A wind turbine fast frequency regulation method based on disturbance adaptive compensation

Abstract

The present application relates to the technical field of device for adjusting, controlling or stabilizing power or frequency in power grid, and particularly relates to a wind turbine fast frequency modulation method based on disturbance adaptive compensation, a dynamic disturbance sensing data set is constructed by collecting frequency deviation, frequency deviation change rate and wind turbine state parameters, a genetic algorithm is used to dynamically optimize the window length and weight coefficient of short-time window sliding mean algorithm, background noise interference of power grid is suppressed and high-precision disturbance features are extracted. Frequency interference of inertia identification is reduced by using reverse test signal. Boundary layer parameters of sliding mode control model are updated through online incremental learning algorithm, output priority is dynamically adjusted based on disturbance energy entropy and power conflict is eliminated. Multi-island genetic algorithm is introduced to optimize mode switching threshold, and frequency modulation safe exit is realized in combination with adaptive power ramp rate limitation. The present application significantly improves frequency response speed and multi-source collaborative efficiency in dynamic multi-disturbance and inertia time-varying scene, and reduces frequency secondary drop risk.

Description

Technical Field

[0001] This invention relates to the field of devices for adjusting, controlling or stabilizing power or frequency in a power grid, and particularly to a method for rapid frequency regulation of wind turbine generators based on disturbance adaptive compensation. Background Technology

[0002] With the high proportion of new energy and distributed power sources connected to the grid, the power system faces complex operating scenarios involving multiple superimposed disturbances and dynamic changes in inertia. Existing wind turbine frequency regulation methods are mostly based on fixed control logic under single disturbances, making it difficult to adjust the frequency regulation strategy in real time under dynamic disturbance environments. Especially under continuous power disturbances or time-varying inertia conditions, traditional methods cannot quickly detect changes in system inertia and the superposition effect of disturbances, leading to delayed frequency regulation response and exacerbating frequency fluctuations. Simultaneously, the lack of dynamic coordination mechanisms among multi-source frequency regulation equipment easily leads to frequency regulation power redundancy or insufficiency, causing resource imbalance and further affecting system frequency stability. There is an urgent need for a wind turbine control method that can adapt to dynamic multi-disturbance scenarios and coordinate frequency regulation resources in real time to improve frequency response accuracy and robustness in complex grid environments. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a rapid frequency regulation method for wind turbines based on disturbance adaptive compensation, which solves the problem of decreased power system frequency stability caused by the lag in frequency regulation response of wind turbines and insufficient coordination of multi-source frequency regulation resources under dynamic multi-disturbance and time-varying inertia conditions.

[0004] To solve the above-mentioned technical problems, the specific technical solution of the present invention is as follows:

[0005] This invention provides a method for rapid frequency regulation of wind turbine generators based on disturbance adaptive compensation, comprising:

[0006] Step S1: Collect the frequency deviation, frequency deviation change rate and wind turbine operating status parameters of the power system to construct a dynamic disturbance sensing dataset. The wind turbine operating status parameters include real-time monitoring data of the rotor kinetic energy reserve of the wind turbine unit.

[0007] Step S2: Based on the dynamic disturbance sensing dataset, the window length and weight coefficient of the short-window sliding mean algorithm are dynamically optimized using a genetic algorithm to generate noise-suppressed disturbance feature data.

[0008] Step S3: Generate a reverse test signal based on the disturbance characteristic data, and estimate the equivalent inertia and damping parameters of the power grid in real time using the recursive least squares method.

[0009] Step S4: Use an online incremental learning algorithm to perform time series modeling on the continuous disturbance events in the dynamic disturbance sensing dataset, and update the boundary layer thickness and switching gain parameters of the preset adaptive sliding mode control model;

[0010] Step S5: Based on the equivalent inertia and damping parameters of the power grid, a compensation power command is generated through a preset fuzzy adaptive proportional controller and a preset integral controller. The compensation power command includes the triggering conditions for the fast power support mode and the droop control mode.

[0011] Step S6: According to the compensation power command, coordinate the distribution of wind turbine rotor kinetic energy reserves and the power response of the grid multi-source frequency regulation equipment through a preset microservice architecture, wherein the grid multi-source frequency regulation equipment includes a virtual synchronous machine and an energy storage device;

[0012] Step S7: In the adaptive sliding mode control model predictive control, the genetic algorithm is introduced to optimize the trigger threshold for dynamic switching, and the fast power support mode and droop control mode are switched according to the frequency deviation change rate.

[0013] Step S8: When the frequency deviation recovers to the preset dead zone range, the microservice exits frequency modulation and switches to maximum power point tracking operation by limiting the adaptive power ramp rate.

[0014] Furthermore, in the wind turbine rapid frequency regulation method based on disturbance adaptive compensation described in this invention, step S2 includes:

[0015] Based on the dynamic perturbation sensing dataset, the crossover probability and mutation probability are dynamically adjusted according to the population fitness distribution. When the population diversity is lower than a set threshold, the mutation probability is increased to the range of 0.2-0.4.

[0016] An elite retention strategy is adopted to directly pass the best individuals of each generation to the next generation, and the distribution characteristics of the perturbation feature data are combined to accelerate parameter convergence.

[0017] By combining genetic algorithms with GPU parallel computing, the window length and weight coefficients of the short-window sliding mean algorithm are evaluated in parallel.

[0018] The test signal amplitude parameters are initialized based on historical disturbance data in the dynamic disturbance sensing dataset, and the frequency interference of inertia estimation is reduced by reverse signal excitation.

[0019] Furthermore, the wind turbine fast frequency regulation method based on disturbance adaptive compensation described in this invention also includes:

[0020] Based on the execution feedback data of the compensation power command, the activation weights of the fuzzy rules are adjusted using a reinforcement learning mechanism;

[0021] The fuzzy intervals of the equivalent inertia parameters of the power grid are dynamically divided by clustering algorithm, and the center point and width parameters of the membership function are optimized.

[0022] Establish the parameter correlation matrix between the adaptive sliding mode control model and the fuzzy controller, and update the boundary layer thickness and input range synchronously;

[0023] When a high-frequency disturbance mode is detected, the high-frequency band membership function parameters of the equivalent inertia parameter of the power grid are optimized first, and the weight allocation of the corresponding rules is enhanced.

[0024] Furthermore, in the wind turbine rapid frequency regulation method based on disturbance adaptive compensation described in this invention, step S6 includes:

[0025] In the containerized deployment framework of the model predictive control, the power allocation ratio between rotor kinetic energy reserve and converter capacity is optimized through dynamic programming microservices.

[0026] The communication protocols of the multi-source frequency regulation equipment in the power grid are integrated through the API gateway to generate a cross-device frequency regulation command synchronization protocol.

[0027] Based on the dynamic disturbance perception dataset, the disturbance energy entropy is calculated, and an event-driven priority weight microservice is designed to dynamically adjust the output ratio.

[0028] When a power conflict is detected, a redundancy elimination algorithm is triggered to reallocate the frequency regulation tasks of the virtual synchronizer and the energy storage device.

[0029] Furthermore, in the wind turbine rapid frequency regulation method based on disturbance adaptive compensation described in this invention, step S7 includes:

[0030] The historical disturbance data and real-time inertia parameters in the dynamic disturbance sensing dataset are input into the rolling time domain optimization model to predict the frequency trend for the next 3-5 sampling periods.

[0031] The initial population of the genetic algorithm is divided using a multi-island genetic algorithm, and elite individuals are periodically migrated to generate differentiated trigger thresholds;

[0032] When the frequency deviation change rate exceeds the start-up threshold, switch to the fast power support mode to release the rotor kinetic energy reserve;

[0033] During the droop control phase, a power slope curve is generated based on the compensation power command. The wind turbine speed recovery rate is calculated by combining the speed change gradient after the release of the wind turbine rotor kinetic energy reserve. The slope limit is then dynamically updated based on the wind turbine speed recovery rate.

[0034] Furthermore, the wind turbine fast frequency regulation method based on disturbance adaptive compensation described in this invention also includes:

[0035] The wind turbine speed recovery rate and frequency deviation are used as hard constraints to construct a gradient search space;

[0036] The optimal power descent gradient parameters are solved within the constraint space using the genetic algorithm described above.

[0037] When a risk of sudden speed change is detected, a dynamic relaxation mechanism is triggered to expand the constraint boundary of the sliding mode control model;

[0038] By initializing the search space with the optimal gradient parameters from historical frequency modulation exit data, the generation time of the safe recovery strategy can be shortened.

[0039] Furthermore, in the wind turbine rapid frequency regulation method based on disturbance adaptive compensation described in this invention, step S8 includes:

[0040] Real-time data streams of the continuous disturbance events are collected through a sliding time window, and principal component analysis is performed for dimensionality reduction.

[0041] The boundary layer parameters of the adaptive sliding mode control model are updated only based on incremental data using the recursive least squares method.

[0042] When data anomalies are detected, a local model rollback mechanism is triggered to restore the parameters to the previous time period.

[0043] The learning rate is dynamically adjusted based on the rate of change of the error predicted by the model, thereby increasing the parameter update speed as the error gradient increases.

[0044] Beneficial effects of this invention;

[0045] This invention significantly improves the noise suppression capability of the short-window moving average algorithm and enhances the accuracy of disturbance feature extraction by constructing a dynamic disturbance sensing dataset and optimizing it with a genetic algorithm. Combined with recursive least squares method for real-time estimation of the equivalent inertia parameter of the power grid, it effectively adapts to the dynamic parameter tracking requirements in inertia-varying scenarios, reducing identification errors. An online incremental learning algorithm and a fuzzy adaptive proportional-integral controller collaboratively optimize the compensation power command generation logic, strengthening the dynamic response capability under high-frequency disturbances. A microservice architecture enables efficient collaboration among multi-source frequency regulation equipment, eliminating power redundancy and conflicts. A genetic algorithm is introduced to optimize the trigger threshold and safe exit mechanism, balancing the switching timing between fast power support mode and droop control, and combined with adaptive power ramp rate limits to achieve a smooth transition during frequency regulation exit. The synergistic effect of these technologies solves the problems of response lag and insufficient resource coordination in traditional methods under dynamic multi-disturbance superposition and inertia-varying scenarios, improving the real-time performance, stability, and multi-source equipment collaboration efficiency of power system frequency regulation, and reducing the risk of secondary frequency drops. Attached Figure Description

[0046] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the accompanying drawings without creative effort.

[0047] Figure 1 The flowchart illustrates a method for rapid frequency regulation of wind turbines based on adaptive disturbance compensation, as provided in an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The technical solutions provided by various embodiments of this invention will be described in detail below with reference to the accompanying drawings. To better understand the objectives of this invention, it will be described in further detail below.

[0049] Please see Figure 1 The present invention provides a method for rapid frequency regulation of wind turbine generators based on disturbance adaptive compensation, comprising:

[0050] Step S1: Collect the frequency deviation, frequency deviation change rate and wind turbine operating status parameters of the power system to construct a dynamic disturbance sensing dataset. The wind turbine operating status parameters include real-time monitoring data of the rotor kinetic energy reserve of the wind turbine unit.

