Network construction type wind turbine generator control method and system based on multidirectional deep learning

Abstract

The invention discloses a multi-directional deep learning-based network construction type wind turbine generator control method and system. The method comprises the following steps: S1, obtaining parameters of a network construction type wind turbine generator; s2, establishing a linearized state space model of the network construction type wind turbine generator according to the parameters of the network construction type wind turbine generator; s3, constructing a multi-directional gating circulation unit neural network, training the multi-directional gating circulation unit neural network, and outputting a model compensation amount to correct the linearized state space model of the network construction type wind turbine generator; and S4, establishing a quadratic programming cost function by minimizing the output frequency deviation and the reactive power deviation of the network construction type wind turbine generator, and obtaining an optimal control instruction of the network construction type wind turbine generator by solving an optimal solution of a quadratic programming problem and performing rolling optimization based on the corrected linearized state space model of the network construction type wind turbine generator. The method has the advantages of improving control precision and dynamic response speed and the like.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of wind power generation, and specifically to a grid-type wind turbine control method and system based on multi-directional deep learning. Background Art

[0002] Traditional grid-following wind turbines are unable to achieve rapid frequency and voltage support to improve the operational stability of new power systems. Grid-connected wind turbines with active support capabilities have attracted attention. However, due to their use of virtual synchronous control, grid-connected wind turbines exhibit slow power response, fluctuating output frequency, and poor power tracking performance under normal operating conditions. Grid-connected wind turbines have numerous control parameters and exhibit strong coupling and nonlinearity. Existing predictive control methods are unable to rapidly calculate optimal control instructions while maintaining high prediction accuracy. Traditional grid-connected wind turbine control methods are unable to address these challenges. Therefore, research is urgently needed to ensure the prediction accuracy of grid-connected wind turbines, rapidly obtain optimal control instructions, and effectively suppress output frequency fluctuations and minimize reactive power deviations for grid-connected wind turbines. Summary of the Invention

[0003] In response to the technical problems existing in the prior art, the present invention provides a control method and system for a grid-type wind turbine generator set based on multi-directional deep learning with high control accuracy and fast dynamic response speed.

[0004] In order to solve the above technical problems, the technical solution proposed by the present invention is: A method for controlling a grid-type wind turbine generator system based on multi-directional deep learning comprises the following steps: S1. Obtaining parameters of grid-type wind turbines; S2. According to the parameters of the grid-type wind turbine, based on the topological structure of the grid-type wind turbine and the virtual synchronous control method, a linearized state space model of the grid-type wind turbine is established; S3. Construct a multi-directional gated recurrent unit neural network, train the multi-directional gated recurrent unit neural network using the parameters of the meshed wind turbine generator set, and output a model compensation to correct the linearized state space model of the meshed wind turbine generator set; S4. A quadratic programming cost function is established to minimize the output frequency deviation and reactive power deviation of the grid-type wind turbines. Based on the modified linearized state space model of the grid-type wind turbines, the optimal solution of the quadratic programming problem is solved, and the optimal control instructions of the grid-type wind turbines are obtained through rolling optimization, so as to effectively reduce the output frequency fluctuation and reactive power deviation of the grid-type wind turbines.

[0005] Preferably, in step S1, the grid-type wind turbine parameters include active power P W , reactive power Q W, virtual damping D , virtual inertia J , voltage integral coefficient K and output frequency F .

[0006] Preferably, the specific process of step S2 is: The grid-connected wind turbine active power-output frequency deviation model and the converter terminal voltage-reactive power deviation model are established, which are:

[0007] Where, and The output active power and active power reference value of the grid-connected wind turbine. and Output reactive power and reactive power reference value for grid-connected wind turbines. and is the output frequency and output frequency reference value of the grid-type wind turbine generator set, 、 J and The virtual damping, virtual inertia and voltage integral coefficient of the grid-type wind turbine; is the voltage at the converter terminal; By Taylor expansion near the operating point, the output frequency deviation linearization model and reactive power deviation linearization model of the grid-type wind turbine are established; the output frequency deviation linearization model is: ; Where, 、 、 and are the output frequency increment, virtual damping increment, virtual inertia increment and converter terminal voltage increment of the grid-type wind turbine. is the intermediate parameter; The reactive power deviation linearization model is:

