Wind power generation system control method and system based on improved recurrent neural network

Abstract

The invention discloses a wind power generation system control method and system based on an improved recurrent neural network (RNN), and belongs to the field of wind power generation system control. The method comprises the following steps: training an improved RNN model by adopting a training data set, and carrying out forward propagation on an input vector to obtain gain data; quantizing a loss value between the gain parameter and an ideal gain parameter, constructing a loss function according to the loss value, obtaining a weight gradient of each layer by using a gradient back propagation algorithm, updating a network weight according to the weight gradient and in combination with a momentum gradient descent algorithm, applying amplitude constraint to a recursive weight, and performing iterative training to obtain an ideal gain parameter. A trained RNN network model is obtained; obtaining a gain parameter of the wind power generation system according to the trained RNN network model, and controlling the operation state of the wind power generation system according to the gain parameter; according to the method, PI controller self-adaption is achieved by introducing the recurrent neural network, the system response time, static errors and overshoot can be remarkably reduced, and therefore the system performance is improved.

Description

Technical Field

[0001] The present invention relates to the field of wind power generation system control, and specifically to a control method and system for a wind power generation system based on an improved recurrent neural network. Background Art

[0002] With the growth of the population and the popularization of household automation devices, the demand for electric energy has increased sharply, while traditional fossil fuel energy faces challenges such as environmental pollution and rising costs. Therefore, renewable energy, especially wind energy, is regarded as one of the ideal alternative solutions for sustainable development due to its environmental friendliness and immunity to fossil fuel price fluctuations. Wind energy has become one of the fastest-growing green energy sources in the power system and the fastest-growing renewable energy source globally. In a wind power generation system, a PI controller is a common type of controller that uses proportional-integral control to regulate the operation of the power generation system. The PI controller combines the advantages of proportional control and integral control and can achieve a good balance between the steady-state and dynamic performance of the system. However, since the wind power generation system is greatly affected by environmental conditions such as wind speed changes and load fluctuations, these factors may cause system parameter changes and external disturbances, and the traditional PI controller performs unstably when dealing with these changes.

[0003] Firstly, in terms of system stability and robustness, the wind power generation system is often disturbed by external environmental factors such as wind speed changes and load fluctuations, and the traditional PI controller is easily affected and produces unstable control results. Secondly, in terms of response speed and overshoot problems, the traditional PI controller has a slow response speed and is prone to overshoot during the adjustment process, which affects the adjustment performance and stability of the system and makes it difficult for the system to quickly adapt to changing working conditions. In addition, insufficient adaptability to complex environments is also a major problem. For example, in the case of grid voltage dips caused by asymmetric phase faults, the traditional PI controller lacks sufficient adaptability and robustness and is difficult to maintain the stability and reliability of the system. Finally, parameter adjustment is also a challenge. The parameter adjustment of the traditional PI controller usually requires experienced engineers to manually adjust, and the adjustment process is cumbersome and time-consuming, lacking self-adaptability and intelligence. Therefore, the current research focus is on finding a new control method that can improve the stability of the control system, increase the response speed, reduce overshoot, and has strong adaptability and self-adaptability to meet the stable operation requirements of the wind power generation system in complex environments. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention provides a control method for a wind power generation system based on an improved recurrent neural network, which is used for controlling a doubly-fed induction generator in the wind power generation system. By introducing a recurrent neural network, the PI controller is made adaptive to cope with situations such as wind speed changes, generator parameter changes, and grid asymmetrical faults. This method can significantly reduce the system response time, static error, and overshoot, thereby improving the system performance.

[0005] The present invention is realized through the following technical solutions: In a first aspect, the present application provides a control method for a wind power generation system based on an improved recurrent neural network, including the following steps: Step 1: Construct a training data set according to the operation data of the wind power generation system. The operation data includes historical control signals, real-time state variables, and gain parameters; Step 2: Use the training data set to train the improved RNN network model. The training method is as follows: Construct an input vector according to the training data and propagate it forward. Generate the current state through the linear combination of the input vector and the historical state and an activation function, and weight the current state to obtain gain data; Determine the deviation between the gain parameter output by the RNN network and the ideal gain parameter. Determine the loss value according to the mean square error and the deviation. Construct a loss function according to the loss value and use the gradient backpropagation algorithm and the chain rule to trace the error contribution layer by layer to obtain the weight gradients of each layer. Update the network weights according to the weight gradients and in combination with the momentum gradient descent algorithm. At the same time, an amplitude constraint is imposed on the recurrent weights. After iterative training until a predetermined condition is reached, obtain the trained RNN network model; Step 3: Obtain the gain parameters of the wind power generation system according to the trained RNN network model, and control the operation state of the wind power generation system according to the gain parameters.

