Active support control method and system for wind turbine generator system based on frequency regulation

Abstract

The present invention discloses a method and system for active support control of wind turbines based on frequency regulation, which belongs to the technical field of wind power generation. The method solves the problem that existing methods cannot effectively extract key features related to frequency shortage, resulting in poor effectiveness, real-time performance and response speed of wind turbine linearization models. The method includes obtaining the day-ahead wind turbine output data and grid power parameters of the power system, identifying and analyzing the wind turbine output data and grid power parameters within the day based on a frequency situation prediction model, and calculating the spare capacity based on overspeed control at the power fluctuation point taking into account energy conversion efficiency and primary frequency regulation as constraints; the present invention realizes frequency situation prediction within the prediction period by constructing a frequency situation prediction model, determines the power fluctuation point and the power fluctuation value at the power fluctuation point within the prediction period based on the frequency prediction set, effectively extracts key features related to frequency shortage, and ensures the primary frequency regulation capability and response speed of the active support control of the wind turbine.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power generation, and in particular relates to a method and system for actively supporting and controlling a wind turbine generator set based on frequency regulation. Background Art

[0002] Currently, wind energy is widely considered the most promising renewable energy source. The penetration rate of renewable energy, represented by wind power, continues to increase. While wind farms, as renewable energy sources, have certain value in energy conservation, emission reduction, and optimizing power supply structures, they are also subject to instability, pulsation, and uncontrollability. Unlike traditional synchronous generators, they cannot respond to power system frequency changes in a timely manner.

[0003] Chinese patent CN114899892B discloses a method for active frequency support control of a wind turbine, comprising: performing overspeed control on the rotor of the wind turbine to reserve mechanical power reserves; constructing an active frequency support controller for the wind turbine consisting of a short-term frequency support controller and a long-term frequency support controller; establishing a wind turbine linearization model containing the active frequency support controller for the wind turbine, and determining the stable operating range of the active frequency support controller parameters through Nyquist plots and modal analysis; however, the feature extraction capability of the wind turbine linearization model in existing methods is relatively limited, and it is impossible to effectively extract key features related to the frequency deficit, resulting in poor effectiveness, real-time performance, and response speed of the wind turbine linearization model. To address the above issues, we propose a method and system for active support control of wind turbines based on frequency regulation. Summary of the Invention

[0004] The purpose of the present invention is to address the shortcomings of the existing technology and provide a method and system for active support control of wind turbines based on frequency regulation, which solves the problem that the feature extraction capability of the wind turbine linearization model in the existing method is relatively limited and the key features related to the frequency deficiency cannot be effectively extracted, resulting in poor effectiveness, real-time performance and response speed of the wind turbine linearization model.

[0005] The present invention is implemented as follows: a method for actively supporting and controlling a wind turbine generator set based on frequency regulation, the method comprising:

[0006] Obtain the day-ahead wind turbine output data and grid power parameters of the power system, and standardize the day-ahead wind turbine output data and grid power parameters;

[0007] Load the day-ahead wind turbine output data and grid power parameters, migrate them to the modeling sample set, traverse the modeling sample set to generate a training set and a test set, pre-build a frequency situation prediction model based on a multi-layer perceptron, iteratively train the frequency situation prediction model using the training set and the test set, and output a converged frequency situation prediction model;

[0008] Real-time collection of daily wind turbine output data and grid power parameters. Using these data as input, the system identifies and analyzes these data based on the frequency situation prediction model, and outputs a frequency prediction set within the prediction period.

[0009] The frequency prediction set is loaded, and the power fluctuation points and the power fluctuation values at the power fluctuation points within the prediction period are determined based on the frequency prediction set. The power fluctuation points and the power fluctuation values at the power fluctuation points are used as prior information, and the energy conversion efficiency and primary frequency regulation are taken into account as constraints to calculate the spare capacity based on overspeed control at the power fluctuation points.

[0010] Obtain the spare capacity at the power fluctuation point, determine the wind turbine load reduction operation point based on the cyclic congestion sorting strategy, and generate the wind turbine active support strategy based on the steady-state frequency deviation at the wind turbine load reduction operation point as a constraint;

[0011] In response to the wind turbine active support strategy, the PID speed regulator triggers an active support instruction based on the wind turbine active support strategy.

[0012] Preferably, the method of iteratively training the frequency situation prediction model using a training set and a test set includes:

[0013] Obtain the training set and test set, and load the initial model of the frequency situation prediction model;

[0014] Set the training batch size, number of training completes, learning rate, activation function, and loss function for the initial model;

[0015] Based on the feedforward neural network, the weights and frequency thresholds between the input layer and the hidden layer are generated. The training set is loaded and normalized by dimensionality reduction. The normalized training sample sequences are used as the columns of the training matrix, and the sample features are used as the rows of the training matrix. The training matrix is constructed and the weights are added to the training matrix.

[0016] Get the training matrix after adding the weight matrix, input the training matrix into the variable convolution attention module to obtain the feature convolution result, perform Fourier frequency analysis on the feature convolution result, and output the frequency value of the feature;

[0017] Determine whether the frequency value of the feature exceeds the preset frequency threshold. If it exceeds the preset frequency threshold, retain the current feature. If it does not exceed the preset frequency threshold, prune and compress the current feature weight.

[0018] Load the initial model after feature weight pruning and compression, retrain the initial model using the training set until convergence, and output the converged initial model;

[0019] Load the test set, use the test set as input, execute the initial model, and the initial model outputs the test results. It is determined whether the test results meet the preset accuracy threshold. If they meet the accuracy threshold, the converged frequency situation prediction model is output.

[0020] Preferably, the frequency situation prediction model uses a multi-layer perceptron model as an initial model, the initial model consists of an input layer, a hidden layer, and an output layer, the hidden layer is located between the input layer and the output layer, and when pre-building the frequency situation prediction model based on the multi-layer perceptron, a sparse autoencoder is introduced between the output layer and the hidden layer, the sparse autoencoder is used to ensure the sparsity of the hidden layer, and the activation function of the sparse autoencoder is a sigmiod activation function;

[0021] The hidden layers of the initial model include the first hidden layer and the second hidden layer. The number of neurons in the first hidden layer is 40. The hidden layers of the initial model are improved by freezing the second hidden layer and replacing it with a variable convolutional attention module. The variable convolutional attention module includes a multi-head attention mechanism, three-layer convolution, and a global average pooling layer. The three-layer convolution includes the first convolution layer, the second convolution layer, and the third convolution layer. The convolution kernel size of the first and second convolution layers is 3×3, and the convolution kernel size of the third convolution layer is 5×5.

[0022] A flexible trigger mechanism is introduced into the input layer of the initial model, and the input layer extracts the frequency deficiency features based on the flexible trigger mechanism.