[0008] Where, 、 is the reactive power output increment and voltage integral coefficient increment of the grid-connected wind turbine, is the intermediate parameter; Therefore, the linearized state space model of the grid-type wind turbine is expressed as:

[0009] Where, 、 、 、 、 is the linearized state space parameter of the grid-type wind turbine, 、 are state quantities and input quantities, is the output, d is the first-order derivative of the state quantity.

[0010] Preferably, in step S3, the specific process of constructing the multi-directional gated recurrent unit neural network is: First, a one-way neural network model of a networked wind turbine is established:

[0011] Where, 、 、 and The update gate, reset gate, activation state and output gate of the unidirectional neural network of the grid-type wind turbine are constructed in k The state of the moment, 、 、 and The update gate, reset gate, activation state and output gate of the unidirectional neural network of the grid-type wind turbine are constructed in k-1 The state of the moment; and Represents the activation functions of the update gate and reset gate of the unidirectional neural network of the grid-type wind turbine; 、 and Represents the weight coefficients of the update gate, reset gate, and activation state, It measures the size of the door opening. is the input of the unidirectional neural network of the grid-type wind turbine generator system. represents vector product; Based on the unidirectional neural network model of the networked wind turbine, a multidirectional neural network model of the networked wind turbine is established by considering the forward and reverse propagation modes of the time series:

[0012] Where, and Represents the activation function of the update gate and reset gate of the multi-directional neural network, and is the activation state of the forward propagation neural network and the back propagation neural network of the networked wind turbine generator set, It is the activation state of the bidirectional neural network model of the grid-type wind turbine.

[0013] Preferably, in step S3, the modified linearized state space model of the grid-type wind turbine generator system is:

[0014] Where, and It is the model compensation output by the multi-directional neural network model.

[0015] Preferably, in step S4, the quadratic programming cost function includes a first control objective and a second control objective, which are: The first control target To minimize the output frequency deviation of the grid-type wind turbine, specifically:

[0016] in, and Predict the step size and number of wind turbines in the grid for the controller, and (k) is the output frequency and output frequency reference value of the k-th step grid-type wind turbine generator set, is the first control target weight coefficient of the controller.

[0017] Preferably, the second control target To minimize the reactive power deviation of grid-connected wind turbine output, the following are the steps:

[0018] in, and (k) is the output reactive power and output reactive power reference value of the k-th step grid-connected wind turbine group, is the second control target weight coefficient of the controller.

[0019] Preferably, the control quantity constraint of the quadratic programming cost function is specifically:

[0020] in, 、 and For the i Virtual damping, virtual inertia and voltage integral coefficient of platform-type wind turbines; 、 and is the minimum value of the wind turbine's virtual damping, virtual inertia, and voltage integral coefficient; 、 and is the maximum value of the virtual damping, virtual inertia and voltage integral coefficient of the wind turbine; The number of grid-type wind turbines.

[0021] Preferably, in step S4, the optimal control instruction includes virtual damping, virtual inertia and voltage integral coefficient.

[0022] The present invention also discloses a meshed wind turbine control system based on multi-directional deep learning, comprising a memory and a processor connected to each other, wherein a computer program is stored on the memory, and when the computer program is run by the processor, the steps of the method described above are executed.

[0023] Compared with the prior art, the advantages of the present invention are: The present invention first establishes a linearized state space model of a grid-type wind turbine according to the parameters and demand parameters of the grid-type wind turbine, based on the topological structure of the grid-type wind turbine and the virtual synchronous control method; then, the linearized state space model is corrected by dynamically learning errors and generating compensation through a multi-directional GRU; and then, the control instructions are solved based on the corrected linearized state space model to improve the control accuracy; the present invention improves the dynamic response speed while maintaining physical interpretability through a digital-analog combination architecture of a linearized state space model and multi-directional deep learning compensation, thereby effectively reducing the output frequency fluctuation and reactive power deviation of the grid-type wind turbine.

