Wind generating set parameter identification method and system based on deep reinforcement learning

Abstract

The invention discloses a wind generating set parameter identification method and system based on deep reinforcement learning. The method comprises the steps that an initial data set is acquired; extracting mechanical characteristics, electrical characteristics and control characteristics of the initial data; establishing a corresponding mechanical subsystem model, an electrical subsystem model and a control subsystem model according to the mechanical characteristics, the electrical characteristics and the control characteristics; analyzing a dynamic coupling relationship among the mechanical subsystem model, the electrical subsystem model and the control subsystem model to obtain a coupling matrix; according to the coupling matrix, coupling the mechanical subsystem model, the electrical subsystem model and the control subsystem model to obtain a comprehensive mathematical model; and performing parameter identification on the wind driven generator according to the comprehensive mathematical model to obtain an identification result. According to the method, accurate modeling and parameter identification of the wind generating set are realized, and important support is provided for optimizing the performance of the set and predictive maintenance.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and in particular relates to a method and system for wind turbine generator set parameter identification based on deep reinforcement learning. Background Art

[0002] In the process of wind turbine parameter identification, how to establish an accurate and comprehensive mathematical model of wind turbine is a key technical issue. A wind turbine is a complex system, including multiple subsystems such as mechanical part, electrical part and control system, each subsystem has its specific parameters and dynamic characteristics. These subsystems are coupled and influence each other, making the mathematical model of the whole system very complex. In addition, the working environment of wind turbine is changeable, and the changes of external factors such as wind speed and wind direction will affect the parameters and performance of the system, further increasing the difficulty of modeling. How to establish an accurate and robust mathematical model of wind turbine based on the dynamic characteristics and mutual coupling relationship of each subsystem, and the model can adapt to the changes of the external environment is a technical problem that needs to be solved urgently. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a wind turbine parameter identification method and system based on deep reinforcement learning, which can establish an accurate and robust mathematical model of the wind turbine, and the model can adapt to changes in the external environment.

[0004] The present invention provides a method for wind turbine generator set parameter identification based on deep reinforcement learning, comprising:

[0005] Get the initial data set;

[0006] extracting mechanical features, electrical features and control features of the initial data;

[0007] According to the mechanical characteristics, electrical characteristics and control characteristics, a corresponding mechanical subsystem model, an electrical subsystem model and a control subsystem model are established;

[0008] Analyzing the dynamic coupling relationship among the mechanical subsystem model, the electrical subsystem model and the control subsystem model to obtain a coupling matrix;

[0009] According to the coupling matrix, the mechanical subsystem model, the electrical subsystem model and the control subsystem model are coupled to obtain a comprehensive mathematical model;

[0010] The wind turbine parameters are identified according to the comprehensive mathematical model to obtain identification results.

[0011] Optionally, obtaining the initial data set includes:

[0012] Obtain historical operation data of wind power groups;

[0013] Preprocessing the historical operation data, and extracting key features related to the operation status and performance of the wind turbine generator set from the preprocessed data;

[0014] The key features are fused to obtain the initial data set.

[0015] Optionally, establishing corresponding mechanical subsystem models, electrical subsystem models, and control subsystem models according to the mechanical characteristics, electrical characteristics, and control characteristics includes:

[0016] A system identification method based on physical models is used to extract mechanical vibration frequency and damping coefficient from the initial data set, and the mechanical parameters are estimated using the least squares method to build a mechanical subsystem model.

[0017] Using data-driven identification technology, the generator speed, output voltage, and current are extracted from the initial data set, and the neural network algorithm is used to fit the generator characteristics to obtain the electrical subsystem model;

[0018] The controller input and output signals are extracted from the initial data set using a state-space based modeling approach, the system state is estimated using the Kalman filter algorithm, and the control subsystem model is established.

[0019] Optionally, build a mechanical subsystem model including:

[0020] The least square method is used to estimate the vibration frequency and damping coefficient of the mechanical system to obtain the estimated values ​​of the mechanical parameters. If there is a large deviation between the estimated values ​​of the mechanical parameters and the actual system, the mechanical parameters are continuously adjusted through an iterative optimization algorithm to obtain the best estimated values.

[0021] constructing a dynamic equation of a mass-spring-damper system based on the optimal estimate;

[0022] According to the dynamic equations of the mass-spring-damper system, a physical model of the mechanical subsystem is established.