[0051] When collecting frequency deviation data from the power system, frequency measurement units deployed at grid nodes acquire the AC voltage frequency signal at the grid's point of common coupling in real time using a preset sampling frequency. These frequency measurement units include a digital signal processing module that converts the original voltage signal into frequency deviation time-series data via Fourier transform. When synchronously collecting the frequency deviation change rate, a differential tracking filter performs first-order differential operations on the frequency deviation signal, and the mean square error of the frequency deviation is statistically calculated using a sliding time window to generate a frequency deviation change rate feature vector. Wind turbine operating status parameters are collected through vibration sensors and speed encoders deployed in the wind turbine drivetrain. These sensors monitor the generator rotor angular acceleration and kinetic energy reserve in real time. The rotor kinetic energy reserve is calculated based on the product of the rotor mass inertia parameter and the square of the measured rotational speed.

[0052] When constructing the dynamic disturbance sensing dataset, the frequency deviation time-series data, frequency deviation change rate feature vector, and rotor kinetic energy reserve monitoring data are aligned according to a unified timestamp, and a sliding window mechanism is used to synchronously fuse the multi-source data. The length of the sliding window is set to 5-10 power frequency cycles based on the typical disturbance duration, and outlier sampling points are removed from the data within the window using an outlier detection algorithm. The synchronously fused dataset is stored in a time-series database, and the data records include timestamp, frequency deviation, frequency deviation change rate, rotor speed, and kinetic energy reserve fields, with additional grid node topology identifiers to distinguish data sources from different regions.

[0053] The real-time update mechanism of the dynamic disturbance sensing dataset adopts an event-driven model. When a frequency deviation exceeding a preset threshold or an abnormal gradient in wind turbine speed change is detected, the data acquisition module is triggered to increase the sampling frequency to 1.5-2 times the preset value. Newly added data is appended to the end of the time-series database after verification, and historical data exceeding the storage capacity is overwritten through a circular buffer management mechanism to maintain the time-series continuity of the dataset. In the data preprocessing stage, principal component analysis is introduced to reduce the dimensionality of multidimensional features, extracting key principal component vectors that characterize the dynamic properties of the system, thereby reducing the computational complexity of subsequent feature extraction algorithms.

[0054] Real-time monitoring data of the wind turbine rotor kinetic energy reserve is acquired through the energy management module of the converter control system. This module integrates a speed-power mapping table and a rotor kinetic energy prediction model. During real-time monitoring, the speed signal collected by the encoder is processed by Kalman filtering and then input into the prediction model. The percentage of available kinetic energy reserve is calculated by combining the current wind speed and pitch angle parameters. The prediction model is trained based on historical operating data and uses a long short-term memory network to establish a nonlinear mapping relationship between speed fluctuations and kinetic energy release capacity. The output result serves as the core state parameters of the dynamic disturbance sensing dataset.

[0055] The construction logic of the dynamic disturbance sensing dataset includes a multi-level data verification mechanism. During the data fusion stage, redundant checksums are used to verify the integrity of data from each sensor, discarding sampling points that fail verification. The timestamp alignment mechanism is based on the IEEE 1588 precise time protocol, synchronizing the clocks of acquisition devices at different nodes to microsecond-level accuracy. The data storage structure uses a columnar database to optimize real-time read / write performance, and deploys multiple data replicas through a distributed architecture to improve the reliability of data access and concurrent processing capabilities.

[0056] The dynamic disturbance sensing dataset provides fundamental data support for closed-loop control, and its construction process forms a feedforward link with the subsequent feature extraction module. Frequency deviation and rate of change data characterize the dynamic characteristics of the power grid, rotor kinetic energy reserve parameters reflect the frequency regulation capability reserve of the wind turbine, and the spatiotemporal correlation characteristics of multidimensional data capture disturbance propagation patterns through a sliding window mechanism. The dynamic update characteristics of the dataset adapt to the time-varying inertia scenario of the power system, providing raw input for the noise suppression algorithm and supporting the real-time optimization of subsequent control strategies.

[0057] Step S2: Based on the dynamic disturbance sensing dataset, the window length and weight coefficient of the short-window sliding mean algorithm are dynamically optimized using a genetic algorithm to generate noise-suppressed disturbance feature data.

[0058] When constructing the initial population based on the dynamic perturbation-aware dataset, the genetic algorithm uses a hybrid binary and real-number encoding method for individual encoding. The window length parameter of the short-window sliding mean algorithm is mapped to integer gene segments, and the weight coefficients are encoded as floating-point gene segments. The population size is set to 50-100 candidate solutions based on the possible value range of the sliding window parameter. The initial individuals are generated using a Monte Carlo random sampling method, covering the typical value range of the sliding window length parameter, and the weight coefficients are uniformly distributed within the normalized interval of 0 to 1. The individual fitness function is calculated based on noise suppression performance evaluation metrics, including a weighted combination of the signal-to-noise ratio improvement rate of the perturbation feature signal and the trend fitting error.

[0059] During population evolution, a roulette wheel selection mechanism selects high-fitness individuals for mating based on their fitness values, while an elite retention strategy directly preserves the best individuals from each generation to the next. Crossover operations employ a multi-point crossover method, setting fixed crossover points within the sliding window length gene segment, and using arithmetic crossover to generate new individuals within the weighted gene segment. Mutation probabilities are dynamically adjusted based on population diversity monitoring results. When the genotypic similarity of individuals exceeds a preset threshold, an adaptive mutation mechanism is triggered, applying a ±10% random perturbation to the sliding window length parameter, and re-initializing the weighted coefficients within the range of 0.1 to 0.9.

[0060] The parameter optimization process of the short-window moving average algorithm is accelerated by combining a GPU parallel computing framework, distributing individuals in the population to different computing units for parallel evaluation. Each thread block independently calculates the noise suppression effect of the corresponding window length and weight coefficient combination, and exchanges intermediate calculation results through shared memory. The parallel computing architecture is implemented on the CUDA platform, dividing the dynamic perturbation-aware dataset into multiple data blocks and loading them into GPU memory to reduce data transmission latency. After summarizing the fitness evaluation results, a tournament selection mechanism is used to select the optimal parameter combination as the real-time configuration parameters for the moving average algorithm.

[0061] During the generation of noise-suppressed disturbance feature data, the optimized sliding window length parameter is applied to real-time data stream processing. The weight coefficients of data points within the window are dynamically weighted according to a Gaussian distribution function, with the center point assigned the highest weight value and the weights of edge data points decreasing progressively. The weighted mean calculation result is then smoothed a second time using a moving average filter to eliminate residual high-frequency noise components. The processed disturbance feature data is stored in a structured data format containing timestamps, frequency deviation correction values, and feature intensities, for subsequent use by the inertia parameter identification module.

[0062] Based on historical disturbance data in the dynamic disturbance sensing dataset, the amplitude parameters of the reverse test signal are initialized using an energy normalization method. The energy distribution characteristics of historical disturbance events are statistically analyzed using a sliding time window, and the 80th percentile of the maximum energy value is extracted as the reference amplitude of the test signal. The signal timing characteristics are synchronously matched with the current disturbance characteristics, and a phase compensation algorithm is used to adjust the time alignment accuracy between the signal waveform and the real-time disturbance. When the reverse test signal is injected into the grid equivalent inertia estimation stage, a bandpass filter is used to limit the signal bandwidth to avoid introducing additional frequency interference.

[0063] The recursive least squares method introduces a forgetting factor during inertia parameter estimation to dynamically adjust the weighting of historical data. The forgetting factor value is dynamically adjusted based on the signal-to-noise ratio of the disturbance feature data; as noise suppression improves, the forgetting factor is increased to above 0.95 to enhance the parameter correction effect of real-time data. The parameter estimation results are iteratively optimized through a covariance matrix update mechanism, calculating only the parameter correction amount of incremental data each time a new data window arrives, reducing computational complexity. The identified equivalent inertia parameters of the power grid serve as input variables for the fuzzy controller, participating in the subsequent generation logic of compensation power commands.

[0064] Step S3: Generate a reverse test signal based on the disturbance characteristic data, and estimate the equivalent inertia and damping parameters of the power grid in real time using the recursive least squares method.

[0065] When generating the reverse test signal, the principal components of the spectrum are extracted based on the noise-suppressed disturbance characteristic data. A band-limited signal reconstruction algorithm is used to generate a reverse waveform with the opposite phase to the original disturbance. The signal amplitude is initialized based on the energy distribution characteristics of historical disturbance events. The 80th percentile of the maximum disturbance energy is statistically analyzed using a sliding time window as the reference amplitude, and the signal strength is dynamically adjusted in conjunction with the current grid frequency fluctuation amplitude. The signal timing characteristics are aligned with the real-time disturbance characteristics through a phase compensation algorithm to eliminate signal superposition errors caused by time delay. When injected into the grid, an anti-aliasing filter limits the signal bandwidth to the 0.5-5Hz range to avoid introducing high-frequency interference components.

[0066] In the estimation of the equivalent inertia parameter of the power grid, the recursive least squares method constructs an input-output matrix using the reverse test signal and the power grid frequency response data. The input matrix contains the time-series amplitude and injection time information of the reverse test signal, while the output matrix represents the observed value of the frequency deviation change rate. The forgetting factor is dynamically adjusted based on the signal-to-noise ratio of the disturbance feature data, gradually increasing to above 0.95 as the noise suppression effect improves, thus strengthening the correction effect of real-time data on parameter estimation. The covariance matrix is ​​iteratively optimized using a rank-one update mechanism, calculating only the correction amount of the incremental data to the covariance matrix each time a new data window arrives, reducing computational complexity.

[0067] When estimating the equivalent inertia parameters of the power grid in real time, a dynamic regression model is constructed by extracting frequency response data after the injection of a reverse test signal through a sliding time window. This model uses the equivalent inertia and damping parameters of the power grid as variables to be identified, and iteratively optimizes the parameter estimates based on the criterion of minimizing prediction error. A regularization term is introduced during the parameter update process to constrain the magnitude of parameter changes and prevent divergence in estimates caused by sudden data changes. The identification results are smoothed using a moving average filter to eliminate instantaneous noise interference, outputting a stable sequence of equivalent inertia parameters.

[0068] The estimation of damping parameters is combined with the analysis of the frequency oscillation attenuation characteristics of the power grid. The frequency free response curve is extracted after the reverse test signal is removed. The attenuation rate of the oscillation amplitude is calculated using an exponential fitting algorithm to derive the estimated value of the equivalent damping coefficient. The fitting process employs a nonlinear least squares optimization algorithm. The initial values ​​are set based on historical operating data, and the iteration step size is dynamically adjusted according to the gradient descent rate to improve the parameter convergence speed. The fitting results are weighted and fused with the damping parameters output by the recursive least squares method to form the final estimated value of the damping parameters.

[0069] The collaborative working mechanism between the reverse test signal and the parameter estimation module is implemented through event-triggered logic. When a frequency deviation exceeding a preset threshold or a significant change in disturbance energy entropy is detected, the signal injection process is initiated. The signal duration is determined based on the autocorrelation function analysis of the disturbance characteristics, covering more than 90% of the main disturbance energy period. The parameter estimation results are transmitted to the fuzzy adaptive controller in real time, serving as the core input variable for generating compensation power commands and supporting the optimization of dynamic frequency modulation strategies under different disturbance scenarios.

[0070] The online identification accuracy of the power grid equivalent inertia parameters is improved through a dual verification mechanism, including historical data backtesting and real-time prediction error monitoring. Historical data backtesting uses a sliding window mechanism to compare parameter estimates with offline simulation results; when the deviation exceeds 5%, a parameter recalibration process is triggered. Real-time prediction error monitoring compares the model output frequency with actual observations, dynamically adjusting the forgetting factor and regularization coefficient to maintain the stability and tracking capability of the parameter estimates. The verified equivalent inertia parameters serve as the basis for power system dynamic characteristic analysis and participate in the collaborative control logic of multi-source frequency regulation equipment.