[0024] Compared with the existing technology, the present invention can simultaneously meet the requirements of accurate output prediction of grid-type wind turbines and rapid calculation of optimal control solutions. By optimizing and adjusting parameters such as virtual damping, virtual inertia, and voltage integral coefficient, the output frequency fluctuations of grid-type wind turbines can be effectively suppressed and reactive power deviations can be minimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of an embodiment of a grid-type wind turbine control method based on multi-directional deep learning of the present invention.

[0026] Figure 2 This is a virtual damping simulation diagram of a grid-type wind turbine under different control methods in the present invention.

[0027] Figure 3 This is a simulation diagram of the virtual inertia of a grid-type wind turbine under different control methods in the present invention.

[0028] Figure 4 This is a simulation diagram of the voltage integral coefficient of the grid-type wind turbine generator set under different control methods in the present invention.

[0029] Figure 5 This is a simulation diagram of the output frequency of a grid-type wind turbine generator system under different control methods in the present invention.

[0030] Figure 6 This is a simulation diagram of reactive power deviation of a grid-connected wind turbine under different control methods in the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0032] like Figure 1 As shown, the embodiment of the present invention provides a grid-type wind turbine control method based on multi-directional deep learning, comprising the steps of: S1. Obtain the parameters of the grid-type wind turbine generator set; the parameters of the grid-type wind turbine generator set include active power P W , reactive power Q W , virtual damping D , virtual inertia J , voltage integral coefficient K and output frequency F wait; S2. According to the parameters of the grid-type wind turbine, based on the topological structure of the grid-type wind turbine and the virtual synchronous control method, a linearized state space model of the grid-type wind turbine is established; S3. Multi-directional deep learning prediction: Construct a multi-directional gated recurrent unit (GRU) neural network, train it using the parameters of the meshed wind turbine, and output compensation to correct the linearized state space model of the meshed wind turbine. S4. A quadratic programming cost function is established to minimize the output frequency deviation and reactive power deviation of the grid-type wind turbines. Based on the modified linearized state space model of the grid-type wind turbines, the optimal control instructions of the grid-type wind turbines are obtained by solving the optimal solution of the quadratic programming problem and performing rolling optimization, so as to effectively reduce the output frequency fluctuation and reactive power deviation of the grid-type wind turbines.

[0033] In a specific embodiment, in step S2, based on the grid-type wind turbine parameters obtained in step S1, taking into account the topology of the grid-type wind turbine and the virtual synchronization control method, a grid-type wind turbine active power-output frequency deviation model and a converter terminal voltage-reactive power deviation model are established, which are respectively:

[0034] Where, and The output active power and active power reference value of the grid-connected wind turbine. and Output reactive power and reactive power reference value for grid-connected wind turbines. and is the output frequency and output frequency reference value of the grid-type wind turbine generator set, 、 J and The virtual damping, virtual inertia and voltage integral coefficient of the grid-type wind turbine; is the voltage at the converter terminal.

[0035] By Taylor expansion near the operating point, the output frequency deviation linearization model and reactive power deviation linearization model of the grid-type wind turbine are established. The output frequency deviation linearization model is: ; Where, 、 、 and are the output frequency increment, virtual damping increment, virtual inertia increment and converter terminal voltage increment of the grid-type wind turbine. is the intermediate parameter; The reactive power deviation linearization model is:

[0036] Where, 、 is the reactive power output increment and voltage integral coefficient increment of the grid-connected wind turbine, is the intermediate parameter; Therefore, the linearized state space model of the grid-type wind turbine can be expressed as:

[0037] Where, 、 、 、 、 is the linearized state space parameter of the grid-type wind turbine; 、 are state quantities and input quantities, is the output, d is the first-order derivative of the state quantity.