[0023] Preferably, the method for extracting frequency deficiency features based on a flexible trigger mechanism at the input layer includes:

[0024] Load the daily wind turbine output data and grid power parameters, and define the frequency deficiency fusion threshold based on the daily wind turbine output data, the maximum frequency change rate of the grid in the grid power parameters, the active power deficiency, and the wind turbine inertia time constant. and mechanism trigger threshold ;

[0025] Construct a square wave function of the maximum frequency change rate of the power grid, the active power shortage, and the wind turbine inertia time constant within the daily sampling interval, and construct the Hessia matrix of the sampling interval based on the square wave function within the sampling interval;

[0026] The square wave function within the sampling interval is expressed as:

[0027] (1)

[0028] The Hessia matrix of the sampling interval is expressed as:

[0029] (2)

[0030] in, represents the square wave function within the sampling interval, represents the sampling interval, is the active power shortage within the sampling interval, represents the fan inertia time constant, is the maximum frequency change rate of the power grid, is the Hessia matrix output representation of the sampling interval, The first element of the square wave function and variables, Represents the square wave function and The second-order partial derivatives of the variables;

[0031] Obtain the Hessia matrix of the sampling interval, calculate the Hessia matrix result using the quasi-Newton method, and determine whether the Hessia matrix result exceeds the missing fusion threshold ;

[0032] If the fusion threshold is exceeded , integrate the Hessia matrix results into an integral space set, calculate the integral of the integral space set, and output the integral calculation results. When the integral calculation results accumulate to the mechanism trigger threshold When , the flexible trigger mechanism is triggered, and the characteristics of the maximum frequency change rate, active power shortage and wind turbine inertia time constant of the power grid within the daily sampling interval are extracted, and the Hessia matrix result is output;

[0033] The integral calculation of the integral space set is expressed as:

[0034] (3)

[0035] in, represents the result of the integral space set integral calculation, are the start and end time of the sampling interval respectively.

[0036] Preferably, the method for identifying and analyzing daily wind turbine output data and grid power parameters based on the frequency situation prediction model includes:

[0037] Obtain daily wind turbine output data and grid power parameters. The input layer extracts frequency deficit features based on a flexible trigger mechanism and outputs the Hessia matrix results.

[0038] Normalize the daily wind turbine output data and grid power parameters after removing the frequency deficiency feature, filter the normalized results, and output the filtered normalized features;

[0039] The sparse autoencoder performs sparse encoding and sparse decoding on the normalized features to obtain the normalized features after adding the gain matrix. Based on the relative entropy, a sparse penalty term is introduced to the normalized features after adding the gain matrix, and the autoencoding result is output;

[0040] Obtain the autoencoding result, perform weighted summation on the autoencoding result in the first hidden layer, and then normalize the weighted summation result using the sigmoid activation function to obtain a weighted autoencoding set;

[0041] The variable convolution attention module obtains the weighted autoencoder set and Hessia matrix results. The variable convolution attention module couples the weighted autoencoder set and Hessia matrix results to the objective function, calculates the frequency prediction results within the prediction period based on the objective function, and outputs the frequency prediction set within the prediction period.

[0042] Preferably, when normalizing the daily wind turbine output data and grid power parameters after removing the frequency deficiency feature, the normalization result is expressed as:

[0043] (4)

[0044] in, is the normalized processing result, Represents the data matrix composed of daily wind turbine output data and grid power parameters, is the data mean, represents the data variance;

[0045] The autoencoding result is expressed as:

[0046] (5)

[0047] (6)

[0048] (7)

[0049] in, represents the autoencoding result, Represents the relative entropy of normalized features, which is used to represent the relative entropy value of two groups of random variables. represents the sparse penalty term, Normalized features The gain matrix, is the number of data, is the characteristic number, represents the sparse ratio, is the activation value of the normalized feature, Represents the maximum value of normalized features;

[0050] When calculating the frequency prediction results within the prediction period based on the objective function, the objective function is defined as:

[0051] (8)

[0052] in, represents the frequency prediction result, are the grid frequency of the previous sampling interval and the intraday frequency mean, respectively.

[0053] Preferably, the calculation of the spare capacity based on overspeed control at the power fluctuation point by taking into account the energy conversion efficiency and the primary frequency regulation as constraints includes:

[0054] Loading power fluctuation points and power fluctuation values at the power fluctuation points, and determining the steady-state deviation of the wind turbine frequency based on the power fluctuation values at the power fluctuation points;

[0055] The steady-state deviation of fan frequency is expressed as:

[0056] (9)

[0057] in, is the steady-state deviation of the fan frequency, is the air density, is the wind speed, represents the leaf coverage area, is the power fluctuation value at the fluctuation point, represents the wind energy utilization coefficient, is the unit adjustment coefficient;

[0058] (10)

[0059] in, is the fan tip speed ratio, represent the fan speed and blade radius respectively;

[0060] Obtain the steady-state deviation of the fan frequency, and calculate the fan primary frequency regulation capability by combining the wind energy utilization coefficient and the steady-state deviation of the fan frequency;

[0061] The calculation expression of fan primary frequency regulation capability is as follows:

[0062] (11)

[0063] Calculate the spare capacity based on overspeed control at the power fluctuation point based on the primary frequency regulation capability and energy conversion efficiency of the wind turbine;

[0064] (12)

[0065] in, Indicates the fan's primary frequency regulation capability. are the mechanical torque and electromagnetic torque of the fan respectively, represents the rotor damping coefficient, is the energy conversion efficiency.

[0066] On the other hand, the present invention further provides a wind turbine active support control system based on frequency regulation, the wind turbine active support control system based on frequency regulation comprising:

[0067] The data acquisition module is used to obtain the wind turbine output data and grid power parameters of the power system on the previous day, and standardize the wind turbine output data and grid power parameters on the previous day;

[0068] The prediction model construction module is used to load the day-ahead wind turbine output data and grid power parameters, migrate the day-ahead wind turbine output data and grid power parameters to the modeling sample set, traverse the modeling sample set to generate a training set and a test set, pre-build a frequency situation prediction model based on a multi-layer perceptron, iteratively train the frequency situation prediction model using the training set and the test set, and output a converged frequency situation prediction model;

[0069] The frequency prediction module is used to collect daily wind turbine output data and grid power parameters in real time. Taking the daily wind turbine output data and grid power parameters as input, the module identifies and analyzes the daily wind turbine output data and grid power parameters based on the frequency situation prediction model, and outputs the frequency prediction set within the prediction period.

