Method and system for continuous ride-through of grid-connected wind turbine based on multi-dimensional double-layer network

Abstract

The application discloses a kind of based on multi-dimension double-layer network's network type wind turbine continuous crossing method and system, method includes: obtaining network type wind turbine key parameters;According to network type wind turbine key parameters, establish network type wind turbine continuous crossing state space model;The network type wind turbine continuous crossing model based on multi-dimension double-layer neural network is constructed, data fitting training is carried out to network type wind turbine continuous crossing model, based on network type wind turbine continuous crossing model real-time compensation minimum network type wind turbine continuous crossing state space model error;With network type wind turbine output power oscillation minimization establishes quadratic programming cost function, based on the network type wind turbine continuous crossing state space model after compensation, by solving quadratic programming problem optimal solution, and rolling optimization obtains network type wind turbine optimal control instruction.The application has obvious advantages such as inhibiting network type wind turbine power oscillation in continuous crossing process.

Description

Technical Field

[0001] This invention mainly relates to the field of wind power generation technology, specifically to a method and system for continuous traversal of a grid-type wind turbine based on a multi-dimensional two-layer network. Background Technology

[0002] Traditional grid-connected wind turbines cannot achieve rapid frequency and voltage support to improve the operational stability of new power systems, thus grid-connected wind turbines with active support capabilities have attracted attention. However, due to the use of virtual synchronous control, grid-connected wind turbines exhibit slow power response, output frequency fluctuations, and poor power point tracking performance under normal operating conditions. Grid-connected wind turbines have numerous control parameters, strong coupling, and nonlinearity. Existing predictive control methods cannot quickly calculate the optimal control command while ensuring high prediction accuracy. Traditional grid-connected wind turbine control methods cannot solve these problems. Therefore, it is urgent to study how to ensure the prediction accuracy of grid-connected wind turbines, quickly obtain the optimal control command, and effectively suppress active power oscillations and minimize reactive power deviations. Summary of the Invention

[0003] To address the technical problems existing in the prior art, this invention provides a method and system for continuous wind turbine crossing based on a multi-dimensional dual-layer network, characterized by high control precision and fast dynamic response.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0005] Firstly, a method for continuous wind turbine crossing based on a multi-dimensional two-layer network is provided, including the following steps:

[0006] S1. Obtain key parameters of grid-connected wind turbine units;

[0007] S2. Based on the key parameters of the grid-type wind turbine, and based on the topology of the grid-type wind turbine and the virtual synchronous control method, establish a continuous traversal state space model of the grid-type wind turbine.

[0008] S3. Construct a continuous crossing model of grid-type wind turbines based on a multi-dimensional two-layer neural network. Use the key parameters of the grid-type wind turbines to perform data fitting training on the continuous crossing model of the grid-type wind turbines. Based on the continuous crossing model of the grid-type wind turbines, compensate and minimize the error of the continuous crossing state space model of the grid-type wind turbines in real time.

[0009] S4. A quadratic programming cost function is established based on the power oscillation of the grid-connected wind turbine. Based on the compensated state space model of the grid-connected wind turbine during continuous crossing, the optimal solution of the quadratic programming problem is solved, and the optimal control command of the grid-connected wind turbine is obtained through rolling optimization, so as to effectively and significantly suppress the power oscillation of the grid-connected wind turbine during continuous crossing.

[0010] Preferably, in step S1, the key parameters of the grid-connected wind turbine include active power P. W Reactive power Q W Virtual damping Virtual inertia Voltage droop coefficient and output angular frequency F.