[0023] Optionally, obtain the electrical subsystem model including:

[0024] Build a neural network model and use grid search to optimize the hyperparameters of the neural network;

[0025] The neural network model is trained using the training set data and the back propagation algorithm, and the model parameters are continuously updated so that the model can fit the voltage-current-speed relationship of the generator to obtain the electrical subsystem model, wherein the training set is the generator speed, output voltage, and current.

[0026] Optionally, building a control subsystem model includes:

[0027] Adopt state space modeling method to construct state equation and output equation of control subsystem;

[0028] Based on the state equation and the output equation, a Kalman filter algorithm is used to estimate the system state of the control subsystem;

[0029] According to the system state, determining whether the control characteristics of the control subsystem meet the preset conditions, and if so, determining the currently established control subsystem model;

[0030] If the estimated system state does not meet the preset conditions, the parameters of the state equation and the output equation are adjusted and the system state is re-estimated until the preset conditions are met.

[0031] Optionally, analyzing the dynamic coupling relationship among the mechanical subsystem model, the electrical subsystem model, and the control subsystem model to obtain a coupling matrix includes:

[0032] According to the mechanical subsystem model, electrical subsystem model and control subsystem model, the system identification method is used to obtain the dynamic characteristic parameters of each subsystem and the dynamic coupling relationship parameters between subsystems;

[0033] According to the acquired dynamic characteristic parameters and dynamic coupling relationship parameters, a state space model is constructed to establish a mapping relationship matrix between state variables and subsystem input and output variables;

[0034] A cross-correlation analysis method is used to calculate the correlation coefficient matrix between the variables of each subsystem, quantitatively describe the degree of mutual influence between the subsystems, and obtain the coupling matrix.

[0035] Optionally, obtain a comprehensive mathematical model including:

[0036] The coupling matrix is ​​combined with each subsystem model, and the state equation and output equation of each subsystem are integrated with the coupling matrix through matrix transformation and equation merging to form a comprehensive state equation and output equation of the wind turbine generator set, namely the comprehensive mathematical model.

[0037] Optionally, performing wind turbine parameter identification according to the comprehensive mathematical model and obtaining identification results includes:

[0038] Using Kalman filtering algorithm to filter the noise in the comprehensive mathematical model;

[0039] The recursive least square method is used to identify the parameters of the filtered comprehensive mathematical model and estimate the unknown parameter values ​​in the model online;

[0040] According to the parameter identification results, the mechanical parameters, electrical parameters and control parameters in the comprehensive mathematical model are updated to obtain an optimized comprehensive mathematical model;

[0041] The optimized comprehensive mathematical model is applied to the real-time control system of the wind turbine generator set to perform wind turbine generator parameter identification and obtain the identification result.

[0042] The present invention also provides a wind turbine generator set parameter identification system based on deep reinforcement learning, comprising: a data acquisition module, a mechanical characteristic identification module, an electrical characteristic identification module, a control characteristic identification module, a coupling analysis module, a model integration module and a parameter identification module;

[0043] The data acquisition module is used to obtain historical operating data of the wind turbine generator set, including mechanical vibration, electrical output, control signal, and wind speed and wind direction environmental parameters, to form an initial data set;

[0044] The mechanical characteristic identification module is used to extract the mechanical vibration frequency and damping coefficient from the initial data set and establish a mechanical subsystem model based on the physical model according to the mechanical characteristics;

[0045] The electrical characteristic identification module is used to extract the generator speed, output voltage and current parameters from the initial data set by using data-driven identification technology to establish an electrical subsystem model;

[0046] The control characteristic identification module is used to extract controller input and output signals from the initial data set and establish a control subsystem model based on the state space modeling method according to the control characteristics;

[0047] The coupling analysis module is used to analyze the dynamic coupling relationship between the subsystems and establish a coupling matrix based on the mechanical subsystem model, the electrical subsystem model and the control subsystem model;

[0048] The model integration module is used to integrate the mechanical subsystem model, the electrical subsystem model, the control subsystem model and the coupling matrix to form a comprehensive mathematical model of the wind turbine generator set;

[0049] The parameter identification module is used to identify the parameters of the wind turbine generator set based on a comprehensive mathematical model to obtain identification result data.

