A method and system for active damping of offshore wind farm oscillations

Abstract

The application provides a kind of offshore wind farm oscillation active inhibition method and system, the method comprises: using offshore wind farm multi-source sensor data to carry out time-frequency analysis and dimension reduction processing, obtain key characteristic quantity set;Key characteristic quantity set is input into Lasso regression algorithm and DeePC framework, establish system nonlinear model, calculate to obtain oscillation source position;According to the oscillation source position, establish local control model, construct target function containing oscillation suppression effect and safety constraint through local control model, calculate target function using Bayesian optimization, obtain control parameter;According to the oscillation source position, carry out offshore wind farm partition operation, implement different control strategies based on control parameters in different regions, realize oscillation suppression.The application realizes adaptive optimization of control parameter by using Safe Barrier Bayesian optimization method, realizes system modeling and oscillation source positioning based on Lasso-DeePC framework, proposes hierarchical and partitioned differential control strategy, improves the efficiency and accuracy of offshore wind farm oscillation suppression.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and in particular to a method and system for actively suppressing oscillations in offshore wind farms, used to solve the broadband oscillation problem that occurs in offshore wind farms during operation. Background Technology

[0002] Offshore wind farms are an important component of clean energy, and their stability control is a key issue in ensuring the safe operation of the power grid. Due to their distributed nature, nonlinear characteristics, and the influence of complex environmental factors, offshore wind farm systems often face broadband oscillation problems.

[0003] Currently, common offshore wind farm oscillation suppression technologies mainly include traditional control methods based on PI controllers and modern control methods based on model predictive control. These methods achieve oscillation suppression by adjusting the active and reactive power of the wind turbine.

[0004] More advanced oscillation suppression technologies employ data-driven control methods. By acquiring real-time operational data from offshore wind farms, a dynamic system model is established, and control strategies are designed based on this model. This technology analyzes system characteristic quantities to identify oscillation sources, thereby achieving oscillation suppression.

[0005] However, existing technologies have the following problems: 1) The lack of safety constraints during the optimization of control parameters may lead to system instability; 2) For strongly nonlinear systems, the control decision-making process lacks interpretability, making it difficult to guarantee the reliability of the control effect; 3) The accuracy of oscillation source localization is not high, affecting the oscillation suppression effect. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for active suppression of offshore wind farm oscillations. It achieves adaptive optimization of control parameters by adopting the Safe Barrier Bayesian optimization method, realizes system modeling and oscillation source localization based on the Lasso-DeePC framework, and proposes a hierarchical and partitioned differentiated control strategy to improve the efficiency and accuracy of offshore wind farm oscillation suppression.

[0007] To achieve the above objectives, the present invention provides a method for actively suppressing oscillations in offshore wind farms, comprising: Time-frequency analysis and dimensionality reduction were performed using multi-source sensor data from offshore wind farms to obtain a set of key feature quantities. The set of key features is input into the Lasso regression algorithm and the DeePC framework to establish a nonlinear model of the system and calculate the location of the oscillation source. A local control model is established based on the location of the oscillation source. An objective function containing oscillation suppression effect and safety constraints is constructed through the local control model. The objective function is calculated using Bayesian optimization to obtain the control parameters. The offshore wind farm is zoned according to the location of the oscillation source, and differentiated control strategies are implemented for different areas based on the control parameters to achieve oscillation suppression.

[0008] Furthermore, the process of using multi-source sensor data from offshore wind farms for time-frequency analysis and dimensionality reduction yields a set of key feature quantities, including: The multi-source sensor data is input into a wavelet transform module to extract the system oscillation frequency components. Principal component analysis is performed using the oscillation frequency components to obtain the main eigenvectors; The mutual information between the feature quantities is calculated using the main feature vectors to generate the key feature quantity set.

[0009] Furthermore, the step of inputting the set of key feature quantities into the Lasso regression algorithm and the DeePC framework to establish a nonlinear model of the system and calculate the location of the oscillation source includes: The set of key features is input into the Hankel matrix construction module, and Lasso regression is performed to obtain the sparse representation of the system. Perturbation analysis is performed using the sparse representation of the system to calculate the oscillation contribution of each node. The location of the oscillation source is determined by performing a positioning operation based on the oscillation contribution.

[0010] Furthermore, a local control model is established based on the location of the oscillation source. An objective function incorporating oscillation suppression effects and safety constraints is constructed using this local control model. Bayesian optimization is then employed to calculate the objective function, yielding control parameters, including: The location of the oscillation source is input into the constraint calculation module to generate system stability constraint parameters; An optimization objective function is constructed using the system stability constraint parameters and oscillation suppression index. The optimization objective function is input into the Bayesian optimization iterator to calculate the control parameters.

[0011] Furthermore, the offshore wind farm is zoned according to the control parameters, including: The spatial distribution of the core region, transition region, and peripheral region is determined using the location of the oscillation source; Calculate the control priority and control intensity of the core area, the transition area and the outer area based on the spatial distribution; The control priority and control strength are input into the partition controller to generate partition control instructions.

[0012] Furthermore, the implementation of differentiated control strategies for different regions based on the control parameters to achieve oscillation suppression includes: Collect the operating status data of the core area wind turbines and calculate the power adjustment range; The power adjustment range is input into the curve optimizer to generate the active power output curve of the wind turbine. Adjust the reactive power control parameters of the core area fan according to the active power output curve to achieve oscillation suppression.

[0013] Furthermore, the implementation of differentiated control strategies for different regions based on the control parameters to achieve oscillation suppression includes: Collect status data of the fans in the transition zone and the outer zone, and calculate power balance parameters; The power response characteristics of the transition zone fan are adjusted using the power balance parameters to construct the oscillation attenuation region of the transition zone fan; The response characteristics of the oscillation attenuation region are input into the parameter tuner to generate the control gain coefficient of the peripheral area fan for suppressing its sensitivity to oscillations.

[0014] Furthermore, it also includes: Collect voltage, power, and oscillation frequency component data at each node of the offshore wind farm; Real-time monitoring calculations are performed using the voltage, power, and oscillation frequency component data to obtain real-time monitoring calculation results; The real-time monitoring calculation result is compared with the system stability boundary value. If the real-time monitoring calculation result exceeds the system stability boundary value, the oscillation suppression controller is triggered.

[0015] Further, the step of comparing the real-time monitoring calculation result with the system stability boundary value, and triggering the oscillation suppression controller if the real-time monitoring calculation result exceeds the system stability boundary value, includes: Use an oscillation amplitude calculator to measure oscillation magnitude indicators; Input the oscillation magnitude index into the control scheme selector to determine the oscillation suppression parameters; The intensity coefficient of the control parameter is adjusted according to the oscillation suppression parameter.

[0016] Furthermore, it also includes: Collect oscillation decay rate and settling time data; The response characteristic index is calculated using the oscillation decay rate and the settling time; The response characteristic index is input into the adaptive regulator to update the control parameters.

[0017] The present invention also provides a system for active suppression of offshore wind farm oscillations, comprising: The feature extraction module is used to acquire multi-source sensor data of offshore wind farms, perform time-frequency analysis and dimensionality reduction on the multi-source sensor data, and obtain a set of key feature quantities. The oscillation source localization module is used to input the set of key feature quantities into the Lasso regression algorithm and the DeePC framework, establish a nonlinear model of the system, and calculate the location of the oscillation source. The parameter optimization module is used to establish a local control model based on the location of the oscillation source, construct an objective function that includes oscillation suppression effect and safety constraints through the local control model, and use Bayesian optimization to calculate the objective function to obtain control parameters. The control execution module is used to perform zoning operations of the offshore wind farm according to the location of the oscillation source, and to implement differentiated control strategies for different areas based on the control parameters to achieve oscillation suppression.

[0018] The beneficial effects of this invention are as follows: 1. The Safe Barrier Bayesian optimization method is adopted to achieve adaptive optimization of control parameters. By introducing a safety barrier function, the system is ensured to remain stable throughout the optimization process, which effectively solves the problem that traditional parameter optimization methods may lead to system instability. 2. System modeling and oscillation source localization are realized based on the Lasso-DeePC framework. By utilizing the sparsity characteristics of the Lasso algorithm and the nonlinear system modeling capability of DeePC, the accuracy of oscillation source localization is improved, while ensuring the interpretability of control decisions. 3. A hierarchical and zoned differentiated control strategy was proposed, which implemented key control based on the location of the oscillation source, thereby improving the efficiency and accuracy of oscillation suppression. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of an active suppression method for offshore wind farm oscillations provided in an embodiment of the present invention; Figure 2 This is a flowchart of multimodal key feature extraction in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the oscillation source tracing and localization process in an embodiment of the present invention; Figure 4This is a flowchart of the Safe Barrier Bayesian optimization control in an embodiment of the present invention; Figure 5 This is a structural block diagram of an active oscillation suppression system for offshore wind farms provided in an embodiment of the present invention. Detailed Implementation

[0021] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments, so that those skilled in the art can better understand and implement the present invention, but the embodiments described do not constitute a limitation of the present invention.

[0022] like Figure 1 As shown in the figure, an active suppression method for offshore wind farm oscillations provided by an embodiment of the present invention includes the following steps: Step S1: Use multi-source sensor data from offshore wind farms for time-frequency analysis and dimensionality reduction to obtain a set of key feature quantities.

[0023] In this step, multi-source sensor data is first collected through the offshore wind farm monitoring system, including operational data such as voltage, current, power, and rotational speed at various nodes of the offshore wind farm. This data covers different types of measurements at different locations within the offshore wind farm, forming a complete description of the offshore wind farm's condition. During data acquisition, the system optimizes the sampling frequency, typically within the range of 10Hz to 100Hz, to ensure the capture of various frequency oscillations that may occur within the offshore wind farm system. The collected data undergoes preprocessing, including noise reduction, outlier handling, and time synchronization, to improve the accuracy of subsequent analysis.

[0024] Next, time-frequency analysis was performed on the collected multi-source sensor data. The purpose of time-frequency analysis is to identify the main oscillation frequency components in the system, which typically correspond to different oscillation modes within the system. In offshore wind farm systems, common oscillation frequencies include subsynchronous oscillations and inter-regional oscillations (0.1~0.7Hz). Through time-frequency analysis, these oscillation frequency components can be clearly identified, providing important information for subsequent oscillation source localization.

[0025] Next, dimensionality reduction is performed. Because offshore wind farm systems have extremely high data dimensionality, directly using the raw data for analysis would result in excessive computation and be susceptible to data redundancy and noise. The purpose of dimensionality reduction is to reduce the data dimensionality while preserving the main information, thereby improving analysis efficiency. The dimensionality reduction process needs to consider the correlation between features, screen out the key features that truly have a significant impact on system oscillations, and eliminate redundant information and noise interference.

[0026] Finally, through the above processing, a set of key characteristic quantities for the offshore wind farm system is obtained. This set contains key parameters that can effectively characterize the system's oscillation characteristics, such as power fluctuations at specific frequencies, voltage change rates, and phase angle differences. These key characteristic quantities will serve as important inputs for subsequent oscillation source localization.

[0027] Step S2: Input the set of key features into the Lasso regression algorithm and the DeePC framework to establish a nonlinear model of the system and calculate the location of the oscillation source.

[0028] In this step, the set of key features obtained in step S1 is first input into the system modeling module. This module uses the Lasso regression algorithm for sparse modeling. Lasso regression is a linear regression method that can automatically select features. Its characteristic is that it adds an L1 norm penalty term to the objective function, which can compress the coefficients of unimportant features to zero, thereby obtaining a sparse system representation. This sparse representation helps to identify key influencing factors in the system and is of great significance for oscillation source localization.

[0029] Next, the sparse system representation obtained from Lasso regression is embedded into the DeePC (Deep equilibrium Predictive Control) framework. DeePC is a method for directly constructing predictive models based on data, without explicit system identification, making it particularly suitable for modeling strongly nonlinear systems such as offshore wind farms. In the DeePC framework, the system solves a convex optimization problem, searching for the optimal linear combination within the Hankel matrix of historical data to predict future system behavior. This method effectively captures the nonlinear dynamic characteristics of the system, improving the model's predictive accuracy.

[0030] Based on the established nonlinear system model, the oscillation source is located and calculated. The location process is mainly achieved by analyzing the contribution of each node in the model to the system oscillation. Specifically, each node is individually perturbed, and its impact on the overall system oscillation is observed, thus quantifying the oscillation contribution of each node. The contribution calculation considers the node's location characteristics, electrical parameters, and coupling relationship with other nodes, enabling a comprehensive assessment of the impact of each node on the system oscillation. By comparing the oscillation contributions of each node, the node with the largest contribution is identified as the oscillation source, achieving precise location.

[0031] Step S3: Establish a local control model based on the location of the oscillation source, construct an objective function that includes oscillation suppression effect and safety constraints through the local control model, and use Bayesian optimization to calculate the objective function to obtain control parameters.

[0032] In this step, a local control model is first established based on the oscillation source location determined in step S2. The local control model focuses on the oscillation source node and its directly related neighboring nodes, including information such as the dynamic characteristics, coupling relationships, and control constraints of these nodes. Establishing a local model can significantly reduce the complexity of control optimization and improve the real-time performance and specificity of the algorithm. The model construction process considers the special characteristics of offshore wind farm systems, such as the power regulation characteristics of wind turbines and grid constraints.

[0033] Next, an optimization objective function is constructed based on the local control model. The objective function consists of two main parts: first, an oscillation suppression performance index, such as the decay rate of the oscillation amplitude and the settling time; and second, a safety constraint term, implemented through a safety barrier function. The safety barrier function is a continuously differentiable function whose value increases rapidly as the system state approaches the safety boundary, thus automatically avoiding unsafe regions during optimization. This design ensures that the system always operates within the safe region during optimization, effectively solving the problem of system instability that may result from traditional optimization methods.

[0034] Then, the objective function is calculated using Bayesian optimization to obtain the optimal control parameters. Bayesian optimization is an efficient global optimization method, particularly suitable for computationally complex optimization problems with high objective function evaluation costs. This method establishes a probabilistic model of the objective function through a Gaussian process and uses a sampling function (such as desired improvement) to balance exploration and exploitation, selecting the optimal parameter update direction. In each iteration, the system selects the optimal parameter combination based on the current probabilistic model, evaluates its effect, and updates the probabilistic model. This process continues until the preset convergence condition is met. Throughout the optimization process, the safety barrier function ensures that the system operates within a safe boundary, avoiding potential system instability risks.

