Method and system for identifying and controlling asymmetric operation of a variable pitch system

Abstract

This application discloses a method and system for identifying and controlling asymmetric operation of a pitch system. It quantifies the stable angle deviation of each blade in real time and aggregates them into average angle deviation and standard deviation of angle deviation. Subsequently, a residual neural network is used to predict the power coefficient residual and thrust coefficient residual based on the aforementioned asymmetric indices. These residuals are used to dynamically correct the reference aerodynamic model within the controller, and calibration is performed by introducing an adaptive correction factor and mapping to the physical boundary, thereby obtaining corrected power and thrust coefficients that accurately reflect the fault state. Finally, based on this corrected model, adaptive control law is calculated to generate adaptive torque and pitch commands to achieve optimized operation under fault conditions.

Description

Technical Field

[0001] This application relates to the field of wind power generation technology, and more specifically, to a method and system for identifying and controlling asymmetric operation of a pitch system. Background Technology

[0002] The pitch system of a large wind turbine is a key component for regulating aerodynamic loads and output power. Ideally, the pitch angles of the three blades should be consistent to achieve symmetrical operation. However, in actual operation, factors such as component wear, sensor drift, or actuator malfunction can cause the actual angles of one or more blades to deviate from the commanded values, resulting in asymmetrical operation. This asymmetrical operation leads to an imbalance in the aerodynamic loads on the wind turbine, increases fatigue damage to critical components, and reduces the unit's power generation efficiency.

[0003] Existing technical solutions typically focus on monitoring and diagnosing asymmetric operational faults in pitch control systems. When such faults are identified, the control system often adopts pre-defined, conservative response strategies, such as common power-reduction operation or shutdown. This approach fails to fully utilize diagnostic information to achieve optimized operation under fault conditions. This is because the global control model relied upon by the wind turbine's main controller, particularly the aerodynamic models used to calculate aerodynamic torque and thrust, such as power coefficient lookup tables, is based on the assumption of symmetrical operation of all blades. When an asymmetric fault occurs, this model cannot handle inputs with inconsistent actual blade angles, leading to a deviation between the control model and the physical reality of the wind turbine. Consequently, the controller cannot accurately estimate the actual aerodynamic performance under fault conditions, making precise control adjustments difficult.

[0004] Therefore, an optimized scheme for identifying and controlling the asymmetric operation of pitch systems is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method and system for identifying and controlling asymmetric operation of a pitch system.

[0006] According to one aspect of this application, a method for identifying and controlling asymmetric operation of a pitch system is provided, comprising: Acquire the pitch command signals of the first to third blades, the actual blade angles of the first to third blades, and the fault types of the first to third blades; Real-time asymmetric state quantization is performed on the pitch command signals of the first to third blades, the actual blade angles of the first to third blades, and the fault types of the first to third blades to obtain the stable angle deviation of the first to third blades. Asymmetric physical metrics were aggregated to obtain the average angle deviation and the standard deviation of the angle deviation for the first to third blades. The aerodynamic coefficient correction prediction based on residual neural network is performed on the mean angle deviation, the standard deviation of the angle deviation, the collective pitch command and the tip speed ratio to obtain the power coefficient residual and the thrust coefficient residual; Based on the collective pitch command and tip speed ratio, the power coefficient residual and thrust coefficient residual are dynamically corrected by the global aerodynamic model to obtain the corrected power coefficient and thrust coefficient. Adaptive control commands based on a modified model are generated using the modified power coefficient, modified thrust coefficient, rotor speed, and air density to obtain adaptive torque commands and adaptive pitch commands. According to another aspect of this application, a control system for identifying and controlling asymmetric operation of a pitch system is provided, comprising: The command signal and fault type acquisition module is used to acquire the first to third blade pitch control command signals, the first to third actual blade angles, and the first to third blade fault types. The real-time asymmetric state quantization module is used to perform real-time asymmetric state quantization on the first to third blade pitch command signals, the first to third actual blade angles and the first to third blade fault types to obtain the first to third blade stable angle deviations. The asymmetric physical metric aggregation module is used to aggregate asymmetric physical metric indicators for the stability angle deviations of the first to third blades to obtain the average angle deviation and the standard deviation of the angle deviation. The aerodynamic coefficient correction prediction module is used to perform aerodynamic coefficient correction prediction based on residual neural network on the mean angle deviation, angle deviation standard deviation, collective pitch command and tip speed ratio to obtain the power coefficient residual and thrust coefficient residual. The global aerodynamic model dynamic correction module is used to dynamically correct the power coefficient residual and thrust coefficient residual based on the collective pitch command and tip speed ratio to obtain the corrected power coefficient and thrust coefficient. The instruction generation module is used to generate adaptive control instructions based on the modified model, based on the modified power coefficient, modified thrust coefficient, rotor speed and air density, to obtain adaptive torque command and adaptive pitch command.

[0007] Compared to existing technologies, to address the mismatch between the control model and physical reality under asymmetric fault conditions in wind turbines, this invention quantifies the stable angle deviation of each blade in real time and aggregates it into average angle deviation and standard deviation of angle deviation. Subsequently, using a residual neural network, the power coefficient residual and thrust coefficient residual are predicted based on the aforementioned asymmetric indices. These residuals are used to dynamically correct the baseline aerodynamic model within the controller, and calibration is performed by introducing an adaptive correction factor and mapping it to the physical boundary, thereby obtaining corrected power and thrust coefficients that accurately reflect the fault state. Finally, based on this corrected model, adaptive control law is calculated to generate adaptive torque and pitch commands for optimized operation under fault conditions. Attached Figure Description

[0008] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 This is a flowchart of a method for identifying and controlling asymmetric operation of a pitch system according to an embodiment of this application; Figure 2 This is a data flow diagram of the identification and control method for asymmetric operation of a pitch system according to an embodiment of this application; Figure 3 This is a flowchart illustrating the real-time asymmetric state quantification of the first to third blade pitch command signals, the first to third actual blade angles, and the first to third blade fault types to obtain the stable angle deviation of the first to third blades, according to the identification and control method for asymmetric operation of the pitch system in an embodiment of this application. Figure 4 The flowchart illustrates the process of dynamically correcting the power coefficient residual and thrust coefficient residual using a global aerodynamic model based on collective pitch command and tip speed ratio in the identification and control method for asymmetric operation of a pitch system according to embodiments of this application, to obtain the corrected power coefficient and thrust coefficient. Figure 5 The flowchart illustrates the process of generating adaptive control commands based on a modified model using the modified power coefficient, modified thrust coefficient, rotor speed, and air density to obtain adaptive torque commands and adaptive pitch commands for the identification and control method of asymmetric operation of a pitch system according to embodiments of this application. Figure 6 This is a block diagram of a control system for identifying and controlling asymmetric operation of a pitch system according to an embodiment of this application. Detailed Implementation

[0010] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0011] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0012] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0013] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0014] Currently, when wind turbine generators face asymmetric operation of their pitch systems, their control systems rely on a global aerodynamic model with symmetric assumptions, leading to a disconnect between the model and physical reality. This makes it impossible for the controller to accurately estimate aerodynamic performance under fault conditions, thus forcing it to adopt conservative strategies such as power reduction or shutdown, making precise adaptive control difficult. Therefore, this application proposes a method for identifying and controlling asymmetric operation of pitch systems. First, it acquires the pitch command signals, actual blade angles, and fault types of the first to third blades, and performs real-time asymmetric state quantification to obtain the stable angle deviation of each blade. Then, it aggregates these stable angle deviations using asymmetric physical metrics to calculate the average angle deviation and the standard deviation of the angle deviation, which serve as physical inputs characterizing the severity of the asymmetry. Next, it inputs the average angle deviation, the standard deviation of the angle deviation, the collective pitch command, and the tip speed ratio into a model based on a residual neural network to perform aerodynamic coefficient correction prediction, obtaining the power coefficient residual and the thrust coefficient residual. The system queries the baseline aerodynamic model based on the collective pitch command and tip speed ratio, and uses the aforementioned predicted residuals to perform global dynamic correction on the baseline model, outputting the corrected power coefficient and corrected thrust coefficient. Finally, based on these corrected aerodynamic coefficients, rotor speed, and air density that are closer to physical reality, the controller generates adaptive control commands based on the corrected model, and calculates adaptive torque commands and adaptive pitch commands, enabling the wind turbine to operate safely and efficiently even under asymmetric faults.

