Method for optimizing frequency control of grid-connected MW synchronous machine by fan variable pitch

Abstract

The invention relates to the technical field of grid-connected control of a wind generating set and frequency collaborative optimization of a synchronous machine, in particular to a frequency control optimization method for a grid-connected MW synchronous machine by variable pitch of a wind turbine, which comprises the following steps of: acquiring multi-source data such as wind speed turbulence intensity, pitch angle position, frequency deviation of a power grid PMU (Power Management Unit) device, rotor inertia and the like; wavelet packet decomposition and moving average filtering are adopted to eliminate high-frequency noise and short-time fluctuation interference respectively, a dynamic time warping algorithm is utilized to align a data sampling rate to generate a multi-source data set with a unified time scale, and a phase lead correction algorithm is utilized to pre-judge a counter-acting force of pitch angle adjustment to generate a feed-forward compensation instruction. Adjusting paths under different disturbances are predicted in combination with a dynamic Bayesian network, a pneumatic torque-frequency response coupling coefficient is dynamically optimized by using an incremental stochastic gradient descent algorithm, and a collaborative optimization system of data driving and closed loop feedback is formed through a similarity matching algorithm. And the frequency recovery efficiency and the power angle stability under the low-frequency crossing working condition are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind turbine grid-connected control and synchronous machine frequency coordination optimization, and particularly relates to a wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method. BACKGROUND

[0002] The wind turbine variable pitch system controls the wind energy captured by the wind wheel by adjusting the blade pitch angle, and then adjusts the generator speed and output power, so as to realize the power balance with the power grid in the wind power grid-connected process. When the wind speed fluctuates, the variable pitch mechanism dynamically adjusts the pitch angle to change the aerodynamic torque, so that the wind energy conversion efficiency matches the demand of the power grid, thereby inhibiting the generator speed from deviating from the synchronous range and maintaining the mechanical-electrical coupling stability of the grid-connected synchronous machine. The synchronous machine relies on the power response characteristics of the variable pitch system to adjust the active output in real time according to the dispatching instructions of the power grid, and compensates the frequency deviation caused by the load change of the power grid through the inertia support and frequency response capability. The dynamic adjustment accuracy of the variable pitch system directly affects the power angle stability and voltage regulation capability of the synchronous machine in the transient process, and is a key control link for wind turbine generators to realize high proportion of renewable energy grid connection.

[0003] In the wind turbine variable pitch and grid-connected MW synchronous machine frequency coordination control, there is a problem of insufficient real-time matching accuracy between the data-driven variable pitch dynamic response model and the synchronous machine frequency regulation demand. Specifically, the existing variable pitch control algorithm relies on historical data static modeling, and it is difficult to real-time fuse the wind speed turbulence characteristics, the power grid frequency fluctuation mode and the synchronous machine rotor inertia dynamic characteristics, resulting in that the variable pitch angle adjustment lags behind the power grid frequency change demand. For example, when the power grid triggers a low frequency ride-through condition due to load mutation, the variable pitch system needs to dynamically adjust the pitch angle according to the synchronous machine rotor kinetic energy release rate to compensate for the active power gap, but the existing control model lacks real-time modeling capability for the coupling relationship between the transient power angle trajectory of the synchronous machine and the pitch angle aerodynamic damping, which easily causes pitch angle overshoot or oscillation, causing the synchronous machine frequency to drop twice or even out-of-step risk. SUMMARY

[0004] In view of the technical problems of the prior art, the present application provides a wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method, which solves the problem of insufficient real-time matching accuracy between the variable pitch dynamic response model driven by multi-source heterogeneous data and the synchronous machine frequency regulation demand.

[0005] To solve the above technical problems, the specific technical solutions of the present application are as follows:

[0006] The wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method provided by the present application comprises:

[0007] Step 1, collect wind speed turbulence intensity, actual position of pitch angle and generator speed, synchronously receive real-time frequency deviation, synchronous machine rotor inertia and power angle dynamic trajectory uploaded by the power grid PMU device, and generate a multi-source data set by preprocessing;

[0008] Step 2, based on the wind speed spectrum features, synchronous machine inertia response delay characteristics and power angle dynamic trajectory in the multi-source data set generated in step 1, the aerodynamic torque transmission characteristics and the electrical inertia response rate are fused through the Kalman filtering algorithm to generate a dynamic weight matrix; the forgetting factor adaptive algorithm is used to update the weight coefficients of the aerodynamic and electrical coupling model in the sliding time window, and the updated weight coefficients are stored in the model parameter library;

[0009] Step 3, according to the direction and amplitude of the real-time frequency deviation, the control interval is divided, the pitch angle compensation amount is calculated under the low-frequency ride-through condition based on the weight coefficients of the aerodynamic and electrical coupling model, the pitch angle adjustment is predicted to the reaction force of the synchronous machine rotor inertia through the phase lead correction algorithm, and the feedforward compensation instruction is generated to correct the pitch angle rate;