[0035] Ultimately, the control parameters obtained through Bayesian optimization can both ensure good oscillation suppression and stable system operation, providing a reliable parameter basis for the next step of control strategy execution.

[0036] Step S4: Perform offshore wind farm zoning operation according to the location of the oscillation source, and implement differentiated control strategies for different areas based on the control parameters to achieve oscillation suppression.

[0037] In this step, the offshore wind farm is first divided into zones based on the location of the oscillation source determined in step S2. The zoning strategy divides the offshore wind farm into three levels according to the degree of oscillation impact: a core zone, a transition zone, and a peripheral zone. The core zone is the directly affected area centered on the oscillation source, typically including the oscillating turbine and its two to three adjacent turbines. The transition zone is the secondary affected area, including a group of turbines with strong electrical coupling to the core zone. The peripheral zone consists of the remaining turbines less affected by the oscillations. This zoning method enables the rational allocation of control resources and improves suppression efficiency.

[0038] Next, differentiated control strategies are implemented for different regions. The core area, as the location of the oscillation source, is the focus of control. The system implements the most intensive control measures for the wind turbines in the core area, mainly including two aspects: first, adjusting the active power output curve of the oscillation source wind turbine to reduce its disturbance impact on the system; second, optimizing the reactive power control strategy to improve the voltage stability of local nodes. The calculation of control quantities is based on the parameters optimized in step S3, and takes into account the physical constraints of the wind turbine, such as power change rate limits and speed range.

[0039] The wind turbines in the transition zone primarily serve an auxiliary control role, with the goal of coordinating with the control behavior in the core zone to prevent oscillation propagation. This is achieved by adjusting the power response characteristics of these turbines to create an oscillation "buffer zone." The wind turbines in the outer zone maintain normal operation, but their control parameters require appropriate sensitivity adjustments to avoid unintentionally amplifying oscillations. The entire coordination process is implemented through a distributed control network, with wind turbines in different zones exchanging necessary status information via a communication network.

[0040] The system also incorporates real-time monitoring and emergency control mechanisms to continuously track key parameters, including voltage amplitude and phase angle at each node, power flow, and oscillation frequency. When an anomaly is detected, such as oscillation amplitude exceeding a preset threshold or suppression effectiveness falling short of expectations, the emergency control scheme will be triggered. The emergency scheme comprises three levels, ranging from increasing control intensity in the core area to temporarily disconnecting severe oscillation sources when necessary. Each level has clearly defined triggering conditions and execution procedures to ensure the system remains stable even under extreme conditions.

[0041] By employing a hierarchical and zoned differentiated control strategy, combined with real-time monitoring and emergency control mechanisms, the system can effectively suppress oscillations in offshore wind farms, thereby improving their stability and reliability.

[0042] like Figure 2 As shown, in one embodiment of the present invention, the step of using multi-source sensor data from offshore wind farms for time-frequency analysis and dimensionality reduction to obtain a set of key feature quantities includes: Step S1.1: Input the multi-source sensor data into the wavelet transform module to extract the system oscillation frequency components.

[0043] In this step, the system first collects various operational data through a multi-source sensor network at the offshore wind farm. These sensors are mainly distributed at key nodes of the offshore wind farm, including wind turbines, power collection lines, and substations, and can collect various electrical and mechanical parameters such as voltage, current, power, and speed in real time. To ensure the validity of the data, the system uses high-precision sampling devices with a sampling accuracy typically better than 0.5%, and the sampling rate is dynamically adjusted within the range of 10Hz to 100Hz to adapt to the oscillation monitoring needs of different frequency ranges. Simultaneously, the system achieves time synchronization of data across the entire field with a synchronization accuracy better than 1 millisecond, ensuring the time consistency of data collected at different locations and laying the foundation for subsequent correlation analysis.

[0044] The acquired raw data, after preprocessing, is input into the wavelet transform module for time-frequency analysis. Wavelet transform is a powerful signal analysis tool, particularly suitable for processing non-stationary signals. Compared to the traditional Fourier transform, wavelet transform offers multi-resolution analysis capabilities, providing information in both the time and frequency domains simultaneously, making it especially suitable for analyzing complex systems like offshore wind farms that exhibit phenomena across multiple time scales. The system employs discrete wavelet transform, primarily using Daubechies wavelets as basis functions to decompose and reconstruct signal components across different frequency ranges.

[0045] Through wavelet transform, the system can clearly extract various oscillation frequency components from the operational data of offshore wind farms. These frequency components typically include subsynchronous oscillations and inter-regional oscillations (0.1~0.7Hz). For each oscillation frequency component, the system calculates its energy distribution characteristics, temporal variation trend, and phase relationship with other signals. This information provides the foundation for subsequent oscillation mode recognition and feature extraction.

[0046] Step S1.2: Perform principal component analysis using the oscillation frequency components to obtain the main eigenvectors.

[0047] In this step, the system organizes the oscillation frequency components extracted in step S1.1 into a high-dimensional data matrix for Principal Component Analysis (PCA). PCA is a commonly used data dimensionality reduction and feature extraction method that maps high-dimensional data to a lower-dimensional space while preserving the main information of the data. In offshore wind farm oscillation analysis, PCA is particularly suitable for handling complex data from multiple measurement points and multiple frequency bands, and can effectively identify the key modes that have the greatest impact on system oscillations.

[0048] The system constructs a high-dimensional data matrix containing information from all measurement points at various oscillation frequency components. Each row of the matrix represents an observation at a given time point, and each column corresponds to the amplitude or phase information of a measurement point at a specific frequency component. This matrix structure can comprehensively capture the spatiotemporal dynamic characteristics of the system. Before performing principal component analysis, the system standardizes the data to eliminate scale differences between different physical quantities, ensuring that the analysis results are not affected by the choice of units.

[0049] When performing principal component analysis (PCA), the system first calculates the covariance matrix of the data matrix, and then solves for the eigenvalues ​​and eigenvectors of this matrix. Eigenvalues ​​represent the variance contribution of each principal component, while eigenvectors define the orientation of the principal components. The system sorts the principal components according to the magnitude of their eigenvalues, retaining the principal components with the largest contributions as the system's principal eigenvectors. In practical applications, a number of principal components with a cumulative variance contribution rate of 85% to 95% are typically selected. This effectively reduces data dimensionality while preserving the system's main information.

[0050] The obtained main eigenvectors have clear physical meanings, representing different oscillation modes in the system, such as local oscillation modes and inter-regional oscillation modes. Each eigenvector contains the participation level and phase relationship of all measuring points in this mode, providing an important basis for oscillation source localization and control strategy design.

[0051] Step S1.3: Calculate the mutual information between the feature quantities using the main feature vectors to generate the key feature quantity set.

[0052] In this step, the system calculates the mutual information between the features based on the main eigenvectors obtained in step S1.2. Mutual information is an indicator in information theory that measures the degree of interdependence between two random variables. Compared with the traditional correlation coefficient, mutual information can capture the nonlinear relationship between variables and is more suitable for the analysis of complex nonlinear systems such as offshore wind farms.

[0053] When calculating mutual information, the system first treats each element in the principal eigenvectors as a random variable, and then estimates the probability distribution of these variables using kernel density estimation or histogram methods. For each pair of features, the system calculates their joint distribution and marginal distributions, and then calculates the mutual information value. A higher mutual information value indicates a stronger dependency between the two features.

[0054] Based on the mutual information matrix, the system further constructs a dependency network among the features. In this network, each node represents a feature, and the connection strength between nodes is determined by the mutual information value. By analyzing the topology of this network, the system can identify key nodes and key connections, which typically represent features that have a significant impact on system oscillations.

[0055] Finally, by setting a mutual information threshold, feature pairs with high mutual information values ​​are selected. Considering both the physical meaning of the features and prior system knowledge, a final set of key features is generated. This set contains key parameters that effectively characterize the system's oscillations, such as power fluctuations at specific frequencies, voltage change rates, and phase angle differences. Compared to the original high-dimensional data, the set of key features significantly reduces the data dimensionality, improving the efficiency and accuracy of subsequent analysis.

[0056] The generated set of key features is not only used for the next step of oscillation source localization, but also provides important indicators for system operation status monitoring and control strategy evaluation. The system will update this set regularly to adapt to changes in the operating conditions of offshore wind farms, ensuring the accuracy and real-time nature of the analysis results.

[0057] like Figure 3 As shown, in one embodiment of the present invention, the step of inputting the set of key feature quantities into the Lasso regression algorithm and the DeePC framework to establish a nonlinear model of the system and calculate the location of the oscillation source includes: Step S2.1: Input the set of key features into the Hankel matrix construction module and perform Lasso regression to obtain the sparse representation of the system.

[0058] In this step, the set of key features obtained in step S1 is first organized into a format suitable for data-driven modeling. This organization is mainly achieved by constructing a Hankel matrix, a special type of matrix where all elements on each antidiagonal are identical. For offshore wind farm oscillation analysis, the construction of the Hankel matrix follows the characteristics of time-series data, arranging the state observations at consecutive time points according to specific rules to form a data structure containing system dynamic information. The constructed Hankel matrix is ​​typically divided into two parts: a past data matrix and a future data matrix. This structure can effectively capture the system's input-output relationship and state transition characteristics.

[0059] The constructed Hankel matrix is ​​input into the Lasso regression algorithm module for processing. Lasso regression is a linear regression method that can automatically perform feature selection; its core is to add an L1 norm penalty term to the objective function. For offshore wind farm oscillation analysis, Lasso regression minimizes the objective function with a regularization term while simultaneously performing parameter estimation and feature selection. The choice of the regularization parameter is crucial, as it determines the sparsity of the model. Cross-validation is used to automatically optimize the regularization parameter value, obtaining sparser solutions while ensuring the model's prediction accuracy.

[0060] Lasso regression can identify the key factors that have the most significant impact on oscillations, compressing the coefficients of less important factors to zero, thus obtaining a sparse representation of the system. This sparse representation has two important characteristics: first, it reduces model complexity and avoids overfitting; second, it improves model interpretability, clearly showing which features have the main influence on system oscillations. In practical applications, sparse representation typically retains only 10% to 30% of the original features, yet it can still explain more than 85% of the system's variation.

[0061] The results of the sparse representation of the system include the magnitude and sign of the non-zero coefficients. The magnitude of the coefficients indicates the degree of influence of the corresponding feature on the oscillation, while the sign of the coefficients indicates the direction of the influence (positive or negative). This information provides important basis for subsequent perturbation analysis and oscillation source localization. Simultaneously, the sparse representation of the system is also the foundation for constructing nonlinear dynamic models and will be integrated into the DeePC framework for more complex system behavior predictions.

[0062] Step S2.2: Perform perturbation analysis using the sparse representation of the system and calculate the oscillation contribution of each node.

[0063] In this step, perturbation analysis is performed based on the sparse representation of the system obtained in step S2.1. The aim is to quantitatively assess the contribution of each node in the offshore wind farm to the system oscillations. Perturbation analysis is a powerful method for studying system behavior. By applying controlled perturbations to specific nodes of the system and observing the overall system response, it helps to understand the roles and influences of different nodes in the system dynamics.

[0064] The first step in perturbation analysis is to construct a complete system dynamic model. The sparse representation obtained in step S2.1 is embedded into the DeePC (Deep equilibrium Predictive Control) framework. The core idea of ​​the DeePC framework is to directly utilize the system's input and output data to construct a predictive model without explicit system identification. This framework solves a convex optimization problem, searching for the optimal linear combination in the Hankel matrix of historical data to predict the system's future behavior. This method is particularly suitable for handling strongly nonlinear systems such as offshore wind farms, and can capture complex system dynamic characteristics, such as modal coupling and nonlinear damping.

[0065] After the model was built, perturbation analysis was performed. During the analysis, a standardized perturbation signal (usually a unit step or unit pulse) was applied sequentially to each node, and the changes in oscillation-related parameters of the system were observed. These parameters included oscillation amplitude, frequency, and decay rate. Perturbation analysis considered the coupling relationships between nodes and could distinguish between direct and indirect contributions. Direct contributions refer to the impact of changes in the node's own parameters on the system oscillations, while indirect contributions refer to the oscillation effects indirectly caused by coupling relationships with other nodes.

[0066] Based on the disturbance analysis results, the oscillation contribution of each node is calculated. The contribution calculation adopts an energy transfer perspective, viewing system oscillations as a process of energy transfer and conversion between different nodes. Specifically, the calculation method analyzes the propagation path and amplification factor of the disturbance energy within the system, quantifying the proportion of each node's contribution to the total oscillation energy. The contribution calculation also considers the time factor, distinguishing the different roles of the initial oscillation source and the oscillation amplifier.

[0067] The final oscillation contribution of each node is a normalized index with a value range of 0 to 1. A higher value indicates a greater contribution to the system oscillation. The oscillation contribution considers not only the node's electrical characteristics but also its position in the network topology, its connection to other nodes, and its operating status, thus comprehensively reflecting the node's role in the system oscillation. This contribution information is a direct basis for determining the location of the oscillation source and provides important reference for subsequent control strategy design.

[0068] Step S2.3: Perform a positioning operation based on the oscillation contribution to determine the location of the oscillation source.

[0069] In this step, the oscillation source localization calculation is performed based on the oscillation contribution of each node calculated in step S2.2. Oscillation source localization is a crucial step in the entire analysis process, and accurate localization results have a decisive impact on the effectiveness of subsequent control strategies. The localization process is not simply about selecting the node with the largest contribution, but rather about comprehensively considering multiple factors and using various algorithms for cross-verification to ensure the accuracy and reliability of the localization results.

[0070] The first step in oscillation localization is to establish a spatial distribution model of the oscillation contribution. This involves correlating the oscillation contribution of each node with its geographical location and electrical topology, creating a spatial distribution map of the contribution. This visualization helps to intuitively identify potential oscillation source regions. In this spatial distribution, nodes with high oscillation contributions typically form distinct clusters, which are the key areas of focus for oscillation source localization.

[0071] Next, the system applies cluster analysis to analyze the contribution of oscillations. Commonly used algorithms include K-means clustering and hierarchical clustering, which aim to group nodes with similar oscillation characteristics together to identify different oscillation modes and potential sources. Cluster analysis is particularly suitable for handling multi-source oscillations, as it can distinguish between primary and secondary oscillation sources, providing a basis for differentiated control strategies.