[0015] Figure 1 This is a flowchart of a method for identifying and controlling asymmetric operation of a pitch system according to an embodiment of this application. Figure 2 This is a data flow diagram illustrating the identification and control method for asymmetric operation of a pitch system according to an embodiment of this application. Figure 1 and Figure 2As shown, the method for identifying and controlling asymmetric operation of a pitch system according to an embodiment of this application includes the following steps: S100, acquiring pitch command signals of the first to third blades, actual blade angles of the first to third blades, and fault types of the first to third blades; S200, performing real-time asymmetric state quantization on the pitch command signals of the first to third blades, actual blade angles of the first to third blades, and fault types of the first to third blades to obtain the stable angle deviation of the first to third blades; S300, aggregating asymmetric physical measurement indices on the stable angle deviation of the first to third blades to obtain the average angle deviation and the standard deviation of the angle deviation. S400: Perform aerodynamic coefficient correction prediction based on residual neural network on the average angle deviation, angle deviation standard deviation, collective pitch command, and tip speed ratio to obtain the power coefficient residual and thrust coefficient residual; S500: Based on the collective pitch command and tip speed ratio, perform global aerodynamic model dynamic correction on the power coefficient residual and thrust coefficient residual to obtain the corrected power coefficient and corrected thrust coefficient; S600: Based on the corrected power coefficient, corrected thrust coefficient, rotor speed, and air density, perform adaptive control command generation based on the correction model to obtain the adaptive torque command and adaptive pitch command.

[0016] Specifically, in step S100, the pitch control command signals of the first to third blades, the actual blade angles of the first to third blades, and the fault types of the first to third blades are acquired. It is understood that subsequent asymmetric state quantization, aerodynamic model correction, and adaptive control calculations all highly depend on an accurate description of the current operating state of the pitch control system. Specifically, the pitch control command signal serves as the benchmark for subsequent deviation calculations, the actual blade angles provide feedback from physical execution, and the fault types are prerequisites for initiating the correction logic. Therefore, in the technical solution of this application, the pitch control command signals of the first to third blades, the actual blade angles of the first to third blades, and the fault types of the first to third blades are acquired to provide the necessary and synchronous raw data input for the subsequent real-time asymmetric state quantization steps. This ensures that the entire identification and control process is executed based on the current accurate operating conditions and diagnostic conclusions of the wind turbine.

[0017] More specifically, in a concrete example of this application, the acquisition process is executed within each control cycle of the wind turbine main controller. First, the target angle setpoints issued by the main controller to the first, second, and third blade drive systems are read via a control network bus, such as a CAN (Controller Area Network) bus, to serve as the first to third blade pitch command signals. Next, the real-time angle values, measured and processed by the angle encoder installed at the blade root, are extracted from the data returned by each blade drive system to serve as the first to third actual blade angles. Simultaneously, the status register of the fault diagnosis unit or its issued diagnostic message is accessed to obtain a predefined code characterizing the operating status of each blade, such as an integer corresponding to a specific fault mode or health state, thus serving as the fault type for the first to third blades. All of these data acquisitions are completed within the same control cycle to ensure temporal consistency of the data used in subsequent processing.

[0018] Specifically, in step S200, real-time asymmetric state quantization is performed on the first to third blade pitch command signals, the first to third actual blade angles, and the first to third blade fault types to obtain the stable angle deviations of the first to third blades. It is understood that the acquired raw blade angle signals and command signals will have instantaneous tracking errors during dynamic pitch control, and the signals themselves may contain measurement noise. Furthermore, the acquired fault types are only qualitative judgments. If these raw data are used directly for subsequent physical measurement aggregation without processing, the aggregated indicators will be distorted, leading to deviations and fluctuations in downstream aerodynamic model corrections. Therefore, in the technical solution of this application, real-time asymmetric state quantization is further performed on the first to third blade pitch command signals, the first to third actual blade angles, and the first to third blade fault types to obtain the stable angle deviations of the first to third blades. This allows for the extraction of the persistent angle deviation value that truly represents the physical fault from the raw data containing noise and transient responses by combining dynamic condition judgment, signal filtering, and fault mode matching. This provides an accurate and robust set of quantitative inputs for the subsequent aggregation of asymmetric physical metrics, ensuring that subsequent aerodynamic model corrections are based on the actual and continuous asymmetric state of the wind turbine.

[0019] Figure 3 This is a flowchart illustrating the real-time asymmetric state quantification of the first to third blade pitch command signals, the first to third actual blade angles, and the first to third blade fault types to obtain the stable angle deviation of the first to third blades, according to the identification and control method for asymmetric operation of the pitch system in an embodiment of this application. Figure 3As shown, step S200 includes: S210, calculating the instantaneous blade angle deviation between the first to third blade pitch command signals and the first to third actual blade angles to obtain the original blade angle deviations; S220, based on the first to third blade pitch command signals, performing deviation signal filtering based on pitch dynamic conditions on the original blade angle deviations to obtain the filtered blade angle deviations; S230, based on the first to third blade fault types, performing fault mode matching on the filtered blade angle deviations to obtain the stable blade angle deviations.

[0020] In step S210, the instantaneous blade angle deviation between the first to third blade pitch command signals and the first to third actual blade angles is calculated to obtain the original blade angle deviations. It is understood that since the acquired first to third blade pitch command signals and the first to third actual blade angles are two independent time-series data, they respectively represent the desired target of the control system and the physical reality of the actuator. To initiate the quantification process of the asymmetric state, a metric characterizing the difference between the two must first be established. Therefore, in the technical solution of this application, the instantaneous blade angle deviation between the first to third blade pitch command signals and the first to third actual blade angles is further calculated to obtain the original blade angle deviations, thereby generating an unprocessed, instantaneous tracking error signal for each blade. This provides the most original deviation data stream containing all dynamic and static information for subsequent deviation signal filtering and fault mode matching steps.

[0021] More specifically, in a particular example of this application, the calculation is performed in parallel for all three blades within each control cycle. In a discrete time step... For the first blade, obtain its pitch control command signal. and its corresponding actual blade angle The original angle deviation of the first blade is obtained by subtracting its corresponding pitch command signal from the actual blade angle. For example, if at time k, the first blade pitch command signal is 10.0 degrees, and the obtained first actual blade angle is 10.2 degrees, then the calculated... It is +0.2 degrees. Similarly, at the same time step k, the signal for the second blade... and And the signal of the third blade and Perform the same subtraction operation to obtain the original angle deviation of the second blade. and the original angle deviation of the third blade These three initial angular deviation values ​​are then passed to the next processing stage.

[0022] In step S220, based on the pitch command signals of the first to third blades, the original angle deviations of the first to third blades are filtered using a deviation signal based on the dynamic conditions of the pitch control to obtain the filtered angle deviations of the first to third blades. It is understood that the original angle deviation signal obtained in the previous step contains both sensor measurement noise and instantaneous tracking errors generated by the pitch control system in response to dynamic commands. If these two components are not removed, especially if normal tracking errors during dynamic pitch control are misjudged as persistent angle deviations, subsequent quantization results will be distorted. Therefore, in the technical solution of this application, the original angle deviations of the first to third blades are further filtered using a deviation signal based on the dynamic conditions of the pitch command signals to obtain the filtered angle deviations of the first to third blades. This utilizes the dynamic characteristics of the pitch command signal as a judgment criterion, performing low-pass filtering on the original deviation only during command stabilization periods, while maintaining the filtered value during dynamic command changes, thereby effectively separating the true steady-state deviation and suppressing noise interference. This provides a smooth and accurate estimate of the angle deviation for subsequent failure mode matching steps, which has eliminated the influence of transient response.