[0010] Step 4, the optimized pitch angle instruction is sent to the preset variable pitch servo system, the feedback signal is generated by monitoring the synchronous machine frequency recovery rate and power angle fluctuation amplitude in real time, the pitch angle adjustment path under different wind speed and power grid disturbance combinations is predicted based on the feedback signal through the dynamic Bayesian network, and the predicted path is obtained;

[0011] Step 5, based on the frequency recovery time and power angle fluctuation peak value in the feedback signal, an evaluation function is constructed, the aerodynamic torque and frequency response coupling coefficient of the aerodynamic and electrical coupling model are dynamically adjusted through the incremental stochastic gradient descent algorithm, similar patterns in historical operating condition data are extracted, and the similar degree matching algorithm is used to filter the reusable parameter set to migrate to a new scene, and the sliding window historical data weight distribution strategy in the model parameter library is updated.

[0012] Further, the wind turbine variable pitch frequency control optimization method for grid-connected MW synchronous machines provided by the application has the steps of:

[0013] Collect wind speed turbulence intensity data and mark time stamp, receive synchronous machine rotor inertia and power angle dynamic trajectory uploaded by the power grid PMU device;

[0014] Perform wavelet packet decomposition algorithm on the wind speed data marked with time stamp to separate high-frequency noise and generate denoised wind speed spectrum features;

[0015] Perform sliding average filtering on the received power grid frequency data to eliminate short-term fluctuation interference and generate filtered power grid frequency fluctuation mode;

[0016] The denoised wind speed spectrum feature and the sampling rate of the filtered power grid frequency fluctuation mode are aligned by a dynamic time warping algorithm to generate a multi-source data set.

[0017] Further, the wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method of the application, the step 2 comprises:

[0018] Based on the denoised wind speed spectrum feature and the filtered power grid frequency fluctuation mode, a dynamic transfer function of the aerodynamic torque transfer characteristic of the aerodynamic and electrical coupling model and the electrical inertia response rate is established;

[0019] The weight coefficient of the aerodynamic and electrical coupling model is updated in the sliding time window using a forgetting factor adaptive algorithm, and recent data is preferentially retained to match the power grid frequency regulation requirement of step 3.

[0020] Further, the wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method of the application further comprises:

[0021] The denoised wind speed spectrum feature and the synchronous machine inertia response delay characteristic are fused by a Kalman filtering algorithm to generate a dynamic weight matrix to balance the contribution degree of aerodynamic damping and electrical inertia, and the dynamic weight matrix is stored in a model parameter library;

[0022] The updated aerodynamic and electrical coupling model weight coefficient is transmitted to the phase lead correction algorithm of step 3 for pre-judging the reaction force of the pitch angle adjustment on the synchronous machine rotor inertia.

[0023] Further, the wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method of the application, the step 3 comprises:

[0024] Based on the frequency recovery time and the power angle fluctuation peak value in the feedback signal generated in step 4, an evaluation function with the two as core indexes is constructed;

[0025] When the frequency secondary drop or power angle overrun event is detected, an incremental stochastic gradient descent algorithm is triggered to dynamically adjust the aerodynamic torque and frequency response coupling coefficient of the aerodynamic and electrical coupling model;

[0026] The optimized aerodynamic torque and frequency response coupling coefficient are returned to the aerodynamic and electrical coupling model parameter library of step 2 to update the model weight coefficient;

[0027] The initial condition of the prediction path in step 4 is updated based on the optimized coupling coefficient by a dynamic Bayesian network to generate a corrected pitch angle adjustment path.

[0028] Further, the wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method of the application, the step 4 comprises:

[0029] Based on the real-time monitoring of the synchronous machine frequency recovery rate and the power angle fluctuation amplitude in step 4, a feedback signal is generated, and the frequency recovery time and the power angle fluctuation peak value are extracted from the feedback signal to construct an evaluation function;

[0030] When the frequency secondary drop or power angle overrun event is detected, an incremental random gradient descent algorithm is triggered to dynamically adjust the aerodynamic torque and frequency response coupling coefficient of the aerodynamic and electrical coupling model;

[0031] The optimized aerodynamic torque and frequency response coupling coefficient are returned to the model parameter library to update the aerodynamic and electrical coupling model weight coefficient;

[0032] The initial conditions of the prediction path in step 4 are updated based on the optimized coupling coefficient through a dynamic Bayesian network to generate a corrected pitch angle adjustment path.

[0033] Further, the wind turbine variable pitch and grid-connected MW synchronous machine frequency control optimization method of the application, the step 5 includes:

[0034] Similar patterns in historical operating condition data are extracted from the screened reusable parameter set;

[0035] The similarity matching algorithm is used to screen the reusable parameter set matched with the current scene;

[0036] The screened reusable parameter set is migrated to a sliding window mechanism to update the sliding window historical data weight distribution strategy of step 2;

[0037] The migrated parameter set and the generated prediction path are verified cooperatively, and the verification result is output to the evaluation function optimization process for dynamically adjusting the aerodynamic and electrical coupling model parameters.