[0072] Propagation path analysis was also employed to verify the location of the oscillation source. This method is based on the propagation law of oscillation energy in the system. By analyzing the time delay, phase relationship, and amplitude attenuation of the oscillation signal, the propagation direction and origin point of the oscillation are inferred. Propagation path analysis typically requires high-time-resolution data and precise time synchronization, which plays a crucial role in verifying the location of the oscillation source.

[0073] After integrating the above analysis results, a weighted decision algorithm was used to determine the final location of the oscillation source. The decision-making process considered factors such as the absolute value and relative distribution of the oscillation contribution, time series characteristics, and prior knowledge of the system. For multiple possible oscillation sources, they were ranked according to their contribution magnitude, and the location of the main oscillation source was determined. In some complex cases, the confidence interval of the oscillation source was also calculated to indicate the range of uncertainty in the location result.

[0074] The final determined location of the oscillation source typically includes the specific wind turbine number or collector line segment, along with the associated electrical circuit information. This location information is directly used for subsequent control model establishment and control strategy design, serving as a crucial basis for achieving precise oscillation suppression. Simultaneously, the system also saves intermediate results and judgment criteria from the location process, providing support for subsequent result interpretation and verification.

[0075] In the oscillation source tracing and localization stage, a data sample matrix is ​​first constructed using the key feature quantities extracted in the previous steps. This matrix contains time-series information about the system's operating state; each row represents a snapshot of the state at a given time point, and the column vectors correspond to different feature quantities, such as voltage amplitude, phase angle, active power, and reactive power. This matrix structure can comprehensively describe the dynamic behavior characteristics of the system.

[0076] Next, the Lasso (Least Absolute Shrinkage and Selection Operator) regression algorithm is introduced to perform sparse modeling of the system. The Lasso algorithm automatically selects features by adding an L1 norm penalty term to the objective function, compressing the coefficients of unimportant influencing factors to zero. This process can be expressed as minimizing the objective function:

[0077] Where Y is the system output, X is the characteristic matrix, β is the coefficient to be determined, and λ is the regularization parameter. This method allows us to identify the key factors that have the most significant impact on system oscillations.

[0078] After obtaining the sparse representation of the system, it is embedded into the DeePC framework to construct a nonlinear dynamic model. The core idea of ​​the DeePC framework is to directly utilize the system's input and output data to build a predictive model without explicit system identification. This framework predicts the system's future behavior by solving a convex optimization problem, searching for the optimal linear combination in the Hankel matrix of historical data. This method is particularly suitable for handling strongly nonlinear systems such as offshore wind farms.

[0079] Finally, by analyzing the influence weights and energy transfer characteristics of each node in the model, the contribution of each node to the system oscillation is calculated. Specifically, this can be achieved through perturbation analysis, which involves applying perturbations to individual nodes in the model and observing their impact on the overall system oscillation, thereby quantifying the oscillation contribution of each node. The node with the largest contribution is identified as the oscillation source.

[0080] like Figure 4 As shown, in one embodiment of the present invention, a local control model is established based on the location of the oscillation source. An objective function containing oscillation suppression effects and safety constraints is constructed using the local control model. Bayesian optimization is then used to calculate the objective function to obtain control parameters, including: In step S3.1, a refined local control model is first constructed based on the oscillation source location determined in step S2.3. This model mainly focuses on the oscillation source node and its directly related neighboring nodes, including the electrical coupling relationship, dynamic characteristics, and control response characteristics between nodes. The construction of the local model adopts a combination of lumped parameters and distributed parameters, which considers both the overall characteristics of the offshore wind farm and accurately describes the detailed behavior of local nodes. This model structure can effectively reduce computational complexity while maintaining the accuracy of key node modeling.

[0081] After the local model is established, it is input into the constraint calculation module. The main task of this module is to determine the system stability constraint parameters that must be followed during the control process. The constraint calculation first considers the physical limitations of the equipment, including the power regulation range of the fan, response speed limits, and control parameter change rate limits. These physical constraints are the basic requirements for ensuring the safe operation of the equipment and must be strictly followed.

[0082] In addition to physical limitations of equipment, constraints on grid safety operation are also considered. These constraints include voltage stability margin requirements, frequency deviation limits, and power flow constraints. Grid safety constraints are formulated based on power system safety and stability operation standards, such as permissible voltage deviation of ±5% of rated voltage and frequency deviation limits. These constraints ensure that control actions do not lead to power grid safety issues and are an important safeguard against system instability during the optimization process.

[0083] Special attention is paid to the stability constraints of oscillation source nodes. For identified oscillation sources, the system sets stricter control constraints to ensure that the control behavior effectively suppresses oscillations rather than exacerbates the problem. These constraints are based on the characteristics of the oscillation source and the results of oscillation mode analysis, including amplitude limits for specific frequency components, phase adjustment ranges, and control gain change rate limits. For example, for a wind turbine with a 1.2Hz oscillation, the system may limit its power fluctuation in the 1.0~1.5Hz frequency band to no more than 2% of the rated power.

[0084] All these constraints are ultimately transformed into mathematical form, represented as a safe region in the control parameter space. This safe region is typically defined by a series of inequality constraints, forming the feasible region of the control parameters. The safety barrier function method is used to transform these hard constraints into continuously differentiable penalty terms, facilitating subsequent optimization. The safety barrier function design ensures that the penalty value increases rapidly when the control parameters approach the constraint boundaries, thereby automatically avoiding unsafe regions during the optimization process.

[0085] The generated system stability constraint parameters are not only used for the subsequent construction of the optimization objective function, but also serve as the basis for safety checks during the execution of the control strategy, ensuring that the control behavior is always within the safety boundary and preventing the risk of system instability due to improper control.

[0086] Step S3.2: Construct an optimization objective function using the system stability constraint parameters and oscillation suppression index.

[0087] In this step, the system stability constraint parameters generated in step S3.1 and the predefined oscillation suppression index are used to construct a comprehensive optimization objective function. The design of the objective function directly affects the control effect and requires balancing multiple aspects such as oscillation suppression effect, system safety, and control cost. It is the core link in the entire optimization process.

[0088] The first part of the objective function consists of oscillation suppression performance indicators. These indicators quantify the impact of control actions on system oscillations and mainly include oscillation decay rate, settling time, and oscillation amplitude reduction rate. The oscillation decay rate refers to the decay coefficient when the oscillation amplitude decays exponentially; this value is generally expected to be as large as possible. The settling time is the time required for the system to recover from a disturbance state to a stable state; this is expected to be as short as possible. The oscillation amplitude reduction rate is the relative percentage reduction in oscillation amplitude before and after control; this is expected to be as high as possible. These indicators are combined using a weighted sum to reflect the overall oscillation suppression effect.

[0089] The second part of the objective function is the safety constraint penalty term. Based on the system stability constraint parameters generated in step S3.1, this part uses a safety barrier function method to transform hard constraints into continuously differentiable penalty terms. A typical safety barrier function increases rapidly near the constraint boundary, forming an "invisible barrier" that automatically prevents the optimization process from entering unsafe regions. The system sets a corresponding safety barrier function for each constraint condition and combines them into the total safety constraint penalty term through a weighted sum.

[0090] The third part of the objective function is the control cost indicator. Control cost considers the economics and implementation difficulty of control actions, including factors such as the amount of change in control parameters, the frequency of control actions, and equipment wear and tear. Larger changes in control parameters mean greater adjustments are needed, potentially leading to higher implementation costs and risks; higher control action frequency may cause accelerated equipment wear and shortened lifespan; some control actions may cause equipment to operate at suboptimal operating points, reducing energy conversion efficiency. By rationally setting control cost indicators, overly aggressive control strategies can be avoided, ensuring the economic efficiency and sustainability of control actions.

[0091] The three indicators mentioned above are combined using a weighted sum to form the final optimization objective function. The weight settings reflect the relative importance of different aspects, with oscillation suppression typically having the highest weight, followed by safety constraints, and control costs having the lowest. In certain special cases, such as when the system approaches the instability boundary, the weight of safety constraints may be dynamically adjusted to the highest level to ensure system safety takes priority.

[0092] The final constructed objective function is a high-dimensional nonlinear function of the control parameters. Its input is a vector of control parameters, and its output is a performance index value that comprehensively considers oscillation suppression, safety constraints, and control costs. This objective function will serve as the optimization object for the next step of Bayesian optimization, finding the optimal combination of control parameters that minimizes (or maximizes) the objective function value through iterative search.

[0093] Step S3.3: Input the optimization objective function into the Bayesian optimization iterator to calculate the control parameters.

[0094] In this step, the objective function constructed in step S3.2 is input into the Bayesian optimization iterator, and the optimal control parameters are obtained through iterative search. Bayesian optimization is an efficient global optimization method, particularly suitable for computationally complex optimization problems with high objective function evaluation costs, such as the optimization of control parameters for oscillation suppression in offshore wind farms. Compared with traditional optimization methods, Bayesian optimization can more effectively utilize historical evaluation information, reduce the number of function evaluations, and accelerate convergence.

[0095] The core of Bayesian optimization is to establish a probabilistic model of the objective function using a Gaussian process. A Gaussian process is a powerful nonparametric model that effectively represents the distribution characteristics of a function. First, an appropriate kernel function, such as the radial basis function or the Matérn kernel function, is selected to determine the covariance structure of the Gaussian process. Then, based on the existing evaluation points (i.e., combinations of control parameters and their corresponding objective function values), a posterior distribution model of the objective function is established. This model not only provides the predicted value for each point but also an estimate of the uncertainty of the prediction, providing a basis for subsequent search strategies.

[0096] After establishing the probabilistic model, an acquisition function is used to balance exploration and exploitation, selecting the next evaluation point. Commonly used acquisition functions include Expected Improvement (EP), Probability of Improvement (POI), and Upper Confidence Bound (PEB). These acquisition functions comprehensively consider predicted values ​​and uncertainties, automatically achieving a balance between in-depth searching (exploitation) of known favorable areas and extensive exploration (exploration) of unknown areas. The system defaults to the EP function, but dynamically adjusts the selection and parameters of the acquisition function based on the specific circumstances of the optimization process.

[0097] The iterative optimization process is as follows: First, several initial points are selected in the control parameter space for evaluation to obtain initial objective function values. These initial points are typically generated using methods such as Latin hypercube sampling to ensure uniform coverage of the parameter space. Then, a Gaussian process model is established based on these initial points, the acquisition function value is calculated, and the point with the largest acquisition function value is selected as the next evaluation point. The objective function value is calculated for the new evaluation point, the Gaussian process model is updated, and the acquisition function is recalculated. This process is repeated until a termination condition is met, such as reaching the maximum number of iterations or the improvement margin of multiple consecutive iterations being less than a preset threshold.

[0098] Several techniques were employed to improve optimization efficiency and robustness during the optimization process. First, a multi-starting-point strategy was used, optimizing from different combinations of initial parameters to avoid getting trapped in local optima. Second, an adaptive learning rate was used, dynamically adjusting the search step size based on the convergence of the optimization process, conducting a large-scale exploration in the early stages and fine-tuning in the later stages. Third, a parallel evaluation strategy was implemented, simultaneously evaluating multiple candidate points to improve computational efficiency. Furthermore, an interruption recovery mechanism for the optimization process was implemented, allowing optimization to resume from a saved intermediate state after an interruption, improving the system's practicality.

[0099] Finally, the point with the optimal objective function value is selected from all evaluation points as the final combination of control parameters. This set of parameters considers oscillation suppression, satisfies system safety constraints, and takes control costs into account, thus achieving effective suppression of oscillations in offshore wind farms. The optimized control parameters include the power control gain, response time constant, and filtering parameters of the oscillation source turbine and surrounding turbines. These parameters will be directly used in the subsequent execution of control strategies.

[0100] In the Safe Barrier Bayesian optimization control phase, a local control model needs to be established based on the determined location of the oscillation source. This model mainly focuses on the oscillation source node and its directly related neighboring nodes, including information such as the dynamic characteristics, coupling relationships, and control constraints of these nodes. Establishing a local model can significantly reduce the complexity of control optimization and improve the real-time performance of the algorithm.

[0101] Next, a safe barrier function is constructed. This function ensures that the system always operates within a safe region during the optimization process. The safe barrier function is typically designed as a continuously differentiable function; its value increases rapidly as the system state approaches the safe boundary, thus automatically avoiding unsafe regions during optimization. Typical safety constraints include voltage stability margin, power limit exceedance, and frequency deviation.

[0102] Next, a Bayesian optimization framework incorporating multiple objectives is designed. The objective function consists of two parts: an oscillation suppression performance index (such as the decay rate of oscillation amplitude) and a safety constraint penalty term (given by a safety barrier function). Bayesian optimization establishes a probabilistic model of the objective function through a Gaussian process and uses a sampling function (such as desired improvement) to balance exploration and exploitation, selecting the optimal parameter update direction.

[0103] Finally, an iterative optimization process is performed. In each iteration, the optimal parameter combination is first selected based on the current probability model. Then, the effect of this parameter set is evaluated in the actual system or simulation model, and the probability model is updated. This process continues until a preset convergence condition is met (such as the improvement in the objective function value being less than a threshold, or reaching the maximum number of iterations). During the optimization process, the safety barrier function always ensures that the system operates within the safety boundary, avoiding the risk of system instability that may occur during the optimization process.

[0104] In one embodiment of the present invention, the offshore wind farm is partitioned according to the control parameters, including: Step S4.1: Determine the spatial distribution of the core region, transition region and peripheral region using the location of the oscillation source.

[0105] In this step, based on the oscillation source location determined in step S2.3, the entire offshore wind farm is systematically partitioned. The purpose of partitioning is to achieve a reasonable allocation of control resources, adopt differentiated control strategies for different areas, and improve the accuracy and efficiency of oscillation suppression. The partitioning process divides the offshore wind farm into three functionally distinct areas based on the electrical distance and electromagnetic coupling strength between the nodes and the oscillation source: the core area, the transition area, and the outer area.