[0023] More specifically, in a specific example of this application, based on the pitch command signals of the first to third blades, the original angle deviations of the first to third blades are filtered by a deviation signal based on the dynamic conditions of pitch to obtain the filtered angle deviations of the first to third blades. This includes: calculating the command change rate of the first to third blades based on the pitch command signals of the first to third blades and the pitch command signals of the first to third blades at the previous moment; determining the command stability state of the command change rate of the first to third blades based on a change rate threshold to obtain a command stability flag of the first to third blades; and performing conditional low-pass filtering on the original angle deviations of the first to third blades based on the command stability flags of the first to third blades to obtain the filtered angle deviations of the first to third blades.

[0024] Accordingly, based on the pitch command signals of the first to third blades and the previous pitch command signals of the first to third blades, the command change rate of the first to third blades is calculated. It is understandable that the subsequent command stability determination step requires a quantitative input to characterize the dynamic characteristics of the pitch command signal. The original command signal only represents the target position and cannot directly reflect whether it is in a rapid change process; therefore, a measure of its dynamics needs to be introduced. Therefore, in the technical solution of this application, the command change rate of the first to third blades is further calculated based on the pitch command signals of the first to third blades and the previous pitch command signals of the first to third blades, thereby obtaining a physical quantity that can instantaneously reflect the drastic degree of command change, namely, the command velocity. This provides a clear and threshold-comparable basis for the subsequent command stability determination.

[0025] More specifically, in a concrete example of this application, the calculation is performed for the first to third blades at each discrete time step k of the control system. Taking the first blade as an example, the first blade pitch command signal at the current time k is obtained, and the first blade pitch command signal at the previous time k-1 is retrieved from the storage unit. By subtracting the command signal value at the previous time from the current command signal value, and then dividing the difference by the length of a control cycle, for example, 0.01 seconds, the first blade command change rate is obtained. For example, if the command at the current time k is 10.15 degrees, the command at time k-1 is 10.13 degrees, and the control cycle is 0.01 seconds, then the calculated first blade command change rate is (10.15-10.13) / 0.01, i.e., 2 degrees / second. The same processing is applied to the second and third blades to obtain the second blade command change rate and the third blade command change rate, respectively.

[0026] Accordingly, the command stability status of the first to third blades is determined based on the rate of change threshold to obtain the command stability flags for the first to third blades. It is understandable that since the blade command rate of change calculated in the previous step is only an instantaneous value, the subsequent conditional low-pass filtering step requires a clear logic switch signal to determine whether to perform a filter update or maintain the value from the previous moment. Without this determination, during dynamic pitch control, the instantaneous tracking error caused by the system's physical delay will be incorrectly included in the filtering calculation, thus contaminating the estimation of the steady-state deviation. Therefore, in the technical solution of this application, the command stability status of the first to third blades is further determined based on the rate of change threshold to obtain the command stability flags for the first to third blades. This allows the blade command state to be divided into dynamic or stable modes by comparing the absolute value of the command rate of change with a preset engineering threshold. This provides a precise Boolean control signal consistent with the physical process for the subsequent conditional low-pass filtering step, ensuring that the filtering operation is only activated when the command is indeed in a stationary or quasi-stationary state.

[0027] More specifically, in a concrete example of this application, the judgment process is executed immediately after the command change rate is obtained. The system internally sets a change rate threshold, for example, 0.5 degrees / second, which is used to define the quasi-static range of the command. Taking the first blade as an example, its command change rate is obtained, assuming a calculated value of 2 degrees / second at time k. The absolute value of this change rate, 2, is compared with the threshold 0.5. Since 2 is greater than 0.5, the first blade is determined to be in a dynamic pitching state, and therefore its corresponding first blade command stability flag is set to False. At another time, assuming the command change rate of the second blade is -0.1 degrees / second, its absolute value 0.1 is less than the threshold 0.5, then the second blade is determined to be in a command stable state, and its second blade command stability flag is set to True. The third blade also performs the same comparison and judgment logic to obtain the third blade command stability flag. These three Boolean flags are then passed to the conditional low-pass filter stage.

[0028] Accordingly, based on the command stabilization flags of the first to third blades, conditional low-pass filtering is performed on the original angle deviations of the first to third blades to obtain the filtered angle deviations. It is understood that the original angle deviation signals of the first to third blades contain random noise and instantaneous errors caused by dynamic tracking, while the command stabilization flags obtained in the previous step only provide the logical timing for filtering. If the actual filtering operation is not performed, the output deviation value will fluctuate, and the instantaneous deviation during dynamic pitch control will be incorrectly propagated downstream, leading to misjudgment of the steady-state deviation. Therefore, in the technical solution of this application, conditional low-pass filtering is further performed on the original angle deviations of the first to third blades based on the command stabilization flags to obtain the filtered angle deviations. This stabilization flag is used as the enable signal for the filter, and the original deviations are weighted and averaged only when the command is stable to smooth noise, while the filter output value is locked during command dynamics. In this way, a set of smooth signals that only reflect the deviation characteristics under steady-state or quasi-steady-state conditions can be generated, effectively eliminating interference from dynamic tracking errors and high-frequency noise.

[0029] More specifically, in a concrete example of this application, the conditional filtering process processes the three blades in parallel at each discrete time step k. Taking the first blade as an example, at time k, the corresponding first blade command stability flag is obtained. If the flag is True, it indicates that the command is in a stable state, and a first-order low-pass filter is activated. This filter obtains the original angle deviation of the first blade at time k, and combines it with the filtered angle deviation value of the first blade stored at the previous time k-1, and performs an exponential moving average calculation according to a preset filtering coefficient (e.g., 0.01) to obtain the filtered angle deviation of the first blade at time k. For example, if the filtered value at time k-1 is +0.5 degrees and the original deviation at time k is +0.7 degrees, then the filtered value at time k is updated to... That is, +0.502 degrees. Conversely, if the stability flag of the first blade command acquired at time k is False, indicating that the command is in dynamic change, the filtering operation is skipped. In this case, the filtered angle deviation of the first blade at time k is directly set to the filtered value at time k-1, i.e., +0.5 degrees, thereby freezing and isolating the instantaneous deviation in the dynamic process (for example, the original deviation at time k may reach -2.0 degrees). The second and third blades also perform the same conditional filtering logic, and finally output the filtered angle deviations of the first to third blades.

[0030] In step S230, based on the fault types of the first to third blades, fault mode matching is performed on the filtered angle deviations of the first to third blades to obtain the stable angle deviations of the first to third blades. It is understandable that, since the filtered angle deviations of the first to third blades output in the previous step are only smoothed values, they cannot distinguish whether the deviation is a true physical deviation caused by sensor zero-point drift or mechanical jamming, or a transient imbalance caused by minor residual noise or unrelated fault types in a healthy state. If all non-zero filtered values ​​are considered valid stable deviations, unnecessary corrections may be made in a healthy state, or incorrect compensation may be made for other fault types. Therefore, in the technical solution of this application, fault mode matching is further performed on the filtered angle deviations of the first to third blades based on the fault types of the first to third blades to obtain the stable angle deviations of the first to third blades. This utilizes the qualitative fault types provided by the upstream diagnostic module as logic gating to arbitrate and confirm the filtered quantitative deviation values. Only when the diagnosed fault mode matches the physical mechanism that generates the stable angle deviation is the deviation value confirmed as a valid output. This ensures that the final output of the first to third blade stability angle deviation is not only numerically smooth, but also physically confirmed by diagnostic information, thus providing a high-confidence input for downstream aerodynamic model correction and avoiding invalid or erroneous correction operations.