[0050] Compared with the prior art, the present invention has the following advantages and technical effects:

[0051] The present invention first obtains the historical operating data of the wind turbine generator set, including mechanical, electrical, control and environmental parameters, to form an initial data set. Then, subsystem models are established for mechanical, electrical and control characteristics respectively, and system identification is performed using physical models, data-driven and state space methods. Next, the dynamic coupling relationship between subsystems is analyzed and a coupling matrix is ​​established. Finally, the subsystem models and the coupling matrix are integrated to form a comprehensive mathematical model of the wind turbine generator set, and parameter identification is performed based on the model. The present invention realizes accurate modeling and parameter identification of the wind turbine generator set by comprehensively considering the mechanical, electrical and control characteristics, as well as the coupling relationship between them, providing important support for optimizing unit performance and predictive maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0053] Figure 1 This is a flow chart of a method for wind turbine generator set parameter identification based on deep reinforcement learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0055] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0056] This embodiment proposes a wind turbine generator parameter identification method based on deep reinforcement learning, such as Figure 1 As shown, the specific steps include:

[0057] Get the initial data set;

[0058] Extract mechanical features, electrical features and control features of initial data;

[0059] According to the mechanical characteristics, electrical characteristics and control characteristics, the corresponding mechanical subsystem model, electrical subsystem model and control subsystem model are established;

[0060] Analyze the dynamic coupling relationship between the mechanical subsystem model, the electrical subsystem model and the control subsystem model to obtain the coupling matrix;

[0061] According to the coupling matrix, the mechanical subsystem model, the electrical subsystem model and the control subsystem model are coupled to obtain a comprehensive mathematical model;

[0062] The wind turbine parameters are identified based on the comprehensive mathematical model to obtain the identification results.

[0063] Furthermore, obtaining the initial data set includes:

[0064] Obtain historical operation data of wind power groups;

[0065] Preprocess the historical operation data and extract key features related to the operation status and performance of the wind turbine from the preprocessed data;

[0066] The key features are fused to obtain the initial data set.

[0067] Specifically, based on the historical operation data of wind turbines, mechanical vibration, electrical output, control signals, and environmental parameters such as wind speed and wind direction are obtained. For the various types of heterogeneous data obtained, data cleaning and preprocessing techniques are used to process missing values, outliers and other noise data to improve data quality and availability. Through feature engineering methods, key features related to the operating status and performance of wind turbines are extracted from the preprocessed historical operation data, including time domain features, frequency domain features, etc., to form standardized feature vectors. Using data fusion technology, the extracted multi-source heterogeneous data such as mechanical vibration, electrical output, control signals and environmental parameters are fused to construct a comprehensive wind turbine operation data set.

[0068] Furthermore, according to the mechanical characteristics, electrical characteristics and control characteristics, the corresponding mechanical subsystem model, electrical subsystem model and control subsystem model are established, including:

[0069] A system identification method based on physical models is used to extract mechanical vibration frequency and damping coefficient from the initial data set, and the mechanical parameters are estimated using the least squares method to build a mechanical subsystem model.

[0070] Using data-driven identification technology, the generator speed, output voltage, and current are extracted from the initial data set, and the neural network algorithm is used to fit the generator characteristics to obtain the electrical subsystem model;

[0071] The controller input and output signals are extracted from the initial data set using a state-space based modeling approach, the system state is estimated using the Kalman filter algorithm, and the control subsystem model is established.

[0072] Furthermore, building a mechanical subsystem model includes:

[0073] The least square method is used to estimate the vibration frequency and damping coefficient of the mechanical system to obtain the estimated values ​​of the mechanical parameters. If there is a large deviation between the estimated values ​​of the mechanical parameters and the actual system, the mechanical parameters are continuously adjusted through an iterative optimization algorithm to obtain the best estimated values.

[0074] Based on the best estimate, the dynamic equations of the mass-spring-damper system are constructed;

[0075] According to the dynamic equations of the mass-spring-damper system, a physical model of the mechanical subsystem is established.

