Inertial damping adaptive regulation method and system for wind farm based on fusion deep q network

Abstract

The application discloses a wind farm inertia damping adaptive regulation method and system fusing a deep Q network, and belongs to the technical field of power systems.The application comprises the following steps: establishing a mathematical model of a doubly-fed wind turbine and a converter and an interactive model of a wind power system connected to a power grid, and analyzing the grid-connected stability of the wind power system; researching dynamic response characteristics of inertia support provided by a VSG system under different conditions, and improving primary frequency modulation capability; selecting rotational inertia and damping coefficients by using a deep Q network, improving a damping inertia adaptive control strategy, realizing the cooperation of wind turbine inertia support, frequency response capability improvement and active power continuous smooth adjustment, improving power grid stability, optimizing wind farm operation performance, realizing dynamic optimal matching of inertia response, primary frequency modulation and unit own operation state, providing auxiliary services while guaranteeing the safety and service life of the wind turbine, and realizing smooth and non-impact adjustment of active power.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method and system for adaptive control of wind farm inertia damping using a fused deep Q-network. Background Technology

[0002] Doubly fed wind turbines operate in grid-connected mode via power electronic converters. Their inherent low inertia and power output fluctuations make it difficult for wind farms to provide natural inertia support and effective frequency regulation response to the power grid like traditional synchronous generators.

[0003] This problem has become increasingly prominent after a high proportion of wind power is integrated into the grid. It not only exacerbates grid frequency fluctuations and reduces voltage stability, but may also trigger safety risks such as power oscillations, severely restricting the absorption capacity of wind power. At the same time, traditional wind farm inertia damping control strategies mostly adopt fixed parameters or empirical control logic, which are difficult to adapt to complex dynamic scenarios such as random wind speed fluctuations and sudden changes in grid load. They cannot achieve a dynamic balance between inertia response, primary frequency regulation, and the safe operation of the turbine itself. This not only affects the quality of grid ancillary services, but may also lead to increased mechanical losses and shortened lifespan of wind turbines due to frequent adjustments, weakening the market competitiveness and economic benefits of wind farms. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for adaptive control of wind farm inertia damping based on a fused deep Q-network, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an adaptive control method for wind farm inertia damping based on deep Q-networks, used for inertia support and damping adjustment after wind farm grid connection, comprising:

[0006] Establish mathematical models of doubly-fed wind turbines and converters, and analyze the parameters of wind turbines, control parameters of the generator-side converter, control parameters of the grid-side converter, and the characteristics of the wind power generation system under different operating conditions.

[0007] Based on the mathematical models of doubly fed wind turbines, converters, and characteristic analysis results, an interactive model for wind power system grid connection is established. Taking into account the actual physical parameters of the system, converter control parameters, wind power system operating status, and grid operating status, the stability of wind power system grid connection is judged by combining time-domain simulation and frequency-domain analysis.

[0008] Based on the active stabilization capability of wind turbine generators, this study investigates the dynamic response characteristics of the VSG system providing inertial support under different conditions, and quantitatively analyzes the inertial response time and frequency regulation amplitude of the VSG system.

[0009] By selecting rotational inertia and damping coefficient based on deep Q-networks, an improved damping inertia adaptive control strategy is developed to achieve synergy between wind turbine inertia support, frequency response capability enhancement, and continuous and smooth active power adjustment.

[0010] Furthermore, a mathematical model for a doubly-fed wind turbine is established, specifically including:

[0011] Based on the electromagnetic coupling principle of a doubly-fed induction generator, voltage equations, flux linkage equations, and torque equations are constructed for the stator and rotor sides.

[0012] A wind turbine aerodynamic model is introduced, and the mechanical power captured by the wind turbine is calculated by taking real-time wind speed as input and combining the tip speed ratio of the wind turbine blades and the wind energy utilization coefficient.

[0013] The transmission relationship between the mechanical power of the wind turbine and the input power of the generator is established by using a gearbox transmission model, and finally a complete mathematical model of the doubly fed wind turbine unit covering the wind turbine, gearbox and generator is formed.

[0014] Furthermore, a mathematical model of the converter is established, specifically including:

[0015] The converter mathematical model includes a machine-side converter mathematical model and a grid-side converter mathematical model;

[0016] The mathematical model of the generator-side converter aims to maintain the stability of the generator stator voltage and track the maximum wind energy capture. Based on the vector control principle in the rotating coordinate system, a dual closed-loop control model with an inner current loop and an outer power loop is constructed.

[0017] The mapping relationship between the generator-side converter control parameters and the generator rotor current and active power output is clarified. The generator-side converter control parameters include current loop PI regulator parameters and power loop PI regulator parameters.

[0018] The mathematical model of the grid-side converter aims to maintain the stability of the DC bus voltage and achieve unity power factor operation. It adopts a vector control strategy in a rotating coordinate system to construct a dual closed-loop control model with an inner voltage loop and an outer current loop.

[0019] This study analyzes the impact of grid-side converter control parameters on the DC bus voltage fluctuation suppression capability and the grid-side reactive power compensation effect.