[0070] A capacity calculation module is used to load a frequency prediction set, determine the power fluctuation points and the power fluctuation values at the power fluctuation points within the prediction period based on the frequency prediction set, use the power fluctuation points and the power fluctuation values at the power fluctuation points as prior information, and calculate the spare capacity based on overspeed control at the power fluctuation points taking into account energy conversion efficiency and primary frequency regulation as constraints;

[0071] The active support module is used to obtain the spare capacity at the power fluctuation point, determine the wind turbine load reduction operation point based on the cyclic congestion sorting strategy, and generate the wind turbine active support strategy based on the steady-state frequency deviation at the wind turbine load reduction operation point as a constraint. In response to the wind turbine active support strategy, the PID speed regulator triggers the active support instruction based on the wind turbine active support strategy.

[0072] Preferably, the active support module comprises:

[0073] An operating point determination unit is used to obtain the spare capacity at the power fluctuation point and determine the wind turbine load reduction operating point based on the cyclic congestion sorting strategy;

[0074] The strategy generation unit generates an active support strategy for the wind turbine using the steady-state frequency deviation at the wind turbine load reduction operation point as a constraint;

[0075] The strategy response unit responds to the active support strategy of the wind turbine generator set, and the PID speed regulator triggers the active support instruction based on the active support strategy of the wind turbine generator set.

[0076] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0077] In an embodiment of the present invention, a frequency situation prediction model is constructed to realize frequency situation prediction within a prediction period, and the power fluctuation point and the power fluctuation value at the power fluctuation point within the prediction period are determined based on the frequency prediction set, thereby providing prior information for the spare capacity based on overspeed control, effectively extracting key features related to frequency deficiency, ensuring the primary frequency regulation capability and response speed of the active support control of the wind turbine, and overcoming the problem that the wind turbine linearization model of the existing method has relatively limited feature extraction capability and cannot effectively extract key features related to frequency deficiency, resulting in poor effectiveness, real-time performance and response speed of the wind turbine linearization model.

[0078] In an embodiment of the present invention, the frequency situation prediction model uses a multi-layer perceptron model as the initial model, and introduces a sparse autoencoder, a variable convolutional attention module, and a flexible trigger mechanism, which helps the model capture more representative features and improve generalization capabilities. The flexible trigger mechanism can respond quickly and trigger active support instructions in a timely manner to help wind turbines quickly adjust output power and maintain the stability of the grid frequency. At the same time, when training the frequency situation prediction model, the feature weights are pruned and compressed by judging whether the frequency value of the feature exceeds a preset frequency threshold, thereby reducing the model's storage space and transmission bandwidth, and improving the model's processing efficiency and speed.

[0079] In an embodiment of the present invention, frequency deficiency characteristics are extracted based on a flexible trigger mechanism, and frequency deficiency characteristics are defined based on the maximum frequency change rate of the power grid, active power deficiency, and wind turbine inertia time constant. This helps to more accurately identify and extract key features related to frequency deficiency, thereby improving the accuracy and efficiency of frequency situation prediction.

[0080] In an embodiment of the present invention, a frequency situation prediction model is used to identify and analyze the daily wind turbine output data and grid power parameters, thereby accurately outputting a frequency prediction set within a prediction period, thereby more accurately determining the required backup capacity size, avoiding resource waste caused by excessive backup capacity or the risk of power shortage caused by too little backup capacity, and also helping to smooth wind power fluctuations, improve the dispatchability of wind power, and thus increase the penetration rate of wind power in the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1It is a schematic diagram of the implementation flow of the active support control method for wind turbines based on frequency regulation provided by the present invention.

[0082] Figure 2 The figure shows a flow chart of the iterative training method for the frequency situation prediction model using training sets and test sets.

[0083] Figure 3 The figure shows a flow chart of the implementation of a method for extracting frequency deficiency features based on a flexible trigger mechanism at the input layer.

[0084] Figure 4 The figure shows a flow chart of the method for identifying and analyzing the daily wind turbine output data and power grid power parameters based on the frequency situation prediction model.

[0085] Figure 5 The present invention shows a schematic diagram of the implementation process of the method for calculating the reserve capacity based on overspeed control at the power fluctuation point taking energy conversion efficiency and primary frequency regulation as constraints.

[0086] Figure 6 It is a structural schematic diagram of the wind turbine active support control system based on frequency regulation provided by the present invention. DETAILED DESCRIPTION

[0087] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of the application are only for the purpose of describing specific embodiments and are not intended to limit this application. The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0088] The existing wind turbine linearization model has relatively limited feature extraction capabilities and cannot effectively extract key features related to frequency deficit, resulting in poor effectiveness, real-time performance, and response speed of the wind turbine linearization model. To address the above problems, we propose a wind turbine active support control method and system based on frequency regulation. In short, when implementing the method, the wind turbine output data and grid power parameters of the power system are first obtained on the day before, and then a frequency situation prediction model based on a multi-layer perceptron is pre-built. The frequency situation prediction model is used to identify and analyze the wind turbine output data and grid power parameters within the day, and a frequency prediction set within the prediction period is output. Based on the frequency prediction set, the power fluctuation points and power fluctuation values at the power fluctuation points within the prediction period are determined. The power fluctuation points and power fluctuation values at the power fluctuation points are used as prior information, and the energy conversion efficiency and primary frequency regulation are taken as constraints to calculate the reserve capacity based on overspeed control at the power fluctuation points. The wind turbine load reduction operation point is determined based on a cyclic congestion sorting strategy. The steady-state frequency deviation at the wind turbine load reduction operation point is used as a constraint to generate a wind turbine active support strategy. In an embodiment of the present invention, a frequency situation prediction model is constructed to realize frequency situation prediction within a prediction period, and the power fluctuation point and the power fluctuation value at the power fluctuation point within the prediction period are determined based on the frequency prediction set, thereby providing prior information for the spare capacity based on overspeed control, effectively extracting key features related to frequency deficiency, ensuring the primary frequency regulation capability and response speed of the active support control of the wind turbine, and overcoming the problem that the wind turbine linearization model of the existing method has relatively limited feature extraction capability and cannot effectively extract key features related to frequency deficiency, resulting in poor effectiveness, real-time performance and response speed of the wind turbine linearization model.

[0089] The embodiment of the present invention provides a method for actively supporting and controlling a wind turbine generator system based on frequency regulation. Figure 1 The following is a schematic diagram of the implementation process of the active support control method for wind turbines based on frequency regulation, wherein the active support control method for wind turbines based on frequency regulation specifically includes:

[0090] S10, obtaining the day-ahead wind turbine output data and grid power parameters of the power system, and standardizing the day-ahead wind turbine output data and grid power parameters;

[0091] It should be noted that the wind turbine output data of the day before includes but is not limited to the day before wind speed data, wind direction data, temperature data, humidity data, power data, rotor diameter, rated power, output voltage, wind turbine speed, and rotor parameters; the grid power parameters include but are not limited to the minimum grid frequency, the time of the minimum value, the steady-state frequency of the grid, the maximum frequency change rate, the active power shortage, the inertia time constant, the dead zone range of the speed regulator, the initial response time, and the average frequency change rate.