[0076] Specifically, the dynamic response performance and anti-interference ability of the system are improved; the operation data of the mechanical system are continuously accumulated; based on the initial data set, the least squares method is used to estimate the key parameters of the mechanical system, such as the vibration frequency and damping coefficient, to obtain the estimated values ​​of the mechanical parameters; based on the estimated mechanical parameters, the dynamic equations of the mass-spring-damping system are constructed, and the physical model of the mechanical subsystem is established; if there is a large deviation between the estimated mechanical parameters and the actual system, the mechanical parameters are continuously adjusted through iterative optimization algorithms such as the gradient descent method, so that the output of the dynamic model fits the measured data and the model accuracy is improved; based on the constructed dynamic model of the mechanical subsystem, the system identification method such as the recursive least squares method is used to estimate the vibration frequency, damping coefficient and other states of the mechanical system in real time, and realize the online identification of the mechanical characteristics; the identified mechanical characteristic parameters are input into the controller, and the adaptive control algorithm such as the model reference adaptive control is used to adjust the controller parameters in real time to realize the robust control of the mechanical system; through the combination of mechanical characteristic identification and adaptive control, a closed-loop control model of the mechanical system is constructed, and the incremental learning algorithm is used to update the mechanical model and controller parameters to realize the continuous optimization and adaptive control of the mechanical characteristics and ensure the long-term stable operation of the system.

[0077] Furthermore, obtaining the electrical subsystem model includes:

[0078] Build a neural network model and use grid search to optimize the hyperparameters of the neural network;

[0079] The neural network model is trained using the training set data and the back propagation algorithm, and the model parameters are continuously updated so that the model can fit the voltage-current-speed relationship of the generator and obtain the electrical subsystem model. The training set is the generator speed, output voltage, and current.

[0080] Specifically, according to the initial data set, data preprocessing technology is used to clean and standardize the electrical parameter data such as generator speed, output voltage, current, etc., remove outliers and noise data, and obtain a high-quality electrical characteristic data set. For the preprocessed electrical characteristic data set, feature engineering methods are used to extract key features that can reflect the electrical characteristics of the generator, such as average speed, peak voltage, active power, etc., to form an optimized feature data set. The optimized feature data set is randomly divided into a training set and a test set. The training set is used to train the neural network model, and the test set is used to evaluate the model performance. If the division ratio is unreasonable, the ratio is adjusted and the data set is re-divided. According to the complexity of the electrical model, the network structure of the neural network model is designed, including the number of neurons in the input layer, hidden layer and output layer, and the selection of the activation function. The hyperparameters of the neural network are optimized by grid search and other methods. Using the training set data, the back propagation algorithm is used to train the neural network model, and the model parameters are continuously updated so that the model can fit the voltage-current-speed relationship of the generator. During the training process, the loss function and accuracy index of the model are monitored. The performance of the trained neural network model is evaluated on the test set, and the evaluation indicators such as the mean square error and determination coefficient of the model are calculated. If the model performance does not meet the requirements, adjust the network structure and hyperparameters and retrain the model. Deploy the trained neural network model to the online system, obtain the speed, voltage, current and other parameters of the generator in real time, input them into the model, predict the electrical characteristics of the generator, and obtain the real-time voltage-current-speed relationship, which provides a basis for monitoring and optimizing the control of the generator.

[0081] Furthermore, establishing the control subsystem model includes:

[0082] Adopt state space modeling method to construct state equation and output equation of control subsystem;

[0083] Based on the state equation and output equation, the Kalman filter algorithm is used to estimate the system state of the control subsystem;

[0084] According to the system status, determine whether the control characteristics of the control subsystem meet the preset conditions. If so, determine the currently established control subsystem model;

[0085] If the estimated system state does not meet the preset conditions, the parameters of the state equation and the output equation are adjusted and the system state is re-estimated until the preset conditions are met.

[0086] Specifically, according to the control characteristics, the input signal and output signal of the controller are extracted. According to the extracted input and output signals, the state space modeling method is used to construct the state equation and output equation of the control subsystem. According to the established state equation and output equation, the Kalman filter algorithm is used to estimate the system state of the control subsystem. According to the estimated system state, it is judged whether the control characteristics of the control subsystem meet the preset conditions. If so, the currently established control subsystem model is determined. If the estimated system state does not meet the preset conditions, the parameters of the state equation and the output equation are adjusted, and the system state is re-estimated until the preset conditions are met. The established control subsystem model is integrated with other subsystem models to construct a complete system model. According to the constructed system model, machine learning algorithms such as support vector machines, neural networks, etc. are used to train and optimize the system to obtain the final control system model.