[0020] Furthermore, the characteristics of the wind power generation system under different operating conditions are analyzed, specifically including:

[0021] Set up operating scenarios for the wind power generation system, including stable operation at rated wind speed, low wind speed fluctuation, and high wind speed cut-out.

[0022] For each operating scenario, real-time operating data of the wind turbine is collected through simulation tools. The real-time operating data includes wind turbine speed, generator stator current, generator stator voltage, active power output, and converter switch duty cycle.

[0023] The collected data are analyzed for trends and fluctuations are calculated to determine the variation patterns of key parameters of wind turbines under different operating scenarios, and to determine the degree of impact of wind speed fluctuations and load changes on the operational stability of the wind power system.

[0024] Furthermore, after establishing an interaction model for wind power system grid connection, grid connection stability is analyzed, specifically including:

[0025] A simulation platform was built based on an interactive model to simulate normal grid operation, grid frequency deviation, and grid voltage drop conditions.

[0026] Calculate the output power fluctuation coefficient, grid frequency recovery time, and voltage sag duration of the wind power system under different operating conditions;

[0027] A grid connection stability evaluation index is set. If the calculation result does not meet the grid connection stability evaluation index, the wind power system is judged to be grid-connected and stable; otherwise, the grid connection is judged to be unstable, and the key influencing parameters that cause the instability are located. The key influencing parameters include the wind speed fluctuation amplitude and the converter control parameter deviation.

[0028] Furthermore, the dynamic response characteristics of the VSG system providing inertia support under different conditions are studied, specifically including:

[0029] A VSG system model was built in the simulation platform, and different test conditions were set, including power grid frequency step disturbance, load change and wind speed change.

[0030] For each test condition, the inertia support response time, frequency regulation rate, and voltage stability of the VSG system were collected.

[0031] A database of dynamic response characteristics of the VSG system is constructed based on the data collected for each test condition. The optimal operating parameters of the VSG system under different test conditions are determined by data fitting, and the optimal parameters are used as the initial input for the parameter optimization of the deep Q network.

[0032] Furthermore, the moment of inertia and damping coefficient are selected based on the deep Q-network, specifically including:

[0033] A deep Q-network model is constructed, which includes an input layer, a hidden layer, and an output layer. The input state vector of the deep Q-network is determined, which includes wind power system operating state parameters, grid operating parameters, and historical regulation errors.

[0034] Define a reward function for a deep Q-network, with the objectives of optimizing inertia support, minimizing frequency deviation, and minimizing active power fluctuation.

[0035] The deep Q-network is trained, and the experience replay mechanism is used to store the input state, action, reward value and the next input state under different operating scenarios. The network parameters are updated by gradient descent until the network converges.

[0036] After training, the real-time collected input state vector is input into the deep Q-network, and the network outputs the optimal moment of inertia and damping coefficient values ​​for the current scenario.

[0037] Furthermore, the damped inertia adaptive control strategy is improved, specifically including:

[0038] Real-time data collection of wind power system operating status and power grid operating parameters is achieved through sensors.

[0039] The collected data is transmitted to the control unit, which then calls the trained deep Q-network to output the optimal moment of inertia and optimal damping coefficient for the current scenario.

[0040] The control unit adjusts the control commands of the VSG system according to the optimal parameters: when the grid frequency deviates from the rated value, the virtual inertia of the VSG system is adjusted to the optimal moment of inertia first, and the damping coefficient is adjusted to the optimal damping coefficient simultaneously.

[0041] The active power output curve is monitored in real time. If the curve fluctuates by more than ±2% for three consecutive acquisition cycles, the deep Q network is called again to update the optimal parameters.

[0042] Furthermore, it also includes a control effect verification step:

[0043] A prototype wind power grid-side inverter was completed. The prototype includes a generator-side converter module, a grid-side converter module, a control module, and a data acquisition module.

[0044] A grid-connected system model was built on a hardware-in-the-loop (HIPL) simulation platform based on RT-LAB. The model includes a wind farm simulation submodule, a power grid simulation submodule, and a data monitoring submodule.

[0045] The experimental prototype was connected to a hardware-in-the-loop simulation platform to simulate typical operating conditions, including rated wind speed operation, wind speed fluctuation, and power grid frequency disturbance. The power grid frequency recovery time, active power fluctuation amplitude, and inertia support response time were collected under different typical operating conditions.

[0046] If the grid frequency recovery time is ≤1.5s, the active power fluctuation amplitude is ≤±2%, and the inertia support response time is ≤0.3s, then the control strategy is deemed effective; otherwise, the cause of the deviation is analyzed, and the reward function weights or training iterations of the deep Q network are optimized until the control effect meets the requirements.

[0047] Furthermore, a wind farm inertia damping adaptive control system integrating deep Q-networks is provided to implement the above-mentioned control method, comprising:

[0048] The model building module is configured to establish mathematical models of doubly fed wind turbines, machine-side converters, grid-side converters, and interactive models of wind power systems connected to the grid, and to store model parameters and simulation data.