[0031] More specifically, in a concrete example of this application, the matching process is executed after obtaining the filtered angle deviation and fault type. Taking the first blade as an example, at time k, the filtered angle deviation of the first blade is obtained, assumed to be +0.502 degrees. Simultaneously, the fault type of the first blade is obtained, assumed to be encoded as an integer 1, and the system predefines integer 1 to represent a fixed angle deviation fault. At this point, the fault mode matching logic determines a match, therefore the +0.502 degrees is taken as a valid value, and the output is the stable angle deviation of the first blade. In another scenario, assuming the fault type of the second blade is 0, representing a healthy state, and its filtered angle deviation is calculated to be -0.015 degrees due to residual sensor noise, the fault mode matching logic determines a mismatch, because a healthy state should not produce a stable deviation requiring correction. Therefore, the system forcibly sets the -0.015 degrees to zero, and the output stable angle deviation of the second blade is 0.0 degrees. Similarly, if the fault type of the third blade is 2, which represents a slow response fault, its filter value may also be non-zero. However, since this fault type does not belong to the angle fixed deviation, the logical judgment is also a mismatch, and the output of the stable angle deviation of the third blade is also 0.0 degrees.

[0032] Specifically, in step S300, the stable angle deviations of the first to third blades are aggregated using asymmetric physical metrics to obtain the average angle deviation and the standard deviation of the angle deviation. It is understandable that the stable angle deviations of the first to third blades output in the previous step are a set of discrete values ​​for individual blades. Directly using these three independent values ​​as input for subsequent aerodynamic coefficient correction prediction would increase the dimensionality of the input features and hinder the model from extracting the overall features of the asymmetry. Therefore, in the technical solution of this application, the stable angle deviations of the first to third blades are further aggregated using asymmetric physical metrics to obtain the average angle deviation and the standard deviation of the angle deviation. This allows for statistical calculations to compress and convert the three independent deviation values ​​into two aggregated metrics with clear physical meanings, representing the overall offset and the severity of the discreteness of the asymmetric deviation, respectively. This provides a low-dimensional, high-information-content feature vector for the subsequent aerodynamic coefficient correction prediction step based on residual neural networks, helping to improve the training efficiency and generalization ability of the prediction model.

[0033] More specifically, in a specific example of this application, the asymmetric physical metrics of the stability angle deviations of the first to third blades are aggregated to obtain the average angle deviation and the standard deviation of the angle deviation, including: calculating the mean of the stability angle deviations of the first to third blades to obtain the average angle deviation; and calculating the standard deviation of the stability angle deviations of the first to third blades to obtain the standard deviation of the angle deviation.

[0034] In other words, more specifically, this aggregation process is executed immediately after the stable angle deviations of the first to third blades are obtained. Assume that at time k, the stable angle deviation of the first blade is +0.502 degrees, the stable angle deviation of the second blade is 0.0 degrees, and the stable angle deviation of the third blade is 0.0 degrees. First, the mean of the stable angle deviations of the first to third blades is calculated to obtain the average angle deviation. These three values ​​are added together (0.502 + 0.0 + 0.0), resulting in a sum of 0.502 degrees. This sum is then divided by 3 to obtain the average angle deviation, which is approximately +0.167 degrees. Next, the standard deviation of the stable angle deviations of the first to third blades is calculated to obtain the standard deviation of the angle deviation. Using the average angle deviation of +0.167 degrees obtained in the previous step, the difference between the stable angle deviation of each blade and this mean is calculated, namely (0.502 - 0.167), (0.0 - 0.167), and (0.0 - 0.167). Then, square each of the three differences, add them together, and divide by 3 to get the variance. Finally, take the square root of the differences to get the standard deviation of the angle deviation. The final output of the mean angle deviation and the standard deviation of the angle deviation will be passed to subsequent processing stages.

[0035] Specifically, in step S400, aerodynamic coefficient correction prediction based on a residual neural network is performed on the average angle deviation, angle deviation standard deviation, collective pitch command, and tip speed ratio to obtain the power coefficient residual and thrust coefficient residual. It is understood that the average angle deviation and angle deviation standard deviation obtained in the previous step only quantify the severity of the asymmetric fault, but the specific impact of this asymmetry on the wind turbine's aerodynamic performance is highly nonlinear and coupled with the current operating conditions of the wind turbine. If the specific magnitude of this impact cannot be known, the baseline aerodynamic model based on the symmetry assumption cannot be effectively corrected. Therefore, in the technical solution of this application, aerodynamic coefficient correction prediction based on a residual neural network is further performed on the average angle deviation, angle deviation standard deviation, collective pitch command, and tip speed ratio to obtain the power coefficient residual and thrust coefficient residual. This utilizes the powerful nonlinear fitting capability of the neural network to calculate in real time the deviation of the power coefficient and thrust coefficient from the symmetric baseline model caused by the combined effects of the current asymmetric state and operating conditions. This provides a quantitative, condition-dependent basis for subsequent dynamic correction steps of the global aerodynamic model, namely the power coefficient residual and the thrust coefficient residual.

[0036] More specifically, in a specific example of this application, the aerodynamic coefficient correction prediction based on a residual neural network is performed on the mean angle deviation, the standard deviation of the angle deviation, the collective pitch command, and the tip speed ratio to obtain the power coefficient residual and the thrust coefficient residual. This includes: arranging the mean angle deviation, the standard deviation of the angle deviation, the collective pitch command, and the tip speed ratio into an input vector; and inputting the input vector into a trained lightweight feedforward neural network to obtain the power coefficient residual and the thrust coefficient residual.

[0037] More specifically, this prediction process is executed within each control cycle. First, the process involves arranging the average angle deviation, standard deviation of the angle deviation, collective pitch command, and tip speed ratio into an input vector. Specifically, the average angle deviation is obtained from the previous aggregation stage, assumed to be +0.167 degrees; the standard deviation of the angle deviation is obtained, assumed to be 0.235 degrees; simultaneously, the current collective pitch command is obtained from the main controller, assumed to be 12.0 degrees; and the current tip speed ratio is obtained from the state estimation unit, assumed to be 7.5. These four values ​​are arranged in a predetermined order, for example, [0.167, 0.235, 12.0, 7.5], to form a four-dimensional input vector. Subsequently, the input vector is fed into a trained lightweight feedforward neural network to obtain the power coefficient residual and thrust coefficient residual. This lightweight feedforward neural network is pre-trained using offline simulation or wind farm test data, and its structure is designed for real-time operation on the wind turbine embedded controller. The previously generated four-dimensional input vector [0.167, 0.235, 12.0, 7.5] is fed into the input layer of the neural network. After multi-layer weight matrix operations and non-linear activation function processing within the network, the neural network generates two scalar values ​​in its output layer. The first output value is defined as the power coefficient residual, for example, -0.008; the second output value is defined as the thrust coefficient residual, for example, +0.005. These two residual values ​​are then output for subsequent dynamic correction of the global aerodynamic model.