[0106] The core area is the directly affected region centered on the oscillation source, typically including the oscillation source fan and its electrically connected adjacent fans. Determining the scope of the core area primarily considers two factors: first, the electrical topology, such as the electrical connections between fans and the routing of the power collection lines; and second, the electromagnetic coupling strength, assessed through calculations of parameters such as the impedance matrix and sensitivity coefficients between nodes. Generally, the core area includes the oscillation source fan and 2-3 directly adjacent fans, which are usually connected to the same power collection line or located on electrically closely coupled adjacent lines.

[0107] The transition zone is a secondary impact area, encompassing a group of wind turbines that have some electrical coupling with the core area. The determination of the transition zone also considers electrical topology and coupling strength, but its scope is larger. A typical transition zone may include wind turbines on several collector lines adjacent to the core area, or areas in the electrical network that are some distance from the core area but still significantly affected by oscillations. The transition zone typically covers 25% to 40% of the entire offshore wind farm, with its exact size depending on the structural characteristics and oscillation propagation properties of the offshore wind farm.

[0108] The outer zone comprises the remaining wind turbines, which are less affected by oscillations. These turbines are typically electrically distant from the oscillation source, and the oscillation signal has significantly attenuated during propagation. The outer zone usually occupies a large portion of the offshore wind farm, covering 50% to 70% of the farm. Although the outer zone is less directly affected by oscillations, it still needs to be considered in control strategies because the control response of the turbines in the outer zone may affect the oscillation source area through grid feedback.

[0109] The determination of spatial distribution considers not only static electrical topology relationships but also dynamic system analysis results. By analyzing the spatial distribution characteristics of oscillation modes, oscillation energy propagation paths, and inter-node coherence, the partition boundaries are further optimized. For example, even if some wind turbines are far from the oscillation source in electrical topology, they may still be classified into the transition zone for focused control if they exhibit high coherence in a specific oscillation mode.

[0110] The final spatial distribution of the core area, transition area, and peripheral area is recorded in the form of turbine numbers or power collection line segments, serving as the basis for subsequent differentiated control strategy design. A visual zoning map is also generated, intuitively displaying the extent and distribution of different areas, facilitating operators' understanding and monitoring of the control strategy's execution.

[0111] Step S4.2: Calculate the control priority and control intensity of the core area, the transition area and the peripheral area based on the spatial distribution.

[0112] In this step, based on the spatial distribution determined in step S4.1, control priorities and control intensities are set for different regions. Control priorities determine the order of control resource allocation, while control intensities determine the magnitude and frequency of adjustments. Properly setting these two parameters is crucial for achieving an efficient differentiated control strategy.

[0113] The core area, as the location of the oscillation source, is assigned the highest control priority. Control priority is typically represented by a value from 1 to 10; the core area's priority is set to 9-10, indicating that under limited resource conditions, the system will prioritize the execution of the core area's control strategy. The control strength of the core area is also set to the highest level, allowing for significant parameter adjustments when necessary. This control strength is manifested in several aspects: firstly, the adjustment range of control parameters; the control parameters of the core area's wind turbines can be adjusted up to the nominal value. Second, response speed: the execution delay of core area control commands must not exceed 100 milliseconds; third, adjustment frequency: control parameters can be adjusted multiple times in a short period of time to cope with rapidly changing oscillations.

[0114] The transition zone is assigned a medium control priority, typically 6-8. The control strength in the transition zone is also correspondingly reduced, and the adjustment range of control parameters is usually limited to the nominal value. Within this range, the response latency requirement is relaxed to 200-300 milliseconds, and the adjustment frequency is also lower than that of the core region. The main purpose of transition region control is to assist the core region in suppressing oscillations and preventing oscillations from spreading to a wider area. Therefore, the control strategy is more conservative to avoid introducing new unstable factors.

[0115] The outer perimeter has the lowest control priority, typically 3-5. Outer perimeter control employs the most conservative strategy, limiting control parameter adjustments to nominal values. Within this range, the response latency is acceptable up to 500 milliseconds, and the adjustment frequency is also minimal. The main purpose of peripheral zone control is to maintain the overall system stability and avoid unintentional amplification of oscillations by peripheral zone fans. Therefore, control operations are mainly preventative, such as appropriately adjusting the sensitivity of the control gain, rather than active intervention.

[0116] The calculation of control priority and control strength also takes into account the dynamic changes in system operating conditions. The system periodically evaluates the oscillation suppression effect and dynamically adjusts the control parameters of different regions based on the actual results. For example, if the control effect in the core region is found to be unsatisfactory, the system may increase the control priority and control strength in the transition region and expand the control range; conversely, if the oscillation is effectively suppressed, the control strength may be gradually reduced to minimize interference with normal operation.

[0117] The calculated control priorities and control strengths are stored in the form of numerical matrices, with each wind turbine corresponding to a set of parameter values. These parameters directly affect the generation and execution of subsequent control commands, ensuring that the system can implement targeted and differentiated control strategies based on the location of the oscillation source.

[0118] Step S4.3: Input the control priority and the control strength into the partition controller to generate partition control instructions.

[0119] In this step, the control priority and control strength calculated in step S4.2 are input into the zone controller to generate specific zone control commands. The zone controller is a specially designed control module that can automatically generate targeted control commands based on the characteristics and control requirements of different areas, thereby achieving coordinated control of offshore wind farms.

[0120] The zone controller first loads the control parameters obtained in step S3. These parameters include optimized baseline control values, such as power control gain, response time constant, and filtering parameters. Then, the controller adjusts the baseline control parameters based on the control priority and control intensity of each zone, generating zone-specific control commands. This adjustment follows the principle of "central focus, peripheral support," ensuring the rational allocation and efficient use of control resources.

[0121] For the core area, the controller generates the most powerful control commands. These commands typically include: first, active power regulation commands, adjusting the active power output curves of the oscillating source fan and surrounding fans to reduce disturbances to the system; second, reactive power control commands, optimizing the voltage stability of local nodes; and third, control parameter adjustment commands, such as adjusting the PID controller parameters of the fan converter and the power angle control gain. Control commands for the core area have the highest execution priority and are transmitted through a dedicated control channel to ensure minimal communication latency and maximum execution reliability.

[0122] For the transition zone, the controller generates auxiliary control commands. These commands mainly include: first, power response characteristic adjustment commands, such as adjusting the speed and depth of the wind turbine's response to system frequency changes; second, damping control commands, enhancing the wind turbine's natural damping capability against system oscillations; and third, coordination control commands, ensuring that the control behavior of the wind turbine in the transition zone synergizes with that in the core zone, rather than canceling each other out. The control commands for the transition zone have a lower priority than those for the core zone, but will still be guaranteed when communication resources are strained.

[0123] For the peripheral zone, the controller generates preventative control commands. These commands are primarily parameter fine-tuning types, such as appropriately reducing the sensitivity of the wind turbine control system to specific frequency disturbances to avoid unintentionally amplifying oscillation signals; adjusting the wind turbine's operating point to increase stability margin; and activating special anti-oscillation function modules when necessary. Control commands for the peripheral zone have the lowest priority and may be delayed in execution when system resources are strained.

[0124] The zone controller also considers the timing coordination of control commands. Since the control behaviors of different zones may influence each other, the controller optimizes the command execution timing to ensure that the control actions of different zones can create a positive synergistic effect. Generally, control commands in the core zone are executed before those in other zones, providing the most direct oscillation suppression effect for the system; followed by control commands in the transition zone, which complement and enhance the control of the core zone; and finally, control commands in the peripheral zone, providing a stable foundation for the overall system.

[0125] The generated zone control commands are stored in a standardized format, containing information such as target equipment, control type, control value, and execution time. These commands are distributed to each wind turbine actuator via the offshore wind farm's SCADA system or a dedicated control network, enabling differentiated control of different areas. The system also records the generation time, content, and target equipment of all control commands, facilitating subsequent performance evaluation and fault analysis.

[0126] In one embodiment of the present invention, the method further includes: implementing differentiated control strategies for different regions based on the control parameters to achieve oscillation suppression, including: Step S4.4: Collect the operating status data of the core area wind turbines and calculate the power adjustment range.

[0127] In this step, detailed operational status data of the wind turbines in the core area are first collected in real time through the offshore wind farm monitoring network. This data is the foundation for precise control and includes two main categories: mechanical status parameters and electrical status parameters. Mechanical status parameters mainly include rotor speed, pitch angle, hub wind speed, and tower vibration; electrical status parameters include generator speed, stator current, power output, and converter status. This data is collected in real time at a high sampling rate (typically above 10Hz), preprocessed, and then transmitted to the control center for analysis.

[0128] During data acquisition, special attention is paid to key oscillation-related indicators, such as active power fluctuation amplitude, reactive power change rate, and voltage fluctuation characteristics. The system uses a specialized oscillation monitoring algorithm to dynamically track and extract features from these parameters, enabling real-time monitoring of the oscillation status of the wind turbines in the core area. Simultaneously, the system also collects information on the wind turbine's operating mode and environmental conditions, such as whether it is currently operating at rated power, whether the wind conditions are stable, and whether it is affected by wake ducts. This information is crucial for accurately assessing the wind turbine's adjustability.

[0129] Based on the collected operational status data, the power adjustment range of the wind turbines in the core area is calculated. The power adjustment range refers to the maximum adjustment range of active and reactive power that the wind turbines can make while ensuring safe and stable operation. The following key factors are considered in the calculation process: First, there are limitations imposed by the physical characteristics of the fans. Different models and capacities of fans have different power regulation capabilities. The system calculates the maximum allowable adjustment range of the mechanical and electrical systems based on the fan model parameters and the current operating status. For example, for variable speed constant frequency fans, the adjustable range varies greatly at different wind speeds; in high wind speed areas, the fan usually operates at its rated power, with a large downward adjustment range but almost no upward adjustment range; in medium wind speed areas, the fan has bidirectional adjustment capabilities.

[0130] Secondly, there are grid constraints. Offshore wind farms operating in conjunction with the grid must meet various grid dispatch requirements, such as voltage level control and power factor requirements. The system determines the power regulation constraints that meet grid requirements based on real-time collected grid status data and dispatch instructions. For example, if the grid voltage is already close to its upper limit, the wind turbine's ability to increase reactive power output will be limited; if the grid is operating under high load, it may require the offshore wind farm to maintain maximum active power output, limiting the scope for downward adjustment.

[0131] Furthermore, the dynamic response characteristics of the wind turbines are also considered. Different wind turbines have varying response speeds and accuracy to control commands, which directly affects their practical usability in oscillation suppression. The system analyzes historical control response data to establish a dynamic response model for the wind turbines and evaluate their actual regulation capabilities at different time scales. For example, some wind turbines may be able to achieve large-scale power regulation, but their response time is long, making them unsuitable for suppressing rapid oscillations; while other wind turbines, although having limited regulation ranges, have rapid responses and are more suitable for direct participation in oscillation control.

[0132] Third, taking into account the above factors, the available power adjustment range for each core area wind turbine is calculated, including the active power adjustment range and the reactive power adjustment range. These adjustment ranges are expressed in the form of power values ​​or percentages, serving as important input parameters for subsequent optimized control strategies. The confidence level of the adjustment ranges is also evaluated to reflect the reliability of the current calculation results, in order to address uncertainties in the operation of offshore wind farms.

[0133] Step S4.5: Input the power adjustment range into the curve optimizer to generate the active power output curve of the wind turbine.

[0134] In this step, the power regulation range calculated in step S4.4 is input into the curve optimizer, with the goal of generating the optimal active power output curve for the wind turbine. The active power output curve refers to the sequence of active power setpoints of the wind turbine over a period of time, which directly determines the wind turbine's effectiveness in suppressing system oscillations and its impact on the power grid. The curve optimizer is a specially designed optimization module capable of generating the optimal power curve that meets oscillation suppression requirements while considering various constraints.

[0135] The curve optimization process begins by defining the optimization objective. For oscillation suppression control, the primary optimization objective is to maximize the damping effect at a specific oscillation frequency. Based on the oscillation characteristics identified in previous steps, particularly the oscillation frequency and oscillation mode, the system sets a targeted optimization objective function. For example, for a 1.2Hz motor-grid oscillation, the system will focus on optimizing the power response characteristics of the wind turbine near that frequency; for a 0.5Hz inter-regional oscillation, a power regulation strategy over a longer timescale needs to be considered.

[0136] Next, the curve optimizer, combined with the dynamic characteristic model of the wind turbine, constructs a model relating power output to system oscillation response. This model describes the mechanism by which changes in wind turbine power affect system oscillation, including direct effects (such as those through the wind turbine's own damping) and indirect effects (such as those through the power grid structure affecting other equipment). Model construction typically employs a combination of simplified linear analysis methods and detailed nonlinear time-domain simulations, ensuring both computational efficiency and accurate representation of key nonlinear characteristics.

[0137] Based on the established model, the curve optimizer begins searching for the optimal power output curve. The optimization process considers multiple constraints, including: first, the power adjustment range constraint calculated in step S4.4, ensuring that the generated power curve is within the physically feasible range of the wind turbine; second, the wind turbine dynamic response constraint, considering the limitations of the wind turbine's response speed and accuracy; third, grid security constraints, ensuring that power adjustment will not lead to grid security problems; and fourth, oscillation mode constraints, designing the most effective suppression strategy for specific oscillation modes.

[0138] Optimization algorithms typically employ model predictive control (MPC) or rolling time-domain optimization. The system first predicts the oscillation trend over a relatively long time window (e.g., 10-30 seconds), then calculates the optimal power curve within this window. In actual execution, the system only applies the most recent few control points (e.g., 1-3 seconds), and then continuously updates the prediction and optimization results over time, achieving rolling optimization control. This method effectively addresses the uncertainties and rapid changes in offshore wind farm operation.

[0139] For the different wind turbines in the core area, the curve optimizer also considers the coordinated control problem. Since multiple wind turbines in the core area jointly affect system oscillations, the optimization process needs to comprehensively consider their joint effects, rather than simply optimizing independently. The system constructs a multi-wind-turbine joint optimization model to calculate the optimal power allocation strategy, achieving coordinated control among different wind turbines. For example, in some cases, it may be necessary for some wind turbines to increase power output while others reduce power output, creating a mutually compensating effect to maximize oscillation suppression.

[0140] The final generated active power output curve of the wind turbine is a time series data, representing the change in the setpoint of the wind turbine's active power over a future period. This curve considers both the need for oscillation suppression and compliance with the physical constraints of the wind turbine and the safety requirements of the power grid, serving as the direct basis for subsequent control execution. The system also generates a confidence assessment of the curve, reflecting the reliability level of prediction and control at different time points, providing decision-making reference for operators.