[0049] The characteristic analysis module connects to the model building module and is configured to analyze the characteristics of wind turbine parameters, converter control parameters, and wind power generation systems under different operating conditions, and output a characteristic analysis report.

[0050] The VSG characteristic research module, connected to the characteristic analysis module, is configured to study the dynamic response characteristics of the VSG system under different conditions and output the optimal initial operating parameters of the VSG system.

[0051] A deep Q-network optimization module is connected to the VSG characteristic research module. The deep Q-network optimization module includes a network training unit and a parameter output unit.

[0052] The network training unit is configured to build and train a deep Q-network, and the parameter output unit is configured to output the optimal moment of inertia and damping coefficient under different operating scenarios.

[0053] An adaptive control module is connected to a deep Q-network optimization module. The adaptive control module includes a data acquisition unit, a control command generation unit, and a parameter adjustment unit.

[0054] Among them, the data acquisition unit is configured to collect wind power system and grid operation data in real time, the control command generation unit is configured to generate VSG system control commands based on optimal parameters, and the parameter adjustment unit is configured to dynamically update parameters according to active power fluctuations.

[0055] The experimental verification module is connected to the adaptive control module. The experimental verification module includes a wind power grid-side inverter experimental prototype and an RT-LAB hardware-in-the-loop simulation platform, configured to build a grid-connected system model to verify the model's accuracy, the correctness of the mechanism analysis, and the effectiveness of the control strategy.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] 1. This invention establishes a doubly-fed induction generator (DFIG) wind turbine, converter, and wind power-grid interaction model. Combined with full-scenario characteristic analysis and multi-dimensional grid connection stability evaluation, it achieves precise control over the interaction characteristics between the wind power system and the grid. This effectively reduces the impact of wind speed fluctuations and grid disturbances on grid operation, significantly improves grid frequency and voltage stability, reduces the impact of active power fluctuations on the grid, and provides a scientific basis for optimizing control strategies by quantifying the influence weights of key parameters. This ensures the safe and stable operation of the grid after a high proportion of wind power is connected, and lays a solid foundation for providing grid auxiliary services for wind farms.

[0058] 2. This invention employs a deep Q-network to optimize rotational inertia and damping coefficient, coupled with an improved damping inertia adaptive control strategy. This achieves dynamic optimal matching between inertia response, primary frequency regulation, and the unit's own operating state. By acquiring data in real time, intelligently outputting optimal parameters, and dynamically adjusting VSG control commands, it ensures smooth and shock-free regulation of active power. While rapidly responding to grid frequency changes and improving primary frequency regulation capability, it effectively suppresses power oscillations, reduces unit mechanical losses, ensures the safety and lifespan of wind turbine operation, and significantly optimizes the overall operating performance of the wind farm.

[0059] 3. This invention, through a verification method combining hardware-in-the-loop simulation and experimental prototypes, ensures the engineering practicality and reliability of the control strategy, enabling wind farms to possess efficient grid ancillary service capabilities. This makes wind farms more competitive in the grid frequency regulation and other ancillary service markets, allowing them to obtain additional economic benefits by providing high-quality ancillary services. Optimized operating performance and stable grid connection reduce operation and maintenance costs and failure risks, improve the power generation efficiency and return on investment of wind farms, and further enhance the core competitiveness and sustainable development capabilities of wind farms in the new energy market. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the overall control process of the present invention;

[0061] Figure 2 This is a schematic diagram of the model construction and characteristic analysis process of the present invention;

[0062] Figure 3 This is a schematic diagram of the deep Q-network optimization and control execution process of the present invention;

[0063] Figure 4 This is a schematic diagram of the control effect verification process of the present invention. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Please see Figure 1-4 The present invention provides the following technical solutions:

[0066] An adaptive control method for wind farm inertia damping based on deep Q-networks is proposed for inertia support and damping adjustment after grid connection of wind farms, including:

[0067] Establish mathematical models of doubly-fed wind turbines and converters, and analyze the parameters of wind turbines, control parameters of the generator-side converter, control parameters of the grid-side converter, and the characteristics of the wind power generation system under different operating conditions such as rated wind speed operation, low wind speed fluctuation, and high wind speed cut-off.

[0068] Based on the mathematical models of doubly-fed induction generator (DFIG) wind turbines, converters, and characteristic analysis results, an interactive model for wind power system grid connection is established. This model comprehensively considers the actual physical parameters of the system (including rotor inertia, gearbox transmission ratio, generator stator / rotor resistance), converter control parameters (including PI regulator proportional coefficient and integral coefficient), wind power system operating status (including real-time wind speed and active power output), and grid operating status (including grid frequency, voltage amplitude, and load power). By combining time-domain simulation and frequency-domain analysis, the stability of the wind power system grid connection is determined.

[0069] Based on the active stabilization capability of wind turbine generators, this study investigates the dynamic response characteristics of the VSG system providing inertial support under different conditions, quantitatively analyzes the inertial response time and frequency regulation amplitude of the VSG system, and improves the primary frequency regulation capability of wind farms.