[0006] Preferably, the historical control signals include the rotor-side converter control signal and the grid-side converter control signal; The real-time state variables include electrical quantities, power quantities, and environmental quantities; The gain parameters include proportional gain and integral gain.

[0007] Preferably, in Step 1, the operation data is preprocessed, and a training data set is constructed according to the preprocessed operation data; The preprocessing method includes filtering, missing value filling, normalization processing, or / and sequence partitioning.

[0008] Preferably, in Step 2, the improved RNN network model includes an input layer, a hidden layer, and an output layer; The input layer constructs an input vector based on the historical control signal, real-time state quantity, and the bias term of the RNN network model; The hidden layer is used to generate the current state through an activation function according to the linear combination of the input vector and the historical state; The output layer sums the weighted values of the current state output by the hidden layer respectively to obtain the gain parameters at the current moment, including the proportional gain and the integral gain .

[0009] Preferably, the output layer generates the gain parameter by accumulating the weighted values output by the hidden layer ;

[0010] Among them, is the weight from the hidden layer to the output layer, is the state of the hidden layer at the current moment;

[0011]

[0012] Among them, is the proportional gain at the current moment, is the integral gain at the current moment, is the state of the hidden layer at the current moment; is the historical state of the hidden layer.

[0013] Preferably, the activation function is activated by the hyperbolic tangent function, and the expression is as follows:

[0014] Among them, is the current state.

[0015] Preferably, the method for controlling the operating state of the wind power generation system according to the gain parameter in step 3 is as follows: Adopt the proportional gain and integral gain output by the RNN network model to replace the fixed parameters of the PI controller in the wind power generation system, and then control the operating state of the wind power generation system.

[0016] In a second aspect, the present application provides a control system for a wind power generation system based on an improved recurrent neural network, including: An acquisition module, configured to construct a training data set according to the operating data of the wind power generation system, where the operating data includes historical control signals, real-time state quantities, and gain parameters; A prediction module, configured to train the improved RNN network model using the training data set, and the training method is as follows: Construct an input vector based on the training data and propagate it forward. Generate the current state through an activation function based on the linear combination of the input vector and the historical state, and obtain the gain data by weighting the current state. Determine the deviation between the gain parameter output by the RNN network and the ideal gain parameter. Determine the loss value based on the mean square error and the deviation. Construct a loss function based on the loss value and use the gradient backpropagation algorithm and the chain rule to trace the error contribution layer by layer to obtain the weight gradients of each layer. Update the network weights according to the weight gradients and in combination with the momentum gradient descent algorithm. At the same time, impose an amplitude constraint on the recurrent weights. After iterative training until a predetermined condition is reached, obtain the trained RNN network model. A control module, configured to obtain the gain parameter of the wind power generation system according to the trained RNN network model, and control the operating state of the wind power generation system according to the gain parameter.

[0017] In a third aspect, the present application provides an electronic device, including: A memory, configured to store a computer program; A processor, configured to implement the steps of the method for controlling a wind power generation system based on an improved recurrent neural network when executing the computer program.

[0018] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for controlling a wind power generation system based on an improved recurrent neural network are implemented.

[0019] Compared with the prior art, the present invention has the following beneficial technical effects: The method for controlling a wind power generation system based on an improved recurrent neural network in the present application fully demonstrates the advantages of the RNN in processing time series data and capturing the dynamic characteristics of the system by introducing the RNN network model. The RNN network model can utilize historical control signals and real-time state variables, and through the recursive structure of its hidden layer, effectively capture complex time series dependencies such as wind speed changes and generator parameter changes in the wind power generation system. This feature enables the method to significantly improve the stability and robustness of the wind power generation system in the face of complex environments, reduce the system response time, static error, and overshoot. At the same time, optimizing the network weights in combination with the gradient backpropagation algorithm and the momentum gradient descent algorithm, and imposing an amplitude constraint on the recurrent weights, further improves the training efficiency and convergence speed of the model, and enhances the stability of the model under complex working conditions. In addition, the method also realizes the adaptive adjustment of the PI controller, eliminating the need for cumbersome manual parameter adjustment, and greatly improving the self-adaptability and intelligent level of the system.