[0087] Furthermore, the dynamic coupling relationship between the mechanical subsystem model, the electrical subsystem model and the control subsystem model is analyzed, and the coupling matrix is ​​obtained including:

[0088] According to the mechanical subsystem model, electrical subsystem model and control subsystem model, the system identification method is used to obtain the dynamic characteristic parameters of each subsystem and the dynamic coupling relationship parameters between subsystems;

[0089] According to the acquired dynamic characteristic parameters and dynamic coupling relationship parameters, a state space model is constructed to establish a mapping relationship matrix between state variables and subsystem input and output variables;

[0090] The cross-correlation analysis method is used to calculate the correlation coefficient matrix between the subsystem variables, quantitatively describe the degree of mutual influence between the subsystems, and obtain the coupling matrix.

[0091] Specifically, according to the models of mechanical subsystem, electrical subsystem and control subsystem, the system identification method is adopted to obtain the dynamic characteristic parameters of each subsystem and the dynamic coupling relationship parameters between subsystems. According to the obtained dynamic characteristic parameters and dynamic coupling relationship parameters, the state space model is constructed, and the mapping relationship matrix between the state variables and the input and output variables of the subsystem is established. The cross-correlation analysis method is adopted to calculate the correlation coefficient matrix between the variables of each subsystem, quantitatively describe the degree of mutual influence between the subsystems, and obtain the coupling matrix. According to the mutual influence coefficient in the coupling matrix, the coupling strength between the subsystems is judged, and the main coupling relationship and the secondary coupling relationship between the subsystems are determined. For the subsystems corresponding to the main coupling relationship, the decoupling control algorithm is adopted to realize the decoupling between the subsystems and reduce the mutual interference between the subsystems. For the subsystems corresponding to the secondary coupling relationship, the adaptive filtering algorithm is adopted to reduce the influence of the coupling interference between the subsystems on the control performance. The results of decoupling control and adaptive filtering are used as feedforward compensation, superimposed with the output of the feedback controller of the subsystem to form a composite control quantity, thereby improving the dynamic performance and robustness of the system.

[0092] Further, obtaining a comprehensive mathematical model includes:

[0093] The coupling matrix is ​​combined with the models of each subsystem, and the state equations and output equations of each subsystem are integrated with the coupling matrix through matrix transformation and equation merging to form the comprehensive state equations and output equations of the wind turbine generator set, that is, the comprehensive mathematical model.

[0094] Specifically, according to the mechanical, electrical and control subsystems of the wind turbine generator set, the respective mathematical models are established, and the state equation and output equation of each subsystem are obtained by using the state space representation method. The coupling relationship between the subsystems is analyzed, and the coupling matrix reflecting the interaction between the subsystems is constructed, and the coupling matrix is ​​combined with the models of each subsystem. Through matrix transformation and equation merging, the state equation and output equation of each subsystem are integrated with the coupling matrix to form the comprehensive state equation and output equation of the wind turbine generator set. The comprehensive state equation and output equation are reduced in dimension, redundant state variables are removed, the model structure is simplified, and the calculation efficiency is improved. According to the comprehensive mathematical model, the state observer and state feedback controller are designed to realize the state estimation and feedback control of the wind turbine generator set. The system identification algorithm, such as the least squares method or Kalman filtering, is used to identify and optimize the parameters of the comprehensive mathematical model to improve the accuracy and adaptability of the model. Based on the comprehensive mathematical model, a simulation platform for wind turbine generator sets is developed to analyze and evaluate the dynamic characteristics and control performance of the system, providing guidance for practical engineering applications.

[0095] Furthermore, the wind turbine parameters are identified based on the comprehensive mathematical model, and the identification results include:

[0096] The Kalman filter algorithm is used to filter the noise in the comprehensive mathematical model;

[0097] The recursive least square method is used to identify the parameters of the filtered comprehensive mathematical model and estimate the unknown parameter values ​​in the model online;

[0098] According to the parameter identification results, the mechanical parameters, electrical parameters and control parameters in the comprehensive mathematical model are updated to obtain an optimized comprehensive mathematical model;

[0099] The optimized comprehensive mathematical model is applied to the real-time control system of the wind turbine generator set to identify the wind turbine parameters and obtain the identification results.

[0100] Specifically, a comprehensive mathematical model is established based on the real-time operation data of the wind turbine generator set, and the model contains unknown parameters such as mechanical parameters, electrical parameters and control parameters. The Kalman filter algorithm is used to filter the noise in the comprehensive mathematical model to improve the accuracy and robustness of the model. The recursive least squares method is used to identify the parameters of the filtered comprehensive mathematical model, and the unknown parameter values ​​in the model are estimated online. According to the parameter identification results, the mechanical parameters, electrical parameters and control parameters in the comprehensive mathematical model are updated to obtain the optimized model. The optimized comprehensive mathematical model is applied to the real-time control system of the wind turbine generator set to optimize the control of the generator set. By comparing the operation data of the wind turbine generator set before and after optimization, the model optimization effect and the improvement of control performance are evaluated. If the optimization effect does not meet the expected goal, the operation data of the wind turbine generator set is re-acquired, and the comprehensive mathematical model and control system are iteratively optimized until the performance requirements are met.