[0070] By selecting rotational inertia and damping coefficient based on deep Q-networks, an improved damping inertia adaptive control strategy is developed to achieve synergy between wind turbine inertia support, frequency response capability enhancement, and continuous and smooth active power regulation, thereby avoiding the impact of active power fluctuations on the stable operation of the power grid.

[0071] In the above embodiments, through multi-model collaborative construction and full-parameter coupling analysis, the interaction characteristics between the wind power system and the power grid are accurately characterized. The deep Q-network and inertia damping control are deeply integrated, solving the problem of parameter adaptive selection caused by the volatility of wind power generation. At the same time, multi-objective synergy of inertia support, frequency response and power smooth regulation is achieved, which significantly improves the stability and flexibility of wind farm grid-connected operation, optimizes wind farm operation performance, and realizes dynamic optimal matching of inertia response, primary frequency regulation and the unit's own operating status. While providing ancillary services, it ensures the safety and lifespan of wind turbine operation, realizes smooth and shock-free regulation of active power, and enhances market competitiveness and economic benefits.

[0072] Establishing a mathematical model for a doubly-fed wind turbine includes:

[0073] Based on the electromagnetic coupling principle of a doubly-fed induction generator, voltage equations, flux linkage equations, and torque equations are constructed for the stator and rotor sides.

[0074] A wind turbine aerodynamic model is introduced, and the mechanical power captured by the wind turbine is calculated by taking real-time wind speed as input and combining the tip speed ratio of the wind turbine blades and the wind energy utilization coefficient.

[0075] The transmission relationship between the mechanical power of the wind turbine and the input power of the generator is established by using a gearbox transmission model, and finally a complete mathematical model of the doubly fed wind turbine unit covering the wind turbine, gearbox and generator is formed.

[0076] In the above embodiments, the mathematical model construction method of the doubly-fed induction generator (DFIG) wind turbine breaks through the limitation of traditional models that only focus on a single component, and realizes dynamic coupling modeling of the entire link of wind turbine-gearbox-generator. By accurately integrating the aerodynamic characteristics of the wind turbine and the transmission mechanism of the gearbox, the model effectively improves the fitting accuracy of the unit's operating state under different wind speed conditions, and solves the problem of large simulation errors caused by neglecting the interaction between components in traditional models. The constructed model can accurately reflect the dynamic change law of key parameters of the unit, providing a high-precision foundation for subsequent grid connection stability analysis and control strategy design, and reducing the deviation between simulation results and actual operating data.

[0077] Establishing a mathematical model for the converter specifically includes:

[0078] The mathematical model of the converter includes the mathematical model of the machine-side converter and the mathematical model of the grid-side converter;

[0079] The mathematical model of the generator-side converter aims to maintain the stability of the generator stator voltage and track the maximum wind energy capture. Based on the vector control principle in the rotating coordinate system, a dual closed-loop control model with an inner current loop and an outer power loop is constructed.

[0080] Clarify the mapping relationship between the generator-side converter control parameters and the generator rotor current and active power output. The generator-side converter control parameters include the current loop PI regulator parameters and the power loop PI regulator parameters.

[0081] The mathematical model of the grid-side converter aims to maintain the stability of the DC bus voltage and achieve unity power factor operation. It adopts a vector control strategy in a rotating coordinate system to construct a dual closed-loop control model with an inner voltage loop and an outer current loop.

[0082] This study analyzes the impact of grid-side converter control parameters on the DC bus voltage fluctuation suppression capability and the grid-side reactive power compensation effect.

[0083] In the above embodiments, the construction of the converter mathematical model solves the problem of blind optimization of control parameters in traditional models by clarifying the mapping relationship between control parameters and output characteristics. This provides a clear parameter adjustment path for subsequent adaptive regulation. The machine-side model focuses on maximizing wind energy capture and stabilizing stator voltage, while the grid-side model focuses on DC bus voltage control and power factor regulation. The two work together to ensure the stable and efficient operation of the converter under different operating conditions, improve voltage fluctuation suppression capability, improve reactive power compensation accuracy, and significantly enhance the grid-connected adaptability of the wind power system.

[0084] The characteristics of wind power generation systems under different operating conditions are analyzed, specifically including:

[0085] Set up operating scenarios for the wind power generation system, including the rated wind speed stable operation scenario (wind speed maintained at the rated wind speed ±0.5m / s), the low wind speed fluctuation scenario (wind speed fluctuates randomly in the range of 3-8m / s, with a fluctuation range ≤2m / s), and the high wind speed cut-out scenario (wind speed increases from the rated wind speed to the cut-out wind speed, with an increase rate ≤1m / s).

[0086] For each operating scenario, real-time operating data of the wind turbine is collected through simulation tools. The real-time operating data includes wind turbine speed, generator stator current, generator stator voltage, active power output, and converter switch duty cycle.

[0087] The collected data are analyzed for trends and fluctuations are calculated to determine the variation patterns of key parameters of wind turbines under different operating scenarios, and to determine the degree of impact of wind speed fluctuations and load changes on the operational stability of the wind power system.