[0038] The application has the following advantages:

[0039] The application significantly improves the real-time matching accuracy of the variable pitch control and the synchronous machine frequency regulation through data fusion and dynamic parameter iteration mechanism. Wavelet packet decomposition and moving average filtering are used to preprocess the wind speed and grid frequency data, eliminate noise interference, and generate time-scale unified multi-source input; a Kalman filter-based aerodynamic-electric coupling model is constructed, and the weight coefficient is dynamically updated combined with the forgetting factor adaptive algorithm to enhance the response ability of the model to the time-varying characteristics of wind speed turbulence and grid disturbance. The pitch angle adjustment path is predicted through a dynamic Bayesian network, and the aerodynamic torque-frequency response coupling coefficient is optimized using an incremental stochastic gradient descent algorithm, forming an iterative link of data acquisition, model updating, and closed-loop feedback. In addition, a similarity matching algorithm is introduced to select the reusable parameter set in the historical working condition, which is migrated to a sliding window mechanism to optimize the weight distribution strategy, breaking through the limitations of traditional static modeling, effectively suppressing the pitch angle overshoot and synchronous machine frequency secondary drop risk, and improving the power angle stability and frequency recovery efficiency under complex working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor based on the drawings.

[0041] Figure 1 A flowchart of a wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method preparation method provided by the embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the present application clearer, the following will combine the specific embodiments of the present application and the corresponding drawings to clearly and completely describe the technical solutions of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application. The technical solutions provided by the embodiments of the present application will be described in detail below with reference to the drawings. In order to better understand the purpose of the present application, the present application will be further described in detail.

[0043] Please refer to Figure 1 The wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method provided by the present application comprises:

[0044] Step 1, collect wind speed turbulence intensity, pitch angle actual position and generator speed, synchronously receive real-time frequency deviation uploaded by grid PMU device, synchronous machine rotor inertia and power angle dynamic trajectory, and generate multi-source data set after preprocessing;

[0045] Step 2, based on the wind speed spectrum features, synchronous machine inertia response delay characteristics, and power angle dynamic trajectory in the multi-source data set generated in step 1, the aerodynamic torque transmission characteristics and electrical inertia response rate are fused through the Kalman filter algorithm to generate a dynamic weight matrix; the aerodynamic and electrical coupling model weight coefficients are updated in a sliding time window using a forgetting factor adaptive algorithm, and the updated weight coefficients are stored in the model parameter library;

[0046] Step 3, according to the real-time frequency deviation direction and amplitude, the control interval is divided, based on the aerodynamic and electrical coupling model weight coefficients, the pitch angle compensation amount is calculated under low frequency ride-through working condition, the pitch angle adjustment on the reaction force of the synchronous machine rotor inertia is predicted through phase lead correction algorithm, and the feedforward compensation instruction is generated to modify the pitch angle rate;

[0047] Step 4, the optimized pitch angle instruction is sent to the preset variable pitch servo system, the feedback signal is generated by real-time monitoring of the synchronous machine frequency recovery rate and power angle fluctuation amplitude, the pitch angle adjustment path under different wind speed and power grid disturbance combinations is predicted based on the feedback signal through dynamic Bayesian network, and the predicted path is obtained;

[0048] Step 5, based on the frequency recovery time and power angle fluctuation peak value in the feedback signal, an evaluation function is constructed, the aerodynamic torque and frequency response coupling coefficient of the aerodynamic and electrical coupling model are dynamically adjusted through the incremental stochastic gradient descent algorithm, the similar patterns in the historical working condition data are extracted, the reusable parameter set is selected through the similarity matching algorithm and migrated to the new scene, and the sliding window historical data weight distribution strategy in the model parameter library is updated.

[0049] In step 1, the wind speed turbulence intensity, pitch angle actual position and generator speed are collected through the wind turbine SCADA system, and the real-time frequency deviation, synchronous machine rotor inertia and power angle dynamic trajectory are received from the power grid PMU device. The wavelet packet decomposition algorithm is performed on the wind speed turbulence data to separate high-frequency noise, and the low-frequency component is retained as the wind speed spectrum feature; the moving average filter is performed on the power grid frequency data to eliminate short-term fluctuation interference, and the steady-state frequency deviation trend is extracted. The sampling rates of wind speed data and power grid data are aligned through dynamic time warping algorithm to eliminate the time difference of multi-source data, and a time-unified multi-source data set is generated. The preprocessed data set contains wind speed spectrum, power grid frequency fluctuation mode and synchronous machine dynamic parameters, providing input basis for subsequent model construction.

[0050] Step 2: Based on the output of step 1, the Kalman filter algorithm is used to fuse the wind speed spectrum characteristics and the inertia response delay characteristics of the synchronous machine, and a dynamic transfer function of the aerodynamic torque transfer characteristics and the electrical inertia response rate is established. In the sliding time window, the forgetting factor adaptive algorithm is applied to dynamically update the weight coefficients of the aerodynamic and electrical coupling model according to the real-time power grid frequency regulation requirements, and the recent data weight is preferentially retained to reflect the time-varying characteristics of the system. The fused dynamic weight matrix is stored in the model parameter library, quantifying the coupling contribution of aerodynamic damping and electrical inertia, and providing parameter benchmarks for control strategy generation.