[0101] This embodiment also provides a wind turbine generator set parameter identification system based on deep reinforcement learning, including: a data acquisition module, a mechanical characteristic identification module, an electrical characteristic identification module, a control characteristic identification module, a coupling analysis module, a model integration module and a parameter identification module;

[0102] The data acquisition module is used to obtain the historical operating data of the wind turbine generator set, including mechanical vibration, electrical output, control signal, and wind speed and wind direction environmental parameters to form an initial data set;

[0103] The mechanical characteristics identification module is used to extract the mechanical vibration frequency and damping coefficient from the initial data set and establish a mechanical subsystem model based on the physical model system identification method for the mechanical characteristics;

[0104] The electrical characteristics identification module is used to extract the generator speed, output voltage, and current parameters from the initial data set using data-driven identification technology to establish an electrical subsystem model;

[0105] The control characteristic identification module is used to extract the controller input and output signals from the initial data set and establish the control subsystem model based on the state space modeling method according to the control characteristics;

[0106] The coupling analysis module is used to analyze the dynamic coupling relationship between the subsystems and establish a coupling matrix based on the mechanical subsystem model, the electrical subsystem model and the control subsystem model;

[0107] Model integration module, used to integrate the mechanical subsystem model, electrical subsystem model, control subsystem model and coupling matrix to form a comprehensive mathematical model of the wind turbine generator set;

[0108] The parameter identification module is used to identify the parameters of the wind turbine generator set based on a comprehensive mathematical model to obtain identification result data.

[0109] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A wind turbine generator parameter identification method based on deep reinforcement learning, characterized in that: include: Get the initial data set; extracting mechanical features, electrical features and control features of the initial data; According to the mechanical characteristics, electrical characteristics and control characteristics, a corresponding mechanical subsystem model, an electrical subsystem model and a control subsystem model are established; Analyzing the dynamic coupling relationship among the mechanical subsystem model, the electrical subsystem model and the control subsystem model to obtain a coupling matrix; According to the coupling matrix, the mechanical subsystem model, the electrical subsystem model and the control subsystem model are coupled to obtain a comprehensive mathematical model; The wind turbine parameters are identified according to the comprehensive mathematical model to obtain identification results.

2. The method for wind turbine generator set parameter identification based on deep reinforcement learning according to claim 1, characterized in that: Obtaining the initial data set includes: Obtain historical operation data of wind power groups; Preprocessing the historical operation data, and extracting key features related to the operation status and performance of the wind turbine generator set from the preprocessed data; The key features are fused to obtain the initial data set.

3. The method for wind turbine generator set parameter identification based on deep reinforcement learning according to claim 1, characterized in that: According to the mechanical characteristics, electrical characteristics and control characteristics, establishing corresponding mechanical subsystem models, electrical subsystem models and control subsystem models includes: A system identification method based on physical models is used to extract mechanical vibration frequency and damping coefficient from the initial data set, and the mechanical parameters are estimated using the least squares method to build a mechanical subsystem model. Using data-driven identification technology, the generator speed, output voltage, and current are extracted from the initial data set, and the neural network algorithm is used to fit the generator characteristics to obtain the electrical subsystem model; The controller input and output signals are extracted from the initial data set using a state-space based modeling approach, the system state is estimated using the Kalman filter algorithm, and the control subsystem model is established.

4. The method for wind turbine generator parameter identification based on deep reinforcement learning according to claim 3 is characterized in that: Building a mechanical subsystem model includes: The least square method is used to estimate the vibration frequency and damping coefficient of the mechanical system to obtain the estimated values ​​of the mechanical parameters. If there is a large deviation between the estimated values ​​of the mechanical parameters and the actual system, the mechanical parameters are continuously adjusted through an iterative optimization algorithm to obtain the best estimated values. constructing a dynamic equation of a mass-spring-damper system based on the optimal estimate; According to the dynamic equations of the mass-spring-damper system, a physical model of the mechanical subsystem is established.