[0088] In the above embodiments, by constructing a typical operating scenario system covering the entire wind speed range, the limitations of traditional analysis, which is limited to stable operating conditions, are overcome. By accurately collecting multi-dimensional operating data and performing quantitative analysis, a comprehensive understanding of the dynamic characteristics of the wind power system is achieved. The changing patterns and influence weights of key parameters under different scenarios are clarified, and the impact of wind speed fluctuations and load changes are quantitatively characterized. This provides data support for the subsequent construction of interactive models and optimization of control strategies, solving the problem of poor adaptability of control strategies caused by the single scenario and lack of data in traditional analysis. This enables subsequent control schemes to accurately match different operating conditions, improving the stability and reliability of wind farm operation.

[0089] After establishing an interaction model for wind power system grid connection, the grid connection stability is analyzed, specifically including:

[0090] A simulation platform was built based on an interactive model to simulate normal grid operation, grid frequency deviation (deviation range ±0.5Hz), and grid voltage drop (drop amplitude 10%-30%).

[0091] Calculate the output power fluctuation coefficient, grid frequency recovery time, and voltage sag duration of the wind power system under different operating conditions;

[0092] Set grid connection stability evaluation indicators. If the output power fluctuation coefficient is ≤5%, the grid frequency recovery time is ≤2s, and the voltage sag duration is ≤0.5s, and the calculation results do not meet the grid connection stability evaluation indicators, the wind power system is judged to be grid-connected stable; otherwise, the grid connection is judged to be unstable, and the key influencing parameters that cause instability are located. The key influencing parameters include wind speed fluctuation amplitude and converter control parameter deviation.

[0093] In the above embodiments, by adopting an analysis method that combines the time domain and frequency domain, a multi-dimensional and quantitative stability evaluation system is constructed. This breaks through the limitations of traditional analysis that relies on only a single indicator. By simulating typical operating conditions such as normal grid operation, frequency deviation, and voltage drop, the system comprehensively verifies the grid-connected adaptability of the wind power system in a complex grid environment. This solves the problem that traditional methods cannot comprehensively evaluate grid-connected stability. It can accurately locate the key influencing parameters of instability, providing a clear direction for subsequent control strategy optimization, improving the accuracy of stability judgment, and effectively reducing the risk of wind power grid connection to the safe operation of the grid.

[0094] The VSG system provides dynamic response characteristics for inertia support under different conditions, specifically including:

[0095] A VSG system model was built in the simulation platform, and different test conditions were set, including grid frequency step disturbance (disturbance value ±0.2Hz), load change (change amplitude ±10% of rated load), and wind speed change (change amplitude ±3m / s).

[0096] For each test condition, the inertia support response time (time from the occurrence of the disturbance to the start of inertia output), frequency regulation rate (the recovery value of the grid frequency per unit time), and voltage stability (the ratio of voltage fluctuation amplitude to rated voltage) of the VSG system were collected.

[0097] A database of dynamic response characteristics of the VSG system is constructed based on the data collected for each test condition. The optimal operating parameters of the VSG system (including the initial values ​​of virtual inertia and damping coefficient) under different test conditions are determined by data fitting. The optimal parameters are used as the initial input for deep Q-network parameter optimization to improve the response speed and regulation accuracy of primary frequency regulation in wind farms.

[0098] In the above embodiments, the VSG system characteristic research method realizes the quantitative analysis of the dynamic response characteristics of inertia support by constructing a test system covering multiple disturbance types on both the power grid and wind power sides. By establishing a characteristic database and fitting the optimal initial parameters, it solves the problems of traditional VSG parameter setting relying on experience and having poor adaptability, provides high-quality initial input for deep Q-network optimization, significantly improves the efficiency and accuracy of subsequent parameter optimization, quantifies key indicators such as inertia response time and frequency regulation rate, clarifies the operating law of VSG system under different disturbance conditions, improves the primary frequency regulation response speed, enhances regulation accuracy, and effectively strengthens the ability of wind farms to cope with complex disturbances.

[0099] The selection of rotational inertia and damping coefficient based on a deep Q-network specifically includes:

[0100] A deep Q-network model is constructed, which includes an input layer, a hidden layer, and an output layer. The input state vector of the deep Q-network is determined, which includes the wind power system operating state parameters (real-time wind speed, active power output value, wind turbine speed), grid operating parameters (grid frequency, voltage amplitude, load power), and historical regulation errors (frequency deviation value and active power fluctuation value of the first 5 sampling periods).

[0101] Define the reward function for the deep Q-network, with the objectives of optimizing the inertia support effect, minimizing the frequency deviation, and minimizing the active power fluctuation.

[0102] The deep Q-network is trained, and the experience replay mechanism is used to store the input state, action, reward value and the next input state under different operating scenarios. The network parameters are updated by gradient descent until the network converges.

[0103] After training, the real-time collected input state vector is input into the deep Q-network, and the network outputs the optimal moment of inertia and damping coefficient values ​​for the current scenario.