[0092] In this embodiment, the standardization processing methods for the wind turbine output data and grid power parameters on the previous day include but are not limited to logarithmic function normalization processing, centering processing, quartile method and Grey Wolf optimized box plot algorithm processing.

[0093] S20, loading the day-ahead wind turbine output data and grid power parameters, migrating the day-ahead wind turbine output data and grid power parameters to a modeling sample set, traversing the modeling sample set to generate a training set and a test set, pre-building a frequency situation prediction model based on a multi-layer perceptron, iteratively training the frequency situation prediction model using the training set and the test set, and outputting a converged frequency situation prediction model;

[0094] In this embodiment, the distribution ratio of the training set and the test set can be 3:1 or 4:1.

[0095] S30, collecting daily wind turbine output data and grid power parameters in real time, using the daily wind turbine output data and grid power parameters as input, identifying and analyzing the daily wind turbine output data and grid power parameters based on a frequency situation prediction model, and outputting a frequency prediction set within a prediction period;

[0096] S40, loading the frequency prediction set, determining the power fluctuation points and the power fluctuation values at the power fluctuation points within the prediction period based on the frequency prediction set, using the power fluctuation points and the power fluctuation values at the power fluctuation points as prior information, taking into account the energy conversion efficiency and primary frequency regulation as constraints, and calculating the spare capacity based on overspeed control at the power fluctuation points.

[0097] S50, obtaining the spare capacity at the power fluctuation point, determining the wind turbine load reduction operation point based on the cyclic congestion sorting strategy, and generating the wind turbine active support strategy with the steady-state frequency deviation at the wind turbine load reduction operation point as a constraint;

[0098] S60 , in response to the wind turbine active support strategy, the PID speed regulator triggers an active support instruction based on the wind turbine active support strategy.

[0099] It's important to note that the PID (Proportional-Integral-Derivative) speed regulator is a classic control algorithm widely used in various industrial control systems, including those for wind turbines. When the grid frequency deviates, the PID speed regulator calculates the required power adjustment based on pre-set control logic and algorithms, and generates corresponding active support commands. These commands guide the wind turbine to adjust its output power to participate in grid frequency regulation.

[0100] In an embodiment of the present invention, a frequency situation prediction model is constructed to realize frequency situation prediction within a prediction period, and the power fluctuation point and the power fluctuation value at the power fluctuation point within the prediction period are determined based on the frequency prediction set, thereby providing prior information for the spare capacity based on overspeed control, effectively extracting key features related to frequency deficiency, ensuring the primary frequency regulation capability and response speed of the active support control of the wind turbine, and overcoming the problem that the wind turbine linearization model of the existing method has relatively limited feature extraction capability and cannot effectively extract key features related to frequency deficiency, resulting in poor effectiveness, real-time performance and response speed of the wind turbine linearization model.

[0101] The embodiment of the present invention provides a method for iteratively training a frequency situation prediction model using a training set and a test set. Figure 2 The following is a schematic diagram of the implementation process of the iterative training method for the frequency situation prediction model using the training set and the test set. The method of iterative training the frequency situation prediction model using the training set and the test set specifically includes:

[0102] Step S101: Obtain a training set and a test set, and load an initial model of a frequency situation prediction model;

[0103] Step S102, setting the training batch size, number of complete training times, learning rate, activation function and loss function of the initial model;

[0104] In this embodiment, the training batch size of the initial model can be 15-20, the number of complete training times can be 150-200 times, the learning rate is set to 0.002, the activation function of the initial model is the softmax function, and the loss function can be the hinge-loss loss function.

[0105] Step S103: Generate weights and frequency thresholds between the input layer and the hidden layer based on the feedforward neural network, load the training set, perform dimensionality reduction and normalization on the training set, use the normalized training sample sequences as the columns of the training matrix, and the sample features as the rows of the training matrix to complete the construction of the training matrix, and add the weights to the training matrix;

[0106] In this embodiment, the weight represents the relative importance of the input features to the prediction results. In frequency situation prediction, different input features (such as historical frequency data, weather conditions, wind turbine status, etc.) have different degrees of influence on the grid frequency. Through the training process, the feedforward neural network automatically learns and adjusts these weights, giving greater weight to features that are more important to the prediction results. During the training process, the frequency threshold can be adjusted as a hyperparameter. Through methods such as cross-validation, the optimal frequency threshold can be found, allowing the model to reasonably control the frequency and amplitude of active support actions while maintaining a high prediction accuracy.

[0107] Step S104: Obtain the training matrix after adding the weight matrix, input the training matrix into the variable convolution attention module to obtain the feature convolution result, perform Fourier frequency analysis on the feature convolution result, and output the frequency value of the feature;

[0108] Step S105, determining whether the frequency value of the feature exceeds a preset frequency threshold;

[0109] In this embodiment, the preset frequency threshold may be 0.3-0.55.

[0110] Step S106: if the frequency exceeds the preset threshold, retain the current feature;

[0111] Step S107: If the frequency does not exceed the preset threshold, the current feature weight is pruned and compressed;

[0112] Step S108, loading the initial model after feature weight pruning and compression processing, retraining the initial model using the training set until convergence, and outputting the converged initial model;

[0113] Step S109: Load the test set, use the test set as input, execute the initial model, and the initial model outputs the test results;

[0114] Step S1010, determining whether the test result meets a preset accuracy threshold. In this embodiment, the preset accuracy threshold may be 0.89-0.91;

[0115] Step S1011: If the accuracy threshold is met, output a converged frequency situation prediction model.

[0116] In an embodiment of the present invention, the frequency situation prediction model uses a multi-layer perceptron model as the initial model, and introduces a sparse autoencoder, a variable convolutional attention module, and a flexible trigger mechanism, which helps the model capture more representative features and improve generalization capabilities. The flexible trigger mechanism can respond quickly and trigger active support instructions in a timely manner to help wind turbines quickly adjust output power and maintain the stability of the grid frequency. At the same time, when training the frequency situation prediction model, the feature weights are pruned and compressed by judging whether the frequency value of the feature exceeds a preset frequency threshold, thereby reducing the model's storage space and transmission bandwidth, and improving the model's processing efficiency and speed.