[0011] Preferably, the specific process of step S2 is as follows:

[0012] Establish models for the active power-output angular frequency deviation of grid-connected wind turbines, and models for the reactive power-converter terminal voltage deviation droop, as follows:

[0013]

[0014] In the formula, and This provides the output active power and active power reference values ​​for grid-connected wind turbines. and This provides the reactive power output and reactive power reference values ​​for grid-connected wind turbines. and This refers to the output angular frequency and reference values ​​for grid-connected wind turbines. , and For grid-connected wind turbines, the virtual damping, virtual inertia, and voltage droop coefficient are used. and These are the converter terminal voltage and terminal voltage reference values;

[0015] A continuous ride-through model for the deviation of active and reactive power output of grid-connected wind turbines is established by Taylor expansion near the operating point; the continuous ride-through model for active power output is as follows:

[0016] ;

[0017] In the formula, , , , and For grid-connected wind turbines, the increments are: output active power, virtual damping, output angular frequency, virtual inertia, and converter terminal voltage. For intermediate parameters;

[0018] The reactive power drooping continuous crossing model is as follows:

[0019]

[0020] In the formula, , and This refers to the increment of reactive power output, voltage droop coefficient, and reactive power reference value for grid-connected wind turbines. For intermediate parameters;

[0021] Therefore, the continuous traversal state-space model of grid-connected wind turbines is represented as:

[0022]

[0023] In the formula, , , , , The state space parameters for continuous traversal of grid-connected wind turbines. , For state variables and input variables, For output quantity, d The first derivative of the state variables is given. The state variables include the output angular frequency, converter terminal voltage, reactive power reference value, output active power and reactive power. The input variables include virtual damping, virtual inertia and voltage droop coefficient. The output variables include active power and reactive power.

[0024] Preferably, in step S3, the specific process of constructing a continuous wind turbine model based on a multi-dimensional two-layer neural network is as follows:

[0025] First, a deep convolutional residual block is established for the continuous traversal model of the grid-type wind turbine:

[0026]

[0027] In the formula, and These are the input and output of the basic depth convolutional block BDCB for the continuous traversal model of grid-type wind turbines, respectively. and These are the weight parameters and bias terms of the basic deep convolutional block of the multi-dimensional two-layer neural network for grid-type wind turbine units. and For the outputs of the first and second stage basic depth convolutional blocks of the continuous traversal model of grid-connected wind turbines, D and These are the input and output of the deep convolutional residual block for the continuous traversal model of the grid-type wind turbine. Here, is a linear rectified function with leakage, and BN is a batch normalization operation. This is a function for splicing features across channels; For linear rectification functions, FCL represents fully connected layer processing;

[0028] Then, establish a multi-dimensional two-layer neural network for the continuous traverse model of the grid-type wind turbine:

[0029]

[0030] In the formula, , , and This refers to the outputs of the first, second, and third fusion layers of the continuous traversal model for grid-connected wind turbines, as well as the final output. This represents the input data for the continuous cross-travel model of grid-connected wind turbines, where B is the batch size, H is the feature index, and T is the time step. and The output of the fusion layer with a kernel size of h is the fth convolutional kernel of the basic deep convolutional block and the deep convolutional residual block of the multidimensional two-layer neural network of the grid-type wind turbine.

[0031] Finally, the weighted average regression layer formula for the continuous ride-through model of grid-connected wind turbines is established:

[0032] ;

[0033] In the formula, for The output result at the t-th time step; For the first The weighting coefficients for each time step are defined as follows: ,in For time decay control parameters, The larger the size, the more emphasis is placed on the recent moment; This is the output after weighted averaging. This is the weight matrix of the fully connected layer. For bias vectors, and To predict the active and reactive power outputs of a continuous cross-traffic model for grid-connected wind turbines, and It refers to the active and reactive power calculated by the continuous traversal state-space model of grid-connected wind turbines. and This is the model compensation amount for the output grid-type wind turbine continuously traversing the state space model.

[0034] Preferably, in step S3, the modified grid-connected wind turbine continuous traversal state space model is as follows:

[0035]

[0036] In the formula, and This refers to the model compensation amount for grid-connected wind turbines continuously traversing the state space model.

[0037] Preferably, in step S4, the quadratic programming cost function includes a first control objective and a second control objective. The first control objective is to minimize the active power output deviation of the grid-connected wind turbine, and the second control objective is to minimize the reactive power output deviation of the grid-connected wind turbine.