[0071] Step S4: Use an online incremental learning algorithm to perform time series modeling on the continuous disturbance events in the dynamic disturbance sensing dataset, and update the boundary layer thickness and switching gain parameters of the preset adaptive sliding mode control model;

[0072] When using an online incremental learning algorithm to model continuous disturbance events over time, a sliding time window is used to capture the real-time data stream from the dynamic disturbance sensing dataset. The window length is dynamically adjusted to 3-5 power frequency cycles based on the periodic characteristics of the disturbance event. Data within the window undergoes principal component analysis for dimensionality reduction, extracting principal component vectors with a contribution rate exceeding 85%, eliminating redundant noise interference, and generating low-dimensional input features suitable for the sliding mode control model. The dimensionality-reduced feature vectors are then standardized to eliminate dimensional differences, forming the input matrix of the time series model and improving the computational efficiency of subsequent parameter updates.

[0073] When updating the boundary layer thickness and switching gain parameters of the adaptive sliding mode control model, a recursive least squares method is used to iteratively optimize the model parameters based solely on incremental data. This recursive algorithm retains the covariance matrix information of historical parameters. Each time a new data window is received, the covariance matrix and parameter vector are updated using the matrix inverse lemma, avoiding recalculation of all data. The update amount of the boundary layer thickness is dynamically adjusted according to the steady-state error gradient of the frequency deviation, and the switching gain parameter is weighted and corrected based on the rate of change of the disturbance characteristics, balancing the speed and smoothness of the control response.

[0074] When an anomaly is detected, a local model rollback mechanism is triggered to restore the valid parameters from the previous time period. The anomaly detection module monitors the statistical characteristics of the input data in real time, including mean drift, variance abrupt changes, and data distribution deviation indices. If the data deviation exceeds a threshold for three consecutive sampling windows, the rollback mechanism calls the historical parameter database to load the sliding mode control parameters that have been verified and are valid from the previous time period, overwriting the update results driven by the current anomaly data. The rolled-back model maintains stable output until subsequent data passes verification, ensuring the continuity of control commands.

[0075] During the dynamic adjustment of the learning rate, the parameter update step size is adaptively adjusted based on the rate of change of the model's prediction error. The gradient of the prediction error is calculated using the error difference within a sliding window, combined with an exponentially weighted average to eliminate instantaneous fluctuations. When the absolute value of the error gradient exceeds a preset threshold, the learning rate is increased to 1.5-2 times the baseline value according to a logarithmic function, accelerating parameter convergence; when the error gradient approaches zero, the learning rate gradually decays to the initial value, refining the parameter search accuracy. This dynamic adjustment mechanism matches the model convergence requirements under different perturbation scenarios, preventing overfitting or underfitting.

[0076] The collaborative working mechanism between the time series modeling module and the sliding mode controller is implemented through event-triggered logic. When the rate of change of frequency deviation exceeds a set threshold or the disturbance energy entropy increases significantly, the incremental learning process is triggered. The model update results are applied to the switching logic of the sliding mode controller in real time, dynamically adjusting the boundary layer thickness to adapt to the control requirements of different disturbance intensities. The updated switching gain parameters are synchronized to the input universe of discourse of the fuzzy controller through the parameter correlation matrix, eliminating the response delay between multiple controllers and improving the efficiency of collaborative control.

[0077] The closed-loop parameter update chain constructed by the online incremental learning algorithm captures dynamic features of disturbances through a sliding window mechanism, reduces computational load through recursive optimization, ensures system robustness through anomaly rollback, and matches the dynamic learning rate to convergence requirements. Data interaction at each technical stage strictly follows temporal logic, forming a complete closed loop from disturbance perception to control parameter optimization. This enhances the adaptive capability of sliding mode control in continuous disturbance scenarios and improves the dynamic response accuracy of wind turbine frequency regulation.

[0078] Step S5: Based on the equivalent inertia and damping parameters of the power grid, a compensation power command is generated through a preset fuzzy adaptive proportional controller and a preset integral controller. The compensation power command includes the triggering conditions for the fast power support mode and the droop control mode.

[0079] When generating compensation power commands, the input variables of the fuzzy adaptive proportional-integral controller include the grid equivalent inertia parameter, real-time frequency deviation, and rate of change. The universe of discourse partitioning of these input variables is dynamically adjusted through a clustering algorithm. Based on historical operating data, the inertia parameter is divided into 3-5 fuzzy intervals, and the frequency deviation and rate of change are divided into overlapping membership functions according to normal distribution characteristics. The center point and width parameters of the membership function are updated through an online learning mechanism. When high-frequency disturbance characteristics are detected, the width of the high-frequency band function is reduced to 60% of the baseline value, improving the rule matching resolution.

[0080] The fuzzy rule base is constructed using a combination of expert experience and data-driven approaches. The initial rule set is defined based on the frequency response characteristics under typical disturbance scenarios. The reinforcement learning mechanism dynamically adjusts the rule activation weights based on the execution feedback of the compensation power command. This feedback data includes frequency fluctuation suppression rate, frequency modulation energy consumption, and power overshoot. The Q-learning algorithm iteratively updates the rule weight matrix, prioritizing the reinforcement of the optimal control path under high-frequency disturbance scenarios. The matching results of historically similar disturbance patterns are weighted and fused into the reward function, improving the dynamic adaptability of the rule base.

[0081] The triggering condition for the rapid power support mode is achieved through a dynamic threshold comparator. A rapid power support command is triggered when the frequency deviation change rate exceeds 120% of a preset threshold or the equivalent inertia parameter falls below a critical value. The threshold parameter is generated through a multi-island genetic algorithm with rolling optimization, and the threshold boundary is dynamically adjusted in conjunction with a real-time inertia parameter prediction model. After the trigger signal is generated, the control command prioritizes releasing 80%-90% of the wind turbine's pre-stored rotor kinetic energy, rapidly injecting compensating power through the converter to suppress further frequency drops.

[0082] The droop control mode is activated based on the steady-state error range of the frequency deviation. When the frequency deviation remains outside the preset dead zone and the rate of change is lower than the fast mode threshold, the system switches to the droop control phase. In this mode, a power slope curve command is generated. The initial slope value is set based on the equivalent damping parameters and the remaining rotor kinetic energy capacity, and the power descent gradient is dynamically adjusted in conjunction with the real-time monitoring value of the speed recovery rate. Dead zone control logic is introduced during the slope adjustment process to shield the influence of minor speed fluctuations on command generation and maintain the continuity of power output.

[0083] The collaborative working mechanism between the fuzzy controller and the sliding mode controller is achieved through a parameter correlation matrix, dynamically linking the sliding mode boundary layer thickness parameter with the fuzzy input universe of discourse. When the sliding mode control model updates the boundary layer parameters, the correlation matrix drives the input range of the fuzzy controller to adjust synchronously, eliminating the response delay between the two controller stages. The output interface of the control command is encapsulated through a standardized protocol, including the power support mode identifier, command amplitude, and slope parameters, ensuring compatibility with the communication protocol of the microservice architecture and guaranteeing the real-time transmission of commands across modules.

[0084] The logic for generating compensation power commands is seamlessly integrated with the closed-loop control link of multi-source frequency modulation equipment. Command parameters are parsed by a dynamic programming microservice and then allocated to the virtual synchronizer and energy storage device. An event-driven priority-weighted microservice dynamically adjusts the output ratio of each device based on the disturbance energy entropy. When a command conflict is detected, a redundancy elimination algorithm is triggered to reallocate tasks. The generated control commands are simultaneously fed back to the online learning module, forming a closed-loop optimization link from parameter estimation and command generation to execution feedback, thereby improving the frequency modulation control accuracy under complex disturbance scenarios.

[0085] Step S6: According to the compensation power command, coordinate the distribution of wind turbine rotor kinetic energy reserves and the power response of the grid multi-source frequency regulation equipment through a preset microservice architecture, wherein the grid multi-source frequency regulation equipment includes a virtual synchronous machine and an energy storage device;

[0086] When coordinating the allocation of kinetic energy reserves in wind turbine rotors, a dynamic programming microservice parses the power demand parameters and time constraints in the compensation power commands to construct a multi-objective optimization model. The model aims to maximize the frequency stability recovery speed and minimize rotor mechanical stress, using the wind turbine rotor kinetic energy reserve capacity, converter output limit, and energy storage device response rate as constraints. During optimization, a linear programming algorithm is used to solve for the optimal combination of rotor kinetic energy release ratio and converter power output, generating a power allocation scheme adapted to dynamic disturbance scenarios. The allocation commands are then distributed to frequency regulation device nodes via a lightweight message queue.

[0087] In the microservice architecture, the API gateway integrates the heterogeneous communication protocols of the virtual synchronizer and energy storage devices, encapsulating Modbus and IEC 61850 protocols into a standardized JSON command format via a RESTful interface. The gateway monitors the communication latency and bandwidth utilization of each frequency modulation device in real time, dynamically adjusting data packet transmission priorities to ensure the timing consistency of frequency modulation commands across devices. The command synchronization protocol includes device status query, power command issuance, and execution feedback confirmation processes, employing a timeout retransmission mechanism to improve command transmission reliability while remaining compatible with the different response characteristics of various devices.

[0088] The event-driven priority-weighted microservice calculates disturbance energy entropy based on a dynamic disturbance perception dataset, quantifying the spatiotemporal energy concentration of disturbance events. When the entropy value exceeds a preset threshold, the output weight of each device is dynamically allocated based on the real-time available capacity of the virtual synchronizer and energy storage device, response rate, and frequency regulation cost coefficient. The weight allocation results are shared in real time through a distributed cache cluster, supporting each frequency regulation node to adjust its output target proportionally, and are synchronously updated to all associated devices through a publish-subscribe mechanism.

[0089] Upon detecting a power conflict, the redundancy elimination algorithm constructs an operating state parameter matrix for the conflicting devices, identifying combinations of devices with overshoot or contradictory directions. The algorithm, based on a genetic algorithm, rapidly solves conflict resolution schemes within a hard-constrained space. Constraints include upper limits on device output, frequency recovery requirements, and mechanical stress limitations. The initial population utilizes historical feasible solutions to accelerate convergence; crossover and mutation operations generate new candidate schemes. After evaluation using a fitness function, the optimal task allocation strategy is output, and the corrected frequency modulation command is reissued via containerized microservices.

[0090] During frequency regulation, the release ratio of the wind turbine rotor kinetic energy reserve is dynamically adjusted through the converter control logic. In the initial stage, 80%-90% of the pre-stored kinetic energy is prioritized for rapid injection into the grid. In subsequent stages, the release rate is gradually reduced based on the speed recovery gradient and the frequency deviation change rate. The power response of the energy storage device adopts a segmented ramp control strategy. In the initial stage, it outputs 120% of the rated power rapidly. After entering steady state, it switches to a power tracking mode in coordination with a virtual synchronous machine to maintain the smoothness of the frequency recovery curve.

[0091] The microservice architecture's collaborative control chain forms a closed-loop optimization mechanism, dynamically planning and generating initial allocation strategies. An API gateway ensures synchronous transmission of instructions across devices, priority and weight are dynamically adjusted to optimize resource utilization efficiency, and a redundancy elimination algorithm resolves execution conflicts. Each module is tightly coupled with event-triggered logic through standardized data interfaces, responding in real-time to changes in grid disturbance characteristics, thereby improving the collaborative efficiency of multi-source frequency regulation equipment and system frequency stability.