[0141] Step S4.6: Adjust the reactive power control parameters of the core area wind turbine according to the active power output curve to achieve oscillation suppression.

[0142] In this step, based on the active power output curve generated in step S4.5, the reactive power control parameters of the core area wind turbines are further adjusted to form a comprehensive suppression strategy that coordinates active and reactive power control. Reactive power control is of particular importance in oscillation suppression, especially for voltage-related oscillation modes, where reasonable reactive power adjustment can often provide a more direct and effective suppression effect.

[0143] The regulation process first analyzes the impact of active power adjustments on system voltage and power balance. Changes in active power will cause changes in system voltage level, power flow direction, and power factor, which may require compensation through reactive power regulation. The system assesses the voltage fluctuations and power factor changes that the active power output curve may cause through power flow analysis and voltage sensitivity calculations, and determines the required reactive power compensation.

[0144] Next, a targeted reactive power control strategy is designed based on the oscillation characteristics. Different types of oscillations require different reactive power control methods. For mechanical oscillation modes (such as oscillations between wind turbines), reactive power control mainly plays an auxiliary role, assisting active power control in providing necessary system damping; for electrical oscillation modes (such as voltage oscillations), reactive power control may become the primary means, suppressing the development of oscillations by directly adjusting system voltage and reactive power flow.

[0145] To achieve effective reactive power control, the system needs to adjust several reactive power control parameters of the wind turbine. These parameters include: first, the reactive power control mode, such as constant power factor mode, constant reactive power mode, or voltage control mode; second, the control setpoints, such as target power factor, target reactive power output, or target voltage level; and third, the control response characteristics, such as control gain, response time constant, and dead zone setting. The system optimizes these parameter settings based on oscillation suppression requirements and wind turbine characteristics to achieve the best reactive power control effect.

[0146] Adjusting reactive power control parameters also requires considering the capacity limitations of the wind turbine converter. The total capacity of the wind turbine converter is finite, and active and reactive power outputs need to be coordinated and distributed within the converter's capacity. Based on the wind turbine type and converter specifications, the system calculates the maximum available reactive power capacity at a given active power level, ensuring that reactive power control commands do not exceed the equipment's physical limitations. Simultaneously, the system also evaluates the converter's dynamic response characteristics, such as reactive power ramp-up limits and overload capacity, to ensure that control commands are also feasible in terms of time.

[0147] In actual oscillation suppression control, the system simultaneously issues active power control commands and reactive power control commands to achieve coordinated control. Timing coordination between active and reactive power control is crucial; the system typically employs a unified control frame structure to ensure the synchronous execution of both types of control commands. In some cases, the system may also dynamically adjust the weights of active and reactive power control based on the real-time characteristics of the oscillation, flexibly switching control strategies at different oscillation stages to achieve the best suppression effect.

[0148] Finally, the oscillation suppression effect and grid conditions are continuously monitored, and the control strategy is adjusted in real time as necessary. If the initial control effect is found to be unsatisfactory, or if the system state changes significantly, the system will recalculate the active and reactive power control parameters and update the control strategy. This closed-loop control mechanism ensures the effectiveness and adaptability of oscillation suppression and can cope with various complex situations in the operation of offshore wind farms.

[0149] In one embodiment of the present invention, the method further includes: implementing differentiated control strategies for different regions based on the control parameters to achieve oscillation suppression, including: Step S4.7: Collect the status data of the wind turbines in the transition zone and the outer zone, and calculate the power balance parameters.

[0150] In this step, status data of wind turbines in the transition and peripheral areas are collected through the offshore wind farm monitoring network to establish operational models for these areas, providing a foundation for subsequent coordinated control. Similar to the core area, this data includes mechanical and electrical status parameters of the wind turbines, but the sampling rate and data accuracy requirements may be reduced to lessen the burden on the communication system.

[0151] State data acquisition focuses particularly on indicators related to overall system balance, such as total power output, power distribution, power reserve, and power change trends in each region. These data reflect the power contribution of the transition zone and outer perimeter areas within the entire offshore wind farm and are crucial for calculating power balance parameters. Simultaneously, the system also monitors the interaction between wind turbines and oscillations, such as the oscillation response characteristics of turbines in different regions and their amplification or suppression effects, to assess the sensitivity of each region to oscillations under the current system state.

[0152] Power balance parameters are a series of indicators describing the power distribution and coordination relationships between different areas of an offshore wind farm. The calculation process first requires clarifying the power balance objectives between regions. In oscillation suppression control, it is generally desirable that power adjustments (especially reductions) in the core area be appropriately compensated by other areas to maintain the overall power output of the offshore wind farm and reduce disturbances to the power grid. Therefore, the system needs to evaluate the power regulation potential and response characteristics of the transition zone and the outer perimeter.

[0153] The calculation of power balance parameters takes into account several factors: First, regional power reserve, which is the amount of additional power that each region's wind turbines can provide under current operating conditions, depending on the wind turbine's operating status and environmental conditions; second, regional response speed, which is the time required for the wind turbine to adjust power from receiving control commands, affecting the dynamic coordination of power balance; third, regional control priority, which is the control priority and control intensity of each region determined in step S4.2, determining the power adjustment allocation strategy; and fourth, regional oscillation sensitivity, which is the degree of influence of each region's wind turbines on system oscillation, determining the effectiveness and risk of power adjustment.

[0154] Based on the above factors, a series of power balance parameters were calculated, including: regional power adjustment ratio coefficient, which represents the proportion that each region should bear when it is necessary to compensate for power changes in the core area; regional power response time constant, which describes the dynamic characteristics of power adjustment in each region; upper and lower limits of regional power adjustment, which limit the maximum magnitude of power adjustment in each region; regional power gradient limit, which ensures that power adjustment is within a safe rate of change; and inter-regional coordination coefficient, which describes the interaction between power adjustments in different regions.

[0155] The calculation of power balance parameters also needs to consider the constraints of grid security and stability. The system uses simplified power flow analysis to assess the impact of different power allocation schemes on the grid, ensuring that power adjustments do not lead to safety issues such as line overload or voltage exceeding limits. In some cases, it may be necessary to adjust the calculated power balance parameters to meet the rigid constraints of grid security.

[0156] The final calculated power balance parameters are represented in matrix or vector form, describing the power coordination relationship between different regions during the oscillation suppression process. These parameters will be used in the subsequent design of control strategies for the transition and peripheral regions to ensure that the overall power balance of the system is maintained while suppressing oscillations, thereby reducing the impact on the power grid.

[0157] Step S4.8: Adjust the power response characteristics of the transition zone fan using the power balance parameters to construct the oscillation attenuation region of the transition zone fan.

[0158] In this step, based on the power balance parameters calculated in step S4.7, the power response characteristics of the wind turbine in the transition zone are adjusted to construct an effective oscillation attenuation area. As the intermediate zone connecting the core area and the outer area, the control strategy of the transition zone needs to both suppress oscillations in conjunction with the core area and prevent oscillations from spreading to the outer area, thus playing the role of a "buffer zone".

[0159] The adjustment process requires first clarifying the functional positioning and control objectives of the transition zone. The main control objectives of the transition zone include: first, coordinating with the control behavior of the core zone to provide a synergistic oscillation suppression effect; second, preventing the diffusion of oscillation energy to the peripheral zone, forming an effective oscillation isolation zone; third, appropriately compensating for the overall power change caused by the power adjustment in the core zone, maintaining system power balance; and fourth, maintaining its own stable operation without generating new oscillation sources. Based on these objectives, the system has designed a special power response characteristic adjustment strategy for the transition zone.

[0160] Adjusting the power response characteristics primarily targets key parameters of the wind turbine control system. These parameters include: the gain coefficient of the power controller, which determines the strength of the wind turbine's response to the input signal; the time constant of the controller, which affects the speed and smoothness of the control response; the parameter settings of the power filter, which determine the wind turbine's response characteristics to different frequency components; the threshold setting of the power limiter, which limits the maximum amplitude of power variation; and the setting of the power change rate limiter, which controls the speed of power adjustment. The system optimizes these parameter settings based on oscillation characteristics and power balance requirements to achieve precise control of the wind turbine in the transition zone.

[0161] For frequency-related characteristic adjustments, the system employs targeted measures. For identified oscillation frequencies (such as 0.8Hz electromechanical oscillations), the system adjusts the characteristics of the power filter to achieve a selective response of the transition zone fan to that frequency component. For example, a specific band-stop filter can be designed to weaken the fan's response to the oscillation frequency, preventing unintentional amplification of oscillations; or a phase compensation circuit can be designed to dampen the fan's power response at a specific frequency, actively suppressing oscillations. These frequency characteristic adjustments are designed in conjunction with the core zone control strategy to ensure that the control behaviors of the two zones have a positive superposition effect, rather than canceling each other out.

[0162] Regarding power balance, the power setpoint and response mode of the transition zone fans are adjusted based on the power balance parameters calculated in step S4.7. If power reduction control is implemented in the core zone, the system may instruct the transition zone to appropriately increase power output to maintain overall power balance; conversely, the system may instruct the transition zone to increase power output appropriately. This power coordination is typically achieved by adjusting the power-frequency characteristic curve or power control dead zone of the transition zone fans, which satisfies power balance requirements without negatively impacting oscillations.

[0163] Constructing the oscillation attenuation zone is a systematic project that requires consideration of the location and characteristics of different wind turbines within the transition zone. The system typically begins at the boundary between the transition zone and the core zone, gradually building the oscillation attenuation zone towards the outer perimeter. Wind turbines closer to the core zone experience larger adjustment amplitudes and require more aggressive control strategies, while those closer to the outer perimeter experience smaller adjustment amplitudes and require more conservative control strategies. This gradual distribution of control intensity creates a smooth oscillation attenuation effect, avoiding the generation of new instabilities at the zone boundaries.

[0164] The final constructed oscillation attenuation region is a specialized area with a clearly defined function and coordinated control. It effectively blocks the propagation of oscillation energy, assists the core region in suppressing oscillation sources, and maintains the system's power balance. The system also incorporates an adaptive adjustment mechanism for this region, which dynamically adjusts the power response characteristics based on the oscillation suppression effect and changes in system state to maintain optimal oscillation attenuation.

[0165] Step S4.9: Input the response characteristics of the oscillation attenuation region into the parameter tuner to generate the control gain coefficient of the peripheral area fan for suppressing its sensitivity to oscillation.

[0166] In this step, based on the response characteristics of the oscillation attenuation region constructed in step S4.8, the control parameters of the peripheral zone fans are further adjusted. The main goal is to suppress the sensitivity of the peripheral zone fans to system oscillations and prevent the oscillations from being amplified and propagated over a wider area. Although the peripheral zone, as the outermost control area, is less directly affected by oscillations, improper setting of its control parameters may unintentionally amplify the oscillations. Therefore, targeted parameter tuning is required.

[0167] The parameter tuning process first analyzes the impact of the response characteristics of the oscillation attenuation region on the peripheral region. The oscillation attenuation region suppresses oscillation propagation through specific power response characteristics, but this response may lead to changes in power flow and voltage distribution, thus affecting the wind turbines in the peripheral region. The system assesses the nature and extent of this impact through simulation analysis or real-time data observation, providing a reference for the parameter tuning of the peripheral region. Particular attention is paid to the attenuation characteristics of the oscillation at the region boundary, such as the oscillation amplitude attenuation rate and phase change, as these characteristics directly determine the oscillation characteristics that the peripheral region needs to address.

[0168] The parameter tuner is a specially designed functional module that automatically calculates the optimal control gain coefficient based on system status and control requirements. The tuner employs a combination of model-based and data-based methods, considering both the theoretical characteristics of the system and the statistical patterns of historical operating data. For peripheral zone fans, the focus of tuning is on the control gain coefficient, including the proportional gain, integral gain, and derivative gain of the power controller (corresponding to the three basic parameters of a PID controller), as well as the gain coefficients of various feedforward and feedback compensation components.

[0169] The optimization objective of the control gain coefficient is to reduce the sensitivity of the peripheral area wind turbine to the oscillation frequency. The system uses frequency domain analysis to calculate the response characteristics of the peripheral area wind turbine at the oscillation frequency and designs appropriate gain adjustment strategies to minimize the turbine's response at that frequency. Typical methods include: reducing the proportional gain to decrease the system's amplification effect on disturbances; increasing the damping effect of the derivative element to suppress high-frequency oscillations; adjusting the time constant of the integral element to avoid resonance at the oscillation frequency; and introducing a band-notch filter to specifically suppress the response at specific frequencies.

[0170] Parameter tuning also needs to consider the normal operating requirements of wind turbines in the surrounding area. Excessively reducing the control gain may affect the normal regulation performance of the wind turbines, such as reducing the tracking accuracy of wind speed changes or weakening the response to grid frequency fluctuations. Therefore, the tuning process needs to find a balance between oscillation suppression and normal operation. The system typically employs a multi-objective optimization method, simultaneously considering multiple indicators such as oscillation suppression effect, power point tracking accuracy, and energy conversion efficiency to find the optimal trade-off.

[0171] To adapt to changing operating conditions in offshore wind farms, the parameter tuner is designed with an adaptive adjustment mechanism. The system continuously monitors the operating status and oscillation response of the peripheral area. When a change in system status is detected (such as changes in wind conditions or grid impedance), the parameter resetting process is automatically triggered to update the control gain coefficient. This adaptive mechanism ensures the timeliness and adaptability of the peripheral area control strategy, maintaining good oscillation suppression under different operating conditions.

[0172] The final generated control gain coefficients are represented in the form of a parameter matrix, with each peripheral wind turbine corresponding to a specific set of control gain values. These parameters are distributed to the controllers of each wind turbine through the offshore wind farm control network, enabling precise adjustment of the wind turbine control system. The system also records the parameter tuning process and results, providing a basis for subsequent performance evaluation and experience accumulation.

[0173] In the active suppression strategy execution phase, a hierarchical and zoned control strategy is formulated based on optimized control parameters. This strategy divides the offshore wind farm into three levels according to the degree of oscillation impact: a core zone, a transition zone, and a peripheral zone. The core zone is the directly affected area centered on the oscillation source, typically including the oscillation source turbine and its two to three adjacent turbines. The transition zone is the secondary affected area, including a group of turbines with strong electrical coupling to the core zone. The peripheral zone comprises the remaining turbines less affected by oscillations. This zoning method enables the rational allocation of control resources and improves suppression efficiency.