[0038] Specifically, in step S500, based on the collective pitch command and tip speed ratio, the power coefficient residual and thrust coefficient residual are dynamically corrected using a global aerodynamic model to obtain the corrected power coefficient and thrust coefficient. It is understandable that in traditional aerodynamic model correction mechanisms, a common practice is to directly superimpose the power coefficient correction and thrust coefficient correction values ​​predicted by the data-driven model with the baseline values ​​obtained by looking up tables based on the symmetry assumption. While this approach is intuitive, it hides a profound technical flaw, stemming from its neglect of the complex physical characteristics and coupling relationships inherent in the aerodynamic system of wind turbine generators. First, this mechanism completely fails to consider the physical boundary constraints of the aerodynamic coefficients. The power coefficient and thrust coefficient are not mathematical variables that can take infinitely many values ​​in the real number domain; they are strictly limited by the law of conservation of energy and aerodynamic principles, possessing definite physical upper and lower bounds. Simple linear addition is mathematically open-ended. When the baseline value is close to the physical limit, or the correction predicted by the neural network is large, the result is very likely to exceed the physically reasonable range, such as producing a power coefficient greater than the Bates limit or a negative thrust coefficient, causing the correction result to lose its physical meaning and potentially threatening the stability of the downstream control system. Secondly, this mechanism fails to reflect the nonlinear sensitivity of aerodynamic efficiency to asymmetric responses. The sensitivity of a wind turbine's aerodynamic characteristics to blade angle deviations varies significantly across different operating conditions. Especially in regions sensitive to aerodynamic performance changes, such as when the tip speed ratio is close to the optimal value or the pitch angle is close to the stall boundary, small asymmetric faults can be nonlinearly amplified, drastically affecting the aerodynamic coefficient. Simple additive correction is a continuation of linear thinking; it cannot capture and express this nonlinear interaction, meaning it cannot intelligently adjust the weight of the correction based on the specific operating point of the wind turbine, leading to potentially serious deviations in corrections under critical operating conditions. Furthermore, this mechanism severs the intrinsic physical relationship between the power coefficient and thrust coefficient corrections. In real aerodynamic processes, the changes in power coefficient and thrust coefficient caused by asymmetric operation are not two isolated events; they are manifestations of the same physical phenomenon in different dimensions, with a strong physical coupling between them. For example, a severe stall of a blade will not only cause a sharp drop in power, but its thrust characteristics will also inevitably change accordingly. Simple independent corrections cannot reflect this coupling relationship, and may lead to physical inconsistencies between the corrected power and thrust, reducing the inherent consistency and credibility of the entire model correction.

[0039] Therefore, in the technical solution of this application, a global aerodynamic model is further dynamically corrected based on the collective pitch command and tip speed ratio to obtain the corrected power coefficient and thrust coefficient, thereby introducing a nonlinear aerodynamic coefficient correction method with adaptive physical boundary constraints. This method utilizes the current operating conditions represented by the collective pitch command and tip speed ratio, introduces an adaptive scaling factor, and combines interactive coupling calibration to nonlinearly weight the residuals. Finally, the correction results are fused with the baseline model through a boundary mapping function that ensures physical consistency. This enables more accurate and robust dynamic correction of the aerodynamic coefficients, ensuring that the final output corrected power coefficient and thrust coefficient are numerically more accurate and maintain physical rationality, robustness, and inherent consistency, providing high-quality model input for the downstream adaptive control command generation step.

[0040] Figure 4 This document describes a flowchart illustrating the process of dynamically correcting the power coefficient residual and thrust coefficient residual using a global aerodynamic model based on collective pitch command and tip speed ratio, according to an embodiment of this application, to obtain the corrected power coefficient and thrust coefficient. Figure 4 As shown, step S500 further includes: S510, determining the reference power coefficient, reference thrust coefficient, power correction scaling factor, and thrust correction scaling factor based on the collective pitch command, tip speed ratio, angle deviation standard deviation, power coefficient residual, and thrust coefficient residual; S520, performing weighted residual correction and interactive coupling calibration on the power coefficient residual and thrust coefficient residual based on the reference power coefficient, reference thrust coefficient, power correction scaling factor, and thrust correction scaling factor to obtain the pre-corrected power coefficient and pre-corrected thrust coefficient; S530, performing physical boundary mapping on the pre-corrected power coefficient and pre-corrected thrust coefficient to obtain the corrected power coefficient and corrected thrust coefficient.

[0041] In step S510, the reference power coefficient, reference thrust coefficient, power correction scaling factor, and thrust correction scaling factor are determined based on the collective pitch command, tip speed ratio, standard deviation of angle deviation, power coefficient residual, and thrust coefficient residual. It is understood that subsequent weighted residual correction steps require a reliable correction reference and a mechanism to determine the reasonableness of adopting the residual correction amount predicted by the neural network under the current specific operating conditions. If the nonlinear sensitivity of the operating conditions is not considered, especially in regions sensitive to aerodynamic performance changes such as when the tip speed ratio is close to the optimal value or the pitch angle is close to the stall boundary, applying the correction amount may cause model oscillations due to over-correction. Therefore, in the technical solution of this application, the reference power coefficient, reference thrust coefficient, power correction scaling factor, and thrust correction scaling factor are further determined based on the collective pitch command, tip speed ratio, and standard deviation of angle deviation to obtain the reference aerodynamic coefficient values ​​under healthy symmetric operating conditions and generate adaptive scaling factors with operating condition awareness. This provides a reliable starting point and an intelligent weight for subsequent correction calculations, ensuring that the correction magnitude matches the nonlinear sensitivity of the current operating condition.

[0042] More specifically, in a particular example of this application, the determination process comprises two parallel computational branches. In the first branch, the current collective pitch command is obtained. Speed ​​ratio of leaf tip And using these two values ​​as input, it queries a pre-defined standard two-dimensional lookup table function. and For example, when the tip speed ratio λ is 8.0 and the collective pitch command is given... When the value is 5.0 degrees, the reference power coefficient can be obtained from the table. The reference thrust coefficient is 0.47. It is 0.82. In the second branch, the tip speed ratio λ and the collective pitch command are... and the standard deviation of the angle deviation characterizing the severity of asymmetry These three values ​​are input together into a pre-trained, lightweight subnetwork or a high-dimensional lookup table. .Should It possesses operational condition awareness capabilities. For example, when the wind turbine is operating at the edge of the stall zone, such as... It is 6.0 and It is 18.0 degrees. It will output a small scaling factor, such as the power coefficient correction scaling factor. A scaling factor of 0.3 and the thrust coefficient. A correction value of 0.4 is applied in a more conservative manner to avoid model oscillations. Conversely, if the model is run at its optimal aerodynamic stability... Near the point, such as It is 8.0 and If the value is 1.0, the possible output is... For 0.9 and The value is 0.85. Finally, the four calculated values ​​are... , , and It is output to the weighted residual correction stage.

[0043] In step S520, based on the reference power coefficient, reference thrust coefficient, power correction scaling factor, and thrust correction scaling factor, weighted residual correction and interactive coupling calibration are performed on the power coefficient residual and thrust coefficient residual to obtain the pre-corrected power coefficient and pre-corrected thrust coefficient. It is understood that since the correction scaling factor determined in the previous step needs to be applied to the original residual output by the neural network, and the changes in power coefficient and thrust coefficient have a strong physical coupling relationship in real aerodynamic processes, applying the two correction values ​​independently would sever this physical connection and reduce the inherent consistency of the correction results. Therefore, in the technical solution of this application, weighted residual correction and interactive coupling calibration are further performed on the power coefficient residual and thrust coefficient residual based on the reference power coefficient, reference thrust coefficient, power correction scaling factor, and thrust correction scaling factor to obtain the pre-corrected power coefficient and pre-corrected thrust coefficient. This allows the adaptive scaling factor to be applied to the original residual, and an interactive coupling calibration function is introduced to simulate the linkage effect between power and thrust in the real world. In this way, the nonlinearity of the working condition can be considered during the application correction process, and the physical coupling between variables can also be incorporated, thereby generating intermediate correction results that are more accurate and physically consistent than simple superposition.

[0044] More specifically, in a particular example of this application, the process first obtains the reference power factor determined in the previous step. and reference thrust coefficient and power correction scaling factor and thrust correction scaling factor Simultaneously, the residual of the original power coefficient predicted by the neural network is obtained. and the original thrust coefficient residual and the standard deviation of the angle deviation Subsequently, a weighted residual correction is performed. For the power factor, the reference power factor is used. With power correction scaling factor Weighted original power coefficient residual Adding them together yields the pre-corrected power coefficient. For the thrust coefficient, an interactive coupling calibration function is introduced. The reference thrust coefficient After thrust correction scaling factor Weighted residuals of the original thrust coefficients ( · )as well as The pre-corrected thrust coefficient is obtained by adding the three output values ​​of the function. The calibration function Able to calculate the residual based on the weighted power coefficient ( · ) and asymmetric severity The weighted thrust coefficient residuals are then fine-tuned a second time. The function simulates the linkage effect between power and thrust. For example, when the calibration function detects a significant loss of power due to severe asymmetry (i.e., · (It is a relatively large negative value), which can synchronously and non-linearly adjust the correction amount of the thrust coefficient. Continuing from the previous data, Calculated as That is, 0.43. Regarding thrust, The function's input is -0.04 (i.e., ... ) and 0.235. If the function data determines that this power loss will be accompanied by an additional thrust decrease, a calibration value will be output. ,but Calculated as ,Right now This achieves a more physically consistent correction.