[0038] In a specific embodiment, in step S3, a forward and backward multi-directional gated recurrent unit neural network is established based on the grid-type wind turbine parameters obtained in step S1. Based on the grid-type wind turbine output training set, a grid-type wind turbine output prediction model based on a multi-directional deep learning method is implemented. The speed and effectiveness of the proposed method are verified by a test set. Specifically, First, a one-way neural network model of a networked wind turbine is established:

[0039] Where, 、 、 and The update gate, reset gate, activation state and output gate of the unidirectional neural network of the grid-type wind turbine are constructed ink The state of the moment, 、 、 and The update gate, reset gate, activation state and output gate of the unidirectional neural network of the grid-type wind turbine are constructed in k-1 The state of the moment; and Represents the activation functions of the update gate and reset gate of the unidirectional neural network of the grid-type wind turbine; 、 and Represents the weight coefficients of the update gate, reset gate, and activation state, It measures the size of the door opening. is the input of the unidirectional neural network of the grid-type wind turbine generator system. represents vector product; Based on the unidirectional neural network model of the networked wind turbine, a multidirectional neural network model of the networked wind turbine is established by considering the forward and reverse propagation modes of the time series:

[0040] Where, and Represents the activation function of the update gate and reset gate of the multi-directional neural network, and is the activation state of the forward propagation neural network and the back propagation neural network of the networked wind turbine generator set, It is the activation state of the bidirectional neural network model of the grid-type wind turbine.

[0041] The parameters of the grid-type wind turbines are used as a training set to train the multi-directional neural network model, and the model compensation is finally output; The linearized state space model is modified to the following form using the model compensation output by the multi-directional neural network model:

[0042] Where, and It is the model compensation output by the multi-directional neural network model.

[0043] In a specific embodiment, in step 4), based on the parameters of the grid-type wind turbine generator set obtained in step 1), a quadratic programming cost function is established to minimize the output frequency deviation and reactive power deviation of the grid-type wind turbine generator set. Based on the corrected linearized state space model, the optimal solution of the quadratic programming problem is solved and rolling optimization is performed to obtain the optimal control instructions for the grid-type wind turbine generator set, thereby effectively reducing the output frequency fluctuation and reactive power deviation of the grid-type wind turbine generator set.

[0044] Specifically, establish the quadratic programming cost function: The first control target To minimize the output frequency deviation of the grid-type wind turbine, specifically:

[0045] in, and Predict the step size and number of wind turbines in the grid for the controller, and (k) is the output frequency and output frequency reference value of the k-th step grid-type wind turbine generator set, is the first control target weight coefficient of the controller; Second control objective To minimize the reactive power deviation of grid-connected wind turbine output, the following are the steps:

[0046] in, and (k) is the output reactive power and output reactive power reference value of the k-th step grid-connected wind turbine group, is the second control target weight coefficient of the controller.

[0047] The specific control quantity constraints are:

[0048] in, 、 and For the i Virtual damping, virtual inertia and voltage integral coefficient of platform-type wind turbines; 、 and is the minimum value of the wind turbine's virtual damping, virtual inertia, and voltage integral coefficient; 、 and is the maximum value of the wind turbine's virtual damping, virtual inertia and voltage integral coefficient; The number of grid-type wind turbines.

[0049] Based on the modified state space model, the optimal solution that meets the first control objective and the second control objective is obtained by rolling solution, and the optimal control instructions of virtual damping, virtual inertia and voltage integral coefficient are output.

[0050] The present invention first establishes a linearized state space model of a grid-type wind turbine according to the parameters of the grid-type wind turbine, based on the topological structure of the grid-type wind turbine and the virtual synchronous control method; then dynamically learns the error through a multi-directional GRU and generates a compensation amount to correct the linearized state space model; then solves the control instructions based on the corrected linear state space model; the present invention improves the dynamic response speed while maintaining physical interpretability through a digital-analog combination architecture of a linearized state space model and multi-directional deep learning compensation, and effectively reduces the output frequency fluctuation and reactive power deviation of the grid-type wind turbine.