[0104] In the above embodiments, by introducing a deep Q-network into the inertia damping control of wind farms, the limitations of traditional parameter selection relying on empirical formulas or fixed rules are overcome. By constructing a multi-dimensional input state vector and a multi-objective reward function, accurate perception of complex operating scenarios and intelligent decision-making on optimal parameters are achieved. This solves the problem of parameter adaptive matching caused by the volatility of wind power generation. The use of an experience playback mechanism improves the stability and generalization ability of network training, ensuring that optimal parameters can be output under different operating scenarios, improving the real-time performance of parameter adjustment, reducing frequency deviation and active power fluctuations, and providing an efficient and intelligent implementation path for adaptive control of inertia damping.

[0105] Improved damped inertia adaptive control strategies include:

[0106] The system collects real-time operating status data of the wind power system (including real-time wind speed, wind turbine speed, and active power output) and grid operating parameters (including grid frequency, voltage amplitude, and load power) through sensors, with a collection cycle of 0.1 seconds.

[0107] The collected data is transmitted to the control unit, which then calls the trained deep Q-network to output the optimal moment of inertia and optimal damping coefficient for the current scenario.

[0108] The control unit adjusts the control commands of the VSG system according to the optimal parameters: when the grid frequency deviates from the rated value, the virtual inertia of the VSG system is adjusted to the optimal rotational inertia first. The inertia supports a rapid response to frequency changes, shortens the frequency recovery time, and the damping coefficient is adjusted to the optimal damping coefficient simultaneously to suppress active power oscillation and keep the active power fluctuation amplitude within ±2% of the rated power.

[0109] The active power output curve is monitored in real time. If the curve fluctuates by more than ±2% for three consecutive acquisition cycles, the deep Q network is called again to update the optimal parameters to ensure that the active power is always continuously and smoothly adjusted.

[0110] In the above embodiments, the deep integration of deep Q network and VSG control achieves real-time optimal matching of inertia and damping coefficient, solving the problem that traditional strategies have difficulty balancing frequency response speed and power smoothness. The introduction of active power fluctuation feedback mechanism can correct parameter deviations in a timely manner, ensuring the continuous optimal control effect, shortening frequency recovery time, improving the control accuracy of active power fluctuation amplitude, effectively enhancing the stability of wind farm grid-connected operation, and providing technical support for the grid to safely accept a high proportion of wind power.

[0111] The wind farm inertia damping adaptive control method and system integrating deep Q-networks is characterized by further including a control effect verification step:

[0112] A prototype wind power grid-side inverter was completed. The prototype includes a generator-side converter module, a grid-side converter module, a control module (including a deep Q-network algorithm deployment unit), and a data acquisition module.

[0113] A grid-connected system model was built on a hardware-in-the-loop simulation platform based on RT-LAB. The model includes a wind farm simulation submodule (containing models of 10 doubly fed wind turbine units), a power grid simulation submodule (containing a 35kV distribution network model and a load model), and a data monitoring submodule.

[0114] The experimental prototype was connected to a hardware-in-the-loop simulation platform to simulate typical operating conditions, including rated wind speed operation, wind speed fluctuation, and power grid frequency disturbance. The power grid frequency recovery time, active power fluctuation amplitude, and inertia support response time were collected under different typical operating conditions.

[0115] If the grid frequency recovery time is ≤1.5s, the active power fluctuation amplitude is ≤±2%, and the inertia support response time is ≤0.3s, then the control strategy is deemed effective; otherwise, the cause of the deviation is analyzed, and the reward function weights or training iterations of the deep Q network are optimized until the control effect meets the requirements.

[0116] In the above embodiments, by building a simulation model that includes multiple wind turbine units and an actual power distribution network, a complex and realistic operating environment is simulated, which solves the problem of the disconnect between traditional verification methods and actual application scenarios. A quantitative verification index system is established, which can accurately locate the causes of deviations in the control strategy and make targeted optimizations, ensuring the practicality and reliability of the control method. Compared with traditional verification methods, the credibility of the verification results is improved, providing solid experimental support for the engineering application of the control strategy.

[0117] An adaptive control system for wind farm inertia damping incorporating deep Q-networks, used to implement the aforementioned control method, includes:

[0118] The model building module is configured to establish mathematical models of doubly fed wind turbines, machine-side converters, grid-side converters, and interactive models of wind power systems connected to the grid, and to store model parameters and simulation data.

[0119] The characteristic analysis module connects to the model building module and is configured to analyze the characteristics of wind turbine parameters, converter control parameters, and wind power generation systems under different operating conditions, and output a characteristic analysis report.

[0120] The VSG characteristic research module, connected to the characteristic analysis module, is configured to study the dynamic response characteristics of the VSG system under different conditions and output the optimal initial operating parameters of the VSG system.

[0121] The deep Q network optimization module connects to the VSG characteristic research module. The deep Q network optimization module includes a network training unit and a parameter output unit.

[0122] The network training unit is configured to build and train a deep Q-network, and the parameter output unit is configured to output the optimal moment of inertia and damping coefficient under different operating scenarios.

[0123] The adaptive control module is connected to the deep Q-network optimization module. The adaptive control module includes a data acquisition unit, a control command generation unit, and a parameter adjustment unit.