[0045] In step S530, the pre-corrected power coefficient and pre-corrected thrust coefficient are physically bounded to obtain the corrected power coefficient and corrected thrust coefficient. It is understood that since the pre-corrected power coefficient and pre-corrected thrust coefficient obtained in the previous step are only weighted and calibrated intermediate values, they may still mathematically exceed the physically reasonable range defined by aerodynamic principles. To address this deficiency and ensure that the aerodynamic coefficients finally output to the controller are absolutely within the effective physical range, thus guaranteeing the stability and safety of the entire control system, the technical solution of this application further performs physical boundary mapping on the pre-corrected power coefficient and pre-corrected thrust coefficient to obtain the corrected power coefficient and corrected thrust coefficient. This is achieved by applying a nonlinear saturation function or a smooth boundary mapping function to forcibly constrain the pre-corrected values ​​within preset physical upper and lower boundaries. This ensures that the final result does not violate basic physical principles, such as eliminating the possibility of negative power coefficients, ensuring that the corrected aerodynamic model is numerically more accurate and physically correct and robust, providing high-quality, high-reliability model input for downstream adaptive control algorithms.

[0046] More specifically, in a particular example of this application, the mapping process is implemented using a nonlinear saturation function as an example. First, based on the specific aerodynamic characteristics of the wind turbine, the physical boundaries of the power coefficient are pre-defined, for example... =0, It is 0.5; and the physical boundary of the thrust coefficient, for example It is 0.05. Set the value to 1.0. Obtain the pre-corrected power coefficient obtained in the previous step. and pre-corrected thrust coefficient .right The value is processed and constrained to a preset physical minimum bound. With the maximum boundary Between, to obtain the final revised version The following formula illustrates this:

[0047] right Perform the same processing to constrain it. and Between, to obtain the final revised version The following formula is used to illustrate this:

[0048] in, , , , These are the physical upper and lower boundaries of the power coefficient and thrust coefficient, respectively. It is a function with maximum value. It is a minimum value function. and This refers to the pre-corrected power coefficient and pre-corrected thrust coefficient. For example, if the calculation obtained in the previous step... The value is -0.1, which exceeds the limit. If the lower bound is 0, then the result of min(0.5,-0.1) is -0.1 and the result of max(0,-0.1) is 0, so the final output is... It was forcibly corrected to 0.0, eliminating the possibility of a negative power coefficient. If another calculated value... It is 0.52, which exceeds the limit. If the upper bound is 0.5, then min(0.5, 0.52) results in 0.5, max(0, 0.5) results in 0.5, and the final output is... It is constrained to 0.5. And if The calculated value is 0.829, which falls within the boundary between 0.05 and 1.0. Therefore, min(1.0, 0.829) results in 0.829, and max(0.05, 0.829) also results in 0.829. The final output is... It remains at 0.829.

[0049] Specifically, in step S600, adaptive control commands based on a modified model are generated using the modified power coefficient, modified thrust coefficient, rotor speed, and air density to obtain adaptive torque and adaptive pitch commands. It is understood that while the modified power coefficient and modified thrust coefficient obtained in the previous step provide an accurate description of the wind turbine's aerodynamic performance under asymmetric faults, the control laws of the wind turbine's main controller, such as the torque control law and the pitch PI controller, were initially tuned based on a baseline model with symmetric assumptions. If these modified coefficients are not applied to the generation of control commands, the control system will still calculate based on an incorrect model, leading to decreased control performance and an inability to achieve optimized operation under fault conditions. Therefore, in the technical solution of this application, adaptive control commands based on a modified model are further generated using the modified power coefficient, modified thrust coefficient, rotor speed, and air density to obtain adaptive torque and adaptive pitch commands. This allows the modified, more physically realistic aerodynamic model to be embedded in the real-time calculation process of the control law, for example, to update the baseline of the optimal torque curve or to adjust the parameters of the pitch controller based on the modified aerodynamic gain. This ensures that the final generated control commands are optimized for the current asymmetric fault state, enabling the wind turbine to maintain stable power output and controlled component loads even under fault conditions, thus achieving precise adaptive control.

[0050] Figure 5 This document describes a flowchart illustrating the process of generating adaptive torque and adaptive pitch commands based on a modified model using a modified power coefficient, modified thrust coefficient, rotor speed, and air density, according to an embodiment of this application, for identifying and controlling asymmetric operation of a pitch system. Figure 5 As shown, step S600 further includes: S610, performing real-time calculation of the corrected aerodynamic load based on the corrected power coefficient, corrected thrust coefficient, rotor speed and air density to obtain the corrected aerodynamic torque; S620, performing adaptive control law calculation based on the operating range on the corrected aerodynamic torque to obtain the target generating torque and the target collective pitch angle; S630, performing final control command synthesis on the target generating torque and the target collective pitch angle to obtain the adaptive torque command and the adaptive pitch command.

[0051] In step S610, the corrected aerodynamic load is calculated in real time based on the corrected power coefficient, corrected thrust coefficient, rotor speed, and air density to obtain the corrected aerodynamic torque. It is understood that since the corrected power coefficient and corrected thrust coefficient output in the previous step are dimensionless coefficients, and the downstream adaptive control law calculation, especially torque control and pitch control, is based on aerodynamic torque in the physical world, the control system cannot utilize this high-precision model information if the corrected coefficients are not converted into actual physical loads. Therefore, in the technical solution of this application, the corrected aerodynamic load is further calculated in real time based on the corrected power coefficient, corrected thrust coefficient, rotor speed, and air density to obtain the corrected aerodynamic torque, thereby outputting the corrected aerodynamic model. and By combining real-time operating and environmental parameters, the instantaneous aerodynamic load values ​​reflecting asymmetric faults are calculated. This provides an accurate and physically meaningful reference input for subsequent adaptive control law calculation steps, ensuring that control commands are based on aerodynamic load estimates that are closest to reality.

[0052] More specifically, in a particular example of this application, the real-time calculation process is executed in each control cycle. First, the corrected power coefficient from the previous model correction stage is obtained. and the corrected thrust coefficient Simultaneously, the real-time rotor speed is acquired from the sensor. and air density from the environmental monitoring unit In addition, the rotor radius is retrieved from the controller parameter library. And obtain the current estimated wind speed from the state observer. Subsequently, to obtain the corrected aerodynamic torque, the corrected aerodynamic power is first calculated. The calculation method is as follows: 0.5, air density... Pi Wind turbine radius squared, estimated wind speed cube and corrected power coefficient Multiply. The corrected aerodynamic power... Then, divide it by the real-time rotor speed. To obtain the final corrected aerodynamic torque. At the same time, using the corrected thrust coefficient Corrected aerodynamic thrust The calculation method is as follows: 0.5, air density... Pi Wind turbine radius squared, estimated wind speed The square of and the corrected thrust coefficient Multiply. Ultimately, the... The value (in Newton-meters) is output to the adaptive control law solution stage, and this The value (in Newtons) is output to downstream control systems such as load monitoring or active damping.