[0174] In the oscillation detection process, pattern recognition is implemented using a Support Vector Machine (SVM) algorithm. A radial basis function (RBF) kernel is chosen, which effectively handles the nonlinear characteristics in offshore wind farm oscillation data. A high penalty parameter is set to reduce misclassification, and the γ parameter of the RDF is optimized through grid search to balance the model's fitting and generalization abilities. The model training data comes from typical oscillation events recorded from offshore wind farms over the past two years. This data is labeled and includes various state classifications such as normal operating conditions, slight oscillations, moderate oscillations, and severe oscillations. Each sample contains multi-dimensional feature data within a specific time window, covering the fluctuation characteristics of key parameters such as voltage, power, and frequency at different frequency bands.

[0175] To enhance the reliability of the detection algorithm, cross-validation was used to evaluate the model's performance. Validation results show that the model can achieve high accuracy in various oscillation scenarios. For special abnormal oscillation patterns, the system combines a single-class support vector machine and a local outlier factor algorithm to construct a two-layer detection mechanism. The former is used to identify known types of oscillation patterns, while the latter is specifically responsible for discovering newly emerging unknown oscillation types. The anomaly threshold is set based on the statistical distribution characteristics of historical data, ensuring both sensitivity to anomalies and avoiding excessive false alarms.

[0176] In real-time application scenarios, considering computational efficiency requirements, a fast detection model based on a lightweight convolutional neural network was also developed. This network employs a multi-layered structure, taking multi-channel time-series data within a fixed time window as input. Convolutional layers extract time-domain and frequency-domain features, pooling layers reduce feature dimensionality while retaining key information, and finally, fully connected layers classify and determine the oscillation state. Network training utilizes batch normalization and dropout techniques to improve generalization ability, with cross-entropy as the loss function, the Adam algorithm as the optimizer, and an adaptive learning rate adjustment strategy. During training, balanced sampling techniques are used for oscillation samples of varying severity to address the data imbalance problem. This lightweight model has low computational complexity, enabling inference to be completed in milliseconds, meeting the response requirements of real-time control systems for offshore wind farms.

[0177] A model update mechanism was also designed to periodically retrain the model based on newly added oscillation event data, enabling the detection algorithm to adapt to long-term changes in offshore wind farm operating conditions and oscillation characteristics. Updates are typically made quarterly, or triggered immediately upon the system detecting a new oscillation pattern. To ensure a smooth transition between model updates, the system employs a strategy of running the old and new models in parallel, only officially replacing the old model when its performance on the validation dataset surpasses that of the old model. This gradual update mechanism ensures the continuous effectiveness and stability of the oscillation detection system, enabling it to cope with various changes and challenges in the offshore wind farm operating environment.

[0178] The core of the entire suppression strategy lies in the focused control of the oscillation source region. Within this core region, the system prioritizes adjusting the control parameters of the oscillating source turbine, including torque control gain and pitch angle control parameters. The control methods mainly involve two aspects: first, adjusting the active power output curve of the oscillating source turbine to reduce its disturbance impact on the system; and second, optimizing the reactive power control strategy to improve the voltage stability of local nodes. The specific control quantity calculations are based on the aforementioned optimized parameters and consider the physical constraints of the turbine, such as power change rate limits and speed range.

[0179] An adaptive coordinated control method was employed to coordinate the response strategies of wind turbines in other areas. Turbines in the transition zone primarily played an auxiliary control role, aiming to coordinate with the control behavior of the core zone and prevent oscillation propagation. This was achieved by adjusting the power response characteristics of these turbines to create an oscillation "buffer zone." Wind turbines in the outer zones maintained normal operation, but their control parameter sensitivity needed appropriate adjustments to avoid unintentionally amplifying oscillations. The entire coordination process was implemented through a distributed control network, with wind turbines in different areas exchanging necessary status information via a communication network.

[0180] Finally, a real-time monitoring and emergency control mechanism was established. The monitoring system continuously tracks key parameters, including voltage amplitude and phase angle at each node, power flow, and oscillation frequency. When an anomaly is detected, such as oscillation amplitude exceeding a preset threshold or suppression effect falling short of expectations, the emergency control plan will be triggered. The emergency plan includes three levels: Level 1 increases the control intensity in the core area; Level 2 expands the control range, including some transition zone wind turbines within the key control range; Level 3 activates the safety protection mechanism, and, if necessary, briefly disconnects the source of severe oscillation. Each level has clearly defined triggering conditions and execution procedures to ensure that the system can maintain stable operation even under extreme conditions.

[0181] To ensure the reliable execution of the control strategy, a real-time evaluation mechanism has been established. This mechanism evaluates the control effect by calculating oscillation suppression performance indicators (such as oscillation decay rate and settling time) and dynamically adjusts the control strategy based on the evaluation results. This closed-loop feedback mechanism ensures the adaptability and robustness of the entire suppression system, enabling it to cope with various uncertainties in the operation of offshore wind farms.

[0182] In one embodiment of the present invention, the method further includes: Step S5: Collect voltage, power, and oscillation frequency component data for each node of the offshore wind farm.

[0183] In this step, operational data from key nodes are continuously collected through the offshore wind farm's monitoring network to establish a real-time monitoring database, providing a basis for the evaluation and adjustment of oscillation suppression control. The purpose of data acquisition is to comprehensively understand the operational status and oscillation development of the offshore wind farm, especially the system response characteristics after the implementation of control strategies.

[0184] Voltage data acquisition covers multiple levels of offshore wind farms, including turbine terminal voltage, collector node voltage, substation bus voltage, and grid connection point voltage. The acquired voltage parameters include not only amplitude but also phase angle, frequency, and harmonic components. This data is obtained through voltage transformers and electronic voltage measurement devices. The sampling rate is dynamically adjusted according to the monitoring objective; for oscillation monitoring, it is typically maintained at 10-20 sampling points per second to ensure accurate capture of the changing characteristics of the main oscillation frequencies (usually within the range of 0.1-5Hz).

[0185] Power data acquisition also covers multiple levels, focusing on single-unit power output, collector line power, substation power, and grid-connected power exchange. The acquired power parameters include active and reactive power, as well as auxiliary information such as power factor and power direction. Power data is obtained through power measurement devices, such as the power metering module built into the wind turbine, current and voltage transformer combinations on the collector lines, and dedicated power analyzers. The sampling rate of power data is consistent with that of voltage data to ensure time synchronization between the two types of data, facilitating subsequent correlation analysis.

[0186] The acquisition of oscillation frequency component data relies on specialized signal processing techniques. The system extracts oscillation-related frequency components by performing real-time spectral analysis on the raw voltage and power data. Commonly used analysis methods include Fast Fourier Transform (FFT), wavelet transform, and mode decomposition. The extracted oscillation frequency component data includes information such as amplitude, phase, and damping ratio at each frequency point, which directly reflects the characteristics and intensity of the oscillation modes present in the system.

[0187] Special emphasis is placed on data quality control during the data acquisition process. The system employs multiple verification mechanisms, including sensor self-diagnosis, data rationality checks, and redundant measurement comparisons, to ensure the accuracy and reliability of the acquired data. For potential data loss or anomalies, the system includes data repair and replacement mechanisms, such as interpolation and data inference, to guarantee the continuity and integrity of the monitoring data.

[0188] All collected data is precisely timestamped, with time synchronization accuracy typically better than 1 millisecond. This high-precision time synchronization is achieved through GPS timing or Network Time Protocol (NTP), ensuring time consistency of data from different locations and devices. Accurate time information is crucial for tasks such as analyzing oscillation propagation paths, calculating phase relationships, and evaluating control delays.

[0189] Finally, the collected data is organized into a structured database, supporting efficient data querying and analysis. The database also stores historical data, facilitating trend analysis and comparative studies. This real-time and historical data together form the data foundation for offshore wind farm oscillation suppression control, providing comprehensive information support for subsequent monitoring calculations and control assessments.

[0190] Step S6: Perform real-time monitoring calculations using the voltage, power, and oscillation frequency component data to obtain real-time monitoring calculation results.

[0191] In this step, based on the voltage, power, and oscillation frequency component data collected in step S5, a series of real-time monitoring calculations are performed to generate comprehensive monitoring results reflecting the current state of the system. These calculations aim to evaluate the effectiveness of oscillation suppression control, provide early warning of potential instability risks, and offer a basis for possible adjustments to control strategies.

[0192] Real-time monitoring and calculation first focus on the changing trends of oscillation characteristics. The system performs time-series analysis on the oscillation frequency component data, calculating the rate of change of oscillation amplitude, the dynamic characteristics of the damping ratio, and the drift of the oscillation frequency. These indicators directly reflect the development trend of the oscillation, such as whether the oscillation is gradually weakening (positive damping), remaining stable (zero damping), or gradually strengthening (negative damping). The system typically uses methods such as exponential fitting or autoregressive moving average (ARMA) models to accurately extract the attenuation characteristics of the oscillation from the noise background and evaluate the dynamic stability of the system.

[0193] Calculate the stability margin indices for critical nodes. Stability margin is the degree to which a system moves away from the instability boundary and is an important indicator for assessing system safety. Commonly used stability margin indices include voltage stability margin (such as the margin coefficient of the power-voltage curve), frequency stability margin (such as the distance between the frequency deviation and the safety limit), and damping margin (such as the difference between the damping ratio of the oscillation mode and the minimum requirement). These margin indices are usually expressed as percentages or safety factors, intuitively reflecting the safety status of the system.

[0194] The system performs control effectiveness evaluation calculations. This part of the calculation mainly compares the changes in system state before and after control to evaluate the actual effect of oscillation suppression control. Evaluation indicators include oscillation suppression rate (the percentage reduction in oscillation amplitude before and after control), settling time (the time required for the system to recover to a steady state), and energy consumption (the control energy required to achieve oscillation suppression). By comparing the actual control effect with the expected target, the system judges the effectiveness of the current control strategy and provides a basis for possible strategy adjustments.

[0195] The system also performs anomaly pattern recognition calculations. This calculation aims to promptly detect potential abnormal oscillation patterns or precursors to instability in the system. The system employs pattern recognition and anomaly detection algorithms, such as Support Vector Machines (SVM), cluster analysis, or neural network models, to analyze system state data in real time and identify abnormal behaviors that deviate from normal operating patterns. Particular attention is paid to newly emerging oscillation frequencies, sudden increases in oscillation amplitude, and abrupt changes in oscillation patterns, as these may be early signals of deteriorating system stability.

[0196] Finally, predictive analysis calculations are performed. Based on the current system state and historical data patterns, the system predicts the oscillation development trend over a future period (typically a few seconds to a few minutes). Predictions employ time series analysis, state-space models, or machine learning methods such as recurrent neural networks (RNNs) or long short-term memory networks (LSTMs). The prediction results include future trends in oscillation amplitude, predicted values ​​of system stability margins, and potential critical state time points, providing forward-looking information for proactive control decisions.

[0197] All these real-time calculations are performed continuously at a high frequency (typically 1-10 times per second), forming a continuous stream of monitoring results. The results are presented in the form of numerical indicators, trend charts, and status indicators, used for both automated control decisions and to provide operators with an intuitive view of the system's status. The system also performs reliability assessments on the calculation results, considering factors such as data quality and model uncertainty, and provides confidence levels for each result to aid in risk assessment during the decision-making process.

[0198] Step S7: Compare the real-time monitoring calculation result with the system stability boundary value. If the real-time monitoring calculation result exceeds the system stability boundary value, trigger the oscillation suppression controller.

[0199] In this step, the real-time monitoring calculation results obtained in step S6 are compared with the preset system stability boundary values ​​to determine whether the current system state is within a safe range. If the monitoring results exceed the safety boundary, the system will automatically trigger the oscillation suppression controller to start or adjust the oscillation suppression control strategy to prevent further deterioration of system stability.

[0200] System stability boundary values ​​are a set of predefined safety thresholds that represent the safety boundaries of system operation. These boundary values ​​are typically determined based on system design specifications, operational experience, and stability analysis results, and include indicators across multiple dimensions. Key stability boundary values ​​include: upper limit of oscillation amplitude (e.g., voltage fluctuations should not exceed ±2% of the rated value, and power fluctuations should not exceed ±5% of the rated value); lower limit of damping ratio (e.g., the damping ratio for all oscillation modes should be greater than 5%); upper limit of oscillation growth rate (e.g., the growth rate of oscillation amplitude should not exceed 10% per second); and lower limit of stability margin (e.g., voltage stability margin should not be less than 20%, and frequency stability margin should not be less than 0.5 Hz).

[0201] The comparison process employs a multi-level triggering mechanism, setting different response levels based on the degree to which the boundary is exceeded. Typical response levels include: Early Warning Level: When the monitoring result is close to but has not yet exceeded the boundary value, the system issues an early warning signal to alert the operator; Slight Exceedance Level: When the boundary value is slightly exceeded, the system initiates a mild control strategy, such as fine-tuning control parameters; Moderate Exceedance Level: When the boundary value is significantly exceeded, the system initiates a standard control strategy, implementing planned oscillation suppression measures; Severe Exceedance Level: When the boundary value is significantly exceeded, the system initiates an emergency control strategy, which may include substantial control adjustments and emergency protection measures.

[0202] The judgment process considers not only the comparison between the current value and the boundary, but also the time factor and the trend of change. Some transient exceedances may be acceptable, while persistent exceedances or rapidly deteriorating trends require immediate response. By setting time windows and using trend analysis algorithms, the system distinguishes between temporary fluctuations and persistent problems, avoiding overreaction to transient phenomena while ensuring timely response to genuine stability issues.

[0203] When the monitoring result exceeds the stability boundary value and meets the trigger condition, the system automatically activates the oscillation suppression controller. The controller activation process includes several key steps: First, confirm the validity of the trigger condition and eliminate possible false alarms or sensor malfunctions; second, select an appropriate control strategy based on the exceedance situation, such as strengthening the existing control, switching to a backup strategy, or activating an emergency plan; third, check the availability of control resources to ensure sufficient control capability to execute the selected strategy; fourth, send a control activation signal to initiate the corresponding control process.