[0117] In this embodiment, the frequency situation prediction model uses a multi-layer perceptron model as the initial model. The initial model consists of an input layer, a hidden layer, and an output layer. The hidden layer is located between the input layer and the output layer. When pre-constructing a frequency situation prediction model based on a multi-layer perceptron, a sparse autoencoder is introduced between the output layer and the hidden layer. The sparse autoencoder is used to ensure the sparsity of the hidden layer, and the sparse autoencoder activation function is a sigmiod activation function. It should be noted that the input layer is connected to the hidden layer, and the hidden layer is connected to the output layer.

[0118] The hidden layers of the initial model include the first hidden layer and the second hidden layer. The number of neurons in the first hidden layer is 40. The hidden layers of the initial model are improved by freezing the second hidden layer and replacing it with a variable convolutional attention module. The variable convolutional attention module includes a multi-head attention mechanism, three-layer convolution and a global average pooling layer. The three-layer convolution includes the first convolution layer, the second convolution layer and the third convolution layer. The convolution kernel size of the first convolution layer and the second convolution layer is 3×3, and the convolution kernel size of the third convolution layer is 5×5. A flexible trigger mechanism is introduced into the input layer of the initial model, and the input layer extracts the frequency deficiency features based on the flexible trigger mechanism.

[0119] The embodiment of the present invention provides a method for extracting frequency deficiency features based on a flexible trigger mechanism at the input layer. Figure 3 The following is a schematic diagram of the implementation process of a method for extracting frequency deficiency features based on a flexible trigger mechanism at the input layer. The method for extracting frequency deficiency features based on a flexible trigger mechanism at the input layer specifically includes:

[0120] S201, loading the daily wind turbine output data and grid power parameters, and defining the frequency deficiency fusion threshold based on the daily wind turbine output data, the maximum frequency change rate of the grid, the active power deficiency, and the wind turbine inertia time constant in the grid power parameters. and mechanism trigger threshold ;

[0121] In this embodiment of the present invention, the frequency deficit characteristic is defined based on the maximum grid frequency change rate, active power deficit, and wind turbine inertia time constant. The maximum grid frequency change rate refers to the maximum rate of change of the grid frequency per unit time and is a key indicator of grid stability. When a grid experiences an active power deficit or surplus, the frequency changes, and the rate of change reflects the grid's ability to respond to this imbalance. Active power deficit is the primary cause of a drop in grid frequency. When the active power deficit is large, the grid frequency drops significantly, potentially even causing serious accidents such as frequency collapse. The wind turbine inertia time constant refers to the time constant for the wind turbine rotor to continue rotating due to mechanical inertia after losing electrical input. It reflects the wind turbine's response speed and regulation capability to grid frequency changes. By comprehensively considering these three factors, a comprehensive frequency deficit characteristic indicator is defined. A larger frequency deficit characteristic value indicates a more severe grid frequency deficit and the need for more proactive measures to support and restore the frequency.

[0122] S202, constructing a square wave function of the maximum frequency change rate of the power grid, the active power shortage, and the wind turbine inertia time constant within the daily sampling interval, and constructing a Hessia matrix of the sampling interval based on the square wave function within the sampling interval;

[0123] The square wave function within the sampling interval is expressed as:

[0124] (1)

[0125] The Hessia matrix of the sampling interval is expressed as:

[0126] (2)

[0127] in, represents the square wave function within the sampling interval, represents the sampling interval, is the active power shortage within the sampling interval, represents the fan inertia time constant, is the maximum frequency change rate of the power grid, is the Hessia matrix output representation of the sampling interval, The first element of the square wave function and variables, Represents the square wave function and The second-order partial derivatives of the variables;

[0128] S203, obtaining the Hessia matrix of the sampling interval, and calculating the Hessia matrix result using the quasi-Newton method;

[0129] It should be noted that Quasi-Newton Methods are a type of iterative algorithm used to solve unconstrained optimization problems. They avoid directly calculating the second-order derivatives by constructing an approximation of the Hessian matrix, thereby improving computational efficiency.

[0130] S204, determine whether the Hessia matrix result exceeds the missing fusion threshold ;

[0131] S205: If the threshold for missing fusion is exceeded , integrate the Hessia matrix results into an integral space set, calculate the integral of the integral space set, and output the integral calculation results. When the integral calculation results accumulate to the mechanism trigger threshold When , the flexible trigger mechanism is triggered, and the characteristics of the maximum frequency change rate, active power shortage and wind turbine inertia time constant of the power grid within the daily sampling interval are extracted, and the Hessia matrix result is output;

[0132] The integral calculation of the integral space set is expressed as:

[0133] (3)

[0134] in, represents the result of the integral space set integral calculation, are the start and end time of the sampling interval respectively.

[0135] In an embodiment of the present invention, frequency deficiency characteristics are extracted based on a flexible trigger mechanism, and frequency deficiency characteristics are defined based on the maximum frequency change rate of the power grid, active power deficiency, and wind turbine inertia time constant. This helps to more accurately identify and extract key features related to frequency deficiency, thereby improving the accuracy and efficiency of frequency situation prediction.

[0136] The embodiment of the present invention provides a method for identifying and analyzing daily wind turbine output data and grid power parameters based on a frequency situation prediction model. Figure 4 The following is a schematic diagram of the implementation process of a method for identifying and analyzing the output data of wind turbines and power parameters of power grids based on a frequency situation prediction model. The method for identifying and analyzing the output data of wind turbines and power parameters of power grids based on a frequency situation prediction model specifically includes:

[0137] S301: Obtain daily wind turbine output data and grid power parameters. The input layer extracts frequency deficit features based on a flexible trigger mechanism and outputs the Hessia matrix result.

[0138] S302, normalizing the daily wind turbine output data and grid power parameters after removing the frequency deficiency feature, filtering the normalized results, and outputting the filtered normalized features;

[0139] In this embodiment, when normalizing the daily wind turbine output data and grid power parameters after removing the frequency deficiency feature, the normalization result is expressed as:

[0140] (4)

[0141] in, is the normalized processing result, Represents the data matrix composed of daily wind turbine output data and grid power parameters, is the data mean, Represents the data variance.

[0142] S303, the sparse autoencoder performs sparse encoding and sparse decoding on the normalized features to obtain normalized features after adding the gain matrix, introduces a sparse penalty term to the normalized features after adding the gain matrix based on relative entropy, and outputs the autoencoding result;

[0143] Among them, the self-encoding result is expressed as:

[0144] (5)

[0145] (6)

[0146] (7)

[0147] in, represents the autoencoding result, Represents the relative entropy of normalized features, which is used to represent the relative entropy value of two groups of random variables. represents the sparse penalty term, Normalized features The gain matrix, is the number of data, is the characteristic number. In this embodiment, the characteristic number is 3-15. represents the sparse ratio, is the activation value of the normalized feature, which can be 0.35-0.68, Represents the maximum value of the normalized feature.