[0092] Step S7: In the adaptive sliding mode control model predictive control, the genetic algorithm is introduced to optimize the trigger threshold for dynamic switching, and the fast power support mode and droop control mode are switched according to the frequency deviation change rate.

[0093] When constructing the trigger threshold optimization module in adaptive sliding mode control model predictive control, the multi-island genetic algorithm divides the search space into 4-6 independently evolving subgroups. Each subgroup uses historical disturbance data captured by a sliding time window and real-time inertia parameters as input to generate differentiated trigger threshold candidate schemes. The initial population is initialized based on feasible threshold combinations under historical frequency modulation scenarios. The fitness function uses frequency stability index as the core evaluation parameter, comprehensively evaluating the weighted score of threshold switching times and frequency recovery rate. After every 5-10 generations of iteration, the top 10% of elite individuals in fitness from each subgroup are migrated to adjacent subgroups, and new threshold combinations are generated through arithmetic crossover operations, expanding the global search range of trigger thresholds.

[0094] The triggering conditions for the dynamic switching logic are tuned in real time using a rolling time-domain optimization model, inputting the predicted frequency trend data for the next 3-5 sampling cycles into a threshold comparator. When the frequency deviation change rate exceeds 120% of the current threshold or the equivalent inertia parameter falls below a critical value, a rapid power support mode switching command is triggered. This critical value is set based on the wind turbine rotor stress analysis model, and the threshold boundary is dynamically adjusted in conjunction with the real-time speed recovery gradient to prevent false triggering due to parameter mutations. After the trigger signal is generated, the control command prioritizes releasing 85%-95% of the pre-stored rotor kinetic energy, and the power output curve is smoothed through the inertia compensation algorithm of the converter control loop.

[0095] A dynamic correlation matrix is ​​established between the boundary layer parameters and trigger thresholds of the sliding mode control model. When the genetic algorithm generates new threshold combinations, the boundary layer thickness is adjusted synchronously through the matrix mapping relationship. The update amount of the boundary layer parameters is calculated based on the steady-state error gradient of the frequency deviation, and the thickness coefficient is iteratively corrected using a recursive least squares method to balance the speed and smoothness of the control response. The switching gain parameter is adjusted by weighting the energy distribution of the disturbance characteristics. In high-frequency disturbance scenarios, the gain value is increased to 1.2-1.5 times the baseline level to enhance the ability to suppress sudden disturbances.

[0096] The mode switching process incorporates a dual verification mechanism. Before issuing control commands, the feasibility of the switching strategy is verified through a virtual simulation module. The simulation module operates based on a real-time power grid equivalent model, simulating frequency response curves under different threshold combinations to screen candidate schemes that meet the constraints of frequency fluctuation suppression rate and mechanical stress. Verified switching commands are encapsulated using a standardized protocol and distributed to the actuators. Simultaneously, they are fed back to the genetic algorithm population database to update the historical feasible solution set, accelerating the subsequent optimization process.

[0097] The anomaly handling mechanism is activated in real time by monitoring the deviation of the sliding surface trajectory. When the control quantity exceeds the preset safety boundary or the frequency deviation poses a secondary drop risk, a local model rollback is triggered. The rollback mechanism calls the three most recent valid parameter combinations from the historical parameter database and uses a majority voting mechanism to select the optimal parameter to cover the current abnormal state. The control model after rollback maintains stable operation until the incremental learning module completes a new round of parameter optimization, forming a closed-loop control link of fault self-healing and parameter update.

[0098] Step S8: When the frequency deviation recovers to the preset dead zone range, the microservice exits frequency modulation and switches to maximum power point tracking operation by limiting the adaptive power ramp rate.

[0099] When the frequency deviation enters the preset dead zone range, the adaptive power ramp rate limiting microservice initiates the frequency regulation exit process, monitoring the steady-state holding time of the frequency deviation through a sliding time window. The dead zone range is dynamically set based on the power grid frequency regulation regulations; a steady-state recovery condition is deemed met when the absolute value of the frequency deviation at five consecutive sampling points within the window is below a threshold. The microservice calls a genetic algorithm to solve for the optimal power descent gradient parameters. Constraints include the upper limit of the wind turbine speed recovery rate and the risk coefficient of a second frequency drop. The initial population of the algorithm utilizes feasible solutions from historical frequency regulation exit scenarios to accelerate convergence, outputting a combination of ramp rate parameters that balances safety and economy.

[0100] During the power ramp rate adjustment phase, the converter control module gradually reduces the compensated power output according to the ramp curve generated by the genetic algorithm. The ramp parameter is dynamically corrected based on the real-time speed recovery gradient. When the speed change rate is detected to be lower than the preset safety threshold, a gradient relaxation mechanism is triggered to temporarily expand the allowable descent range to prevent overload of the mechanical transmission chain. The corrected ramp command is encapsulated in a standardized protocol and distributed to the virtual synchronizer and energy storage device through the API gateway to coordinate the synchronous reduction of output by multiple source devices and maintain grid power balance.

[0101] When switching to maximum power point tracking (MPPT) operation, the wind turbine control system employs a gradual mode transition strategy. Initially, 10%-15% of the rotor kinetic energy is retained as an inertial buffer. The pitch angle and converter reference power are dynamically adjusted using a speed-power mapping table to ensure a smooth transition of output power along the MPPT curve. During the transition, the frequency deviation trend is continuously monitored. If the deviation exceeds the dead zone again, the process is immediately interrupted and the frequency regulation mode is reactivated, forming a closed-loop protection mechanism for mode switching.

[0102] The anomaly handling module identifies potential speed change risks by monitoring frequency fluctuations and equipment status parameters during power degradation. When the second derivative of the speed recovery rate exceeds the mechanical strength limit, a dynamic relaxation mechanism is triggered to temporarily expand the constraint boundaries of the genetic algorithm and re-solve the safe exit strategy. Simultaneously, the local model rollback function is invoked to load the previously validated ramp rate parameters, overwriting the current anomaly-driven optimization results and maintaining the stability of the control process.

[0103] The adaptive power ramp rate limiting microservice constructs a complete frequency regulation exit closed loop, with dead zone monitoring triggering the optimization process, a genetic algorithm generating safety parameters, multi-source devices collaboratively executing power ramp control, and anomaly handling ensuring transition reliability. Each technical component achieves state synchronization through an event bus and data pipeline, dynamically adjusting command parameters and device response logic to achieve seamless switching from frequency regulation mode to maximum power point tracking mode, thus improving the long-term stability of the power grid frequency.

[0104] The fast frequency regulation method for wind turbines based on disturbance adaptive compensation provided by this invention includes the following steps:

[0105] Real-time acquisition of power system frequency deviation, frequency deviation rate of change, and wind turbine operating status parameters constructs a dynamic disturbance sensing dataset. Data acquisition is achieved through a sensor network deployed at key nodes of the power grid. Sensor types include frequency measurement units, phasor measurement units, and wind turbine status monitoring modules. The sampling frequency is dynamically adjusted according to the characteristics of power grid disturbances. After timestamp alignment and outlier removal preprocessing, the acquired data is stored as a data structure containing a time-series sequence of frequency deviation, statistical characteristics of the frequency deviation rate of change, and the wind turbine speed-power mapping relationship, forming a dynamic disturbance sensing dataset that provides input for subsequent feature extraction and control model updates.

[0106] Based on a dynamic disturbance sensing dataset, a genetic algorithm is used to dynamically optimize the window length and weight coefficients of a short-window sliding mean algorithm to generate noise-suppressed disturbance feature data. The initial population of the genetic algorithm is composed of a combination of sliding window parameters, and the fitness function uses the signal-to-noise ratio and trend preservation ability of the disturbance features as evaluation indicators. During population evolution, a roulette wheel selection mechanism is used to select individuals, and the crossover and mutation probabilities are dynamically adjusted based on the population diversity index. GPU parallel computing accelerates fitness evaluation; each thread independently calculates the noise suppression effect of different window parameter combinations, outputting the optimal window length and weight coefficient combination to suppress the interference of power grid background noise on the disturbance features.

[0107] A reverse test signal is generated based on the noise-suppressed disturbance characteristic data, and the equivalent inertia and damping parameters of the power grid are estimated in real time using the recursive least squares method. The amplitude of the reverse test signal is initialized based on the historical disturbance energy distribution, and the timing parameters are adjusted synchronously with the current disturbance characteristics to reduce additional frequency interference. The equivalent inertia and damping parameters of the power grid are identified online using the recursive least squares method. During the identification process, a forgetting factor is introduced to optimize the weight of historical disturbance data, improving the dynamic tracking capability of parameter estimation. The identification results serve as input parameters for the fuzzy controller, supporting the subsequent generation of compensation power commands.

[0108] An online incremental learning algorithm is employed to model the time series of continuous disturbance events in a dynamic disturbance sensing dataset, updating the boundary layer thickness and switching gain parameters of the adaptive sliding mode control model. The incremental learning algorithm extracts real-time data streams of continuous disturbance events through a sliding time window, performs dimensionality reduction using principal component analysis, and then inputs these data into the sliding mode control model. Recursive least squares is used to update boundary layer parameters based solely on incremental data, reducing computational complexity. When data anomalies are detected, a local model rollback mechanism is triggered to load valid parameters from the previous time period, maintaining control stability. The learning rate is dynamically adjusted based on the gradient of the model prediction error; parameter updates are accelerated as the error increases, and the search accuracy is refined when the error converges.

[0109] Based on real-time estimated equivalent inertia and damping parameters of the power grid, a fuzzy adaptive proportional-integral controller generates compensation power commands. The membership function parameters of the fuzzy controller are dynamically partitioned using a clustering algorithm, and the activation weights of the fuzzy rule base are iteratively optimized by a reinforcement learning mechanism based on historical frequency regulation effects. The controller's input variables include equivalent inertia, frequency deviation, and rate of change, and the output is a multi-level compensation power command. In high-frequency disturbance scenarios, the resolution of the high-frequency band membership function is adjusted first to enhance the weight allocation of corresponding rules and improve the dynamic adaptability of the compensation commands.

[0110] Based on the compensation power command, a microservice architecture coordinates the allocation of wind turbine rotor kinetic energy reserves and the power response of multi-source frequency regulation equipment in the power grid. This microservice architecture adopts a containerized deployment model, dynamically planning microservices to parse compensation commands and optimize the rotor kinetic energy release ratio and converter power output. The API gateway integrates heterogeneous communication protocols of virtual synchronous machines, energy storage devices, and traditional units, generating a standardized frequency regulation command format. Event-driven priority-weighted microservices dynamically adjust equipment output weights based on disturbance energy entropy. When a power conflict is detected, a redundancy elimination algorithm reallocates frequency regulation tasks, optimizing the collaborative efficiency of multi-source equipment.

[0111] In model predictive control, a genetic algorithm is introduced to optimize the trigger threshold for dynamic switching, switching between fast power support mode and droop control mode based on the frequency deviation change rate. The rolling time-domain optimization model takes historical disturbance data and real-time inertia parameters as input to predict the frequency trend of future sampling periods. A multi-island genetic algorithm divides the initial population and migrates elite individuals to generate differentiated trigger threshold candidate schemes. When the frequency deviation change rate exceeds the preset threshold, the system switches to fast power support mode to release pre-stored rotor kinetic energy. During the droop control phase, the power slope curve is dynamically adjusted in conjunction with the fan speed recovery rate to balance the frequency modulation exit speed and mechanical stress limitations.