[0053] In step S620, the modified aerodynamic torque is calculated using an adaptive control law based on the operating range to obtain the target generating torque and the target collective pitch angle. It is understood that the modified aerodynamic torque obtained in the previous step is only a precise estimate of the current aerodynamic load, while the wind turbine has drastically different control objectives and control laws in different operating ranges, namely the variable speed operating range and the constant power operating range. If the original control law, tuned based on the symmetry assumption, is not adjusted accordingly based on the modified torque information, the controller will still execute suboptimal control and will not be able to leverage the advantages of model correction. Therefore, in the technical solution of this application, the modified aerodynamic torque is further calculated using an adaptive control law based on the operating range to obtain the target generating torque and the target collective pitch angle. This allows the wind turbine to determine its operating range based on the real-time rotor speed and activate specific adaptive logic in different ranges: in the variable speed range, the optimal torque tracking curve is adjusted based on the modified aerodynamic model; in the constant power range, the parameters of the pitch controller are tuned online based on the modified aerodynamic gain. This ensures that the target generating torque generated in the variable speed region can maximize the actual power capture under asymmetric faults, while the target collective pitch angle generated in the constant power region can stably maintain the rated power, thereby improving the system's robustness and power generation efficiency under fault conditions.

[0054] More specifically, in a concrete example of this application, the calculation process first obtains the corrected aerodynamic torque from the previous step, and then obtains the current rotor speed from the sensor and compares it with the stored rated rotor speed. If the current rotor speed (1.0 rad / s) is less than the rated rotor speed (1.2 rad / s), the wind turbine is determined to be in the variable speed operation range. Within this range, the target collective pitch angle is maintained at the optimal pitch angle, for example, 0 degrees. The control objective at this time is maximum power point tracking. Based on the new optimal tip speed ratio implied by the corrected power coefficient under the current asymmetric fault, the optimal torque coefficient is compared with the baseline. After correction, an optimal torque coefficient is obtained. The target power generation torque was then calculated as... The product of the rotor speed and the square of the rotational speed. For example, if the reference... The corrected value is 15.0 kNm / (rad / s)². The target generating torque is set to 14.5 kNm / (rad / s)². That is, 14.5 kNm. At another moment, if the acquired rotor speed is 1.21 rad / s, which is greater than the rated rotor speed of 1.2 rad / s, then the wind turbine is determined to be in the constant power operation zone. In this zone, the target generating torque is clamped at the rated torque value, for example, 20.0 kNm. The target collective pitch angle is calculated by the pitch PI controller based on the speed error. At this time, the corrected aerodynamic gain at the current operating point (determined by the tip speed ratio and the current pitch angle) is calculated online using the corrected power coefficient. The corrected gain is compared with the baseline aerodynamic gain obtained during the design based on the baseline model. Compare the baseline PI parameters. and It will be adaptively adjusted based on the ratio of the two; for example, the adjusted Kp_adapt will equal... Multiply and The ratio. The pitch PI controller then uses the adjusted value. and By combining the rotational speed error of 0.01 rad / s, the target collective pitch angle is calculated, for example, 10.5 degrees. Finally, the calculated target generator torque and target collective pitch angle are output to the final control command synthesis stage.

[0055] In step S630, the target generating torque and target collective pitch angle are synthesized into adaptive torque and adaptive pitch commands. It is understood that the target generating torque and target collective pitch angle calculated in the previous step primarily carry the power and speed regulation targets of the wind turbine under asymmetric faults. However, the wind turbine control system often includes individual pitch control for suppressing unbalanced loads or additional control functions for suppressing tower vibration. These functions generate additional pitch modulation in parallel. Without effectively synthesizing these parallel control commands, the complete control objective cannot be achieved. Therefore, in the technical solution of this application, the target generating torque and target collective pitch angle are further synthesized into adaptive torque and adaptive pitch commands. This superimposes the collective command carrying the main control objective with the individual modulation command carrying the load suppression objective, and applies slope and amplitude limits to the torque command. This generates a set of final, executable control commands that not only achieve adaptive power and speed control based on the modified model but also consider component load management, ensuring the integrity and safety of the control system.

[0056] More specifically, in a concrete example of this application, the synthesis process is divided into two branches: torque command synthesis and pitch command synthesis. In torque command synthesis, the target generating torque calculated in the previous stage is obtained, for example, 14.5 kNm. This target value is first processed by a slope limiter to ensure that its rate of change does not exceed the maximum allowable torque change rate of the converter, for example, 5 kNm / s. Subsequently, the slope-limited torque value is passed through an amplitude limiter to ensure that it is within the minimum and maximum torque range for safe generator operation. The final constrained torque value is determined as the adaptive torque command and sent to the converter for execution. In pitch command synthesis, the target collective pitch angle calculated in the previous stage is obtained, for example, 10.5 degrees. Simultaneously, the individual modulation amounts of the first, second, and third blades are obtained from the parallel individual pitch control units, for example, +0.2 degrees, -0.1 degrees, and -0.1 degrees, respectively. For the first blade, its final adaptive pitch command is calculated as the sum of the target collective pitch angle and the individual modulation amount of the first blade. The same additive synthesis is performed for the second blade. The same additive synthesis is performed for the third blade. This set of instructions, containing three independent angle values, together constitutes the adaptive pitch command, which is then sent to the drive systems of the first, second, and third blades, respectively.

[0057] In summary, the identification and control method for asymmetric operation of the pitch system according to the embodiments of this application is explained. To address the mismatch between the control model and physical reality under asymmetric fault conditions in wind turbines, the stable angle deviation of each blade is quantified in real time and aggregated into an average angle deviation and an angle deviation standard deviation. Subsequently, a residual neural network is used to predict the power coefficient residual and thrust coefficient residual based on the aforementioned asymmetric indices. This residual is used to dynamically correct the reference aerodynamic model within the controller, and calibration is performed by introducing an adaptive correction factor and mapping to the physical boundary, thereby obtaining corrected power and thrust coefficients that accurately reflect the fault state. Finally, based on this corrected model, adaptive control law is calculated to generate adaptive torque and pitch commands to achieve optimized operation under fault conditions.

[0058] Furthermore, a recognition and control system for asymmetric operation of a pitch system is also provided.

[0059] Figure 6 This is a block diagram of a control system for identifying and controlling asymmetric operation of a pitch system according to an embodiment of this application. Figure 6As shown, the pitch system asymmetric operation identification and control system 100 according to an embodiment of this application includes: a command signal and fault type acquisition module 110, used to acquire the first to third blade pitch command signals, the first to third actual blade angles, and the first to third blade fault types; a real-time asymmetric state quantization module 120, used to perform real-time asymmetric state quantization on the first to third blade pitch command signals, the first to third actual blade angles, and the first to third blade fault types to obtain the first to third blade stable angle deviations; and an asymmetric physical metric aggregation module 130, used to aggregate the first to third blade stable angle deviations into asymmetric physical metric indicators to obtain the average angle deviation and the angle deviation. The aerodynamic coefficient correction prediction module 140 is used to perform aerodynamic coefficient correction prediction based on residual neural networks on the average angle deviation, angle deviation standard deviation, collective pitch command, and tip speed ratio to obtain the power coefficient residual and thrust coefficient residual; the global aerodynamic model dynamic correction module 150 is used to perform global aerodynamic model dynamic correction on the power coefficient residual and thrust coefficient residual based on the collective pitch command and tip speed ratio to obtain the corrected power coefficient and corrected thrust coefficient; the command generation module 160 is used to generate adaptive control commands based on the corrected power coefficient, corrected thrust coefficient, rotor speed, and air density to obtain adaptive torque command and adaptive pitch command.

[0060] As described above, the pitch system asymmetric operation identification and control system 100 according to embodiments of this application can be implemented in various types of computing devices or control units. For example, it can be deployed in the main controller of a wind turbine generator, a dedicated controller for the pitch system, or an industrial computer for wind farm-level monitoring. In one possible implementation, the pitch system asymmetric operation identification and control system 100 according to embodiments of this application can be integrated into the computing device as a software module and / or a hardware module. For example, the pitch system asymmetric operation identification and control system 100 can be a software module in the control firmware of the computing device or control unit, or it can be a dedicated control algorithm program developed for the computing device or control unit. Of course, the pitch system asymmetric operation identification and control system 100 can also be one of many hardware modules of the computing device or control unit, such as being implemented as dedicated digital signal processor logic, field-programmable gate array circuits, or application-specific integrated circuits.