[0051] Step 3: According to the real-time frequency deviation direction and amplitude, the control interval is divided, and the pitch angle compensation amount is calculated based on the weight coefficients of the aerodynamic and electrical coupling model under low-frequency ride-through conditions. Through the phase lead correction algorithm, the coupling relationship between the synchronous machine rotor inertia and the pitch angle adjustment is analyzed, the influence of the reaction force on the power angle stability is predicted, and the feedforward compensation instruction is generated and superimposed to the pitch angle rate control signal. The feedforward compensation amount and the basic compensation amount jointly constitute the comprehensive control instruction, which can suppress power angle oscillation and improve the frequency recovery efficiency of the synchronous machine.

[0052] Step 4: The optimized pitch angle instruction is sent to the variable pitch servo system, and the feedback signal is generated by real-time monitoring of the synchronous machine frequency recovery rate and power angle fluctuation amplitude. Based on the feedback signal, a dynamic Bayesian network is constructed to predict the optimal pitch angle adjustment path under different wind speed and power grid disturbance combinations. The predicted path is embedded in the cooperative control logic of step 3 to form a feedforward and feedback composite control architecture, which dynamically corrects the execution trajectory of the pitch angle instruction.

[0053] Step 5: Based on the frequency recovery time and power angle fluctuation peak value in the feedback signal, an evaluation function is constructed, and the aerodynamic torque-frequency response coupling coefficient is dynamically adjusted by the incremental stochastic gradient descent algorithm. Similar patterns in historical operating data are extracted, and a similarity matching algorithm is used to select reusable parameter sets that match the new scenario, which are migrated to the sliding window mechanism to update the historical data weight distribution strategy. The migration parameters and the predicted path are verified cooperatively, and the verification results are output to the evaluation function optimization process.

[0054] Specifically, the wind turbine variable pitch pitch control optimization method for grid-connected MW synchronous machines according to the present application comprises the following steps:

[0055] Collect wind speed turbulence intensity data and mark time stamps, and receive synchronous machine rotor inertia and power angle dynamic trajectories uploaded by the power grid PMU device;

[0056] Perform wavelet packet decomposition algorithm on the wind speed data marked with time stamps to separate high-frequency noise and generate denoised wind speed spectrum characteristics;

[0057] Perform sliding average filtering on the received power grid frequency data to eliminate short-term fluctuation interference and generate filtered power grid frequency fluctuation patterns;

[0058] The denoised wind speed spectrum features and the filtered power grid frequency fluctuation patterns are aligned in sampling rate by a dynamic time warping algorithm to generate a multi-source data set.

[0059] When collecting wind speed turbulence intensity data, wind speed signals are obtained in real time through a fan SCADA system and time stamps are marked, and synchronous machine rotor inertia and power angle dynamic trajectories uploaded by a power grid PMU device are received synchronously to ensure the spatiotemporal correlation of multi-source data. Time stamp marking is implemented by using an NTP protocol to achieve cross-system clock synchronization, thereby providing a unified time sequence reference for subsequent data fusion.

[0060] A wavelet packet decomposition algorithm is performed on the wind speed data marked with time stamps, a specific decomposition layer is selected to separate high-frequency noise components, and low-frequency spectrum features reflecting wind speed turbulence characteristics are retained. The decomposed low-frequency components represent wind speed energy distribution and dynamic change trends, avoiding interference of high-frequency noise with the quality of model input data.

[0061] A moving average filtering algorithm is used for power grid frequency data, an adaptive window length is set to eliminate short-term fluctuation interference, and power grid frequency steady-state deviation and low-frequency fluctuation patterns are extracted. The filtered data retain the frequency gradual change characteristics caused by power grid load disturbance, thereby providing input for analyzing synchronous machine inertia response delay characteristics.

[0062] The denoised wind speed spectrum features and the filtered power grid frequency fluctuation patterns are aligned in nonlinear time series by a dynamic time warping algorithm, thereby eliminating the time sequence misalignment problem caused by the sampling rate difference between wind speed and power grid data. The aligned data set integrates aerodynamic and electrical dynamic parameters to generate a multi-source data set with unified time scale, which serves as an input basis for constructing an aerodynamic and electrical coupling model.

[0063] Specifically, the wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method described in the present application comprises the following steps:

[0064] Based on the denoised wind speed spectrum features and the filtered power grid frequency fluctuation patterns, a dynamic transfer function of the aerodynamic torque transmission characteristics and the electrical inertia response rate of the aerodynamic and electrical coupling model is established.

[0065] In a sliding time window, a forgetting factor adaptive algorithm is used to update the weight coefficients of the aerodynamic and electrical coupling model, and recent data is preferentially retained to match the power grid frequency regulation requirements of step 3.