[0148] S304, obtaining the autoencoding result, performing weighted summation processing on the autoencoding result in the first hidden layer, and then normalizing the weighted summation result using a sigmoid activation function to obtain a weighted autoencoding set;

[0149] In this embodiment, when the first hidden layer performs a weighted summation on the autoencoder results, each neuron in the first hidden layer performs a weighted summation on the received inputs. This means that each input feature is multiplied by the corresponding weight, and then all the products are added together to obtain the total input of the neuron. The result of this weighted summation is the total input of the neuron. The sigmoid activation function compresses the input value to between 0 and 1, making the output more stable and conducive to probabilistic interpretation or binary classification tasks.

[0150] S305, the variable convolution attention module obtains the weighted autoencoder set and the Hessia matrix result. The variable convolution attention module unifies the coupling constraints of the weighted autoencoder set and the Hessia matrix result into the objective function, calculates the frequency prediction result within the prediction period based on the objective function, and outputs the frequency prediction set within the prediction period.

[0151] When calculating the frequency prediction results within the prediction period based on the objective function, the objective function is defined as:

[0152] (8)

[0153] in, represents the frequency prediction result, are the grid frequency of the previous sampling interval and the intraday frequency mean, respectively.

[0154] In an embodiment of the present invention, a frequency situation prediction model is used to identify and analyze the daily wind turbine output data and grid power parameters, thereby accurately outputting a frequency prediction set within a prediction period, thereby more accurately determining the required backup capacity size, avoiding resource waste caused by excessive backup capacity or the risk of power shortage caused by too little backup capacity, and also helping to smooth wind power fluctuations, improve the dispatchability of wind power, and thus increase the penetration rate of wind power in the power grid.

[0155] The embodiment of the present invention provides a method for calculating the spare capacity based on overspeed control at the power fluctuation point by taking energy conversion efficiency and primary frequency modulation as constraints. Figure 5 A schematic diagram of a method for calculating the spare capacity based on overspeed control at a power fluctuation point taking into account energy conversion efficiency and primary frequency regulation as constraints is shown. The method specifically includes:

[0156] S401, loading a power fluctuation point and a power fluctuation value at the power fluctuation point, and determining a wind turbine frequency steady-state deviation based on the power fluctuation value at the power fluctuation point;

[0157] The steady-state deviation of fan frequency is expressed as:

[0158] (9)

[0159] in, is the steady-state deviation of the fan frequency, is the air density, is the wind speed, represents the leaf coverage area, is the power fluctuation value at the fluctuation point, represents the wind energy utilization coefficient, is the unit adjustment coefficient;

[0160] (10)

[0161] in, is the fan tip speed ratio, represent the fan speed and blade radius respectively;

[0162] It should be noted that the steady-state frequency deviation of a wind turbine refers to the difference between the output frequency of a wind turbine and the rated frequency of the grid when the wind turbine is operating in a steady state. The steady-state frequency deviation of a wind turbine can reflect the system's ability to resist active power disturbances through primary frequency regulation.

[0163] S402, obtaining a steady-state deviation of the wind turbine frequency, and calculating the primary frequency regulation capability of the wind turbine by combining the wind energy utilization coefficient and the steady-state deviation of the wind turbine frequency;

[0164] The calculation expression of fan primary frequency regulation capability is as follows:

[0165] (11)

[0166] S403, calculating the spare capacity based on overspeed control at the power fluctuation point based on the primary frequency regulation capability and energy conversion efficiency of the wind turbine;

[0167] (12)

[0168] in, Indicates the fan's primary frequency regulation capability. are the mechanical torque and electromagnetic torque of the fan respectively, Represents the rotor damping coefficient. The rotor damping coefficient is an important parameter to describe the vibration stability of the generator rotor. It refers to the ratio of the rotor vibration attenuation rate to the vibration frequency. The larger this coefficient is, the faster the rotor vibration attenuation rate is, the smaller the speed fluctuation range is, and the more stable the rotor is. Conversely, the smaller the damping coefficient is, the larger the speed fluctuation range is, and the more unstable the rotor is. The energy conversion efficiency can be 35-45%.

[0169] The embodiment of the present invention provides a wind turbine active support control system based on frequency regulation. Figure 6The structure diagram of the wind turbine active support control system based on frequency regulation is shown. The wind turbine active support control system based on frequency regulation specifically includes:

[0170] The data acquisition module 100 is used to obtain the wind turbine output data and grid power parameters of the power system on the previous day, and to perform standardization processing on the wind turbine output data and grid power parameters on the previous day;

[0171] The prediction model construction module 200 is used to load the wind turbine output data and grid power parameters of the day before, migrate the wind turbine output data and grid power parameters to the modeling sample set, traverse the modeling sample set to generate a training set and a test set, pre-build a frequency situation prediction model based on a multi-layer perceptron, iteratively train the frequency situation prediction model using the training set and the test set, and output a converged frequency situation prediction model;

[0172] The frequency prediction module 300 is used to collect daily wind turbine output data and grid power parameters in real time. Taking the daily wind turbine output data and grid power parameters as input, the module identifies and analyzes the daily wind turbine output data and grid power parameters based on the frequency situation prediction model, and outputs a frequency prediction set within the prediction period.

[0173] Capacity calculation module 400 is configured to load a frequency prediction set, determine power fluctuation points and power fluctuation values at the power fluctuation points within a prediction period based on the frequency prediction set, calculate the reserve capacity based on overspeed control at the power fluctuation points using the power fluctuation points and the power fluctuation values at the power fluctuation points as prior information, and taking into account energy conversion efficiency and primary frequency regulation as constraints;

[0174] The active support module 500 is used to obtain the spare capacity at the power fluctuation point, determine the wind turbine load reduction operation point based on the cyclic congestion sorting strategy, and generate the wind turbine active support strategy based on the steady-state frequency deviation at the wind turbine load reduction operation point as a constraint. In response to the wind turbine active support strategy, the PID speed regulator triggers the active support instruction based on the wind turbine active support strategy.

[0175] It should be noted that the data acquisition module 100, the prediction model construction module 200, the frequency prediction module 300, the capacity calculation module 400, and the active support module 500 are connected by a local area network or DTU communication, and the frequency-adjusted wind turbine active support control system provided in the embodiment of the present invention corresponds to the above-mentioned frequency-adjusted wind turbine active support control method. The explanations, examples, beneficial effects, etc. of the relevant contents can refer to the corresponding contents in the frequency-adjusted wind turbine active support control method, and will not be repeated here.