[0204] Once triggered by the controller, the system enters a special monitoring state, monitoring system response with higher frequency and precision, evaluating control effectiveness, and dynamically adjusting control strategies as needed. Simultaneously, the system records detailed information about the triggered event, including trigger time, trigger conditions, system status, and response measures, providing a basis for subsequent analysis and improvement.

[0205] The system also includes a manual intervention mechanism, allowing operators to manually trigger or suppress controller activation under special circumstances. This human-machine collaboration mechanism enhances the system's flexibility and adaptability, enabling it to handle complex situations that automated decision-making systems may not be able to fully address. Manual intervention actions are recorded in the system log, serving as an important reference for experience accumulation and system optimization.

[0206] In one embodiment of the present invention, comparing the real-time monitoring calculation result with the system stability boundary value, and triggering the oscillation suppression controller if the real-time monitoring calculation result exceeds the system stability boundary value, includes: Step S7.1: Use an oscillation amplitude calculator to measure the oscillation magnitude index.

[0207] In this step, the oscillation amplitude calculator is activated first to perform precise quantitative analysis of the detected oscillations. The oscillation amplitude calculator is a specially designed functional module capable of accurately extracting characteristic quantities of oscillation signals from complex system operating data, providing a quantitative basis for subsequent control strategy selection. The oscillation magnitude index is a comprehensive parameter characterizing the severity of oscillations, reflecting the potential threat posed by the oscillations to the stable operation of the system.

[0208] Measuring the oscillation amplitude requires signal processing of the raw data. The system employs digital filtering technology to separate the oscillation frequency-related components from the noisy measurement signal. Based on the identified oscillation frequency, a bandpass filter is designed so that its passband center frequency matches the oscillation frequency, and the passband width is appropriately set to effectively capture the oscillation signal while suppressing interference from other frequencies. The filtered signal is cleaner, with more pronounced oscillation characteristics, facilitating subsequent amplitude extraction.

[0209] The specific calculation of the oscillation amplitude employs a combination of peak detection and statistical analysis. The system identifies the peak and trough values ​​of the signal within one oscillation cycle, calculating the peak-to-peak value as the basic measurement of the oscillation amplitude. Considering the potential fluctuations and irregularities in real-world signals, the system typically performs measurements over multiple cycles, then uses statistical methods (such as averaging, median, or maximum values) to obtain a more reliable amplitude estimate. For particularly complex cases, the system also employs advanced signal processing methods such as envelope analysis and Hilbert transform to extract the instantaneous amplitude variation characteristics of the oscillating signal.

[0210] Besides the amplitude itself, the oscillation magnitude index includes several auxiliary parameters, which together constitute a comprehensive assessment of the severity of the oscillation. These auxiliary parameters include: relative amplitude index, which is the ratio of the oscillation amplitude to the normal operating value, reflecting the significance of the oscillation relative to the basic operating state of the system; amplitude change rate, which represents the speed at which the oscillation amplitude changes over time, and is an important basis for judging the development trend of the oscillation; duration index, which indicates the length of time the oscillation has lasted, and long-term oscillation, even if the amplitude is not large, may cause cumulative damage to the equipment; and frequency stability index, which represents the stability of the oscillation frequency, and frequency instability usually indicates complex changes in the dynamic characteristics of the system.

[0211] The calculation of the oscillation magnitude index also considers the differentiated characteristics of different types of oscillations. For voltage oscillations, the system focuses on the ratio of voltage deviation to the allowable fluctuation range; for power oscillations, the system focuses on the ratio of power fluctuation to rated capacity; for frequency oscillations, the system focuses on the relationship between frequency deviation and system control performance indicators. This differentiated treatment ensures that the oscillation magnitude index can accurately reflect the actual impact of different types of oscillations.

[0212] Finally, by comprehensively considering all parameters, a standardized oscillation magnitude index is calculated, typically expressed as a percentage or a numerical scale of 0-10. This index intuitively reflects the severity of the oscillation, facilitating comparison with preset thresholds and guiding subsequent control strategy selection. The system also provides a confidence assessment for this index, reflecting the impact of uncertainties during the calculation process and providing a risk reference for decision-making.

[0213] Step S7.2: Input the oscillation magnitude index into the control scheme selector to determine the oscillation suppression parameters.

[0214] In this step, the oscillation magnitude index calculated in step S7.1 is input into the control scheme selector. Based on the severity and characteristics of the oscillation, the most suitable oscillation suppression scheme for the current situation is selected or combined from a preset control strategy library, and the specific suppression parameters are determined. The control scheme selector is the core decision-making module of the oscillation suppression system, and its function is to transform the oscillation assessment results into specific executable control strategies.

[0215] The control scheme selection is based on the classification of oscillation magnitude indicators. The system presets multiple control response levels, corresponding to different oscillation severity. Typical classification standards may include: slight oscillation (magnitude indicator less than 30%), employing a conservative control strategy, mainly involving fine-tuning parameters; moderate oscillation (magnitude indicator 30%-60%), employing a standard control strategy, implementing planned oscillation suppression measures; severe oscillation (magnitude indicator 60%-80%), employing an enhanced control strategy, which may include significant control adjustments; and emergency state (magnitude indicator greater than 80%), activating emergency control strategies to prioritize system safety, which may include load shedding or emergency power reduction of wind turbines.

[0216] In addition to the magnitude of the oscillation, the selection of a control scheme also considers the dynamic characteristics of the oscillation and the system's operating conditions. The rate of increase of the oscillation directly affects the urgency of the control strategy; rapidly increasing oscillations may require more timely and forceful intervention. The duration of the oscillation affects the persistence of the control; long-term oscillations may require continuous suppression measures and a gradual adjustment strategy. Operating conditions such as the system's load level and the status of adjustable resources also influence the selection of specific strategies. For example, under high load conditions, a more conservative control strategy may be needed to avoid additional system stress.

[0217] The control scheme selector employs a decision-making approach that combines rule-based and model-based methods. The rule-based component incorporates control strategy selection logic summarized from expert experience, such as "if the oscillation magnitude exceeds a critical value and the growth rate exceeds a threshold, then activate the emergency control mode." The model-based component utilizes the system's dynamic model and historical control performance data to predict the possible effects of different control strategies and select the scheme with the best expected results. This hybrid approach combines the reliability of expert experience with the adaptability of data models, enabling it to handle various complex oscillation situations.

[0218] Once the control scheme type is selected, the specific oscillation suppression parameters are further determined. These parameters provide concrete guidance for the execution of the control strategy and include several aspects: control target parameters, such as the desired oscillation attenuation rate and settling time; control strength parameters, such as power adjustment range and control gain adjustment; control range parameters, such as the number of wind turbines to be controlled and the control area; and control timing parameters, such as execution order, duration, and exit conditions. The setting of these parameters considers both the effectiveness of oscillation suppression and the constraints of system safety and equipment protection.

[0219] The control scheme selector also possesses adaptive learning capabilities. It records the selection, execution process, and suppression effect of each control strategy, establishing a strategy-effect database. By analyzing this historical data, the system can identify the most effective control strategy under different conditions, continuously optimize the decision-making logic, and improve the accuracy and efficiency of control scheme selection. This self-learning mechanism enables the system to adapt to constantly changing operating conditions and oscillation characteristics of offshore wind farms, maintaining long-term control effectiveness.

[0220] The finalized oscillation suppression parameters are output in a structured format, containing all necessary control command information, providing complete guidance for subsequent control parameter adjustments. The system also generates a strategy selection report, recording the decision-making process and its basis, facilitating subsequent auditing and improvement.

[0221] Step S7.3: Adjust the intensity coefficient of the control parameter according to the oscillation suppression parameter.

[0222] In this step, based on the oscillation suppression parameters determined in step S7.2, the intensity coefficient of the control parameters is dynamically adjusted to achieve precise adjustment of the control behavior. The intensity coefficient of the control parameters is an important adjustment factor that determines the strength and response characteristics of the control behavior; its proper setting directly affects the effectiveness of oscillation suppression and the stability of the system.

[0223] The adjustment process first establishes a mapping relationship between oscillation suppression parameters and control parameter intensity coefficients. This mapping relationship is typically a set of nonlinear functions that convert suppression parameters (such as the desired oscillation decay rate, control urgency, etc.) into specific intensity coefficient values. The design of the mapping function takes into account the system's dynamic characteristics, equipment response characteristics, and control effect experience, enabling it to provide the most suitable intensity coefficient settings under different oscillation conditions.

[0224] The adjustment of control parameter intensity coefficients encompasses multiple control stages and various control types. For the power control stage, the system adjusts the gain coefficient, response time constant, and limiting parameters of the power controller, affecting the magnitude and speed of power adjustment. For the voltage control stage, the system adjusts the proportional gain and integral time constant of the voltage regulator, affecting the sensitivity and stability of the voltage response. For the frequency response stage, the system adjusts the slope and dead zone parameters of the frequency-power characteristic curve, affecting the wind turbine's response characteristics to changes in system frequency. These adjustments, taken together, constitute a comprehensive optimization of the wind turbine control system.

[0225] The intensity coefficient adjustment also considers the need for differentiated regional control. Based on the zoning results determined in step S4, the system adopts different intensity coefficient adjustment strategies for fans in different regions. Fans in the core area typically use the highest intensity coefficient to directly and strongly suppress oscillation sources; fans in the transition area use a medium intensity coefficient to assist in core area control and prevent oscillation propagation; fans in the outer areas use a lower intensity coefficient, mainly playing an auxiliary and preventative role. This differentiated intensity coefficient setting ensures the rational allocation of control resources and optimal utilization.

[0226] In actual adjustment, a gradual adjustment strategy is adopted to avoid additional shocks to the system caused by sudden parameter changes. An initial intensity coefficient is applied, and then fine-tuning is performed based on the system response until the expected control effect is achieved. This adjustment process typically consists of several steps: the first step is basic adjustment, setting the initial intensity coefficient according to the oscillation suppression parameters; the second step is response evaluation, observing the system's response to the initial adjustment; the third step is fine-tuning optimization, appropriately adjusting the intensity coefficient based on the response evaluation results; and the fourth step is stability maintenance, maintaining parameter stability after achieving the expected effect and continuously monitoring the system status.

[0227] It also implements real-time adaptive control of the intensity coefficient. When changes in oscillation characteristics, system operating states, or control effects are detected, the system automatically triggers recalculation and adjustment of the intensity coefficient. This adaptive adjustment mechanism enables the control system to flexibly respond to various changes and uncertainties, maintaining optimal oscillation suppression.

[0228] The adjusted control parameter strength coefficients are distributed to the controllers of each wind turbine through the control network, enabling timely adjustments to control behavior. The system also records the history of all parameter adjustments, including adjustment time, reason for adjustment, parameter values ​​before and after adjustment, and system response, providing a basis for subsequent effect analysis and experience accumulation.

[0229] In one embodiment of the present invention, the method further includes: Step S8: Collect oscillation decay rate and settling time data.

[0230] In this step, after the oscillation suppression control strategy is implemented, system response data is continuously collected, with a focus on two key indicators: oscillation decay rate and settling time. These indicators directly reflect the effectiveness of the control strategy and are important bases for evaluating the oscillation suppression effect and optimizing control parameters.

[0231] The oscillation attenuation rate is acquired based on continuous time-series data obtained from the system's real-time monitoring network. The system collects dynamic change data of voltage, power, and other relevant parameters from various key nodes of the offshore wind farm, paying particular attention to signal components related to the oscillation frequency. A high sampling rate (typically 50-100 times per second) is used during the acquisition process to ensure accurate capture of detailed changes in the oscillation waveform, providing high-quality raw data for attenuation rate calculation.

[0232] After preprocessing, the collected raw data enters the oscillation analysis module to calculate the oscillation decay rate. The calculation process uses filtering and signal decomposition techniques to separate the time series of specific oscillation modes from the complex system response. Then, the system employs a decreasing exponential fitting method to fit the envelope of the oscillation waveform into an exponential decay curve. By analyzing the parameters of the fitted curve, the system can accurately calculate the oscillation decay rate, i.e., the percentage decrease in oscillation amplitude per unit time.

[0233] Simultaneously, settling-time data is collected and calculated. Settling-time refers to the time required for the system to recover from an oscillation state to a stable state, and is another important indicator for measuring control effectiveness. The determination of settling-time is based on preset stability criteria, such as "oscillation amplitude decreases to less than 5% of the initial value" or "oscillation amplitude changes do not exceed a threshold within multiple consecutive cycles." The system continuously monitors changes in oscillation amplitude, and when the stability criteria are met, the time interval from the start of the control strategy to reaching a stable state is recorded as the measured value of settling-time.

[0234] During the acquisition of oscillation attenuation rate and settling time, the system considered the needs of multi-point monitoring and multi-mode analysis. Since offshore wind farms are widely distributed systems, oscillation responses may differ at different locations. The system simultaneously acquires data at multiple key nodes, calculates local and global oscillation attenuation characteristics, and comprehensively evaluates the spatial distribution effect of oscillation suppression. Simultaneously, the system can also identify and analyze various possible oscillation modes, calculating the attenuation rate and settling time for each mode, providing a more detailed assessment of control effectiveness.

[0235] To enhance data reliability, the system employs a multi-layered verification mechanism. On one hand, it cross-verifies the calculated attenuation rate and settling time by comparing the oscillation responses at different measurement points and with different parameters (such as voltage, power, and frequency). On the other hand, it uses multiple algorithms (such as exponential fitting, the Plenyi index method, and the energy ratio method) to calculate the attenuation index in parallel, and improves the accuracy of the estimation through result consistency checks. For situations with noise interference or measurement uncertainties, the system also provides confidence intervals for the results, reflecting a quantitative assessment of data reliability.

[0236] The collected oscillation decay rate and settling time data are stored in time series format, recording the dynamic changes in the system response during the control process. This data is not only used for real-time evaluation of the current control effect but also forms an important part of the historical database, providing fundamental data support for long-term control performance analysis and algorithm optimization.

[0237] Step S9: Calculate the response characteristic index using the oscillation decay rate and the settling time.

[0238] In this step, based on the oscillation decay rate and settling time data collected in step S8, a comprehensive response characteristic index is calculated to fully evaluate the performance and effectiveness of the control strategy. The response characteristic index is a set of parameters that can quantitatively characterize the dynamic response characteristics of the system and serves as a direct basis for guiding the adjustment of control parameters.