[0066] Based on the denoised wind speed spectrum characteristics and the filtered power grid frequency fluctuation mode, the dynamic transfer function of the aerodynamic and electrical coupling model is established by Kalman filtering algorithm to fuse the aerodynamic torque transmission characteristics and the electrical inertia response rate. The dynamic transfer function quantifies the time-varying influence of wind speed turbulence intensity on aerodynamic torque, and correlates the response delay characteristics of synchronous machine rotor inertia to power grid frequency deviation, forming the coupling mapping relationship of aerodynamic side and electrical side parameters.

[0067] In the sliding time window, the weight coefficient of the aerodynamic and electrical coupling model is dynamically adjusted by using the forgetting factor adaptive algorithm. The length of the sliding window is adaptively adjusted according to the power grid frequency fluctuation period, and the forgetting factor dynamically changes with the frequency deviation amplitude, preferentially attenuating the weight of historical data and enhancing the contribution of recent data. The updated weight coefficient is stored in the model parameter library, which reflects the dynamic coupling strength of aerodynamic damping and electrical inertia in real time, and provides parameter input for the compensation amount calculation of step 3 in the low-frequency ride-through working condition.

[0068] The model weight coefficient is combined with the dynamic weight matrix through Kalman filtering algorithm to balance the real-time coupling relationship between wind speed spectrum characteristics and power grid frequency fluctuation mode. The update mechanism of the dynamic weight matrix cooperates with the forgetting factor strategy of the sliding window to suppress the interference of old data on the prediction accuracy of the model, and improve the matching efficiency of the aerodynamic and electrical coupling model to the power grid frequency regulation demand.

[0069] Specifically, the wind turbine variable pitch to grid-connected MW synchronous machine frequency control optimization method described in the application further comprises:

[0070] The dynamic weight matrix is generated by Kalman filtering algorithm to balance the contribution of aerodynamic damping and electrical inertia, and the dynamic weight matrix is stored in the model parameter library.

[0071] The updated weight coefficient of the aerodynamic and electrical coupling model is transmitted to the phase lead correction algorithm of step 3 for predicting the reaction force of pitch angle adjustment on the synchronous machine rotor inertia.

[0072] The state space model of the coupling relationship between aerodynamic damping and electrical inertia is constructed by Kalman filtering algorithm to fuse the denoised wind speed spectrum characteristics and the synchronous machine inertia response delay characteristics. The Kalman filtering algorithm predicts the aerodynamic torque fluctuation trend based on the time series of wind speed spectrum characteristics, calculates the dynamic coupling strength of the aerodynamic side and the electrical side by combining the hysteresis characteristics of the synchronous machine inertia response, and generates the dynamic weight matrix reflecting the interaction between aerodynamic and electrical. The dynamic weight matrix is stored in the model parameter library, which provides quantitative basis for aerodynamic damping suppression and electrical inertia compensation of the control strategy.

[0073] The updated aerodynamic and electrical coupling model weight coefficients are transmitted to the phase lead correction algorithm of step 3 to analyze the coupling proportion relationship between the aerodynamic torque and the electrical inertia in the weight coefficients. The phase lead correction algorithm dynamically adjusts the prediction time window according to the weight coefficients, and quantifies the amplitude and phase offset of the reaction force of the pitch angle adjustment on the rotor inertia of the synchronous machine. The adjusted prediction parameters generate a feedforward compensation instruction to suppress the synchronous machine power angle oscillation caused by pitch angle overshoot and improve the frequency recovery stability under low-frequency ride-through conditions.

[0074] A real-time data interface is established between the model parameter library and the phase lead correction algorithm, and the update of the dynamic weight matrix triggers the synchronous iteration of the control instruction. The coupling strength parameters output by the Kalman filtering algorithm and the prediction logic of the phase lead correction algorithm form a closed-loop feedback to realize the dynamic adaptation of the aerodynamic and electrical coupling model and the control strategy.

[0075] Specifically, the wind turbine variable pitch frequency control optimization method for grid-connected MW synchronous machines provided by the present application comprises the following steps:

[0076] Based on the frequency recovery time and the power angle fluctuation peak value in the feedback signal generated in step 4, an evaluation function is constructed with the two as core indicators;

[0077] When the frequency secondary drop or power angle overrun event is detected, an incremental stochastic gradient descent algorithm is triggered to dynamically adjust the aerodynamic torque and frequency response coupling coefficient of the aerodynamic and electrical coupling model;

[0078] The optimized aerodynamic torque and frequency response coupling coefficient are returned to the aerodynamic and electrical coupling model parameter library of step 2 to update the model weight coefficient;

[0079] The initial conditions of the prediction path in step 4 are updated based on the optimized coupling coefficient through the dynamic Bayesian network to generate a corrected pitch angle adjustment path.