[0176] In this embodiment, the active support module 500 includes:

[0177] An operating point determination unit 510 is configured to obtain spare capacity at a power fluctuation point and determine a load shedding operating point for the wind turbine based on a cyclic congestion sorting strategy;

[0178] A strategy generating unit 520 generates an active support strategy for the wind turbine generator set based on the steady-state frequency deviation at the wind turbine load reduction operation point as a constraint;

[0179] The strategy response unit 530 responds to the wind turbine active support strategy, and the PID speed regulator triggers an active support instruction based on the wind turbine active support strategy.

[0180] On the other hand, an embodiment of the present invention further provides a computer device, which includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the method of any one of the above embodiments is implemented.

[0181] As a non-volatile computer-readable storage medium, the memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as the program instructions / modules corresponding to the frequency-regulated wind turbine active support control method in the embodiment of the present application. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created by the use of the frequency-regulated wind turbine active support control method, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely located relative to the processor, and these remote memories may be connected to the local module via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0182] In summary, the present invention provides a method and system for active support control of wind turbines based on frequency regulation. In an embodiment of the present invention, a frequency situation prediction model is constructed to realize frequency situation prediction within a prediction period, and the power fluctuation point and the power fluctuation value at the power fluctuation point within the prediction period are determined based on the frequency prediction set, thereby providing prior information for the spare capacity based on overspeed control, effectively extracting key features related to frequency deficiency, ensuring the primary frequency regulation capability and response speed of the active support control of the wind turbine, and overcoming the problem that the wind turbine linearization model of the existing method has relatively limited feature extraction capability and cannot effectively extract key features related to frequency deficiency, resulting in poor effectiveness, real-time performance and response speed of the wind turbine linearization model.

[0183] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the embodiments described are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also fall within the scope of protection of the present invention.

Claims

1. A wind turbine active support control method based on frequency regulation, characterized in that: include: Obtain the day-ahead wind turbine output data and grid power parameters of the power system, and standardize the day-ahead wind turbine output data and grid power parameters; Load the day-ahead wind turbine output data and grid power parameters, migrate them to the modeling sample set, traverse the modeling sample set to generate a training set and a test set, pre-build a frequency situation prediction model based on a multi-layer perceptron, iteratively train the frequency situation prediction model using the training set and the test set, and output a converged frequency situation prediction model. The frequency situation prediction model uses a multi-layer perceptron model as the initial model. The initial model consists of an input layer, a hidden layer, and an output layer. A sparse autoencoder is introduced between the output layer and the hidden layer. The input layer extracts frequency deficiency features based on a flexible trigger mechanism. Real-time collection of daily wind turbine output data and grid power parameters. Using these data as input, the system identifies and analyzes these data based on the frequency situation prediction model, and outputs a frequency prediction set within the prediction period. The frequency prediction set is loaded, and the power fluctuation points and power fluctuation values at the power fluctuation points within the prediction period are determined based on the frequency prediction set. The power fluctuation points and power fluctuation values at the power fluctuation points are used as prior information, and the energy conversion efficiency and primary frequency regulation are taken into account as constraints to calculate the reserve capacity based on overspeed control at the power fluctuation points. The calculation of the reserve capacity based on overspeed control at the power fluctuation point taking into account energy conversion efficiency and primary frequency regulation as constraints includes: Loading power fluctuation points and power fluctuation values at the power fluctuation points, and determining the steady-state deviation of the wind turbine frequency based on the power fluctuation values at the power fluctuation points; The steady-state deviation of the fan frequency is expressed as: （9） in, is the steady-state deviation of the fan frequency, is the air density, is the wind speed, represents the leaf coverage area, is the power fluctuation value at the fluctuation point, represents the wind energy utilization coefficient, is the unit adjustment coefficient; （10） in, is the fan tip speed ratio, represent the fan speed and blade radius respectively; Obtain the steady-state deviation of the fan frequency, and calculate the fan primary frequency regulation capability by combining the wind energy utilization coefficient and the steady-state deviation of the fan frequency; The calculation expression of fan primary frequency regulation capability is as follows: （11） Calculate the spare capacity based on overspeed control at the power fluctuation point based on the primary frequency regulation capability and energy conversion efficiency of the wind turbine; （12） in, Indicates the fan's primary frequency regulation capability. are the mechanical torque and electromagnetic torque of the fan respectively, represents the rotor damping coefficient, is the energy conversion efficiency.

2. The method for controlling active support of a wind turbine generator system based on frequency regulation according to claim 1, wherein: The wind turbine active support control method based on frequency regulation further includes: Obtain the spare capacity at the power fluctuation point, determine the wind turbine load reduction operation point based on the cyclic congestion sorting strategy, and generate the wind turbine active support strategy based on the steady-state frequency deviation at the wind turbine load reduction operation point as a constraint; In response to the wind turbine active support strategy, the PID speed regulator triggers an active support instruction based on the wind turbine active support strategy.

3. The wind turbine active support control method based on frequency regulation according to claim 1, characterized in that: The method for iteratively training the frequency situation prediction model using a training set and a test set includes: Obtain the training set and test set, and load the initial model of the frequency situation prediction model; Set the training batch size, number of training completes, learning rate, activation function, and loss function for the initial model; Based on the feedforward neural network, the weights and frequency thresholds between the input layer and the hidden layer are generated. The training set is loaded and normalized by dimensionality reduction. The normalized training sample sequences are used as the columns of the training matrix, and the sample features are used as the rows of the training matrix. The training matrix is constructed and the weights are added to the training matrix. Get the training matrix after adding the weight matrix, input the training matrix into the variable convolution attention module to obtain the feature convolution result, perform Fourier frequency analysis on the feature convolution result, and output the frequency value of the feature; Determine whether the frequency value of the feature exceeds the preset frequency threshold. If it exceeds the preset frequency threshold, retain the current feature. If it does not exceed the preset frequency threshold, prune and compress the current feature weight. Load the initial model after feature weight pruning and compression, retrain the initial model using the training set until convergence, and output the converged initial model; Load the test set, use the test set as input, execute the initial model, and the initial model outputs the test results. It is determined whether the test results meet the preset accuracy threshold. If they meet the accuracy threshold, the converged frequency situation prediction model is output.

4. The wind turbine active support control method based on frequency regulation according to claim 3, characterized in that: The frequency situation prediction model uses a multi-layer perceptron model as an initial model. The initial model consists of an input layer, a hidden layer, and an output layer. The hidden layer is located between the input layer and the output layer. When pre-building the frequency situation prediction model based on the multi-layer perceptron, a sparse autoencoder is introduced between the output layer and the hidden layer. The sparse autoencoder is used to ensure the sparsity of the hidden layer, and the activation function of the sparse autoencoder is a sigmiod activation function. The hidden layers of the initial model include the first hidden layer and the second hidden layer. The number of neurons in the first hidden layer is 40. The hidden layers of the initial model are improved by freezing the second hidden layer and replacing it with a variable convolutional attention module. The variable convolutional attention module includes a multi-head attention mechanism, three-layer convolution, and a global average pooling layer. The three-layer convolution includes the first convolution layer, the second convolution layer, and the third convolution layer. The convolution kernel size of the first and second convolution layers is 3×3, and the convolution kernel size of the third convolution layer is 5×5. A flexible trigger mechanism is introduced into the input layer of the initial model, and the input layer extracts the frequency deficiency features based on the flexible trigger mechanism.