5. The method for wind turbine generator set parameter identification based on deep reinforcement learning according to claim 3, characterized in that: Access to electrical subsystem models includes: Build a neural network model and use grid search to optimize the hyperparameters of the neural network; The neural network model is trained using the training set data and the back propagation algorithm, and the model parameters are continuously updated so that the model can fit the voltage-current-speed relationship of the generator to obtain the electrical subsystem model, wherein the training set is the generator speed, output voltage, and current.

6. A method for wind turbine generator set parameter identification based on deep reinforcement learning according to claim 3, characterized in that: Building a control subsystem model includes: Adopt state space modeling method to construct state equation and output equation of control subsystem; Based on the state equation and the output equation, a Kalman filter algorithm is used to estimate the system state of the control subsystem; According to the system state, determining whether the control characteristics of the control subsystem meet the preset conditions, and if so, determining the currently established control subsystem model; If the estimated system state does not meet the preset conditions, the parameters of the state equation and the output equation are adjusted and the system state is re-estimated until the preset conditions are met.

7. The method for wind turbine generator set parameter identification based on deep reinforcement learning according to claim 1, characterized in that: Analyzing the dynamic coupling relationship between the mechanical subsystem model, the electrical subsystem model and the control subsystem model to obtain the coupling matrix includes: According to the mechanical subsystem model, electrical subsystem model and control subsystem model, the system identification method is used to obtain the dynamic characteristic parameters of each subsystem and the dynamic coupling relationship parameters between subsystems; According to the acquired dynamic characteristic parameters and dynamic coupling relationship parameters, a state space model is constructed to establish a mapping relationship matrix between state variables and subsystem input and output variables; A cross-correlation analysis method is used to calculate the correlation coefficient matrix between the variables of each subsystem, quantitatively describe the degree of mutual influence between the subsystems, and obtain the coupling matrix.

8. The method for wind turbine generator set parameter identification based on deep reinforcement learning according to claim 7, characterized in that: Get comprehensive mathematical models including: The coupling matrix is ​​combined with each subsystem model, and the state equation and output equation of each subsystem are integrated with the coupling matrix through matrix transformation and equation merging to form a comprehensive state equation and output equation of the wind turbine generator set, namely the comprehensive mathematical model.

9. The method for wind turbine generator set parameter identification based on deep reinforcement learning according to claim 8, characterized in that: The wind turbine parameters are identified according to the comprehensive mathematical model, and the identification results are obtained including: Using Kalman filtering algorithm to filter the noise in the comprehensive mathematical model; The recursive least square method is used to identify the parameters of the filtered comprehensive mathematical model and estimate the unknown parameter values ​​in the model online; According to the parameter identification results, the mechanical parameters, electrical parameters and control parameters in the comprehensive mathematical model are updated to obtain an optimized comprehensive mathematical model; The optimized comprehensive mathematical model is applied to the real-time control system of the wind turbine generator set to perform wind turbine generator parameter identification and obtain the identification result.

10. A wind turbine generator parameter identification system based on deep reinforcement learning, characterized in that: include: Data acquisition module, mechanical characteristic identification module, electrical characteristic identification module, control characteristic identification module, coupling analysis module, model integration module and parameter identification module; The data acquisition module is used to obtain historical operating data of the wind turbine generator set, including mechanical vibration, electrical output, control signal, and wind speed and wind direction environmental parameters, to form an initial data set; The mechanical characteristic identification module is used to extract the mechanical vibration frequency and damping coefficient from the initial data set and establish a mechanical subsystem model based on the physical model according to the mechanical characteristics; The electrical characteristic identification module is used to extract the generator speed, output voltage and current parameters from the initial data set by using data-driven identification technology to establish an electrical subsystem model; The control characteristic identification module is used to extract controller input and output signals from the initial data set and establish a control subsystem model based on the state space modeling method according to the control characteristics; The coupling analysis module is used to analyze the dynamic coupling relationship between the subsystems and establish a coupling matrix based on the mechanical subsystem model, the electrical subsystem model and the control subsystem model; The model integration module is used to integrate the mechanical subsystem model, the electrical subsystem model, the control subsystem model and the coupling matrix to form a comprehensive mathematical model of the wind turbine generator set; The parameter identification module is used to identify the parameters of the wind turbine generator set based on a comprehensive mathematical model to obtain identification result data.

Images