[0080] Based on the feedback signal of step 4, the frequency recovery time and the power angle fluctuation peak value are extracted, and a multi-dimensional evaluation function is generated by weighted combination. The frequency recovery time represents the recovery efficiency of the power grid frequency deviation, and the power angle fluctuation peak value reflects the transient stability of the rotor inertia of the synchronous machine, both of which quantize the adjustment effect of the control strategy. The evaluation function eliminates the dimensional difference through normalization processing to provide a unified evaluation benchmark for parameter optimization.

[0081] When the frequency secondary drop or power angle overrun event is detected, an incremental stochastic gradient descent algorithm is triggered based on a preset threshold. The algorithm iteratively adjusts the aerodynamic torque-frequency response coupling coefficient according to the gradient direction of the evaluation function to real-time correct the dynamic characteristics of the aerodynamic and electrical coupling model. The incremental update mechanism only adjusts local parameters, avoiding the computational burden of global model reconstruction and improving optimization efficiency.

[0082] The optimized coupling coefficient is returned to the aerodynamic and electrical coupling model parameter library of step 2 through a data interface, replaces the historical weight coefficient and updates the model prediction logic. The parameter library stores multiple sets of weight coefficients using a version control mechanism, supports dynamic switching and backtracking verification, and guarantees the robustness of model iteration.

[0083] The dynamic Bayesian network loads the optimized coupling coefficient, updates the initial condition and state transition probability matrix of the prediction path. The network recalculates the pitch angle adjustment path under different combinations of wind speed and grid disturbance based on the corrected initial condition, generates a prediction result that adapts to the latest model parameters. The corrected path is embedded into the execution logic of step 4 through a control instruction mapping relationship.

[0084] Specifically, the wind turbine variable pitch and grid-connected MW synchronous machine frequency control optimization method described in the present application, step 4 includes:

[0085] Based on the real-time monitored synchronous machine frequency recovery rate and power angle fluctuation amplitude in step 4, a feedback signal is generated, and the frequency recovery time and power angle fluctuation peak value are extracted from the feedback signal to construct an evaluation function;

[0086] When detecting a frequency secondary drop or power angle overrun event, an incremental stochastic gradient descent algorithm is triggered to dynamically adjust the aerodynamic torque and frequency response coupling coefficient of the aerodynamic and electrical coupling model;

[0087] The optimized aerodynamic torque and frequency response coupling coefficient is returned to the model parameter library to update the weight coefficient of the aerodynamic and electrical coupling model;

[0088] Through the dynamic Bayesian network, the initial condition of the prediction path in step 4 is updated based on the optimized coupling coefficient, and a corrected pitch angle adjustment path is generated.

[0089] Based on the real-time monitored synchronous machine frequency recovery rate and power angle fluctuation amplitude in step 4, a feedback signal is generated, and the frequency recovery time and power angle fluctuation peak value are extracted from the feedback signal to construct a multi-dimensional evaluation function. The frequency recovery time reflects the recovery efficiency of the grid frequency deviation, and the power angle fluctuation peak value represents the transient stability of the synchronous machine rotor inertia, and the two are normalized and weighted to form a comprehensive evaluation index, which quantifies the adjustment effect of the control strategy.

[0090] When detecting a frequency secondary drop or power angle overrun event, an incremental stochastic gradient descent algorithm is triggered based on a preset threshold. The algorithm takes the evaluation function as the optimization target, iteratively adjusts the aerodynamic torque-frequency response coupling coefficient of the aerodynamic and electrical coupling model along the gradient direction, and realizes local parameter dynamic optimization. The incremental update mechanism only corrects the model parameters associated with the current operating condition, avoiding the computational overhead of global model reconstruction, and improving the optimization efficiency.

[0091] The optimized coupling coefficients are returned to the model parameter library through the real-time data interface, replace the historical weight coefficients and update the prediction logic of the aerodynamic and electrical coupling model. The parameter library adopts a version control and rolling storage mechanism, retains multiple sets of weight coefficients to support dynamic switching and abnormal rollback, and enhances the robustness of model iteration.

[0092] The dynamic Bayesian network loads the optimized coupling coefficients, updates the state transition probability matrix and initial condition parameters of the prediction path. The network recalculates the pitch angle adjustment path probability distribution under different wind speed and grid disturbance combinations based on the corrected initial conditions, generates a prediction result that adapts to the latest model parameters. The corrected path is embedded in the control logic of the variable pitch servo system through the instruction mapping relationship.

[0093] Specifically, the wind turbine variable pitch frequency control optimization method for grid-connected MW synchronous machines, step 5 includes:

[0094] Extract similar patterns in historical operating condition data from the screened reusable parameter set;

[0095] Screen the reusable parameter set that matches the current scenario through a similarity matching algorithm;

[0096] Migrate the screened reusable parameter set to the sliding window mechanism and update the sliding window historical data weight distribution strategy of step 2;

[0097] Co-verify the migrated parameter set and the generated prediction path, output the verification result to the evaluation function optimization process, and use it to dynamically adjust the aerodynamic and electrical coupling model parameters.