[0112] When the frequency deviation recovers to the preset dead zone range, the microservice exits frequency regulation and switches to maximum power point tracking (MPPT) operation via an adaptive power ramp rate constraint. The power ramp rate parameter is optimized using a genetic algorithm within a hard-constrained space, with constraints including the turbine speed recovery rate and the frequency deviation range. Feasible solutions from historical frequency regulation exit data are used to initialize the search space, shortening the optimization time. A dynamic relaxation mechanism temporarily expands the constraint boundaries and re-solves the safe exit strategy when a sudden speed change is detected, achieving a seamless switch from frequency regulation mode to MPPT mode.

[0113] The above steps, through a logical closed loop of dynamic disturbance perception, parameter optimization, control command generation, and multi-source collaborative execution, improve the frequency regulation response speed and stability of wind turbines in complex power grid environments. Data flow and control commands are tightly coupled between each step; genetic algorithms and incremental learning collaboratively optimize model parameters; and a microservice architecture ensures real-time synchronization of commands across devices, ultimately achieving high-precision frequency control of the power system.

[0114] Specifically, in the wind turbine rapid frequency regulation method based on disturbance adaptive compensation described in this invention, step S2 includes:

[0115] Based on the dynamic perturbation sensing dataset, the crossover probability and mutation probability are dynamically adjusted according to the population fitness distribution. When the population diversity is lower than a set threshold, the mutation probability is increased to the range of 0.2-0.4.

[0116] An elite retention strategy is adopted to directly pass the best individuals of each generation to the next generation, and the distribution characteristics of the perturbation feature data are combined to accelerate parameter convergence.

[0117] By combining genetic algorithms with GPU parallel computing, the window length and weight coefficients of the short-window sliding mean algorithm are evaluated in parallel.

[0118] The test signal amplitude parameters are initialized based on historical disturbance data in the dynamic disturbance sensing dataset, and the frequency interference of inertia estimation is reduced by reverse signal excitation.

[0119] Based on a dynamic perturbation-aware dataset, the crossover and mutation probabilities are dynamically adjusted according to the population fitness distribution. The population fitness distribution is quantified by calculating the ratio of the standard deviation to the mean of individual fitness values. When the population diversity index is below a preset threshold, the mutation probability is adaptively increased to the range of 0.2-0.4. The fitness function uses the signal-to-noise ratio of perturbation features extracted by the short-window moving average algorithm and the trend fitting error as evaluation indicators. A roulette wheel selection mechanism is combined to screen high-fitness individuals, balancing the needs of global search and local optimization.

[0120] An elite retention strategy is employed to directly pass the best individuals from each generation to the next, accelerating the parameter convergence process. The window length and weight coefficient combination of elite individuals are optimized based on the distribution characteristics of perturbation feature data, including the fluctuation amplitude and gradient of frequency deviation. A tournament selection mechanism is used to select the top 10% of individuals in terms of fitness from the current population, and their window parameters are used as the initial solution for the moving average algorithm to guide the direction of subsequent iterations and improve feature extraction efficiency.

[0121] This paper combines genetic algorithms with GPU parallel computing to perform parallel evaluation of the window length and weight coefficients of the short-window moving average algorithm. The parallel computing task is divided into multiple thread blocks, each independently calculating the noise suppression effect of different parameter combinations. The fitness scores of multiple parameter combinations are processed synchronously using the high-speed read / write capabilities of GPU memory. After integrating the evaluation results of each thread block, the optimal window parameter combination is selected. The parallel computing framework is implemented based on a unified computing device architecture, significantly reducing the time per iteration.

[0122] The amplitude parameters of the test signal are initialized based on historical disturbance data from the dynamic disturbance sensing dataset, generating a reverse test signal opposite to the current disturbance direction. The timing parameters of the test signal are dynamically adjusted according to historical disturbance patterns, including disturbance duration and energy distribution characteristics. The reverse signal excitation is superimposed onto the equivalent inertia estimation stage of the power grid to suppress frequency interference during inertia parameter identification. The recursive least squares method introduces a forgetting factor during identification, reducing the impact of historical noise on real-time estimation results and improving parameter identification accuracy.

[0123] The above steps form a complete chain for the genetic algorithm to optimize the short-window sliding mean parameters: population diversity monitoring drives dynamic adjustment of crossover and mutation probabilities, elite retention strategy accelerates convergence direction guidance, GPU parallel computing improves evaluation efficiency, and historical perturbation data supports the initialization of test signal parameters. Through the collaboration of multiple mechanisms, dynamic optimization and noise suppression of the sliding window parameters are achieved, providing high-precision perturbation feature data for subsequent inertia estimation and frequency modulation control.

[0124] Specifically, the wind turbine fast frequency regulation method based on disturbance adaptive compensation described in this invention further includes:

[0125] Based on the execution feedback data of the compensation power command, the activation weights of the fuzzy rules are adjusted using a reinforcement learning mechanism;

[0126] The fuzzy intervals of the equivalent inertia parameters of the power grid are dynamically divided by clustering algorithm, and the center point and width parameters of the membership function are optimized.

[0127] Establish the parameter correlation matrix between the adaptive sliding mode control model and the fuzzy controller, and update the boundary layer thickness and input range synchronously;

[0128] When a high-frequency disturbance mode is detected, the high-frequency band membership function parameters of the equivalent inertia parameter of the power grid are optimized first, and the weight allocation of the corresponding rules is enhanced.

[0129] Based on the execution feedback data of the compensation power command, a reinforcement learning mechanism is used to adjust the activation weights of the fuzzy rules. The execution feedback data includes frequency fluctuation suppression rate, frequency modulation energy consumption, and power overshoot, which are normalized to construct a state-action reward matrix. The Q-learning algorithm iteratively updates the rule weight coefficients, prioritizing the strengthening of rule activation paths in high-frequency disturbance scenarios. Historical disturbance pattern matching results are used to weight the reward values ​​of similar scenarios, improving the dynamic adaptability of the fuzzy controller and optimizing the response efficiency of the compensation command.

[0130] The fuzzy intervals of the equivalent inertia parameters of the power grid are dynamically divided using a clustering algorithm, optimizing the center point and width parameters of the membership function. The clustering algorithm constructs feature vectors based on the mean, variance, and gradient of the inertia parameters obtained through sliding window statistics, and uses an improved K-means algorithm to calculate the optimal cluster centers. When an inertia parameter distribution offset exceeds a preset threshold, the membership function parameters are recalibrated, and the input range and rule matching threshold of the fuzzy controller are updated synchronously to adapt to the dynamic characteristics of the power grid inertia.

[0131] A parameter correlation matrix is ​​established between the adaptive sliding mode control model and the fuzzy controller to achieve synchronous updates of the boundary layer thickness and the input range. The correlation matrix defines the mapping relationship between the sliding mode switching gain parameter and the fuzzy input universe of discourse, and dynamically corrects the matrix weight coefficients through real-time frequency deviation gradient. When the boundary layer parameters of the sliding mode model are updated, the correlation matrix drives the adjustment of the fuzzy controller's input range, eliminating response delay between the two models and avoiding control command oscillations.

[0132] When a high-frequency disturbance pattern is detected, the high-frequency membership function parameters of the equivalent inertia parameters of the power grid are optimized first, and the weight allocation of the corresponding rules is enhanced. Fast Fourier Transform analysis is used to analyze the spectral characteristics of the inertia parameters to identify high-frequency disturbance events where the rate of change of frequency deviation exceeds a preset threshold. For high-frequency disturbances, the membership function width is reduced to 60% of the baseline value to improve rule matching resolution. The activation weights of the corresponding high-frequency bands in the fuzzy rule base are enhanced through a weighted summation mechanism, prioritizing the triggering of high-frequency compensation commands and accelerating power response speed.

[0133] The above steps form a closed-loop optimization chain: the execution effect of compensation instructions is fed back to the reinforcement learning module to optimize rule weights; real-time inertia parameter clustering divides fuzzy intervals to support precise control; sliding mode and fuzzy model parameters are updated in association to improve collaborative efficiency; high-frequency disturbance features drive priority adjustment to achieve scenario adaptation. Through multi-level technical collaboration, the frequency adjustment accuracy under complex disturbances is enhanced, and frequency modulation resource consumption and equipment impact are reduced.

[0134] Specifically, in the wind turbine fast frequency regulation method based on disturbance adaptive compensation described in this invention, step S6 includes:

[0135] In the containerized deployment framework of the model predictive control, the power allocation ratio between rotor kinetic energy reserve and converter capacity is optimized through dynamic programming microservices.

[0136] The communication protocols of the multi-source frequency regulation equipment in the power grid are integrated through the API gateway to generate a cross-device frequency regulation command synchronization protocol.

[0137] Based on the dynamic disturbance perception dataset, the disturbance energy entropy is calculated, and an event-driven priority weight microservice is designed to dynamically adjust the output ratio.

[0138] When a power conflict is detected, a redundancy elimination algorithm is triggered to reallocate the frequency regulation tasks of the virtual synchronizer and the energy storage device.

[0139] In the containerized deployment framework of the model predictive control, the power allocation ratio between rotor kinetic energy reserve and converter capacity is optimized through a dynamic programming microservice. This dynamic programming microservice parses compensation power commands, inputting the wind turbine rotor kinetic energy reserve capacity, converter output upper limit, and grid frequency recovery requirements into a multi-objective optimization model. Based on frequency stability constraints in the rolling time domain, a linear programming algorithm is used to solve for the optimal combination of rotor kinetic energy release ratio and converter power output, generating power allocation commands adapted to dynamic disturbance scenarios. During the optimization process, the constraint boundaries are dynamically adjusted in conjunction with real-time inertia parameters, and the allocation commands are distributed to frequency regulation device nodes via a lightweight message queue.

[0140] A cross-device frequency regulation command synchronization protocol is generated by integrating heterogeneous communication protocols of multi-source frequency regulation equipment in the power grid through an API gateway. The API gateway defines a standardized command interaction interface based on a RESTful architecture, encapsulating the proprietary communication protocols of virtual synchronous machines, energy storage devices, and traditional generating units into a unified JSON format. The gateway monitors the communication latency and bandwidth utilization of each device in real time, dynamically adjusting the data packet transmission priority to ensure the timing consistency of cross-device frequency regulation commands. The command synchronization protocol includes equipment status query, power command issuance, and execution feedback confirmation processes, and improves command transmission reliability through a timeout retransmission mechanism.

[0141] Based on a dynamic disturbance sensing dataset, disturbance energy entropy is calculated, and an event-driven priority-weighted microservice is designed to dynamically adjust the output ratio. The disturbance energy entropy quantifies the concentration of disturbance energy by statistically analyzing the spatiotemporal distribution characteristics of frequency deviation, rate of change, and power deficit through a sliding time window. When the entropy value exceeds a preset threshold, the priority-weighted microservice dynamically allocates the output weights of the virtual synchronizer and energy storage device based on the real-time available capacity, response rate, and frequency regulation cost coefficient of the multi-source frequency regulation equipment. The weight allocation results are shared through a distributed cache of the microservice cluster, supporting proportional synchronous adjustment of output targets by each frequency regulation node.