[0051] Compared with the existing technology, the present invention can simultaneously meet the requirements of accurate output prediction of grid-type wind turbines and rapid calculation of optimal control solutions. By optimizing and adjusting parameters such as virtual damping, virtual inertia, and voltage integral coefficient, the output frequency fluctuations of grid-type wind turbines can be effectively suppressed and reactive power deviations can be minimized.

[0052] Figure 2-Figure 4 They are respectively the simulation diagrams of virtual damping, virtual inertia and voltage integral coefficient of the grid-type wind turbine under different control methods in the present invention. The virtual damping, virtual inertia and voltage integral coefficient are optimized and adjusted to achieve rapid suppression of output frequency and reactive power deviation of the grid-type wind turbine.

[0053] Figure 5 This is a simulation diagram of the output frequency of the grid-type wind turbine under different control methods in the present invention. Compared with the existing control method, the control method proposed in the present invention effectively suppresses the output frequency fluctuation of the grid-type wind turbine by optimizing and adjusting the virtual damping and virtual inertia parameters.

[0054] Figure 6 This is a simulation diagram of the reactive power deviation of the grid-type wind turbine under different control methods in the present invention. Compared with the existing control method, the control method proposed in the present invention effectively reduces the output reactive power deviation of the grid-type wind turbine by optimizing and adjusting the voltage integral coefficient.

[0055] An embodiment of the present invention further discloses a grid-based wind turbine control system based on multi-directional deep learning, comprising an interconnected memory and a processor, wherein the memory stores a computer program that, when executed by the processor, executes the steps of the above-described method. The control system of the present invention corresponds to the above-described control method and similarly possesses the advantages described above.

[0056] The present invention can implement all or part of the process steps in the above-described method embodiments through hardware associated with computer program instructions. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above-described method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. Computer-readable storage media include any entity or device capable of carrying computer program code, recording media, USB flash drives, removable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. 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 accessing data stored in the memory. The memory may include a high-speed random access memory and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0057] 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 based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A grid-type wind turbine control method based on multi-directional deep learning, characterized in that: Including steps: S1. Obtaining parameters of grid-type wind turbines; S2. According to the parameters of the grid-type wind turbine, based on the topological structure of the grid-type wind turbine and the virtual synchronous control method, a linearized state space model of the grid-type wind turbine is established; S3. Construct a multi-directional gated recurrent unit neural network, train the multi-directional gated recurrent unit neural network using the parameters of the meshed wind turbine generator set, and output a model compensation to correct the linearized state space model of the meshed wind turbine generator set; S4. A quadratic programming cost function is established to minimize the output frequency deviation and reactive power deviation of the grid-type wind turbines. Based on the modified linearized state space model of the grid-type wind turbines, the optimal solution of the quadratic programming problem is solved, and the optimal control instructions of the grid-type wind turbines are obtained through rolling optimization, so as to effectively reduce the output frequency fluctuation and reactive power deviation of the grid-type wind turbines.

2. The method for controlling a grid-type wind turbine generator system based on multi-directional deep learning according to claim 1, characterized in that: In step S1, the grid-type wind turbine parameters include active power P W , reactive power Q W , virtual damping D , virtual inertia J , voltage integral coefficient K and output frequency F .