[0038] Preferably, the first control target To minimize the active power deviation of grid-connected wind turbines, the specific steps are as follows:

[0039]

[0040] in, and To predict the step size and the number of grid-connected wind turbines for the controller, Here are the active power and active power reference values ​​for the i-th grid-connected wind turbine at step k. The weight coefficient for the first control objective of the controller.

[0041] Preferably, the second control target To minimize the reactive power deviation of grid-connected wind turbines, the specific steps are as follows:

[0042]

[0043] in, For the k-th step, the i-th step is the output reactive power and the reference value of the output reactive power of the grid-type wind turbine. The weighting coefficient for the second control objective of the controller.

[0044] Preferably, the control constraints of the quadratic programming cost function are as follows:

[0045]

[0046] in, , and Let represent the virtual damping, virtual inertia, and voltage droop coefficient of the i-th grid-connected wind turbine. , and This represents the minimum values ​​of the virtual damping, virtual inertia, and voltage droop coefficient of the wind turbine. , and This represents the maximum value of the virtual damping, virtual inertia, and voltage droop coefficient of the wind turbine. The number of grid-connected wind turbine units.

[0047] Preferably, in step S4, the optimal control command includes virtual damping, virtual inertia, and voltage droop coefficient.

[0048] The present invention also discloses a grid-type wind turbine continuous ride control system based on a multi-dimensional two-layer network, including an interconnected memory and a processor. The memory stores a computer program, which executes the steps of the method described above when run by the processor.

[0049] Compared with the prior art, the advantages of the present invention are as follows:

[0050] This invention first establishes a continuous traversal state-space model of a grid-connected wind turbine based on its parameters, topology, and virtual synchronous control method. Then, it uses a multi-dimensional, two-layer neural network for data fitting and training to minimize the error in the continuous traversal state-space model in real time. Finally, it solves for control commands based on the corrected continuous traversal state-space model, improving control accuracy. This invention, through a combined numerical and analog architecture of the continuous traversal state-space model and multi-dimensional, two-layer neural network compensation, improves dynamic response speed while maintaining physical interpretability, effectively reducing active power oscillations and reactive power deviations in grid-connected wind turbines.

[0051] Compared with existing technologies, this invention can simultaneously satisfy the requirements of accurate prediction of grid-connected wind turbine output and rapid calculation of optimal control solutions. By optimizing and adjusting the virtual damping, virtual inertia and voltage droop coefficient of the wind turbine grid parameters, it can effectively suppress the oscillation of active power output and minimize the reactive power deviation of the grid-connected wind turbine during continuous crossing. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of the continuous crossing method for grid-type wind turbines based on a multi-dimensional two-layer network according to the present invention;

[0054] Figure 2 This is an architecture diagram of the continuous crossing model of the networked wind turbine based on a multi-dimensional two-layer neural network in this invention;

[0055] Figure 3 This is a virtual damping simulation diagram of the grid-type wind turbine network parameters under different control methods in this invention;

[0056] Figure 4 This is a simulation diagram of the virtual inertia of the grid-type wind turbine network parameters under different control methods in this invention;

[0057] Figure 5 This is a simulation diagram of the voltage droop coefficient of the grid-connected wind turbine under different control methods in this invention;

[0058] Figure 6 The above are simulation diagrams of active power oscillation of grid-type wind turbines under different control methods in this invention.

[0059] Figure 7 This is a simulation diagram of reactive power deviation of grid-type wind turbines under different control methods in this invention. Detailed Implementation

[0060] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0061] like Figure 1 As shown, the continuous wind turbine crossing method based on a multi-dimensional two-layer network provided in this embodiment of the invention includes the following steps:

[0062] S1. Obtain key parameters of grid-connected wind turbines, including active power P. W Reactive power Q W Virtual damping Virtual inertia Voltage droop coefficient and output angular frequency F;

[0063] S2. Based on the key parameters of the grid-type wind turbine, and based on the topology of the grid-type wind turbine and the virtual synchronous control method, establish a continuous traversal state space model of the grid-type wind turbine.