[0061] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method of identification and control of asymmetric operation of a variable pitch system, characterized in that, include: Acquire the pitch command signals of the first to third blades, the actual blade angles of the first to third blades, and the fault types of the first to third blades; Real-time asymmetric state quantization is performed on the pitch command signals of the first to third blades, the actual blade angles of the first to third blades, and the fault types of the first to third blades to obtain the stable angle deviation of the first to third blades. Asymmetric physical metrics were aggregated to obtain the average angle deviation and the standard deviation of the angle deviation for the first to third blades. The aerodynamic coefficient correction prediction based on residual neural network is performed on the mean angle deviation, the standard deviation of the angle deviation, the collective pitch command and the tip speed ratio to obtain the power coefficient residual and the thrust coefficient residual; Based on the collective pitch command and tip speed ratio, the power coefficient residual and thrust coefficient residual are dynamically corrected by the global aerodynamic model to obtain the corrected power coefficient and thrust coefficient. Based on the corrected power coefficient, corrected thrust coefficient, rotor speed, and air density, adaptive control commands based on the corrected model are generated to obtain adaptive torque commands and adaptive pitch commands.

2. The method for identifying and controlling asymmetric operation of a pitch system according to claim 1, characterized in that, Real-time asymmetric state quantization is performed on the pitch command signals of the first to third blades, the actual blade angles of the first to third blades, and the fault types of the first to third blades to obtain the stable angle deviations of the first to third blades, including: Calculate the instantaneous blade angle deviation between the first to third blade pitch command signals and the first to third actual blade angles to obtain the original blade angle deviations; Based on the pitch control command signals of the first to third blades, the original angle deviations of the first to third blades are filtered based on the dynamic conditions of pitch control to obtain the filtered angle deviations of the first to third blades. Based on the fault types of the first to third blades, fault mode matching is performed on the filtered angle deviations of the first to third blades to obtain the stable angle deviations of the first to third blades.

3. The method for identifying and controlling asymmetric operation of a pitch system according to claim 2, characterized in that, Based on the pitch command signals of the first to third blades, the original angle deviations of the first to third blades are filtered according to the dynamic conditions of pitch control to obtain the filtered angle deviations of the first to third blades, including: Based on the pitch control command signals of the first to third blades and the pitch control command signals of the first to third blades at the previous moment, calculate the command change rate of the first to third blades. Based on the rate of change threshold, the command stability status of the first to third blades is determined to obtain the command stability flags of the first to third blades. Based on the command stabilization flags of the first to third blades, the original angle deviations of the first to third blades are subjected to conditional low-pass filtering to obtain the filtered angle deviations of the first to third blades.

4. The method for identifying and controlling asymmetric operation of a pitch system according to claim 1, characterized in that, The stability angle deviations of the first to third blades were aggregated using asymmetric physical metrics to obtain the mean angle deviation and the standard deviation of the angle deviation, including: Calculate the mean of the stability angle deviations of the first to third blades to obtain the average angle deviation; Calculate the standard deviation of the stability angle deviation of the first to third blades to obtain the standard deviation of the angle deviation.

5. The method for identifying and controlling asymmetric operation of a pitch system according to claim 1, characterized in that, Aerodynamic coefficient correction predictions based on residual neural networks are performed on the mean angle deviation, standard deviation of angle deviation, collective pitch command, and tip speed ratio to obtain the power coefficient residual and thrust coefficient residual, including: The average angular deviation, standard deviation of angular deviation, collective pitch command, and tip speed ratio are arranged into an input vector; The input vector is fed into a trained lightweight feedforward neural network to obtain the power coefficient residual and the thrust coefficient residual.

6. The method for identifying and controlling asymmetric operation of a pitch system according to claim 1, characterized in that, Based on the collective pitch command and tip speed ratio, the power coefficient residual and thrust coefficient residual are dynamically corrected using a global aerodynamic model to obtain the corrected power coefficient and thrust coefficient, including: Based on the collective pitch command, tip speed ratio, standard deviation of angle deviation, power coefficient residual and thrust coefficient residual, the reference power coefficient, reference thrust coefficient, power correction scaling factor and thrust correction scaling factor are determined. Based on the reference power coefficient, reference thrust coefficient, power correction scaling factor, and thrust correction scaling factor, the power coefficient residual and thrust coefficient residual are weighted residual correction and interactive coupling calibration to obtain the pre-corrected power coefficient and pre-corrected thrust coefficient. Physical boundary mapping is performed on the pre-corrected power coefficient and pre-corrected thrust coefficient to obtain the corrected power coefficient and corrected thrust coefficient.

7. The method for identifying and controlling asymmetric operation of a pitch system according to claim 6, characterized in that, To obtain the corrected power coefficient and the corrected thrust coefficient by performing physical boundary mapping on the pre-corrected power coefficient and the pre-corrected thrust coefficient, the following steps are taken: Physical boundary mapping is performed on the pre-corrected power coefficient and the corrected thrust coefficient using the following formula: in, , , , These are the physical upper and lower boundaries of the power coefficient and thrust coefficient, respectively. It is a function with maximum value. It is a minimum value function. and These are the pre-corrected power coefficient and the pre-corrected thrust coefficient.

8. The method for identifying and controlling asymmetric operation of a pitch system according to claim 1, characterized in that, Based on the corrected power coefficient, corrected thrust coefficient, rotor speed, and air density, adaptive control commands based on the modified model are generated to obtain adaptive torque commands and adaptive pitch commands, including: Based on the corrected power coefficient, corrected thrust coefficient, rotor speed and air density, the corrected aerodynamic load is calculated in real time to obtain the corrected aerodynamic torque. The modified aerodynamic torque is solved by an adaptive control law based on the operating range to obtain the target generator torque and the target collective pitch angle; The target generator torque and the target collective pitch angle are used to synthesize the final control commands to obtain the adaptive torque command and the adaptive pitch command.

9. A system for identifying and controlling asymmetric operation of a pitch system, characterized in that, include: The command signal and fault type acquisition module is used to acquire the first to third blade pitch control command signals, the first to third actual blade angles, and the first to third blade fault types. The real-time asymmetric state quantization module is used to perform real-time asymmetric state quantization on the first to third blade pitch command signals, the first to third actual blade angles and the first to third blade fault types to obtain the first to third blade stable angle deviations. The asymmetric physical metric aggregation module is used to aggregate asymmetric physical metric indicators for the stability angle deviations of the first to third blades to obtain the average angle deviation and the standard deviation of the angle deviation. The aerodynamic coefficient correction prediction module is used to perform aerodynamic coefficient correction prediction based on residual neural network on the mean angle deviation, angle deviation standard deviation, collective pitch command and tip speed ratio to obtain the power coefficient residual and thrust coefficient residual. The global aerodynamic model dynamic correction module is used to dynamically correct the power coefficient residual and thrust coefficient residual based on the collective pitch command and tip speed ratio to obtain the corrected power coefficient and thrust coefficient. The instruction generation module is used to generate adaptive control instructions based on the modified model, based on the modified power coefficient, modified thrust coefficient, rotor speed and air density, to obtain adaptive torque command and adaptive pitch command.

10. The identification and control system for asymmetric operation of a pitch system according to claim 9, characterized in that, The instruction generation module includes: The corrected aerodynamic torque calculation unit is used to perform real-time calculation of the corrected aerodynamic load based on the corrected power coefficient, corrected thrust coefficient, rotor speed and air density to obtain the corrected aerodynamic torque. The adaptive control law calculation unit is used to perform adaptive control law calculation based on the operating range on the corrected aerodynamic torque to obtain the target generator torque and the target collective pitch angle. The final control command synthesis unit is used to synthesize the target generator torque and the target collective pitch angle into an adaptive torque command and an adaptive pitch command.

Images