[0020] The present application also provides a control method system for a wind power generation system based on an improved recurrent neural network, an electronic device, and a computer storage medium, which have all the advantages of the above-mentioned control method for a wind power generation system based on an improved recurrent neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] To more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 is the structural diagram of the improved RNN network model of the present invention; Figure 2 is the control flow chart of the MPPT controller of the present invention; Figure 3 is the control flow chart of the rotor side converter of the present invention; Figure 4 is the control flow chart of the GSC and DC bus voltage of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Usually, the components of the embodiments of the present application described and illustrated herein can be arranged and designed in various different configurations.

[0024] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the present application to be protected, but merely represents the selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts fall within the scope of protection of the present application.

[0025] Refer to Figure 1 , a control method for a wind power generation system based on an improved recurrent neural network, includes the following steps: Step 1: Obtain the operation data of the wind power generation system, where the operation data includes historical control signals, real-time state variables, and gain parameters; 1. Historical control signals, that is, physical control instructions output by the controller for driving actuators (such as converters, power switches). The historical control signals include rotor side converter (RSC) control signals and grid side converter (GSC) control signals: The control signals of the rotor-side converter (RSC) include (the reference value of the d-axis rotor current), which is used to regulate the reactive power; (the reference value of the q-axis rotor current), which is used to regulate the active power.

[0026] The control signals of the grid-side converter (GSC) include (the reference value of the d-axis grid current), which is used to control the injection of active power; (the reference value of the q-axis grid current), which is used to control the reactive power support; the control signal of the DC bus voltage ( ), which is used to maintain the stability of the DC capacitor voltage.

[0027] Definition of the control signal: the reference value output by the PI controller, which is used to drive the converter to generate the PWM waveform.

[0028]

[0029] 2. The real-time state variables include mechanical variables, electrical variables, power variables, and environmental variables.

[0030] Mechanical variables: rotor speed ( ), pitch angle ( β ).

[0031] Electrical variables: stator / rotor current and voltage ( ), DC bus voltage ( ), grid voltage and current ( ) Power variables: mechanical power ( ), grid active / reactive power ( ) Environmental variables: real-time wind speed ( ) 3. The gain parameters include the proportional gain ( T P ) and the integral gain ( T I ); the gain parameters can be obtained through the simulation of the wind power generation simulation system or expert experience.

[0032] Step 2: Preprocess the operation data of the wind power generation system in the historical time period, and construct a training data set according to the preprocessed operation data.

[0033] Optionally, the preprocessing methods include filtering, missing value filling, normalization processing, or / and sequence partitioning.

[0034] In this embodiment, filtering, missing value filling, normalization processing, and sequence partitioning are sequentially used to process the operation data.

[0035] S2.1. Filter the operating data to remove high-frequency noise and abnormal interference in the data and retain the effective low-frequency components of the system dynamic characteristics.

[0036] Filtering methods include low-pass filtering and sliding average filtering.

[0037] For example, low-pass filtering is used to reduce the noise of the running data as follows: Filter type: fourth-order Butterworth low-pass filter (Butterworth Filter).

[0038] Cut-off frequency: 10 Hz (set according to the dynamic response characteristics of the wind power generation system).

[0039] Since the mechanical inertia of wind turbines is large and the main dynamic response frequency is lower than 5 Hz, the 0-10 Hz signal is retained to cover the effective frequency band.

[0040] The difference equation for low-pass filtering:

[0041] Among them, the coefficient The bilinear transformation design ensures that the passband is flat and the stopband attenuation is steep.

[0042] For another example, a sliding average filter is used to reduce the noise of the operating data, suppress short-term fluctuations (such as converter switching noise), and retain trend changes. The method is as follows: Window width: 50 sampling points (corresponding to 0.5 seconds of data, sampling rate 100 Hz).