[0064] S3. Construct a continuous crossing model of grid-type wind turbines based on a multi-dimensional two-layer neural network. Train the continuous crossing model of grid-type wind turbines by fitting data through the parameters of the grid-type wind turbines. Based on the continuous crossing model of grid-type wind turbines, compensate and minimize the error of the continuous crossing state space model of grid-type wind turbines in real time.

[0065] S4. A quadratic programming cost function is established based on the power oscillation of the grid-connected wind turbine. Based on the compensated state space model of the grid-connected wind turbine during continuous crossing, the optimal solution of the quadratic programming problem is solved, and the optimal control command of the grid-connected wind turbine is obtained through rolling optimization, so as to effectively and significantly suppress the power oscillation of the grid-connected wind turbine during continuous crossing.

[0066] In a specific embodiment, in step S2, based on the key parameters of the grid-connected wind turbine obtained in step S1, considering the topology of the grid-connected wind turbine and the virtual synchronous control method, an active power-output angular frequency deviation model and a reactive power-converter terminal voltage deviation droop model of the grid-connected wind turbine are established, respectively:

[0067]

[0068] In the formula, and This provides the output active power and active power reference values ​​for grid-connected wind turbines. and This provides the reactive power output and reactive power reference values ​​for grid-connected wind turbines. and This refers to the output angular frequency and reference values ​​for grid-connected wind turbines. , and For grid-connected wind turbines, the virtual damping, virtual inertia, and voltage droop coefficient are used. and These are the converter terminal voltage and terminal voltage reference values;

[0069] By Taylor expansion near the operating point, continuous active power output model and continuous reactive power drooping model of grid-connected wind turbines are established; the continuous active power output model is as follows:

[0070] ;

[0071] In the formula, , , , and For grid-connected wind turbines, the increments are: output active power, virtual damping, output angular frequency, virtual inertia, and converter terminal voltage. For intermediate parameters;

[0072] The reactive power drooping continuous crossing model is as follows:

[0073]

[0074] In the formula, , and This refers to the increment of reactive power output, voltage droop coefficient, and reactive power reference value for grid-connected wind turbines. For intermediate parameters;

[0075] Therefore, the continuous traversal state-space model of grid-connected wind turbines is represented as:

[0076]

[0077] In the formula, , , , , The state space parameters for continuous traversal of grid-connected wind turbines. , For state variables and input variables, For output quantity, d The first derivative of the state variables is given. The state variables include the output angular frequency, converter terminal voltage, reactive power reference value, output active power and reactive power. The input variables include virtual damping, virtual inertia and voltage droop coefficient. The output variables include active power and reactive power.

[0078] In a specific embodiment, in step S3, based on the key parameters of the grid-type wind turbine obtained in step S1, a continuous crossing model of the grid-type wind turbine based on a multi-dimensional two-layer neural network is established. Data fitting training is performed on the continuous crossing model of the grid-type wind turbine based on the multi-dimensional two-layer neural network. The error of the continuous crossing state-space model of the grid-type wind turbine is minimized in real time based on the continuous crossing model of the grid-type wind turbine. The speed and effectiveness of the proposed method are verified through a test set. Figure 2 The diagram shows the architecture of a network-type wind turbine continuous crossing model based on a multi-dimensional two-layer neural network.

[0079] Specifically, the first step is to establish a deep convolutional residual block for the continuous traversal model of the grid-type wind turbine:

[0080] ;