[0142] When a power conflict is detected, a redundancy elimination algorithm is triggered to reallocate the frequency regulation tasks of the virtual synchronizer and the energy storage device. The conflict detection module monitors the output superposition effect of multiple source devices in real time, identifying power overshoot or contradictory output directions. The redundancy elimination algorithm constructs an optimization model based on the real-time operating status parameters of the conflicting devices, prioritizing the adjustment of the output ratio of devices with response delays exceeding a threshold. A genetic algorithm initializes the population with a set of historical feasible solutions, quickly solves conflict resolution schemes within a hard-constrained space, and redistributes the corrected frequency regulation commands through containerized microservices to achieve consistent collaborative output from multiple source devices.

[0143] The above steps form a complete closed loop for multi-source collaborative frequency regulation: dynamic programming generates the initial allocation strategy, the API gateway ensures synchronous transmission of commands across devices, disturbance energy entropy drives dynamic priority adjustment, and redundancy elimination algorithms resolve output conflicts. Through containerized deployment and the elastic scalability of a microservice architecture, it adapts to the frequency regulation needs of power grids of different sizes, improving control command response efficiency and system frequency stability under multi-disturbance scenarios.

[0144] Specifically, the wind turbine rapid frequency regulation method based on disturbance adaptive compensation described in this invention includes step S7 as follows:

[0145] The historical disturbance data and real-time inertia parameters in the dynamic disturbance sensing dataset are input into the rolling time domain optimization model to predict the frequency trend for the next 3-5 sampling periods.

[0146] The initial population of the genetic algorithm is divided using a multi-island genetic algorithm, and elite individuals are periodically migrated to generate differentiated trigger thresholds;

[0147] When the frequency deviation change rate exceeds the start-up threshold, switch to the fast power support mode to release the rotor kinetic energy reserve;

[0148] During the droop control phase, a power slope curve is generated based on the compensation power command. The wind turbine speed recovery rate is calculated by combining the speed change gradient after the release of the wind turbine rotor kinetic energy reserve. The slope limit is then dynamically updated based on the wind turbine speed recovery rate.

[0149] The historical disturbance data and real-time inertia parameters are input into a rolling time-domain optimization model to predict the frequency trend over the next 3-5 sampling periods. The rolling time-domain optimization model constructs a prediction input vector by extracting historical frequency deviation, rate of change, and inertia parameter time-series data through a sliding time window. An autoregressive integral moving average algorithm is used to generate frequency trend curves for future sampling periods. The prediction results include frequency fluctuation extremes, recovery time windows, and deviation integral characteristics. The predicted data is used to optimize trigger threshold parameters, supporting the generation of dynamic switching strategies and improving the foresight of mode switching.

[0150] A multi-island genetic algorithm is used to divide the initial population of the genetic algorithm, and elite individuals are periodically migrated to generate differentiated trigger thresholds. The multi-island genetic algorithm divides the search space into multiple independently evolving subgroups, and each subgroup independently optimizes the combination of trigger thresholds. Every preset number of iterations, the top 10% of elite individuals in terms of fitness are migrated to adjacent subgroups, and candidate solutions for differentiated thresholds are generated through crossover operations. The fitness function uses the frequency stability index as the core evaluation parameter, combined with the constraint of the number of switching actions, to screen the optimal threshold combination, balancing response speed and system stability.

[0151] When the frequency deviation rate exceeds a preset activation threshold, the system switches to a rapid power support mode to release the rotor's kinetic energy reserves. The activation threshold is dynamically adjusted based on historical disturbance data and real-time inertia parameters, and a threshold optimization algorithm reduces the risk of false triggering. In rapid power support mode, 80%-90% of the pre-stored rotor kinetic energy is released first, and compensation power is rapidly injected through the converter. The release ratio is dynamically adjusted according to the frequency recovery demand predicted in the rolling time domain, and combined with an inertial compensation algorithm, the power output curve is smoothed, reducing secondary impacts on the power grid.

[0152] During the droop control phase, a power slope curve is generated based on the compensated power command, and the slope limit is dynamically updated in conjunction with the fan speed recovery rate. The initial value of the power slope is set according to the frequency recovery target, and the power reduction amplitude is dynamically adjusted by monitoring the gradient change of the fan speed recovery rate in real time. When the speed recovery rate is lower than the safety threshold, the power slope is reduced to reduce mechanical stress; when the rate stabilizes, the slope is gradually increased to accelerate the exit from frequency regulation. Dead-zone control logic is introduced in the slope adjustment process to shield the interference of small speed fluctuations on parameter updates and ensure the continuity of control commands.

[0153] The above steps generate trigger thresholds through rolling prediction and multi-island optimization, enabling seamless switching between rapid power support and droop control modes. Historical disturbance data drives the predictive model to improve strategy foresight, while the multi-island genetic algorithm enhances the globality of threshold search. Dynamic slope adjustment balances frequency regulation exit speed and equipment safety. Data flows and control commands at each stage are tightly coupled, forming a closed-loop control link that adapts to time-varying inertia and multi-disturbance scenarios, significantly improving the frequency regulation response accuracy of wind turbines and system frequency stability.

[0154] Specifically, the wind turbine fast frequency regulation method based on disturbance adaptive compensation described in this invention further includes:

[0155] The wind turbine speed recovery rate and frequency deviation are used as hard constraints to construct a gradient search space; the optimal power descent gradient parameters are solved within the constraint space using the genetic algorithm.

[0156] When a risk of sudden speed change is detected, a dynamic relaxation mechanism is triggered to expand the constraint boundary of the sliding mode control model;

[0157] By initializing the search space with the optimal gradient parameters from historical frequency modulation exit data, the generation time of the safe recovery strategy can be shortened.

[0158] The wind turbine speed recovery rate and frequency deviation are used as hard constraints to construct a gradient search space. These hard constraints are set based on the wind turbine's mechanical strength threshold and the grid frequency dead zone, defining the boundary of the multidimensional feasible solution region. The upper limit of the speed recovery rate constraint is determined through a wind turbine rotor stress analysis model, while the frequency deviation constraint is dynamically adjusted according to the grid frequency regulation regulations. The coupling relationship between speed and frequency is transformed into mathematical constraints through hyperplane equations, limiting the search range of the genetic algorithm and ensuring the physical feasibility of the power reduction strategy.

[0159] The optimal power descent gradient parameters are solved within a constrained space using a genetic algorithm. The initial population of the genetic algorithm is generated based on feasible solutions from historical frequency modulation exit data. The fitness function uses the suppression effect of frequency second-order drop during the power descent process and the smoothness of speed recovery as evaluation indicators. A constraint violation penalty mechanism is used to screen candidate solutions that meet the hard constraints. Through crossover and mutation operations, iterative optimization is performed to output a power ramp rate parameter combination that balances safety and economy, supporting a smooth transition during the frequency modulation exit phase.

[0160] When a risk of sudden speed change is detected, a dynamic relaxation mechanism is triggered to expand the constraint boundary of the sliding mode control model. This risk is identified by real-time monitoring of the absolute value of the speed change gradient. When the gradient value exceeds 80% of a preset safety threshold, the constraint boundary layer thickness is adjusted. The dynamic relaxation mechanism temporarily expands the feasible solution region boundary, allowing the genetic algorithm to resolve within the expanded search space. Combined with the speed recovery trend, the relaxed constraint range is gradually narrowed until the risk of sudden change is eliminated, restoring the original constraint conditions.

[0161] The search space is initialized using the optimal gradient parameters from historical frequency modulation exit data, shortening the generation time of the safe recovery strategy. The historical disturbance data is stored categorized by disturbance scenario, including frequency deviation range, inertia parameters, and rotational speed recovery characteristics. When a current disturbance mode matches a historical record, the optimal solution from the same scenario is preferentially loaded as the initial individual for the genetic algorithm. A similarity-weighted mechanism is used to correct the fitness scores of historical solutions, accelerating population convergence. Historical disturbance data is dynamically maintained using a sliding time window update mechanism, eliminating records with insufficient timeliness to ensure the real-time effectiveness of the initialization strategy.

[0162] The above steps form a closed loop for safe recovery during the frequency regulation exit phase: hard constraints define and ensure the safety of the strategy, a genetic algorithm searches for the optimal solution within the feasible space, dynamic relaxation handles sudden anomalies, and historical disturbance data is injected to improve optimization efficiency. Through the synergy of multiple mechanisms, the generation time of the power reduction strategy is shortened, the risk of mechanical damage caused by sudden speed changes is reduced, and secondary frequency drops are suppressed, enabling a smooth transition of the wind turbine from frequency regulation mode to maximum power point tracking mode.

[0163] Specifically, the wind turbine rapid frequency regulation method based on disturbance adaptive compensation described in this invention includes step S8, which comprises:

[0164] Real-time data streams of the continuous disturbance events are collected through a sliding time window, and principal component analysis is performed for dimensionality reduction.

[0165] The boundary layer parameters of the adaptive sliding mode control model are updated only based on incremental data using the recursive least squares method.

[0166] When data anomalies are detected, a local model rollback mechanism is triggered to restore the parameters to the previous time period.

[0167] The learning rate is dynamically adjusted based on the rate of change of the error predicted by the model, thereby increasing the parameter update speed as the error gradient increases.

[0168] Real-time data streams of the continuous disturbance events are acquired through a sliding time window, and principal component analysis (PCA) is performed for dimensionality reduction. The length of the sliding time window is dynamically adjusted according to the duration of a typical disturbance, and the window includes time-series data on frequency deviation, rate of change, and power deficit. PCA calculates the eigenvalue distribution of the covariance matrix, filters principal components with contribution rates exceeding a preset threshold, extracts low-dimensional eigenvectors characterizing the disturbance, eliminates redundant noise interference, and generates an input dataset suitable for the sliding mode control model, thereby improving the efficiency of subsequent parameter updates.

[0169] The boundary layer parameters of the adaptive sliding mode control model are updated using a recursive least squares method based solely on incremental data. This recursive update process retains the covariance matrix and residual information of historical parameters, and calculates parameter corrections using only incremental data each time a new data window arrives. The boundary layer thickness and switching gain parameters are updated via iterative formulas, avoiding the computational overhead of retraining with full data. The updated parameters directly affect the switching frequency and tracking accuracy of the sliding mode control, balancing control response speed and system stability.

[0170] When an anomaly is detected, a local model rollback mechanism is triggered to restore the parameters to the previous time period. Anomaly detection is achieved by monitoring the statistical characteristics of the data stream, including mean drift, variance mutations, and outlier identification exceeding the confidence interval. The rollback mechanism calls the historical parameter database, loads the sliding mode control parameters that have been verified to be valid in the previous time period, and overwrites the update results of the current anomaly window. The rolled-back model maintains stable output until subsequent continuous window data passes integrity verification and the incremental update process resumes, ensuring the continuity of control commands.

[0171] The learning rate is dynamically adjusted based on the rate of change of the model's prediction error, increasing the parameter update speed as the error gradient increases. The rate of change of error is calculated using the difference in prediction errors within a sliding window, combined with an exponentially weighted average to eliminate instantaneous fluctuations. When the absolute value of the error gradient exceeds a preset threshold, the learning rate increases according to a logarithmic function, accelerating parameter convergence; when the error gradient approaches zero, the learning rate gradually decreases to a baseline value, refining the parameter search accuracy. This dynamic adjustment mechanism balances parameter update speed and stability, avoiding overfitting or underfitting.

