflow angle offset window is constructed. Based on the time length of the offset window, the maximum offset angle, the probability of occurrence and the rate of change, the predictive yaw delay level is output.

5. The wind power generation anomaly data analysis method according to claim 4, characterized in that, Predictive yaw delay levels are divided into four levels: Level 1 indicates a low risk of yaw lag, Level 2 indicates a certain probability of continuous inflow angle deviation, Level 3 indicates a high probability of deviation caused by lag, and Level 4 indicates a high risk of delay that could lead to yaw control disorder and abnormal stress on the entire machine. These levels serve as quantitative control indicators for adjusting the accuracy of subsequent target yaw angle corrections and for actively scheduling the servo drive response frequency.

6. The wind power generation anomaly data analysis method according to claim 1, characterized in that, Injecting predictive yaw delay levels into the dynamic target yaw angle generation process includes the following steps: The predictive yaw delay level is received and used as a weighting factor in the calculation of the target yaw angle, and the nacelle azimuth, rotor rotation center azimuth and blade attitude parameters are obtained in real time. Based on the nacelle azimuth angle and the predicted wind direction angle, combined with the blade attitude and inertial response boundary conditions, the expected aerodynamic inflow angle change trend curve within the prediction time window is calculated. The compensation coefficient corresponding to the delay level is nonlinearly superimposed with the trend curve to generate a target yaw angle sequence containing hysteresis compensation factor. Over-correction is avoided by smoothing filtering and lead limit. The difference between the target yaw angle sequence and the current position of the cabin is calculated to obtain the instantaneous yaw angle adjustment. After energy consumption optimization, a control command is generated and output to the yaw drive actuator.

7. The wind power generation anomaly data analysis method according to claim 6, characterized in that, The algorithm for triggering the bearing transient torque response equalization based on the feedforward yaw angle adjustment includes the following steps: By receiving the feedforward yaw angle adjustment and combining it with the wind turbine main shaft speed, wind speed change rate and predicted wind direction change intensity and direction, a starting torque demand matrix is ​​established to determine the starting inertia release strategy. The servo drive start-up inertia release curve is calculated based on the start-up inertia release strategy. The curve includes an acceleration up section, a stable section and a gradual descent section, and is matched with the predicted wind direction change characteristics. During the execution process, the nacelle angular acceleration and bearing reaction torque are collected in real time, and the difference between them and the release curve is calculated and the drive input is adjusted in a closed loop to make the actual angular acceleration aligned with the target curve. Before completing the target yaw angle adjustment, energy consumption optimization is performed based on the fit between the release curve and the predicted wind direction change trend, and the high torque output time is controlled to reduce energy consumption and mechanical shock.

8. The method for analyzing abnormal wind power generation data according to claim 7, characterized in that, The establishment of an adaptive closed-loop correction path based on the trend of inertial moment variation and the actual execution error of yaw angle includes the following steps: During the yaw operation, the inertial moment data of the wind turbine main shaft and the real-time yaw angle of the nacelle are collected simultaneously to calculate the trend curve of the inertial moment change and the actual execution error. By mapping the trend curve of the change in inertial moment to the execution error input nonlinear mapping function, a nonlinear difference reflecting the coupling characteristics of inertial change and error is obtained; Based on the nonlinear difference, an adaptive closed-loop correction path is dynamically generated to determine the servo drive response frequency and target angular velocity of each correction node and refresh them in each yaw execution cycle. During the execution of the closed-loop path, when the difference converges to the tolerance range, the response frequency and angular velocity are reduced to enter a steady-state maintenance mode, and the drive energy consumption is reduced through energy optimization.

9. The wind power generation anomaly data analysis method according to claim 8, characterized in that, The self-updating of yaw response parameters by fusing multi-source data in the adaptive closed-loop correction path and based on short-term memory and error reinforcement learning includes the following steps: Collect and fuse wind direction change rate, wind direction change probability, inflow angle prediction curve, target yaw angle, servo drive response frequency, angular velocity setpoint and execution error curve to construct a synchronous input data matrix; The input data and yaw execution results of the most recent control cycles are stored in a sliding window dataset to form a time series feature set. An error reinforcement learning decision model is built based on the time series feature set and the current real-time data to output the optimal control parameter adjustment scheme. In each control cycle, the feedforward yaw angle compensation value, servo drive response frequency and angular velocity limit range are updated, and the execution results are fed back to the short-term memory window to achieve continuous optimization.

10. A wind power generation anomaly data analysis system, used to implement the wind power generation anomaly data analysis method according to any one of claims 1-9, characterized in that, It includes a wind direction change enhanced perception module, a high-frequency yaw disturbance prediction module, a dynamic target yaw angle generation module, a bearing transient torque response equalization module, an adaptive closed-loop correction module, and a dynamic control and scheduling module. The wind direction change enhancement perception module constructs a wind direction change rate enhancement perception model. Based on the ultra-short timescale wind direction change trend curve and wind speed fluctuation coupling index collected from multiple measurement points, it analyzes potential precursors of sudden wind direction changes and generates risk warning labels for wind turbine pointing deviation. The high-frequency yaw disturbance prediction module, based on risk warning labels, executes a high-frequency yaw disturbance prediction mechanism. It uses nonlinear time series evolution trends and short-term wind direction reversal probability curves to predict the continuous inflow angle offset window caused by yaw response lag and outputs the corresponding predictive yaw delay level. The dynamic target yaw angle generation module injects the predictive yaw delay level into the dynamic target yaw angle generation process. Combining the current wind turbine azimuth information, blade attitude parameters and inertial response boundary conditions, it calculates the feedforward yaw angle adjustment amount including the hysteresis compensation factor, so that the control command can be corrected in advance to follow the wind direction change trend. The bearing transient torque response equalization module triggers the bearing transient torque response equalization algorithm based on the feedforward yaw angle adjustment. By adjusting the starting inertia release curve of the servo drive, it makes the rotational starting angle acceleration of the yaw system match the direction and intensity of the predicted wind direction change, while optimizing drive energy consumption. The adaptive closed-loop correction module continuously monitors the nonlinear difference between the changing trend of the wind turbine's inertial torque and the actual execution error of the yaw angle. Based on this nonlinear difference, an adaptive closed-loop correction path is established to dynamically adjust the response frequency and angular velocity of the servo drive and suppress the cumulative amplification effect of the positioning deviation. The dynamic control and scheduling module integrates wind direction perception results, yaw control parameters, and execution error trajectories in the adaptive closed-loop correction path to construct a dynamic control and scheduling engine with short-term memory and error reinforcement learning capabilities. This enables the self-updating of yaw response parameters in each control cycle, thereby maintaining the long-term stability of the blade inflow angle and optimal aerodynamic conditions.

Images