[0081] In the formula, and These are the input and output of the basic depth convolutional block BDCB for the continuous traversal model of grid-type wind turbines, respectively. and These are the weight parameters and bias terms of the basic deep convolutional block of the multi-dimensional two-layer neural network for grid-type wind turbine units. and For the outputs of the first and second stage basic depth convolutional blocks of the continuous traversal model of grid-connected wind turbines, D and These are the input and output of the deep convolutional residual block for the continuous traversal model of the grid-type wind turbine. Here, is a linear rectified function with leakage, and BN is a batch normalization operation. This is a function that concatenates features across channels. This indicates that the features of channels 1 to H are spliced ​​together; For linear rectification functions, FCL represents fully connected layer processing. The basic deep convolutional block (BDCB) of the continuous crossing model of grid-type wind turbines consists of a batch normalization layer (BN), a deep convolutional layer (DWconv), and a corrected linear layer (Leaky ReLU) connected in sequence. The deep convolutional residual block (DWCRB) consists of a first BDCB, a second BDCB, a fully connected layer (FCL), and an activation layer (ReLU) connected in sequence. After the original input of the deep convolutional residual block is the first BDCB, its output can include two branches. One branch outputs directly, and the other branch is fused with the original input and input into the second BDCB. The output of the second BDCB is fused with the original input and then input into the fully connected layer and the activation layer.

[0082] A multi-dimensional two-layer neural network is then constructed for the continuous wind turbine crossing model. This network includes small-scale branches, medium-scale branches, large-scale branches, and a skip connection. Multi-scale features are extracted from different kernel sizes using convolution and residual blocks to obtain multi-scale information and promote network diversity. Cross-branch connections are utilized to strengthen information fusion between different branches and levels, thereby facilitating long-term dependency modeling. The multi-dimensional two-layer neural network is represented as follows:

[0083] ;

[0084] In the formula, , , and This refers to the outputs of the first, second, and third fusion layers of the continuous traversal model for grid-connected wind turbines, as well as the final output. This represents the input data (including active power P) for the continuous ride-through model of grid-connected wind turbines. W Reactive power Q W Virtual damping Virtual inertia Voltage droop coefficient And the output angular frequency F), where B is the batch size, H is the characteristic exponent, and T is the time step. and This is the output of the fusion layer with a kernel size of h for the basic deep convolutional block and the deep convolutional residual block of the multidimensional two-layer neural network of the grid-type wind turbine. Figure 2The lower part is the multi-dimensional two-layer neural network, which includes three "+" symbols. Each "+" represents a fusion layer. They are located at the output convergence point of all branches in each column (layer). f taking 0 means the 0th fusion layer, which is the part to the left of the first "+". The second "+" represents the second fusion layer, and so on. This represents the output of the 1×1DWCRB residual block in the diagram. This represents the output of the first BDCB in the 1×1DWCRB residual block shown in the figure.

[0085] Finally, the weighted average regression layer formula for the continuous ride-through model of grid-connected wind turbines is established:

[0086] ;

[0087] In the formula, for The output result at the t-th time step; For the first The weighting coefficients for each time step are defined as follows: ,in For time decay control parameters, The larger the size, the more emphasis is placed on the recent moment; This is the output after weighted averaging. This is the weight matrix of the fully connected layer. For bias vectors, and To predict the active and reactive power outputs of a continuous cross-traffic model for grid-connected wind turbines, and It refers to the active and reactive power calculated by the continuous traversal state-space model of grid-connected wind turbines. and This is the model compensation amount for the output grid-type wind turbine continuously traversing the state space model.

[0088] The parameters of the grid-type wind turbine are used as the training set to train the continuous crossing model of the grid-type wind turbine based on a multi-dimensional two-layer neural network. The model compensation amount is obtained by subtracting the predicted value of the continuous crossing model of the grid-type wind turbine from the calculated value of the continuous crossing state space model of the grid-type wind turbine.

[0089] During training, a large number of training samples were obtained through simulation experiments to construct a training set. The loss function was constructed with the minimum reactive power deviation, and the gradient descent algorithm was used to train the continuous crossing model of the grid-type wind turbine based on a multi-dimensional two-layer neural network.

[0090] The continuous travel state-space model of grid-connected wind turbines fits some nonlinear characteristics to linear values, resulting in errors. Therefore, the active and reactive power of the continuous travel model of grid-connected wind turbines based on a multi-dimensional two-layer neural network are predicted to calculate the model compensation, thereby correcting the continuous travel state-space model of grid-connected wind turbines. The corrected continuous travel state-space model of grid-connected wind turbines is as follows:

[0091]

[0092] In the formula, and This refers to the model compensation amount for grid-connected wind turbines continuously traversing the state space model.