[0098] When extracting similar patterns in historical operating condition data from the screened reusable parameter set, a time series-based clustering algorithm is used to analyze historical operating condition data, identify similar operating scenarios with current wind speed turbulence characteristics and grid frequency fluctuation trends. Similar patterns are aligned on the time axis through a dynamic time warping algorithm, extract common features of aerodynamic torque and electrical inertia coupling, and form a migratable model parameter template.

[0099] Calculate the multi-dimensional feature distance between the current scenario and the historical template through a similarity matching algorithm, set an adaptive threshold to screen the reusable parameter set with a matching degree higher than the preset value. The similarity matching algorithm combines wind speed spectral distribution, grid frequency fluctuation period, and synchronous machine inertia response delay characteristics to quantify the dynamic coupling similarity between scenarios, and selects a subset of model parameters that adapts to the current operating condition.

[0100] The reusable parameter set screened is migrated to the sliding window mechanism of step 2, and the historical data weight distribution strategy is dynamically adjusted. The migration parameter updates the forgetting factor distribution in the sliding window through a weighted fusion mechanism, enhances the weight of the historical data associated with the current scene, and attenuates the contribution of irrelevant data. The updated weight distribution strategy optimizes the response accuracy of the aerodynamic and electrical coupling model to the real-time power grid frequency regulation demand.

[0101] The migrated parameter set and the predicted path generated in step 4 are verified cooperatively, the deviation between the predicted path and the actual control effect is compared, the verification result is generated and the effectiveness of the parameter migration is quantified. The verification result is input into the evaluation function optimization process through normalization processing, driving the incremental stochastic gradient descent algorithm to iteratively correct the parameters of the aerodynamic and electrical coupling model.

[0102] The present application improves the real-time matching accuracy of the variable pitch dynamic response model and the synchronous machine frequency regulation demand through multi-source heterogeneous data fusion and dynamic model iterative optimization. First, based on wavelet packet decomposition and moving average filtering, the wind speed turbulence intensity and power grid frequency data are denoised to eliminate high-frequency noise and short-term fluctuation interference, and a unified time scale multi-source data set is generated. The dynamic time warping algorithm is used to align the time sequence difference of wind speed and power grid data, integrate the dynamic parameters of the aerodynamic side and the electrical side, and provide input basis for the construction of the coupled model.

[0103] Secondly, the Kalman filter algorithm is used to fuse the wind speed spectrum characteristics and the synchronous machine inertia response delay characteristics, and the dynamic transfer function of the aerodynamic-electrical coupling model is established to quantify the real-time interaction of aerodynamic torque and electrical inertia. In the sliding time window, the forgetting factor adaptive algorithm is applied to dynamically update the model weight coefficient to preferentially retain recent data, and the dynamic Bayesian network is used to predict the pitch angle adjustment path under different disturbance combinations to form a feedforward and feedback composite control architecture. The incremental stochastic gradient descent algorithm constructs an evaluation function according to the frequency recovery time and the power angle fluctuation peak value, dynamically optimizes the aerodynamic torque-frequency response coupling coefficient, and realizes the closed-loop iterative correction of the model parameters.

[0104] Finally, the similarity matching algorithm is used to screen the reusable parameter set in the historical working condition, and the sliding window mechanism is migrated to update the historical data weight distribution strategy. The migration parameter and the predicted path are verified cooperatively, and the verification result is output to the evaluation function optimization process to drive the dynamic adjustment of the model parameters. This mechanism combines real-time updating and historical experience migration driven by data, breaks through the limitations of traditional static modeling, and enhances the generalization ability and regulation accuracy of the model in complex working conditions.

Claims

1. A method for optimizing the frequency control of a wind turbine pitch control system for a grid-connected MW synchronous machine, characterized in that, include: Step 1: Collect wind speed turbulence intensity, actual position of blade pitch angle and generator speed, and simultaneously receive real-time frequency deviation, synchronous machine rotor inertia and dynamic trajectory of power angle uploaded by the power grid PMU device, and perform preprocessing to generate multi-source datasets. Step 2: Based on the wind speed spectrum characteristics, synchronous machine inertia response delay characteristics and power angle dynamic trajectory in the multi-source dataset generated in Step 1, the aerodynamic torque transmission characteristics and electrical inertia response rate are fused using the Kalman filter algorithm to generate a dynamic weight matrix; within the sliding time window, the forgetting factor adaptive algorithm is used to update the weight coefficients of the aerodynamic and electrical coupling model, and the updated weight coefficients are stored in the model parameter library. Step 3: Divide the control interval according to the direction and amplitude of the real-time frequency deviation. Based on the weight coefficient of the aerodynamic and electrical coupling model, calculate the pitch angle compensation amount under the low-frequency crossing condition. Predict the reaction force of the pitch angle adjustment on the inertia of the synchronous machine rotor through the phase lead correction algorithm, and generate the feedforward compensation command to correct the pitch angle rate. Step 4: Send the optimized pitch angle command to the preset pitch servo system, monitor the synchronous machine frequency recovery rate and power angle fluctuation amplitude in real time to generate feedback signals, and predict the pitch angle adjustment path under different wind speed and power grid disturbance combinations through a dynamic Bayesian network based on the feedback signals to obtain the predicted path. Step 5: Construct an evaluation function based on the frequency recovery time and power angle fluctuation peak in the feedback signal. Dynamically adjust the aerodynamic torque and frequency response coupling coefficients of the aerodynamic and electrical coupling model using the incremental stochastic gradient descent algorithm. Extract similar patterns from historical operating data. Select reusable parameter sets to migrate to new scenarios using a similarity matching algorithm. Update the sliding window historical data weight allocation strategy in the model parameter library.