[0043] formula:

[0044] S2.2. Fill missing values ​​in the filter operation data to solve the problem of data point loss due to communication interruption or sensor failure. The method is as follows: Set short-term missing (≤ 5 consecutive points): Using linear interpolation:

[0045] Set long-term missing (>5 consecutive points): Mark the data segment as abnormal and directly remove the data in this time period to avoid introducing erroneous information.

[0046] Supplementary rules: If the missing segment contains a critical event (such as a sudden change in wind speed), the data must be reconstructed through simulation.

[0047] S2.3. Normalize the operation data filled with missing values. By normalization, eliminate the unit differences of voltage, current, speed, etc., so that they are distributed in the interval [−1, 1], ensure the balanced update of the model weights, and avoid the influence of dimensional differences on the training effect. Adopt the maximum - minimum normalization method:

[0048] S2.4. Divide the normalized operation data into fixed - length segments according to the continuous time series, so that the data segments are adapted to the time - series learning structure of the RNN, increase the amount of training data, and improve the generalization ability of the model.

[0049] Divide the operation data into multiple segments according to the time series (for example, each segment contains m time steps) for the time - series learning of the RNN.

[0050] Set the division parameters, the time window length (m): 100 time steps (corresponding to 1 - second data, sampling rate 100 Hz).

[0051] Based on the fact that the dynamic response time constant of the wind power generation system is about 0.5 - 2 seconds, a 1 - second window can cover the main dynamic processes.

[0052] Step 3. Construct an improved RNN network model and train it using the training data set in Step 1.

[0053] The improved RNN network model includes an input layer, a hidden layer, and an output layer, and its training method is as follows: S3.1. Construct an input vector according to the input data and propagate it forward. According to the linear combination of the input vector and the historical state, and generate the current state through an activation function, and weight the current state to obtain the gain data.

[0054] 1) The output layer contains 3 neurons for receiving different input data. Input the training data set into the input layer of the RNN network model, and construct an input vector according to the input data. The input data includes the control signal and the real - time state quantity , and splice with the bias term of the RNN network model to construct the input vector. The expression of the input vector is as follows: .

[0055] Among them, is the control signal at the previous moment, the state quantity at the current moment (such as speed, current, voltage, etc.), and b is the bias term b = 1: used to enhance the expression ability of the network.

[0056] 2) The hidden layer is composed of dynamic recurrent neurons. The input of each neuron includes the weighted combination of the current input vector and the historical hidden layer output. The hidden layer captures the temporal dependence relationship through a non-linear activation function and retains the dynamic characteristics of the system.

[0057] The hidden layer receives the input vector from the input layer, generates the current state according to the linear combination of the input vector and the historical state, and passes it through the activation function. ;

[0058] Among them, is the weight from the input layer to the hidden layer, is the recurrent weight, is the weight from the hidden layer to the output layer, is the state of the hidden layer at the previous moment.

[0059] The activation function is the hyperbolic tangent function activation, and the expression is as follows:

[0060] 3) The output layer weights and sums the current states output by the hidden layer respectively to obtain the gain parameters at the current moment, which include the proportional gain and the integral gain .

[0061] The output layer neurons generate the gain parameters by accumulating the weighted values output by the hidden layer:

[0062] Among them, is the weight from the hidden layer to the output layer, directly corresponds to and .

[0063]

[0064]

[0065] Among them, is the proportional gain at the current moment, is the integral gain at the current moment, is the state of the hidden layer at the current moment; is the historical state of the hidden layer, which is a zero vector at the initial moment.

[0066] The proportional gain is directly generated by the weighted sum of the hidden layer states, reflecting the model's demand for a rapid response to the current error. The integral gain achieves the gradual elimination of the steady-state error by explicitly accumulating historical values.

[0067] Forward propagation is the core link of the RNN model. Its goal is to calculate the proportional gain and integral gain in real time through the input data, replacing the fixed parameters of the traditional PI controller. Through the time-series modeling ability of the hidden layer, this process dynamically captures the non-linear characteristics of the wind power generation system (such as sudden changes in wind speed and grid disturbances), and generates adaptive gain parameters based on the current control signal, system state, and historical hidden state, providing a basis for dynamic adjustment in real-time control.