[0172] The above steps form a complete closed loop for online adaptive sliding mode control: sliding window data acquisition supports feature extraction, recursive updates reduce computational load; anomaly rollback ensures model robustness, and dynamic adjustment of the learning rate optimizes the convergence process. Through multi-mechanism collaboration, control parameters are adapted to dynamic disturbance characteristics in real time, improving the response speed and anti-interference capability of wind turbine frequency regulation control in complex scenarios, and enhancing the frequency stability of the power system.

[0173] The technical features of this invention are explained as follows:

[0174] Dynamic Disturbance Sensing Dataset: This dataset, comprised of frequency measurement units, phasor measurement units, and wind turbine condition monitoring modules deployed at grid nodes, collects data in real-time, including frequency deviation, rate of change of frequency deviation, and wind turbine speed-power mapping parameters. The raw data is timestamped and then outliers are removed using a sliding window mechanism, resulting in a structured dataset containing both temporal and statistical features. This dataset provides the foundation for subsequent disturbance feature extraction and parameter identification, and its dynamic update characteristics adapt to the time-varying nature of grid disturbances.

[0175] The genetic algorithm optimizes the short-window sliding mean algorithm: The genetic algorithm constructs an initial population using the sliding window length and weight coefficients as optimization variables. The fitness function evaluates individual performance based on the signal-to-noise ratio of perturbation features and trend fitting error. A roulette wheel selection mechanism is used to select high-fitness individuals, dynamically adjusting the crossover and mutation probabilities (increasing the mutation probability to the 0.2-0.4 range when population diversity is insufficient). GPU parallel computing is combined to accelerate the evaluation of noise suppression effects of window parameter combinations, outputting the optimal parameter combination to suppress the interference of power grid background noise on perturbation feature extraction.

[0176] Recursive least squares inertia parameter estimation: A reverse test signal is generated based on the noise-suppressed disturbance characteristic data. The signal amplitude is initialized according to the historical disturbance energy distribution, and the time-series parameters are adjusted synchronously with the current disturbance characteristics. The recursive least squares method introduces a forgetting factor (0.95-0.98) to optimize the weight ratio of historical disturbance data. The estimated values ​​of the power grid equivalent inertia and damping parameters are corrected in real time through incremental updates, thereby improving the tracking accuracy of parameters in time-varying inertia scenarios.

[0177] Online incremental learning updates sliding mode control parameters: A sliding time window captures the data stream of continuous disturbance events, which is then dimensionality-reduced using principal component analysis and input into the sliding mode control model. Recursive least squares is used to update the boundary layer thickness and switching gain parameters based solely on incremental data, reducing computational complexity. When a mean shift or variance mutation is detected, a local model rollback mechanism is triggered, loading parameters that have been validated in the previous time period to maintain control stability.

[0178] Fuzzy Adaptive Proportional-Integral Controller: The membership function is dynamically partitioned using the K-means clustering algorithm, with the number of cluster centers set to 3-5 based on the inertia parameter distribution characteristics. The reinforcement learning mechanism uses frequency modulation energy consumption and frequency overshoot as reward indicators to iteratively update the activation weights of fuzzy rules. In high-frequency disturbance scenarios, the width of the high-frequency membership function is reduced to 60% of the baseline value, enhancing the weight allocation of corresponding rules and improving the dynamic response speed of compensation commands.

[0179] Multi-source collaborative control in a microservice architecture: Dynamically programmed microservices parse compensation power commands, using frequency stability as a constraint to optimize the rotor kinetic energy release ratio and converter power output. The API gateway encapsulates the heterogeneous communication protocols of the virtual synchronizer and energy storage devices into standardized JSON commands, ensuring cross-device command synchronization through a timeout retransmission mechanism. Disturbance energy entropy quantifies the concentration of disturbance energy, triggering event-driven priority weight adjustments. A redundancy elimination algorithm, based on a genetic algorithm, reallocates frequency modulation tasks for conflicting devices within 10-20 seconds.

[0180] Genetic algorithm optimization mode switching threshold: The rolling time-domain optimization model takes historical disturbance data and real-time inertia parameters as input and predicts the frequency trend over the next 3-5 sampling periods. The multi-island genetic algorithm divides the search space into 4-6 subgroups and periodically migrates elite individuals to generate differentiated trigger thresholds. When the frequency deviation change rate exceeds 120% of the dynamic tuning threshold, it switches to the fast power support mode to release 80%-90% of the pre-stored rotor kinetic energy; during the droop control phase, the power slope curve is dynamically adjusted according to the speed recovery rate gradient, with the initial slope set to 5% / second of the rated power.

[0181] Adaptive power ramp rate exit mechanism: The genetic algorithm uses a speed recovery rate ≤2% / second and a frequency deviation ≤0.05Hz as hard constraints to solve for the optimal power descent gradient parameters within a search space initialized with historical feasible solutions. The dynamic relaxation mechanism temporarily expands the constraint boundary by 20%-30% when the speed gradient abruptly exceeds the safety threshold of 80%, combining historical frequency modulation exit data to accelerate the optimization process and achieve a seamless switch from frequency modulation mode to maximum power point tracking mode.

[0182] Construction of a dynamic disturbance sensing dataset: Frequency deviation, frequency deviation rate of change, and wind turbine speed-power mapping parameters are collected in real time through frequency measurement units, phasor measurement units, and wind turbine status monitoring modules deployed at power grid nodes. After timestamp alignment and outlier removal via a sliding window, the collected data forms a structured dataset containing time-series sequences and statistical features. This dataset adapts to power grid disturbance characteristics through a dynamic update mechanism, providing fundamental data support for subsequent noise suppression and parameter identification.

[0183] A genetic algorithm is used to optimize the short-window sliding mean algorithm: An initial population is constructed and individual fitness is evaluated using the sliding window length and weight coefficients as optimization variables. The fitness function uses the signal-to-noise ratio of perturbation features and trend fitting error as evaluation metrics, employing roulette wheel selection and elite retention strategies to select high-fitness individuals. When the population diversity index falls below a threshold, the mutation probability dynamically increases to the 0.2-0.4 range. GPU parallel computing is used to accelerate parameter combination evaluation, outputting the optimal window parameters and suppressing the interference of power grid background noise on perturbation feature extraction.

[0184] Recursive least squares inertia parameter estimation: A reverse test signal is generated based on the disturbance characteristics after noise suppression, and the signal amplitude is initialized according to the 80th percentile of the historical disturbance energy distribution. The equivalent inertia and damping parameters of the power grid are updated online by recursive least squares method. A forgetting factor (0.95-0.98) is introduced to reduce the weight of historical noise, and the parameter estimates are corrected in real time to improve the identification accuracy in time-varying inertia scenarios.

[0185] Online incremental learning updates sliding mode control parameters: A sliding time window is used to capture continuous disturbance event data streams, which are then input into the sliding mode control model after dimensionality reduction via principal component analysis. The recursive least squares method updates the boundary layer thickness and switching gain parameters based solely on incremental data. When a mean shift in the data is detected to exceed twice the standard deviation, a local model rollback mechanism is triggered to restore the effective parameters from the previous time period, maintaining the continuity and stability of the control commands.

[0186] Fuzzy Adaptive Proportional-Integral Controller: The fuzzy intervals of the equivalent inertia parameters of the power grid are dynamically divided using the K-means clustering algorithm, with 3-5 cluster centers, optimizing the center points and width of the membership function. A reinforcement learning mechanism uses frequency regulation energy consumption and frequency overshoot as reward indicators to iteratively update the activation weights of the fuzzy rules. Under high-frequency disturbance scenarios, the width of the high-frequency membership function is reduced to 60% of the baseline value, enhancing the priority of corresponding rules and improving the dynamic response speed of compensation commands.

[0187] Multi-source collaborative control in a microservice architecture: Dynamic programming microservices parse compensation power commands, using frequency stability as a constraint, and employing linear programming algorithms to optimize the rotor kinetic energy release ratio and converter power output. The API gateway encapsulates the Modbus and IEC 61850 protocols of virtual synchronizers and energy storage devices into standardized JSON commands, ensuring cross-device command synchronization through a timeout retransmission mechanism. Disturbance energy entropy calculates the concentration of disturbance energy, triggering event-driven priority weight adjustments. A redundancy elimination algorithm, based on a genetic algorithm, reallocates frequency modulation tasks for conflicting devices within 10-20 seconds.

[0188] Genetic algorithm optimization mode switching threshold: The rolling time-domain optimization model takes historical disturbance data and real-time inertia parameters as input and predicts the frequency trend over the next 3-5 sampling periods. The multi-island genetic algorithm divides the search space into 4-6 subgroups, and generates differentiated trigger thresholds every 10 generations of migration of elite individuals. When the frequency deviation change rate exceeds 120% of the dynamic tuning threshold, it switches to a fast power support mode to release 80%-90% of the pre-stored rotor kinetic energy, and combines this with an inertial compensation algorithm to smooth the power output curve.

[0189] Adaptive power ramp rate exit mechanism: With hard constraints of speed recovery rate ≤ 2% / second and frequency deviation ≤ 0.05Hz, the genetic algorithm solves for the optimal power descent gradient parameters within the search space initialized by the historical feasible solution set. Dynamic relaxation mechanism: When the speed gradient abruptly exceeds the safety threshold of 80%, the constraint boundary is temporarily expanded by 20%-30%, and the power ramp rate is adjusted in conjunction with a recursive update strategy to achieve a seamless switch from frequency modulation mode to maximum power point tracking mode.

[0190] The specific implementation of this invention is based on the frequency regulation requirements under dynamic multi-disturbance and time-varying inertia scenarios, and achieves rapid frequency regulation and multi-source collaborative control of wind turbine units through the following technical solutions:

[0191] During the dynamic disturbance sensing phase, frequency measurement units, phasor measurement units, and wind turbine status monitoring modules deployed at key power grid nodes collect frequency deviation, frequency deviation change rate, and wind turbine speed-power mapping parameters in real time. The sampling frequency is dynamically adjusted to 5-10 times / second based on the intensity of the power grid disturbance. After timestamp alignment and outlier removal via a sliding window, the collected raw data is used to construct a multidimensional dynamic disturbance sensing dataset containing temporal and statistical features. The window length and weight coefficients of the short-window sliding mean algorithm are dynamically optimized using a genetic algorithm. The initial population size is set to 50-100 parameter combinations, and the mutation probability is increased to the range of 0.2-0.4 when the population diversity is below a threshold. Combined with GPU parallel computing to accelerate fitness evaluation, the window parameters with the optimal signal-to-noise ratio are selected to generate noise-suppressed disturbance feature data. This data drives the generation of reverse test signals. The signal amplitude is initialized based on the 80th percentile of the historical disturbance energy distribution. Combined with the recursive least squares method, the equivalent inertia parameters of the power grid are identified online. The forgetting factor is set to 0.95-0.98 to balance the weight of historical disturbance data and improve the parameter tracking accuracy in time-varying inertia scenarios.