[0093] In a specific embodiment, in step S4, based on the key parameters of the grid-type wind turbine obtained in step S1, a quadratic programming cost function is established to minimize the active power deviation and reactive power deviation of the grid-type wind turbine. Based on the modified continuous traversal state space model, the optimal control command of the grid-type wind turbine is obtained by solving the optimal solution of the quadratic programming problem and rolling optimization, which effectively reduces the active power deviation and reactive power deviation of the grid-type wind turbine.

[0094] Specifically, we establish the quadratic programming cost function:

[0095] First control objective To minimize the active power deviation of grid-connected wind turbines, the specific steps are as follows:

[0096]

[0097] in, and To predict the step size and the number of grid-connected wind turbines for the controller, and (k) represents the active power and active power reference value of the i-th grid-type wind turbine in the k-th step. The weight coefficient for the first control objective of the controller.

[0098] Second control objective To minimize the reactive power deviation of grid-connected wind turbines, the specific steps are as follows:

[0099]

[0100] in, The output reactive power and the reference value of the output reactive power of the i-th grid-type wind turbine in step k are given. The weighting coefficient for the second control objective of the controller.

[0101] The control constraints are specifically as follows:

[0102]

[0103] in, , and Let represent the virtual damping, virtual inertia, and voltage droop coefficient of the i-th grid-connected wind turbine. , and This represents the minimum values ​​of the virtual damping, virtual inertia, and voltage droop coefficient of the wind turbine. , and This represents the maximum values ​​of the virtual damping, virtual inertia, and voltage droop coefficient of the wind turbine.

[0104] Based on the modified state-space model, the optimal solution that satisfies the first and second control objectives is solved in a rolling manner, and the output includes the optimal control command including the virtual damping, virtual inertia and voltage droop coefficient of the wind turbine network parameters.

[0105] This invention first establishes a continuous traversal state-space model of a grid-connected wind turbine based on its parameters, topology, and virtual synchronous control method. Then, it corrects the continuous traversal state-space model by dynamically learning errors and generating compensation quantities through a multi-dimensional, two-layer neural network. Finally, it solves for control commands based on the corrected continuous traversal state-space model. This invention, through a combined numerical and analog architecture of the continuous traversal state-space model and multi-dimensional, two-layer neural network compensation, improves dynamic response speed while maintaining physical interpretability, effectively reducing active power oscillations and reactive power deviations of the grid-connected wind turbine during continuous traversal.

[0106] Compared with existing technologies, this invention can simultaneously meet the requirements of accurate prediction of grid-connected wind turbine output and rapid calculation of optimal control solutions. By optimizing and adjusting parameters such as virtual damping, virtual inertia, and voltage droop coefficient of the wind turbine grid, it can effectively suppress active power oscillations and minimize reactive power deviations of the grid-connected wind turbine during continuous traversal.

[0107] Figures 3-5 The figures show simulation diagrams of virtual damping, virtual inertia, and voltage droop coefficient of grid-connected wind turbines under different control methods in this invention. The virtual damping, virtual inertia, and voltage droop coefficient of the grid-connected wind turbines are optimized and adjusted to achieve rapid suppression of active and reactive power deviations in grid-connected wind turbines.

[0108] Figure 6The figures show the active power deviation of the grid-connected wind turbine under different control methods in this invention. Compared with the existing droop control method, the control method proposed in this invention effectively suppresses the active power oscillation of the grid-connected wind turbine by optimizing and adjusting the virtual damping and virtual inertia of the wind turbine grid parameters.

[0109] Figure 7 The figures show the reactive power deviation of the grid-connected wind turbine under different control methods in this invention. Compared with existing control methods, the control method proposed in this invention effectively reduces the reactive power deviation of the grid-connected wind turbine by optimizing and adjusting the voltage droop coefficient of the grid-connected wind turbine.