5. The wind turbine active support control method based on frequency regulation according to claim 4, characterized in that: The method for extracting frequency deficiency features based on a flexible trigger mechanism in the input layer includes: Load the daily wind turbine output data and grid power parameters, and define the frequency deficiency fusion threshold based on the daily wind turbine output data, the maximum frequency change rate of the grid in the grid power parameters, the active power deficiency, and the wind turbine inertia time constant. and mechanism trigger threshold ; Construct a square wave function of the maximum frequency change rate of the power grid, the active power shortage, and the wind turbine inertia time constant within the daily sampling interval, and construct the Hessia matrix of the sampling interval based on the square wave function within the sampling interval; The square wave function within the sampling interval is expressed as: （1） The Hessia matrix of the sampling interval is expressed as: （2） in, represents the square wave function within the sampling interval, represents the sampling interval, is the active power shortage within the sampling interval, represents the fan inertia time constant, is the maximum frequency change rate of the power grid, is the Hessia matrix output representation of the sampling interval, The first element of the square wave function and variables, Represents the square wave function and The second-order partial derivatives of the variables; Obtain the Hessia matrix of the sampling interval, calculate the Hessia matrix result using the quasi-Newton method, and determine whether the Hessia matrix result exceeds the missing fusion threshold ; If the fusion threshold is exceeded , integrate the Hessia matrix results into an integral space set, calculate the integral of the integral space set, and output the integral calculation results. When the integral calculation results accumulate to the mechanism trigger threshold When , the flexible trigger mechanism is triggered, and the characteristics of the maximum frequency change rate, active power shortage and wind turbine inertia time constant of the power grid within the daily sampling interval are extracted, and the Hessia matrix result is output; The integral calculation of the integral space set is expressed as: （3） in, represents the result of the integral space set integral calculation, are the start and end time of the sampling interval respectively.

6. The wind turbine active support control method based on frequency regulation according to claim 5, characterized in that: The method for identifying and analyzing daily wind turbine output data and grid power parameters based on a frequency situation prediction model includes: Obtain daily wind turbine output data and grid power parameters. The input layer extracts frequency deficit features based on a flexible trigger mechanism and outputs the Hessia matrix results. Normalize the daily wind turbine output data and grid power parameters after removing the frequency deficiency feature, filter the normalized results, and output the filtered normalized features; The sparse autoencoder performs sparse encoding and sparse decoding on the normalized features to obtain the normalized features after adding the gain matrix. Based on the relative entropy, a sparse penalty term is introduced to the normalized features after adding the gain matrix, and the autoencoding result is output; Obtain the autoencoding result, perform weighted summation on the autoencoding result in the first hidden layer, and then normalize the weighted summation result using the sigmoid activation function to obtain a weighted autoencoding set; The variable convolution attention module obtains the weighted autoencoder set and Hessia matrix results. The variable convolution attention module couples the weighted autoencoder set and Hessia matrix results to the objective function, calculates the frequency prediction results within the prediction period based on the objective function, and outputs the frequency prediction set within the prediction period.

7. The wind turbine active support control method based on frequency regulation according to claim 6, characterized in that: When normalizing the daily wind turbine output data and grid power parameters after removing the frequency deficiency feature, the normalization result is expressed as: （4） in, is the normalized processing result, Represents the data matrix composed of daily wind turbine output data and grid power parameters, is the data mean, represents the data variance; The autoencoding result is expressed as: （5） （6） （7） in, represents the autoencoding result, Represents the relative entropy of normalized features, which is used to represent the relative entropy value of two groups of random variables. represents the sparse penalty term, Normalized features The gain matrix, is the number of data, is the characteristic number, represents the sparse ratio, is the activation value of the normalized feature, Represents the maximum value of normalized features; When calculating the frequency prediction results within the prediction period based on the objective function, the objective function is defined as: （8） in, represents the frequency prediction result, are the grid frequency of the previous sampling interval and the intraday frequency mean, respectively.

8. A wind turbine active support control system based on frequency regulation, used to implement the wind turbine active support control method based on frequency regulation according to any one of claims 1 to 7, characterized in that: The wind turbine active support control system based on frequency regulation includes: The data acquisition module is used to obtain the wind turbine output data and grid power parameters of the power system on the previous day, and standardize the wind turbine output data and grid power parameters on the previous day; The prediction model construction module is used to load the day-ahead wind turbine output data and grid power parameters, migrate the day-ahead wind turbine output data and grid power parameters to the modeling sample set, traverse the modeling sample set to generate a training set and a test set, pre-build a frequency situation prediction model based on a multi-layer perceptron, iteratively train the frequency situation prediction model using the training set and the test set, and output a converged frequency situation prediction model; The frequency prediction module is used to collect daily wind turbine output data and grid power parameters in real time. Taking the daily wind turbine output data and grid power parameters as input, the module identifies and analyzes the daily wind turbine output data and grid power parameters based on the frequency situation prediction model, and outputs the frequency prediction set within the prediction period. A capacity calculation module is used to load a frequency prediction set, determine the power fluctuation points and the power fluctuation values at the power fluctuation points within the prediction period based on the frequency prediction set, use the power fluctuation points and the power fluctuation values at the power fluctuation points as prior information, and calculate the spare capacity based on overspeed control at the power fluctuation points taking into account energy conversion efficiency and primary frequency regulation as constraints; The active support module is used to obtain the spare capacity at the power fluctuation point, determine the wind turbine load reduction operation point based on the cyclic congestion sorting strategy, and generate the wind turbine active support strategy based on the steady-state frequency deviation at the wind turbine load reduction operation point as a constraint. In response to the wind turbine active support strategy, the PID speed regulator triggers the active support instruction based on the wind turbine active support strategy.

9. The wind turbine active support control system based on frequency regulation according to claim 8, characterized in that: The active support module comprises: An operating point determination unit is used to obtain the spare capacity at the power fluctuation point and determine the wind turbine load reduction operating point based on the cyclic congestion sorting strategy; The strategy generation unit generates an active support strategy for the wind turbine using the steady-state frequency deviation at the wind turbine load reduction operation point as a constraint; The strategy response unit responds to the active support strategy of the wind turbine generator set, and the PID speed regulator triggers the active support instruction based on the active support strategy of the wind turbine generator set.

Images