[0192] During the control strategy optimization phase, an online incremental learning algorithm is used to model continuous disturbance events. The sliding time window length is set to 3-5 disturbance cycles. Principal component analysis retains principal components with a contribution rate exceeding 85%, generating dimensionality-reduced feature vectors that are input into the sliding mode control model. Recursive least squares updates the boundary layer thickness and switching gain parameters based solely on incremental data. When a mean shift exceeding twice the standard deviation is detected, a local model rollback mechanism is triggered to load valid parameters from the previous time period. The membership function of the fuzzy adaptive proportional-integral controller is dynamically partitioned using K-means clustering, with 3-5 cluster centers. The reinforcement learning mechanism uses frequency modulation energy consumption and frequency overshoot as reward indicators, iteratively updating the rules to activate weights. In the microservice architecture, the dynamic programming microservice solves the optimal allocation ratio of rotor kinetic energy and converter power with frequency stability as a constraint. The API gateway integrates the Modbus and IEC 61850 protocols of the virtual synchronous machine and energy storage device to generate cross-device frequency modulation commands in JSON format. When the disturbance energy entropy exceeds the threshold, the priority weight microservice is triggered to dynamically adjust the output ratio. The redundancy elimination algorithm generates a conflict resolution scheme within 10-20 seconds based on the genetic algorithm.

[0193] During the mode switching and safe exit phases, the rolling time-domain optimization model predicts the frequency trend for the next 3-5 sampling cycles using historical disturbance data and real-time inertia parameters. A multi-island genetic algorithm divides the system into 4-6 subgroups, generating differentiated trigger thresholds every 10 generations of migration of elite individuals. When the frequency deviation change rate exceeds 120% of the dynamic tuning threshold, the system switches to a fast power support mode to release 80%-90% of the pre-stored rotor kinetic energy. During the droop control phase, the power slope curve is dynamically adjusted based on the speed recovery rate gradient, with the initial slope value set to 5% / second of the rated power. Upon frequency tuning exit, the genetic algorithm solves for the optimal power ramp rate with hard constraints of a speed recovery rate ≤2% / second and a frequency deviation ≤0.05Hz. The population size is initialized with 30-50 groups of historical feasible solutions. A dynamic relaxation mechanism temporarily expands the constraint boundary by 20%-30% when the speed gradient abruptly exceeds the safety threshold of 80%, achieving a seamless switch to maximum power point tracking operation.

[0194] The above implementation addresses the response lag issue of traditional methods in scenarios with time-varying inertia and multiple superimposed disturbances through a closed-loop link of dynamic sensing, parameter optimization, collaborative control, and safe exit. Data interaction and command coordination at each technical stage strictly match the features in the claims, improving the dynamic adaptability of frequency regulation commands and the collaborative efficiency of multi-source devices, ultimately achieving high-precision and stable frequency control of the power system.

[0195] The technical solution of this invention addresses the problem of decreased power system frequency stability caused by the lag in frequency regulation response of wind turbines and insufficient coordination of multi-source frequency regulation resources under dynamic multi-disturbance and time-varying inertia conditions through the following methods:

[0196] First, by constructing a dynamic disturbance sensing dataset and optimizing it with a genetic algorithm, the accuracy of disturbance feature extraction and the efficiency of inertia parameter identification are improved. The dynamic disturbance sensing dataset is constructed by real-time acquisition of power system frequency deviation, frequency deviation change rate, and wind turbine operating status parameters. Based on a genetic algorithm, the window length and weight coefficients of the short-window sliding mean algorithm are dynamically optimized to suppress background noise interference from the power grid and generate disturbance feature data with a high signal-to-noise ratio. Combined with the recursive least squares method, the equivalent inertia and damping parameters of the power grid are estimated in real time, providing accurate input for subsequent control models. The synergistic effect of these technologies overcomes the problems of perception lag and insufficient parameter identification accuracy in dynamic disturbance scenarios using traditional methods, thus shortening the frequency regulation response time.

[0197] Secondly, an online incremental learning algorithm and fuzzy adaptive control are used for collaborative optimization to enhance the dynamic adaptability of frequency regulation commands and the coordination capability of multi-source resources. Continuous disturbance event data streams are collected through a sliding time window, and the boundary layer thickness and switching gain parameters of the sliding mode control model are updated using the recursive least squares method to adapt to changes in disturbance characteristics in real time. Compensated power commands are generated based on a fuzzy adaptive proportional-integral controller, and the weights of fuzzy rules are dynamically adjusted using a reinforcement learning mechanism to optimize the command response speed under high-frequency disturbances. A microservice architecture is used to coordinate the allocation of wind turbine rotor kinetic energy reserves and the output of multi-source frequency regulation equipment, and an event-driven priority weight dynamic adjustment strategy is designed to eliminate power redundancy and conflicts. These technologies achieve self-optimization of control parameters and cross-device command synchronization, improving the collaborative efficiency of multi-source frequency regulation resources.

[0198] Finally, a genetic algorithm is introduced to optimize the dynamic switching threshold and safe exit mechanism, balancing the frequency regulation mode switching speed and system stability. In model predictive control, a rolling time-domain optimization model predicts future frequency trends, and a multi-island genetic algorithm generates differentiated trigger thresholds. The fast power support mode and droop control mode are switched based on the frequency deviation change rate. The turbine speed recovery rate and frequency deviation are used as hard constraints, and a genetic algorithm is employed to solve for the optimal power descent gradient parameters, combined with a dynamic relaxation mechanism to address the risk of sudden speed changes. An adaptive power ramp rate limiter ensures the smooth exit of the microservice from frequency regulation, switching to maximum power point tracking operation. These technologies guarantee the safety and economy of the frequency regulation strategy in time-varying inertia scenarios, suppress secondary frequency drops, and ultimately improve the overall frequency stability of the power system.

Claims

1. A method for rapid frequency regulation of wind turbine generators based on disturbance adaptive compensation, characterized in that, include: Step S1: Collect the frequency deviation, frequency deviation change rate and wind turbine operating status parameters of the power system to construct a dynamic disturbance sensing dataset. The wind turbine operating status parameters include real-time monitoring data of the rotor kinetic energy reserve of the wind turbine unit. Step S2: Based on the dynamic disturbance sensing dataset, the window length and weight coefficient of the short-window sliding mean algorithm are dynamically optimized using a genetic algorithm to generate noise-suppressed disturbance feature data. Step S3: Generate a reverse test signal based on the disturbance characteristic data, and estimate the equivalent inertia and damping parameters of the power grid in real time using the recursive least squares method. Step S4: Use an online incremental learning algorithm to perform time series modeling on the continuous disturbance events in the dynamic disturbance perception dataset, and update the boundary layer thickness and switching gain parameter of the preset adaptive sliding mode control model. The boundary layer thickness is used to balance the speed and smoothness of the control response, and the switching gain parameter is used to improve the ability to suppress high-frequency sudden disturbances. Step S5: Based on the equivalent inertia and damping parameters of the power grid, a compensation power command is generated through a preset fuzzy adaptive proportional controller and a preset integral controller. The compensation power command includes the triggering conditions for the fast power support mode and the droop control mode. Step S6: According to the compensation power command, coordinate the distribution of wind turbine rotor kinetic energy reserves and the power response of the grid multi-source frequency regulation equipment through a preset microservice architecture, wherein the grid multi-source frequency regulation equipment includes a virtual synchronous machine and an energy storage device; Step S7: In the adaptive sliding mode control model predictive control, the genetic algorithm is introduced to optimize the trigger threshold for dynamic switching, and the fast power support mode and droop control mode are switched according to the frequency deviation change rate. Step S8: When the frequency deviation recovers to the preset dead zone range, the microservice exits frequency modulation and switches to maximum power point tracking operation by limiting the adaptive power ramp rate. Step S2 includes: Based on the dynamic perturbation sensing dataset, the crossover probability and mutation probability are dynamically adjusted according to the population fitness distribution. When the population diversity is lower than a set threshold, the mutation probability is increased to the range of 0.2-0.

4. An elite retention strategy is adopted to directly pass the best individuals of each generation to the next generation, and the distribution characteristics of the perturbation feature data are combined to accelerate parameter convergence. By combining genetic algorithms with GPU parallel computing, the window length and weight coefficients of the short-window sliding mean algorithm are evaluated in parallel. The test signal amplitude parameters are initialized based on historical disturbance data in the dynamic disturbance sensing dataset, and the frequency interference of inertia estimation is reduced by reverse signal excitation.

2. The method for rapid frequency regulation of wind turbine generators based on disturbance adaptive compensation according to claim 1, characterized in that, Also includes: Based on the execution feedback data of the compensation power command, the activation weights of the fuzzy rules are adjusted using a reinforcement learning mechanism; The fuzzy intervals of the equivalent inertia parameters of the power grid are dynamically divided by clustering algorithm, and the center point and width parameters of the membership function are optimized. Establish the parameter correlation matrix between the adaptive sliding mode control model and the fuzzy controller, and update the boundary layer thickness and input range synchronously; When a high-frequency disturbance mode is detected, the high-frequency band membership function parameters of the equivalent inertia parameter of the power grid are optimized first, and the weight allocation of the corresponding rules is enhanced.

3. The method for rapid frequency regulation of wind turbine generators based on disturbance adaptive compensation according to claim 1, characterized in that, Step S6 includes: In the containerized deployment framework of the model predictive control, the power allocation ratio between rotor kinetic energy reserve and converter capacity is optimized through dynamic programming microservices. The communication protocols of the multi-source frequency regulation equipment in the power grid are integrated through the API gateway to generate a cross-device frequency regulation command synchronization protocol. Based on the dynamic disturbance perception dataset, the disturbance energy entropy is calculated, and an event-driven priority weight microservice is designed to dynamically adjust the output ratio. When a power conflict is detected, a redundancy elimination algorithm is triggered to reallocate the frequency regulation tasks of the virtual synchronizer and the energy storage device.

4. The method for rapid frequency regulation of wind turbine generators based on disturbance adaptive compensation according to claim 1, characterized in that, Step S7 includes: The historical disturbance data and real-time inertia parameters in the dynamic disturbance sensing dataset are input into the rolling time domain optimization model to predict the frequency trend for the next 3-5 sampling periods. The initial population of the genetic algorithm is divided using a multi-island genetic algorithm, and elite individuals are periodically migrated to generate differentiated trigger thresholds; When the frequency deviation change rate exceeds the start-up threshold, switch to the fast power support mode to release the rotor kinetic energy reserve; During the droop control phase, a power slope curve is generated based on the compensation power command. The wind turbine speed recovery rate is calculated by combining the speed change gradient after the release of the wind turbine rotor kinetic energy reserve. The slope limit is then dynamically updated based on the wind turbine speed recovery rate.

5. The method for rapid frequency regulation of wind turbine generators based on disturbance adaptive compensation according to claim 4, characterized in that, Also includes: The wind turbine speed recovery rate and frequency deviation are used as hard constraints to construct a gradient search space; The optimal power descent gradient parameters are solved within the constraint space using the genetic algorithm described above. When a risk of sudden speed change is detected, a dynamic relaxation mechanism is triggered to expand the constraint boundary of the sliding mode control model; By initializing the search space with the optimal gradient parameters from historical frequency modulation exit data, the generation time of the safe recovery strategy can be shortened.

6. The method for rapid frequency regulation of wind turbine generators based on disturbance adaptive compensation according to claim 1, characterized in that, Step S8 includes: Real-time data streams of the continuous disturbance events are collected through a sliding time window, and principal component analysis is performed for dimensionality reduction. The boundary layer parameters of the adaptive sliding mode control model are updated only based on incremental data using the recursive least squares method. When data anomalies are detected, a local model rollback mechanism is triggered to restore the parameters to the previous time period. The learning rate is dynamically adjusted based on the rate of change of the error predicted by the model, thereby increasing the parameter update speed as the error gradient increases.

Images