[0068] S3.2. Determine the deviation between the gain parameters output by the RNN network and the ideal gain parameters. Determine the loss value based on the mean square error and the deviation, using this as a standard to measure the control performance. Construct a loss function based on the loss value and use the gradient backpropagation algorithm and the chain rule to trace the error contribution layer by layer to obtain the weight gradients of each layer. Update the network weights according to the weight gradients and in combination with the momentum gradient descent algorithm, thereby optimizing the network weights and improving the control accuracy.

[0069] Using the gain parameters output by the RNN network (including the proportional gain and integral gain) and the ideal gain labels based on expert experience as inputs, calculate the mean square error (MSE) to quantify the deviation between the RNN output and the ideal gain parameters, using this as a standard to measure the control performance. The mean square error (MSE) measures the control performance deviation, balancing the dynamic and steady-state requirements. The expression of the loss value is as follows:

[0070] where is the ideal gain parameter, is the gain parameter output by the RNN network, including the proportional gain and integral gain.

[0071] Using the gradient backpropagation algorithm, trace the error source layer by layer, from the output layer to the hidden layer, and calculate the gradients of the weights of each layer. These gradient information is used to guide the update of the network weights.

[0072] The weight gradients include

[0073] The weight gradient of the output layer (direct error transmission):

[0074]

[0075] The weight gradient of the hidden layer (chain rule combined with the derivative of the activation function):

[0076]

[0077] S3.3. Using the current weights and the gradients calculated through gradient backpropagation For the input data, through the momentum gradient descent algorithm, combined with an appropriate learning rate = 0.01 and a momentum constant = 0.9 for weight update to accelerate convergence and suppress oscillations. At the same time, to prevent the problem of gradient explosion in the long-term time series dependence of the RNN model, an amplitude constraint is imposed on the recurrent weights. After this series of operations, the weights of the RNN model are incrementally optimized, the control accuracy is improved, and the stability of the model under complex working conditions is guaranteed.

[0078] The expression of the current weight is as follows:

[0079] The updated weight , the weight update amount

[0080] The method of updating the weight by the momentum gradient descent algorithm is as follows:

[0081] Recurrent weight constraint (to prevent gradient explosion):

[0082] Weight update is a key operation for the model to learn from errors. By combining the gradient descent algorithm with the momentum term, it accelerates convergence and suppresses oscillations. At the same time, an amplitude constraint is imposed on the recurrent weights to ensure the stability of the RNN in long-term time series dependence and avoid abnormal control instructions caused by gradient explosion. This step deeply integrates mathematical optimization with engineering stability requirements to ensure the robustness of the model under complex working conditions.

[0083] S3.4. Repeat S3.1 - S3.3 to iteratively train the improved RNN network model until the training stops when reaching the predetermined conditions (number of iterations, error reaches the target), and obtain the trained RNN network model.

[0084] Iterative training gradually approaches the optimal model parameters through the closed-loop process of repeating "forward propagation → error feedback → weight update". It balances the computational cost and the model performance requirements, preventing overfitting and ensuring that the control accuracy meets the standard.

[0085] Step 4. According to the trained RNN network model, predict the control parameters at the current moment, and input the predicted control parameters into the PI controller to control the wind power generation system.

[0086] According to the above steps, by improving the adaptive mechanism of the RNN, the PI controller can respond in real time to dynamic disturbances such as wind speed changes and grid faults, significantly reducing the response time, static error, and overshoot. At the same time, it enhances the robustness of the system under asymmetric phase faults, ensuring the efficient and stable operation of the wind power generation system.

[0087] Figure 2 For the control flow design of the MPPT controller, the hidden layer recurrent neural network method is used to adjust the gains of the PI regulator. The improved RNN-MPPT controller design has 3 neurons in the input layer, 7 neurons in the hidden layer, and 2 neurons in the output layer (3-7-2 structure).

[0088] Figure 3 For the control flow design of the rotor side converter, this design is based on an improved RNN method for adaptive PI regulator settings. The improved RNN-RSC 1 and 2 controllers are designed with a 3-10-2 structure.