[0110] This invention also discloses a continuous ride control system for a grid-type wind turbine based on a multi-dimensional two-layer network, including an interconnected memory and a processor. The memory stores a computer program, which, when run by the processor, executes the steps of the method described above. The control system of this invention corresponds to the control method described above and also possesses the advantages described therein.

[0111] The present invention can implement all or part of the processes in the methods of the above embodiments, or it can be implemented by hardware related to computer program instructions. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above method embodiments. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium includes: any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. The memory is used to store computer programs and / or modules. The processor implements various functions by running or executing the computer programs and / or modules stored in the memory, and by calling data stored in the memory. The memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0112] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A continuous wind turbine crossing method based on a multi-dimensional two-layer network, characterized in that, Including the following steps: S1. Obtain key parameters of grid-connected wind turbine units; S2. Based on the key parameters of the grid-type wind turbine, and based on the topology of the grid-type wind turbine and the virtual synchronous control method, establish a continuous traversal state space model of the grid-type wind turbine. S3. Construct a continuous crossing model of grid-type wind turbines based on a multi-dimensional two-layer neural network. Use the key parameters of the grid-type wind turbines to perform data fitting training on the continuous crossing model of the grid-type wind turbines. Based on the continuous crossing model of the grid-type wind turbines, compensate and minimize the error of the continuous crossing state space model of the grid-type wind turbines in real time. S4. Establish a quadratic programming cost function based on the output power oscillation of the grid-type wind turbine. Based on the modified grid-type wind turbine continuous traversal state space model, solve the optimal solution of the quadratic programming problem and obtain the optimal control command of the grid-type wind turbine through rolling optimization.

2. The continuous wind turbine crossing method based on a multi-dimensional two-layer network according to claim 1, characterized in that, In step S1, the key parameters of the grid-connected wind turbine include active power P. W Reactive power Q W Virtual damping Virtual inertia Voltage droop coefficient and output angular frequency F.

3. The continuous crossing method for grid-type wind turbines based on a multi-dimensional two-layer network according to claim 1, characterized in that, The specific process of step S2 is as follows: Establish models for the active power-output angular frequency deviation of grid-connected wind turbines, and models for the reactive power-converter terminal voltage deviation droop, as follows: ； In the formula, and This provides the output active power and active power reference values ​​for grid-connected wind turbines. and This provides the reactive power output and reactive power reference values ​​for grid-connected wind turbines. and This refers to the output angular frequency and reference values ​​for grid-connected wind turbines. , and For grid-connected wind turbines, the virtual damping, virtual inertia, and voltage droop coefficient are used. and These are the converter terminal voltage and terminal voltage reference values; By Taylor expansion near the operating point, continuous active power output model and continuous reactive power drooping model of grid-connected wind turbines are established; the continuous active power output model is as follows: ； In the formula, , , , and For grid-connected wind turbines, the increments are: output active power, virtual damping, output angular frequency, virtual inertia, and converter terminal voltage. For intermediate parameters; The reactive power drooping continuous crossing model is as follows: ； In the formula, , and This refers to the increment of reactive power output, voltage droop coefficient, and reactive power reference value for grid-connected wind turbines. For intermediate parameters; Therefore, the continuous traversal state-space model of grid-connected wind turbines is represented as: ； In the formula, , , , , The state space parameters for continuous traversal of grid-connected wind turbines. , For state variables and input variables, For output quantity, d The first derivative of the state variables is given. The state variables include the output angular frequency, converter terminal voltage, reactive power reference value, output active power and reactive power. The input variables include virtual damping, virtual inertia and voltage droop coefficient. The output variables include active power and reactive power.