[0239] The calculation of response characteristic indicators requires normalization of the original oscillation decay rate and settling time data. The normalization process considers the differences under various oscillation conditions, converting the original data into a comparable standard form. For example, for the oscillation decay rate, the system calculates the relative decay rate (the ratio of the current decay rate to the historical average decay rate) or the normalized decay rate (the ratio to the theoretical optimal decay rate); for the settling time, the system calculates the relative settling time (the ratio of the current settling time to the oscillation period) or the efficiency index (the ratio to the expected settling time). This normalization process allows for direct comparison of response characteristics under different conditions, facilitating comprehensive evaluation and trend analysis.

[0240] Next, response characteristic indicators across multiple dimensions are calculated to comprehensively reflect different aspects of the control effect. The main indicator categories include: effectiveness indicators, such as oscillation suppression rate (the percentage reduction in oscillation amplitude before and after control) and stability improvement rate (the percentage increase in system stability margin before and after control), reflecting the direct impact of the control strategy on system stability; efficiency indicators, such as control response speed (the time from control initiation to the appearance of a significant suppression effect) and control energy efficiency ratio (the oscillation attenuation effect per unit of control energy), reflecting the execution efficiency of the control strategy; and adaptability indicators, such as control robustness (the ability to remain effective under different operating conditions) and anti-interference capability (the ability to maintain stable control in the face of external disturbances), reflecting the adaptability of the control strategy.

[0241] The system also calculates distribution characteristic indicators in both time and spatial dimensions. Time-dimensional indicators analyze the changing characteristics of control effectiveness over time, such as initial response strength (the effect in the short term after control begins) and sustained stability (the stability of control effectiveness over long-term operation). Spatial-dimensional indicators analyze the distribution characteristics of control effectiveness in different areas of the offshore wind farm, such as the suppression strength in the core area, the coordination effect in the transition area, and the stability in the outer area. These distribution characteristic indicators can reveal the advantages and disadvantages of the control strategy in both time and space, providing a basis for differentiated parameter adjustments.

[0242] For complex control strategies, the system calculates component contribution indices to assess the contribution of different control components to the overall effect. For example, when using active power control and reactive power control in combination, the system analyzes the contribution ratio of each control method; in multi-regional coordinated control, the system evaluates the degree of influence of the control behavior of each region on the overall suppression effect. This component analysis helps identify key links and potential bottlenecks in the control strategy, providing guidance for optimizing the allocation of control resources.

[0243] Ultimately, the system uses a weighted fusion approach to synthesize the indicators from various dimensions into an overall response characteristic indicator. The weight settings reflect the relative importance of different indicators and can be adjusted according to specific control objectives and system characteristics. For example, in scenarios where rapid suppression is emphasized, the weight of the response speed indicator can be increased; in scenarios where long-term stable operation is prioritized, the weight of the continuous stability indicator can be increased. This flexible weight adjustment allows the response characteristic indicator to adapt to different evaluation needs and control scenarios.

[0244] The calculated response characteristics are presented in the form of numerical scores, classifications, or radar charts, intuitively reflecting the performance characteristics and effectiveness of the control strategy. The system also provides historical comparative analysis, comparing the current response characteristics with historical best performance, industry standards, or theoretical optimal values ​​to identify potential improvement areas and optimization directions. These comprehensive evaluation results are crucial for subsequent control parameter adjustments and serve as fundamental data for long-term system performance optimization.

[0245] Step S10: Input the response characteristic index into the adaptive regulator to update the control parameters.

[0246] In this step, the response characteristic index calculated in step S9 is input into the adaptive regulator. Based on the evaluation results of the control effect, the oscillation suppression control parameters are dynamically updated to achieve continuous optimization and self-adjustment of the control strategy. The adaptive regulator is the core module for system self-learning and optimization, capable of automatically adjusting the control strategy according to the actual control effect to adapt to constantly changing system characteristics and operating environment.

[0247] The adaptive control process analyzes the difference between response characteristic indicators and expected targets. The system sets expected response characteristic targets for each control scenario, such as the ideal oscillation decay rate and the desired settling time. The adaptive controller compares the actual indicators with the expected targets, calculates the performance gap, and identifies optimization directions. This target-based analysis considers both absolute performance (e.g., whether a specific threshold requirement has been met) and relative performance (e.g., whether there has been improvement compared to the previous control), providing clear target guidance for parameter updates.

[0248] Parameter updates employ a gradient-based optimization method. The adaptive regulator establishes a sensitivity model between the response characteristic index and the control parameters, analyzing the degree and direction of the impact of different parameter changes on the system response. Based on this sensitivity analysis, the system adjusts parameters along the direction of most significant index improvement, achieving rapid performance optimization. The sensitivity model is continuously updated using an online learning approach to adapt to dynamic changes in system characteristics, maintaining the model's accuracy and timeliness.

[0249] The control parameter update process considers multiple constraints to ensure that the adjusted parameters remain within a safe and feasible range. These constraints include: physical limitations of the equipment, such as hard constraints like adjustment range and response speed; system stability constraints to ensure that parameter adjustments do not introduce new instabilities; control smoothness constraints to limit the magnitude and rate of parameter changes and avoid drastic adjustments that could impact the system; and operational safety constraints to ensure that parameter adjustments comply with operating procedures and safety standards. The adaptive regulator employs a constraint optimization algorithm to find the optimal parameter settings while satisfying all constraints.

[0250] Adaptive adjustment also implements a multi-timescale parameter optimization mechanism. For different types of control parameters and varying degrees of performance differences, the system employs different periodic adjustment strategies: for parameters with rapid response and direct impact (such as power control gain), more frequent fine-tuning can be performed to quickly respond to performance changes; for parameters with far-reaching and stable effects (such as the basic structure of the control strategy), longer adjustment periods are used to ensure the consistency and predictability of system behavior. This hierarchical adjustment strategy balances response speed and system stability, maintaining optimal control performance in dynamic environments.

[0251] During parameter updates, special attention is paid to learning efficiency and convergence. To avoid over-adjustment and parameter oscillations, the adaptive regulator employs several techniques: adaptive learning rate, which dynamically adjusts the step size of parameter changes based on performance improvement trends, using larger step sizes initially or when performance gaps are large, and automatically reducing the step size as it approaches the optimum; memory mechanism, which saves historical optimal parameter settings, allowing it to revert to known good states when new parameters are ineffective; exploration strategy, which performs limited random exploration in the parameter space to avoid getting trapped in local optima; and diversity preservation, which maintains appropriate diversity in parameter settings to enhance adaptability to environmental changes.

[0252] The updated control parameters will not be applied immediately across the board, but rather gradually rolled out using a phased deployment strategy. New parameters are piloted on a small scale to evaluate their effectiveness; once proven effective, their application is gradually expanded until the entire control area is covered. This cautious deployment strategy reduces the risk of parameter adjustments and ensures the system remains stable during parameter updates.

[0253] The adaptive regulator not only updates specific parameter values ​​but also optimizes the structure and logic of the control strategy based on long-term operating experience. By analyzing a large amount of historical data, the system identifies the most effective control modes and parameter combinations under different operating conditions and establishes a condition-strategy mapping library. This structural optimization enables the system to automatically select the most suitable control strategy according to specific situations, greatly improving the intelligence and adaptability of oscillation suppression.

[0254] Ultimately, the updated control parameters are configured to each control unit through the control network, realizing the update of the oscillation suppression control strategy for the entire offshore wind farm. The system records the parameter update process and basis in detail, including initial state, update reason, optimization objective, adjustment process, and final result, forming a complete parameter management file to provide a basis for subsequent auditing and improvement.

[0255] like Figure 5 As shown, the present invention also provides a system for active suppression of offshore wind farm oscillations, comprising: a feature extraction module 1, an oscillation source localization module 2, a parameter optimization module 3, and a control execution module 4.

[0256] Feature extraction module 1 is used to acquire multi-source sensor data of offshore wind farms, perform time-frequency analysis and dimensionality reduction processing on the multi-source sensor data, and obtain a set of key feature quantities; The oscillation source localization module 2 is used to input the set of key feature quantities into the Lasso regression algorithm and the DeePC framework to establish a nonlinear model of the system and calculate the location of the oscillation source. Parameter optimization module 3 is used to establish a local control model based on the location of the oscillation source, construct an objective function that includes oscillation suppression effect and safety constraints through the local control model, and use Bayesian optimization to calculate the objective function to obtain control parameters; The control execution module 4 is used to perform zoning operations of the offshore wind farm according to the location of the oscillation source, and implement differentiated control strategies for different areas based on the control parameters to achieve oscillation suppression.

[0257] The present invention provides a method and system for active oscillation suppression in offshore wind farms. It employs multimodal data analysis to extract key feature quantities, utilizes the Lasso-DeePC framework for precise oscillation source localization, employs the Safe Barrier Bayesian optimization method for adaptive optimization of control parameters considering safety constraints, and implements a hierarchical and zone-based differentiated control strategy based on the oscillation source location. This method improves the accuracy of oscillation source localization, ensures the safety of the control process, and enhances the efficiency and accuracy of oscillation suppression, which is of great significance for improving the stability of offshore wind farm systems.

[0258] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for actively suppressing oscillations in offshore wind farms, characterized in that, include: Time-frequency analysis and dimensionality reduction were performed using multi-source sensor data from offshore wind farms to obtain a set of key feature quantities. The set of key features is input into the Lasso regression algorithm and the DeePC framework to establish a nonlinear model of the system and calculate the location of the oscillation source. A local control model is established based on the location of the oscillation source. An objective function containing oscillation suppression effect and safety constraints is constructed through the local control model. The objective function is calculated using Bayesian optimization to obtain the control parameters. The offshore wind farm is zoned according to the location of the oscillation source, and differentiated control strategies are implemented for different areas based on the control parameters to achieve oscillation suppression.

2. The method according to claim 1, characterized in that, The method utilizes multi-source sensor data from offshore wind farms for time-frequency analysis and dimensionality reduction to obtain a set of key feature quantities, including: The multi-source sensor data is input into a wavelet transform module to extract the system oscillation frequency components. Principal component analysis is performed using the oscillation frequency components to obtain the main eigenvectors; The mutual information between the feature quantities is calculated using the main feature vectors to generate the key feature quantity set.

3. The method according to claim 1, characterized in that, The process of inputting the set of key features into the Lasso regression algorithm and the DeePC framework to establish a nonlinear model of the system and calculate the location of the oscillation source includes: The set of key features is input into the Hankel matrix construction module, and Lasso regression is performed to obtain the sparse representation of the system. Perturbation analysis is performed using the sparse representation of the system to calculate the oscillation contribution of each node. The location of the oscillation source is determined by performing a positioning operation based on the oscillation contribution.

4. The method according to claim 1, characterized in that, A local control model is established based on the location of the oscillation source. An objective function incorporating oscillation suppression and safety constraints is constructed using this model. Bayesian optimization is then employed to calculate the objective function, yielding control parameters, including: The location of the oscillation source is input into the constraint calculation module to generate system stability constraint parameters; An optimization objective function is constructed using the system stability constraint parameters and oscillation suppression index. The optimization objective function is input into the Bayesian optimization iterator to calculate the control parameters.

5. The method according to claim 1, characterized in that, The offshore wind farm is divided into zones according to the control parameters, including: The spatial distribution of the core region, transition region, and peripheral region is determined using the location of the oscillation source; Calculate the control priority and control intensity of the core area, the transition area and the outer area based on the spatial distribution; The control priority and control strength are input into the partition controller to generate partition control instructions.

6. The method according to claim 5, characterized in that, The method further includes: implementing differentiated control strategies for different regions based on the control parameters to achieve oscillation suppression, including: Collect the operating status data of the core area wind turbines and calculate the power adjustment range; The power adjustment range is input into the curve optimizer to generate the active power output curve of the wind turbine. Adjust the reactive power control parameters of the core area fan according to the active power output curve to achieve oscillation suppression.

7. The method according to claim 5, characterized in that, The method further includes: implementing differentiated control strategies for different regions based on the control parameters to achieve oscillation suppression, including: Collect status data of the fans in the transition zone and the outer zone, and calculate power balance parameters; The power response characteristics of the transition zone fan are adjusted using the power balance parameters to construct the oscillation attenuation region of the transition zone fan; The response characteristics of the oscillation attenuation region are input into the parameter tuner to generate the control gain coefficient of the peripheral area fan for suppressing its sensitivity to oscillations.

8. The method according to claim 1, characterized in that, Also includes: Collect voltage, power, and oscillation frequency component data at each node of the offshore wind farm; Real-time monitoring calculations are performed using the voltage, power, and oscillation frequency component data to obtain real-time monitoring calculation results; The real-time monitoring calculation result is compared with the system stability boundary value. If the real-time monitoring calculation result exceeds the system stability boundary value, the oscillation suppression controller is triggered.

9. The method according to claim 8, characterized in that, The step of comparing the real-time monitoring calculation result with the system stability boundary value, and triggering the oscillation suppression controller if the real-time monitoring calculation result exceeds the system stability boundary value, includes: Use an oscillation amplitude calculator to measure oscillation magnitude indicators; Input the oscillation magnitude index into the control scheme selector to determine the oscillation suppression parameters; The intensity coefficient of the control parameter is adjusted according to the oscillation suppression parameter.

10. The method according to claim 1, characterized in that, Also includes: Collect oscillation decay rate and settling time data; The response characteristic index is calculated using the oscillation decay rate and the settling time; The response characteristic index is input into the adaptive regulator to update the control parameters.

11. A system for actively suppressing oscillations in offshore wind farms, characterized in that, include: The feature extraction module is used to acquire multi-source sensor data of offshore wind farms, perform time-frequency analysis and dimensionality reduction on the multi-source sensor data, and obtain a set of key feature quantities. The oscillation source localization module is used to input the set of key feature quantities into the Lasso regression algorithm and the DeePC framework, establish a nonlinear model of the system, and calculate the location of the oscillation source. The parameter optimization module is used to establish a local control model based on the location of the oscillation source, construct an objective function that includes oscillation suppression effect and safety constraints through the local control model, and use Bayesian optimization to calculate the objective function to obtain control parameters. The control execution module is used to perform zoning operations of the offshore wind farm according to the location of the oscillation source, and to implement differentiated control strategies for different areas based on the control parameters to achieve oscillation suppression.

Images