[0089] Figure 4 For the GSC and DC bus voltage control process based on the improved RNN method, this method aims to optimize the gains of the PI regulator. The GSC controller is designed with a 3-10-2 architecture. On the other hand, the improved RNN DC link controller is based on a 3-8-2 structure.

[0090] This application adapts the parameters of the PI controller by improving the recurrent neural network (RNN), thereby providing better response time and minimizing overshoot. The improved RNN method performs well under various harsh conditions, including wind speed changes, generator parameter changes, and grid voltage dips caused by asymmetric phase faults. Compared with traditional methods, this technology can track the reference value more effectively and ensure the stable operation of the system in various challenging environments. Through this innovation, we can improve the stability and reliability of the energy system, bringing greater benefits and sustainable development to the energy industry.

[0091] Correspondingly, this application also provides a wind power generation system control system based on an improved recurrent neural network, including: An acquisition module, used to construct a training data set according to the operation data of the wind power generation system, and the operation data includes historical control signals, real-time state quantities, and gain parameters; A prediction module, used to train the improved RNN network model using the training data set, and the training method is as follows: Construct an input vector according to the training data and propagate it forward. According to the linear combination of the input vector and the historical state, and generate the current state through the activation function, and weight the current state to obtain the gain data; Determine the deviation between the gain parameter output by the RNN network and the ideal gain parameter, determine the loss value according to the mean square error and the deviation, construct a loss function based on the loss value, and use the gradient backpropagation algorithm and the chain rule to trace the error contribution layer by layer to obtain the weight gradients of each layer. Update the network weights according to the weight gradients and in combination with the momentum gradient descent algorithm. At the same time, an amplitude constraint is imposed on the recurrent weights. After iterative training until a predetermined condition is reached, a trained RNN network model is obtained; A control module, configured to obtain the gain parameter of the wind power generation system according to the trained RNN network model, and control the operating state of the wind power generation system according to the gain parameter.

[0092] It should be noted that in the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of each module is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules can be combined or integrated into another device, or some features can be ignored or not executed. The modules described as separate components may or may not be physically separated. The components shown as modules may be one physical unit or multiple physical units, that is, they may be located in one place, or may be distributed to multiple different places. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0093] In addition, in each embodiment of the present invention, each module can be integrated in a processing unit, or each module can exist physically alone, or two or more modules can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0094] An electronic device provided by an embodiment of the present application includes a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the steps of the wind power generation system control method based on the improved recurrent neural network described in any of the above embodiments are implemented.

[0095] Another electronic device provided by an embodiment of the present application may further include: an input port connected to a processor, configured to transmit multi-modal data collected by an external acquisition device to the processor; a display unit connected to the processor, configured to display the processing result of the processor to the outside; and a communication module connected to the processor, configured to implement communication between the electronic device and the outside. The display unit may be a display panel, a laser scanning display, etc.; the communication methods adopted by the communication module include but are not limited to Mobile High-Definition Link technology (HML), Universal Serial Bus (USB), High-Definition Multimedia Interface (HDMI), wireless connection (including Wireless Fidelity technology (WiFi), Bluetooth communication technology, Bluetooth Low Energy communication technology, communication technology based on IEEE802.11s).

[0096] A computer-readable storage medium provided by an embodiment of the present application stores a computer program, and when the computer program is executed by a processor, it implements the steps of the wind power generation system control method based on an improved recurrent neural network described in any of the above embodiments.

[0097] For the description of the relevant parts in the control system, electronic device, and computer-readable storage medium of the wind power generation system based on an improved recurrent neural network provided by an embodiment of the present application, please refer to the detailed description of the corresponding parts in the wind power generation system control method based on an improved recurrent neural network provided by an embodiment of the present application, and will not be elaborated here. In addition, in the above technical solutions provided by an embodiment of the present application, the parts that are the same as the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0098] The above content is only for explaining the technical idea of the present invention, and cannot be used to limit the protection scope of the present invention. Any modification made on the basis of the technical solution according to the technical idea proposed by the present invention falls within the protection scope of the claims of the present invention.