4. The continuous wind turbine crossing method based on a multi-dimensional two-layer network according to any one of claims 1-3, characterized in that, In step S3, the specific process of constructing a continuous wind turbine model based on a multi-dimensional two-layer neural network is as follows: First, a deep convolutional residual block is established for the continuous traversal model of the grid-type wind turbine: ； In the formula, and These are the input and output of the basic depth convolutional block BDCB for the continuous traversal model of grid-type wind turbines, respectively. and These are the weight parameters and bias terms of the basic deep convolutional block of the multi-dimensional two-layer neural network for grid-type wind turbine units. and For the outputs of the first and second stage basic depth convolutional blocks of the continuous traversal model of grid-connected wind turbines, D and These are the input and output of the deep convolutional residual block for the continuous traversal model of the grid-type wind turbine. Here, is a linear rectified function with leakage, and BN is a batch normalization operation. This is a function for splicing features across channels; For linear rectification functions, FCL represents fully connected layer processing; Then, establish a multi-dimensional two-layer neural network for the continuous traverse model of the grid-type wind turbine: ； In the formula, , , and This refers to the outputs of the first, second, and third fusion layers of the continuous traversal model for grid-connected wind turbines, as well as the final output. This represents the input data for the continuous cross-travel model of grid-connected wind turbines, where B is the batch size, H is the feature index, and T is the time step. and The output of the fusion layer with a kernel size of h is the fth convolutional kernel of the basic deep convolutional block and the deep convolutional residual block of the multidimensional two-layer neural network of the grid-type wind turbine. Finally, the weighted average regression layer formula for the continuous ride-through model of grid-connected wind turbines is established: ； In the formula, for The output result at the t-th time step; For the first The weighting coefficients for each time step are defined as follows: ,in This is the time decay control parameter; This is the output after weighted averaging. This is the weight matrix of the fully connected layer. For bias vectors, and To predict the active and reactive power outputs of a continuous cross-traffic model for grid-connected wind turbines, and It refers to the active and reactive power calculated by the continuous traversal state-space model of grid-connected wind turbines. and This is the model compensation amount for the output grid-type wind turbine continuously traversing the state space model.

5. The continuous crossing method for grid-type wind turbines based on a multi-dimensional two-layer network according to claim 3, characterized in that, In step S3, the corrected grid-connected wind turbine continuous traversal state space model is as follows: ； In the formula, and This refers to the model compensation amount for grid-connected wind turbines continuously traversing the state space model.

6. The continuous wind turbine crossing method based on a multi-dimensional two-layer network according to any one of claims 1-3, characterized in that, In step S4, the quadratic programming cost function includes a first control objective and a second control objective. The first control objective is to minimize the active power output deviation of the grid-connected wind turbine, and the second control objective is to minimize the reactive power output deviation of the grid-connected wind turbine.

7. The continuous crossing method for grid-type wind turbines based on a multi-dimensional two-layer network according to claim 6, characterized in that, First control objective To minimize the active power deviation of grid-connected wind turbines, the specific steps are as follows: ； in, and To predict the step size and the number of grid-connected wind turbines for the controller, and The output active power and active power reference value of the i-th grid-connected wind turbine in step k are given. The weighting coefficient for the first control objective of the controller; Second control objective To minimize the reactive power deviation of grid-connected wind turbines, the specific steps are as follows: ； in, and The output reactive power and the reference value of the output reactive power of the i-th grid-type wind turbine in step k are given. The weighting coefficient for the second control objective of the controller.

8. The continuous crossing method for grid-type wind turbines based on a multi-dimensional two-layer network according to claim 6, characterized in that, The control constraints of the quadratic programming cost function are as follows: ； in, , and Let represent the virtual damping, virtual inertia, and voltage droop coefficient of the i-th grid-connected wind turbine. , and This represents the minimum values ​​of the virtual damping, virtual inertia, and voltage droop coefficient of the wind turbine. , and This represents the maximum value of the virtual damping, virtual inertia, and voltage droop coefficient of the wind turbine. The number of grid-connected wind turbine units.

9. The continuous wind turbine crossing method based on a multi-dimensional two-layer network according to claim 6, characterized in that, In step S4, the optimal control command includes virtual damping, virtual inertia, and voltage droop coefficient.

10. A continuous wind turbine ride-through system based on a multi-dimensional, two-layer network, comprising an interconnected memory and a processor, wherein the memory stores a computer program, characterized in that, The computer program, when run by a processor, performs the steps of the method as described in any one of claims 1-9.

Images