[0124] Among them, the data acquisition unit is configured to collect wind power system and grid operation data in real time, the control command generation unit is configured to generate VSG system control commands based on optimal parameters, and the parameter adjustment unit is configured to dynamically update parameters according to active power fluctuations.

[0125] The experimental verification module is connected to the adaptive control module. The experimental verification module includes a wind power grid-side inverter experimental prototype and the RT-LAB hardware-in-the-loop simulation platform. It is configured to build a grid-connected system model to verify the accuracy of the model, the correctness of the mechanism analysis, and the effectiveness of the control strategy.

[0126] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A wind farm inertia damping adaptive control method integrating deep Q-networks, characterized in that, Used for inertia support and damping adjustment after wind farms are connected to the grid, including: Establish mathematical models of doubly-fed wind turbines and converters, and analyze the parameters of wind turbines, control parameters of the generator-side converter, control parameters of the grid-side converter, and the characteristics of the wind power generation system under different operating conditions. Based on the mathematical models of doubly fed wind turbines, converters, and characteristic analysis results, an interactive model for wind power system grid connection is established. Taking into account the actual physical parameters of the system, converter control parameters, wind power system operating status, and grid operating status, the stability of wind power system grid connection is judged by combining time-domain simulation and frequency-domain analysis. Based on the active stabilization capability of wind turbine generators, this study investigates the dynamic response characteristics of the VSG system providing inertial support under different conditions, and quantitatively analyzes the inertial response time and frequency regulation amplitude of the VSG system. By selecting rotational inertia and damping coefficient based on deep Q-networks, an improved damping inertia adaptive control strategy is developed to achieve synergy between wind turbine inertia support, frequency response capability enhancement, and continuous and smooth active power adjustment.

2. The wind farm inertia damping adaptive control method based on fused deep Q-networks as described in claim 1, characterized in that, Establishing a mathematical model for a doubly-fed wind turbine includes: Based on the electromagnetic coupling principle of a doubly-fed induction generator, voltage equations, flux linkage equations, and torque equations are constructed for the stator and rotor sides. A wind turbine aerodynamic model is introduced, and the mechanical power captured by the wind turbine is calculated by taking real-time wind speed as input and combining the tip speed ratio of the wind turbine blades and the wind energy utilization coefficient. The transmission relationship between the mechanical power of the wind turbine and the input power of the generator is established by using a gearbox transmission model, and finally a complete mathematical model of the doubly fed wind turbine unit covering the wind turbine, gearbox and generator is formed.

3. The wind farm inertia damping adaptive control method based on fused deep Q-networks as described in claim 1, characterized in that, Establishing a mathematical model for the converter specifically includes: The converter mathematical model includes a machine-side converter mathematical model and a grid-side converter mathematical model; The mathematical model of the generator-side converter aims to maintain the stability of the generator stator voltage and track the maximum wind energy capture. Based on the vector control principle in the rotating coordinate system, a dual closed-loop control model with an inner current loop and an outer power loop is constructed. The mapping relationship between the generator-side converter control parameters and the generator rotor current and active power output is clarified. The generator-side converter control parameters include current loop PI regulator parameters and power loop PI regulator parameters. The mathematical model of the grid-side converter aims to maintain the stability of the DC bus voltage and achieve unity power factor operation. It adopts a vector control strategy in a rotating coordinate system to construct a dual closed-loop control model with an inner voltage loop and an outer current loop. This study analyzes the impact of grid-side converter control parameters on the DC bus voltage fluctuation suppression capability and the grid-side reactive power compensation effect.

4. The wind farm inertia damping adaptive control method based on fused deep Q-networks as described in claim 1, characterized in that, The characteristics of wind power generation systems under different operating conditions are analyzed, specifically including: Set up operating scenarios for the wind power generation system, including stable operation at rated wind speed, low wind speed fluctuation, and high wind speed cut-out. For each operating scenario, real-time operating data of the wind turbine is collected through simulation tools. The real-time operating data includes wind turbine speed, generator stator current, generator stator voltage, active power output, and converter switch duty cycle. The collected data are analyzed for trends and fluctuations are calculated to determine the variation patterns of key parameters of wind turbines under different operating scenarios, and to determine the degree of impact of wind speed fluctuations and load changes on the operational stability of the wind power system.

5. The adaptive control method for wind farm inertia damping based on fused deep Q-networks as described in claim 1, characterized in that, After establishing an interaction model for wind power system grid connection, the grid connection stability is analyzed, specifically including: A simulation platform was built based on an interactive model to simulate normal grid operation, grid frequency deviation, and grid voltage drop conditions. Calculate the output power fluctuation coefficient, grid frequency recovery time, and voltage sag duration of the wind power system under different operating conditions; A grid connection stability evaluation index is set. If the calculation result does not meet the grid connection stability evaluation index, the wind power system is judged to be grid-connected and stable; otherwise, the grid connection is judged to be unstable, and the key influencing parameters that cause the instability are located. The key influencing parameters include the wind speed fluctuation amplitude and the converter control parameter deviation.