2. The method for optimizing the frequency control of a wind turbine pitch control system for a grid-connected MW synchronous machine according to claim 1, characterized in that, Step 1 includes: Collect wind speed and turbulence intensity data and mark the timestamp; receive the synchronous machine rotor inertia and power angle dynamic trajectory uploaded by the power grid PMU device; Wavelet packet decomposition algorithm is performed on the wind speed data marked with timestamps to separate high-frequency noise and generate denoised wind speed spectrum features; The received power grid frequency data is subjected to moving average filtering to eliminate short-term fluctuation interference and generate a filtered power grid frequency fluctuation pattern. The sampling rate of the denoised wind speed spectrum features is aligned with that of the filtered power grid frequency fluctuation pattern using a dynamic time warping algorithm to generate a multi-source dataset.

3. The method for optimizing the frequency control of a wind turbine pitch control system for a grid-connected MW synchronous machine according to claim 1, characterized in that, Step 2 includes: Based on the denoised wind speed spectrum characteristics and the filtered power grid frequency fluctuation pattern, a dynamic transfer function of aerodynamic torque transmission characteristics and electrical inertia response rate is established for the aerodynamic and electrical coupling model. Within the sliding time window, the weight coefficients of the aerodynamic and electrical coupling model are updated using an adaptive forgetting factor algorithm, prioritizing the retention of recent data to match the power grid frequency regulation requirements of step 3.

4. The method for optimizing the frequency control of a wind turbine pitch control system for a grid-connected MW synchronous machine according to claim 3, characterized in that, Also includes: By fusing the denoised wind speed spectrum characteristics and the synchronous machine inertia response delay characteristics through the Kalman filter algorithm, a dynamic weight matrix is ​​generated to balance the contributions of aerodynamic damping and electrical inertia, and the dynamic weight matrix is ​​stored in the model parameter library. The updated aerodynamic and electrical coupling model weight coefficients are passed to the phase lead correction algorithm in step 3 to predict the reaction force of pitch angle adjustment on the rotor inertia of the synchronous machine.

5. The method for optimizing the frequency control of a wind turbine pitch control system for a grid-connected MW synchronous machine according to claim 1, characterized in that, Step 3 includes: Based on the frequency recovery time and power angle fluctuation peak value in the feedback signal generated in step 4, an evaluation function with these two as the core indicators is constructed. When a frequency double drop or power angle over-limit event is detected, the incremental stochastic gradient descent algorithm is triggered to dynamically adjust the coupling coefficient of aerodynamic torque and frequency response of the aerodynamic and electrical coupling model. The optimized aerodynamic torque and frequency response coupling coefficients are fed back to the aerodynamic and electrical coupling model parameter library in step 2 to update the model weight coefficients. The initial conditions of the predicted path in step 4 are updated using a dynamic Bayesian network based on the optimized coupling coefficient, thereby generating a corrected pitch angle adjustment path.

6. The method for optimizing the frequency control of a wind turbine pitch control system for a grid-connected MW synchronous machine according to claim 1, characterized in that, Step 4 includes: Based on the synchronous frequency recovery rate and power angle fluctuation amplitude monitored in step 4, a feedback signal is generated, and the frequency recovery time and power angle fluctuation peak value are extracted from the feedback signal to construct an evaluation function. When a frequency double drop or power angle over-limit event is detected, the incremental stochastic gradient descent algorithm is triggered to dynamically adjust the coupling coefficient of aerodynamic torque and frequency response of the aerodynamic and electrical coupling model. The optimized aerodynamic torque and frequency response coupling coefficients are fed back to the model parameter library to update the weight coefficients of the aerodynamic and electrical coupling model. The initial conditions of the predicted path in step 4 are updated using a dynamic Bayesian network based on the optimized coupling coefficient, thereby generating a corrected pitch angle adjustment path.

7. The method for optimizing the frequency control of a wind turbine pitch control system for a grid-connected MW synchronous machine according to claim 1, characterized in that, Step 5 includes: Extract similar patterns from historical operating condition data from the selected set of reusable parameters; The set of reusable parameters that matches the current scene is selected using a similarity matching algorithm; The filtered reusable parameter set is migrated to the sliding window mechanism, and the sliding window historical data weight allocation strategy in step 2 is updated. The migrated parameter set is co-validated with the generated prediction path, and the validation results are output to the evaluation function optimization process for dynamically adjusting the parameters of the aerodynamic and electrical coupling model.

Images