3. The method for controlling a grid-type wind turbine generator system based on multi-directional deep learning according to claim 1, characterized in that: The specific process of step S2 is: The grid-connected wind turbine active power-output frequency deviation model and the converter terminal voltage-reactive power deviation model are established, which are: Where, and The output active power and active power reference value of the grid-connected wind turbine. and Output reactive power and reactive power reference value for grid-connected wind turbines. and is the output frequency and output frequency reference value of the grid-type wind turbine generator set, 、 J and The virtual damping, virtual inertia and voltage integral coefficient of the grid-type wind turbine; is the voltage at the converter terminal; By Taylor expansion near the operating point, the output frequency deviation linearization model and reactive power deviation linearization model of the grid-type wind turbine are established; the output frequency deviation linearization model is: ； Where, 、 、 and are the output frequency increment, virtual damping increment, virtual inertia increment and converter terminal voltage increment of the grid-type wind turbine. is the intermediate parameter; The reactive power deviation linearization model is: Where, 、 is the reactive power output increment and voltage integral coefficient increment of the grid-connected wind turbine, is the intermediate parameter; Therefore, the linearized state space model of the grid-type wind turbine is expressed as: Where, 、 、 、 、 is the linearized state space parameter of the grid-type wind turbine, 、 are state quantities and input quantities, is the output, d is the first-order derivative of the state quantity.

4. The method for controlling a grid-type wind turbine generator system based on multi-directional deep learning according to claim 1, 2 or 3, wherein: In step S3, the specific process of constructing a multi-directional gated recurrent unit neural network is as follows: First, a one-way neural network model of a networked wind turbine is established: Where, 、 、 and The update gate, reset gate, activation state and output gate of the unidirectional neural network of the grid-type wind turbine are constructed in k The state of the moment, 、 、 and The update gate, reset gate, activation state and output gate of the unidirectional neural network of the grid-type wind turbine are constructed in k-1 The state of the moment; and Represents the activation functions of the update gate and reset gate of the unidirectional neural network of the grid-type wind turbine; 、 and Represents the weight coefficients of the update gate, reset gate, and activation state, It measures the size of the door opening. is the input of the unidirectional neural network of the grid-type wind turbine generator system. represents vector product; Based on the unidirectional neural network model of the networked wind turbine, a multidirectional neural network model of the networked wind turbine is established by considering the forward and reverse propagation modes of the time series: Where, and Represents the activation function of the update gate and reset gate of the multi-directional neural network, and is the activation state of the forward propagation neural network and the back propagation neural network of the networked wind turbine generator set, It is the activation state of the bidirectional neural network model of the grid-type wind turbine.

5. The method for controlling a networked wind turbine generator system based on multi-directional deep learning according to claim 3, wherein: In step S3, the modified linearized state space model of the grid-type wind turbine is: Where, and It is the model compensation output by the multi-directional neural network model.

6. The method for controlling a grid-type wind turbine generator system based on multi-directional deep learning according to claim 1, 2 or 3, characterized in that: In step S4, the quadratic programming cost function includes a first control objective and a second control objective, which are: The first control target To minimize the output frequency deviation of the grid-type wind turbine, specifically: in, and Predict the step size and number of wind turbines in the grid for the controller, and (k) is the output frequency and output frequency reference value of the k-th step grid-type wind turbine generator set, is the first control target weight coefficient of the controller.

7. The method for controlling a grid-type wind turbine generator system based on multi-directional deep learning according to claim 6, characterized in that: Second control objective To minimize the reactive power deviation of grid-connected wind turbine output, the following are the steps: in, and (k) is the output reactive power and output reactive power reference value of the k-th step grid-connected wind turbine group, is the second control target weight coefficient of the controller.

8. The multi-directional deep learning-based grid-type wind turbine control method according to claim 6, characterized in that: The control quantity constraints of the quadratic programming cost function are as follows: in, 、 and For the i Virtual damping, virtual inertia and voltage integral coefficient of platform-type wind turbines; 、 and is the minimum value of the wind turbine's virtual damping, virtual inertia, and voltage integral coefficient; 、 and is the maximum value of the wind turbine's virtual damping, virtual inertia and voltage integral coefficient; The number of grid-type wind turbines.

9. The method for controlling a grid-type wind turbine generator system based on multi-directional deep learning according to claim 6, wherein: In step S4, the optimal control instruction includes virtual damping, virtual inertia and voltage integral coefficient.

10. A grid-type wind turbine control system based on multi-directional deep learning, comprising a memory and a processor connected to each other, wherein a computer program is stored in the memory, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 9.

Images