6. The wind farm inertia damping adaptive control method based on fused deep Q-networks as described in claim 1, characterized in that, The study investigates the dynamic response characteristics of the VSG system providing inertia support under different conditions, specifically including: A VSG system model was built in the simulation platform, and different test conditions were set, including power grid frequency step disturbance, load change and wind speed change. For each test condition, the inertia support response time, frequency regulation rate, and voltage stability of the VSG system were collected. A database of dynamic response characteristics of the VSG system is constructed based on the data collected for each test condition. The optimal operating parameters of the VSG system under different test conditions are determined by data fitting, and the optimal parameters are used as the initial input for the parameter optimization of the deep Q network.

7. The adaptive control method for wind farm inertia damping based on fused deep Q-networks as described in claim 1, characterized in that, The selection of rotational inertia and damping coefficient based on a deep Q-network specifically includes: A deep Q-network model is constructed, which includes an input layer, a hidden layer, and an output layer. The input state vector of the deep Q-network is determined, which includes wind power system operating state parameters, grid operating parameters, and historical regulation errors. Define a reward function for a deep Q-network, with the objectives of optimizing inertia support, minimizing frequency deviation, and minimizing active power fluctuation. The deep Q-network is trained, and the experience replay mechanism is used to store the input state, action, reward value and the next input state under different operating scenarios. The network parameters are updated by gradient descent until the network converges. After training, the real-time collected input state vector is input into the deep Q-network, and the network outputs the optimal moment of inertia and damping coefficient values ​​for the current scenario.

8. The adaptive control method for wind farm inertia damping based on fused deep Q-networks as described in claim 1, characterized in that, Improved damped inertia adaptive control strategies include: Real-time data collection of wind power system operating status and power grid operating parameters is achieved through sensors. The collected data is transmitted to the control unit, which then calls the trained deep Q-network to output the optimal moment of inertia and optimal damping coefficient for the current scenario. The control unit adjusts the control commands of the VSG system according to the optimal parameters: when the grid frequency deviates from the rated value, the virtual inertia of the VSG system is adjusted to the optimal moment of inertia first, and the damping coefficient is adjusted to the optimal damping coefficient simultaneously. The active power output curve is monitored in real time. If the curve fluctuates by more than ±2% for three consecutive acquisition cycles, the deep Q network is called again to update the optimal parameters.

9. The adaptive control method for wind farm inertia damping based on fused deep Q-networks as described in claim 1, characterized in that, It also includes a control effect verification step: A prototype wind power grid-side inverter was completed. The prototype includes a generator-side converter module, a grid-side converter module, a control module, and a data acquisition module. A grid-connected system model was built on a hardware-in-the-loop (HIPL) simulation platform based on RT-LAB. The model includes a wind farm simulation submodule, a power grid simulation submodule, and a data monitoring submodule. The experimental prototype was connected to a hardware-in-the-loop simulation platform to simulate typical operating conditions, including rated wind speed operation, wind speed fluctuation, and power grid frequency disturbance. The power grid frequency recovery time, active power fluctuation amplitude, and inertia support response time were collected under different typical operating conditions. If the grid frequency recovery time is ≤1.5s, the active power fluctuation amplitude is ≤±2%, and the inertia support response time is ≤0.3s, then the control strategy is deemed effective; otherwise, the cause of the deviation is analyzed, and the reward function weights or training iterations of the deep Q network are optimized until the control effect meets the requirements.

10. A wind farm inertia damping adaptive control system integrating deep Q-networks, characterized in that, The method for adaptive control of wind farm inertia damping for implementing the fused deep Q-network according to any one of claims 1-9 includes: The model building module is configured to establish mathematical models of doubly fed wind turbines, machine-side converters, grid-side converters, and interactive models of wind power systems connected to the grid, and to store model parameters and simulation data. The characteristic analysis module connects to the model building module and is configured to analyze the characteristics of wind turbine parameters, converter control parameters, and wind power generation systems under different operating conditions, and output a characteristic analysis report. The VSG characteristic research module, connected to the characteristic analysis module, is configured to study the dynamic response characteristics of the VSG system under different conditions and output the optimal initial operating parameters of the VSG system. A deep Q-network optimization module is connected to the VSG characteristic research module. The deep Q-network optimization module includes a network training unit and a parameter output unit. The network training unit is configured to build and train a deep Q-network, and the parameter output unit is configured to output the optimal moment of inertia and damping coefficient under different operating scenarios. An adaptive control module is connected to a deep Q-network optimization module. The adaptive control module includes a data acquisition unit, a control command generation unit, and a parameter adjustment unit. Among them, the data acquisition unit is configured to collect wind power system and grid operation data in real time, the control command generation unit is configured to generate VSG system control commands based on optimal parameters, and the parameter adjustment unit is configured to dynamically update parameters according to active power fluctuations. The experimental verification module is connected to the adaptive control module. The experimental verification module includes a wind power grid-side inverter experimental prototype and an RT-LAB hardware-in-the-loop simulation platform, configured to build a grid-connected system model to verify the model's accuracy, the correctness of the mechanism analysis, and the effectiveness of the control strategy.

Images