Claims

1. A wind power generation system control method based on an improved recurrent neural network, characterized in that: The following steps are involved: Step 1: construct a training data set based on the operation data of the wind power generation system, where the operation data includes historical control signals, real-time state quantities and gain parameters; Step 2: Use the training data set to train the improved RNN network model. The training method is as follows: Construct an input vector based on the training data and propagate it forward. Generate the current state based on the linear combination of the input vector and the historical state through the activation function, and weight the current state to obtain the gain data. Determine the deviation between the gain parameter output by the RNN network and the ideal gain parameter, determine the loss value according to the mean square error and the deviation, construct a loss function according to the loss value, and use the gradient back-propagation algorithm and the chain rule to trace the error contribution layer by layer to obtain the weight gradient of each layer. Update the network weight according to the weight gradient and combine the momentum gradient descent algorithm. At the same time, impose amplitude constraints on the recursive weights. After iterative training, until the predetermined conditions are met, the trained RNN network model is obtained. Step 3: Obtain the gain parameters of the wind power generation system according to the trained RNN network model, and control the operating state of the wind power generation system according to the gain parameters.

2. A wind power generation system control method based on improved recurrent neural network according to claim 1, characterized in that: The historical control signals include rotor-side converter control signals and grid-side converter control signals; The real-time state quantities include electrical quantities, power quantities and environmental quantities; The gain parameters include proportional gain and integral gain.

3. A wind power generation system control method based on improved recurrent neural network according to claim 1, characterized in that: In step 1, the operating data is preprocessed, and a training data set is constructed based on the preprocessed operating data; The preprocessing method includes filtering, missing value filling, normalization and / or sequence division.

4. A wind power generation system control method based on improved recurrent neural network according to claim 1, characterized in that: The improved RNN network model described in step 2 includes an input layer, a hidden layer and an output layer; The input layer constructs an input vector according to the historical control signal, the real-time state quantity and the bias term of the RNN network model; The hidden layer is used to generate the current state according to the linear combination of the input vector and the historical state through the activation function; The output layer performs weighted summation on the current state of the hidden layer output to obtain the gain parameters at the current moment, including the proportional gain and integral gain .

5. A wind power generation system control method based on improved recurrent neural network according to claim 4, characterized in that: The output layer generates a gain parameter by accumulating the weighted value output by the hidden layer ; in, is the weight from the hidden layer to the output layer, is the state of the hidden layer at the current moment; in, is the proportional gain at the current moment, is the integral gain at the current moment, is the hidden layer state at the current moment; is the historical state of the hidden layer.

6. A wind power generation system control method based on improved recurrent neural network according to claim 1, characterized in that: The activation function is a hyperbolic tangent function activation, and the expression is as follows: in, is the current status.

7. A wind power generation system control method based on improved recurrent neural network according to claim 1, characterized in that: The method for controlling the operating state of the wind power generation system according to the gain parameter in step 3 is as follows: The proportional gain and integral gain output by the RNN network model are used to replace the fixed gain of the PI controller in the wind power generation system, and then the operating state of the wind power generation system is controlled.

8. A wind power generation system control system based on an improved recurrent neural network, characterized in that: include: An acquisition module is used to construct a training data set based on the operating data of the wind power generation system, where the operating data includes historical control signals, real-time state quantities and gain parameters; The prediction module is used to train the improved RNN network model using the training data set. The training method is as follows: Construct an input vector based on the training data and propagate it forward. Generate the current state based on the linear combination of the input vector and the historical state through the activation function, and weight the current state to obtain the gain data. Determine the deviation between the gain parameter output by the RNN network and the ideal gain parameter, determine the loss value according to the mean square error and the deviation, construct a loss function according to the loss value, and use the gradient back-propagation algorithm and the chain rule to trace the error contribution layer by layer to obtain the weight gradient of each layer. Update the network weight according to the weight gradient and combine the momentum gradient descent algorithm. At the same time, impose amplitude constraints on the recursive weights. After iterative training, until the predetermined conditions are met, the trained RNN network model is obtained. The control module is used to obtain the gain parameters of the wind power generation system according to the trained RNN network model, and control the operating state of the wind power generation system according to the gain parameters.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, used to implement the steps of the wind power generation system control method based on the improved recurrent neural network as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the wind power generation system control method based on the improved recurrent neural network as described in any one of claims 1 to 7 are implemented.

Images