Method and device for controlling stability of wind power generation system

Abstract

The invention provides a wind power generation system stability control method and device, and the method comprises the steps: obtaining the state data of a wind power generation system, and obtaining a probability density function through a kernel density estimation method based on the state data; based on the probability density function, determining an optimal parameter value of a fan pitch angle of the wind power generation system through a Bayesian reinforcement learning method; and based on the optimal parameter value, the fan pitch angle of the wind power generation system is controlled, and the stability of the wind power generation system is recovered. Therefore, the calculation cost is reduced, and the reliability and adaptability of the system in practical application are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation systems, and in particular to a method and device for controlling the stability of a wind power generation system. Background Art

[0002] In view of the significant uncertainty and randomness inherent in wind power generation systems, probabilistic stability analysis methods have received extensive attention and research. Currently, many studies have proposed methods for dealing with uncertainty, such as scenario optimization and Monte Carlo methods, probabilistic methods such as maximum entropy theory and Gaussian process simulators, and reinforcement learning methods combined with models such as deep neural networks.

[0003] Although scenario optimization and Monte Carlo methods can handle uncertainty, the limited sampling of scenario optimization and the high computational cost of Monte Carlo limit their application in large-scale systems. Secondly, although probabilistic methods such as maximum entropy theory and Gaussian process simulators perform well in specific scenarios, their adaptability and accuracy are still insufficient in complex wind power systems. In addition, traditional stability analysis methods gradually show limitations when dealing with the inherent variability and dynamic changes of wind power generation, and it is difficult to effectively capture small signal disturbances. Although reinforcement learning shows potential in wind power control, existing methods have high computational complexity and lack verification in large-scale systems. Summary of the invention

[0004] The present invention provides a wind power generation system stability control method and device, which are used to solve the defects of the prior art that the wind power generation control stability process has high calculation complexity and lacks verification in large-scale systems, thereby reducing the calculation cost and improving the reliability and adaptability of the system in practical applications.

[0005] The present invention provides a wind power generation system stability control method, comprising the following steps: Acquire state data of the wind power generation system, and obtain a probability density function based on the state data by a kernel density estimation method; Based on the probability density function, determining the optimal parameter value of the wind turbine pitch angle of the wind power generation system by using a Bayesian reinforcement learning method; Based on the optimal parameter value, the pitch angle of the wind turbine of the wind power generation system is controlled to restore the stability of the wind power generation system.

[0006] According to a wind power generation system stability control method provided by the present invention, based on the probability density function, the optimal parameter value of the wind turbine pitch angle of the wind power generation system is determined by a Bayesian reinforcement learning method, specifically comprising: Modeling is performed through a Markov decision process, taking the current output power of the wind power generation system as a state, taking the change of the wind turbine pitch angle as an action, and determining an action space, a state transfer function, and a reward function; Based on the probability density function, determining a prior distribution; A deep Q network is used as an agent, and a Bayesian reinforcement learning model is trained based on the prior distribution, the action space, the state transfer function and the reward function until a termination condition is reached, thereby obtaining an optimal parameter value of a wind turbine pitch angle of the wind power generation system.

[0007] According to a wind power generation system stability control method provided by the present invention, the reward function is determined according to the following formula: In the formula, is the reward value, is the current output power, Indicates rated output power, variable Indicates the allowable power error range, Indicates the adjustment range of the wind turbine pitch angle, It is the maximum allowable adjustment range of the wind turbine pitch angle.

[0008] According to a wind power generation system stability control method provided by the present invention, the state transfer function is determined according to the following formula: In the formula, The time step is The output power, The time step is The output power, is the control coefficient of the wind turbine pitch angle, The time step is The action selected.

[0009] According to a wind power generation system stability control method provided by the present invention, the value range of the action space is .

[0010] According to a wind power generation system stability control method provided by the present invention, based on the state data, a probability density function is obtained by a kernel density estimation method, which specifically includes: Preprocessing the state data and calculating active output power data to obtain preprocessed state data; Based on the preprocessed state data, a probability density function is obtained through a kernel density estimation algorithm.

[0011] The present invention also provides a wind power generation system stability control device, comprising the following modules: A processing module, used for acquiring state data of the wind power generation system, and obtaining a probability density function based on the state data by a kernel density estimation method; A determination module, used to determine the optimal parameter value of the wind turbine pitch angle of the wind power generation system through a Bayesian reinforcement learning method based on the probability density function; A control module is used to control the pitch angle of the wind turbine of the wind power generation system based on the optimal parameter value to restore the stability of the wind power generation system.

[0012] According to a wind power generation system stability control device provided by the present invention, the determination module is specifically used for: Modeling is performed through a Markov decision process, taking the current output power of the wind power generation system as a state, taking the change of the wind turbine pitch angle as an action, and determining an action space, a state transfer function, and a reward function; Based on the probability density function, determining a prior distribution; A deep Q network is used as an agent, and a Bayesian reinforcement learning model is trained based on the prior distribution, the action space, the state transfer function and the reward function until a termination condition is reached, thereby obtaining an optimal parameter value of a wind turbine pitch angle of the wind power generation system.

[0013] According to a wind power generation system stability control device provided by the present invention, the reward function is determined according to the following formula: In the formula, is the reward value, is the current output power, Indicates rated output power, variable Indicates the allowable power error range, Indicates the adjustment range of the wind turbine pitch angle, It is the maximum allowable adjustment range of the wind turbine pitch angle.

[0014] According to a wind power generation system stability control device provided by the present invention, the state transfer function is determined according to the following formula: In the formula, The time step is The output power, The time step is The output power, is the control coefficient of the wind turbine pitch angle, The time step is The action selected.

[0015] According to a wind power generation system stability control device provided by the present invention, the value range of the action space is .

[0016] According to a wind power generation system stability control device provided by the present invention, the processing module is specifically used for: Preprocessing the state data and calculating active output power data to obtain preprocessed state data; Based on the preprocessed state data, a probability density function is obtained through a kernel density estimation algorithm.

[0017] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a wind power generation system stability control method as described above is implemented.

[0018] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the stability control method of a wind power generation system as described in any one of the above is implemented.

[0019] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for controlling the stability of a wind power generation system.

[0020] The wind power generation system stability control method and device provided by the present invention obtain the probability density function of the wind power generation system state through the kernel density estimation method, accurately quantify the uncertainty, and then determine the optimal parameter value of the wind turbine pitch angle of the wind power generation system through the Bayesian reinforcement learning method. Then, according to the optimal parameter value, the wind turbine pitch angle control can be used to optimize the system stability, which not only reduces the calculation cost, but also improves the reliability and adaptability of the system in practical applications, and makes up for the limitations of the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 It is a flow chart of the wind power generation system stability control method provided by the present invention.

[0023] Figure 2It is a flow chart of the probabilistic stability and optimization control method under small signal interference provided by the present invention.

[0024] Figure 3 It is a flow chart of propeller angle control based on BRL algorithm provided by the present invention.

[0025] Figure 4 It is a schematic diagram of the main features of the simulation data set provided by the present invention.

[0026] Figure 5 It is a PDF schematic diagram of the frequency data provided by the present invention.

[0027] Figure 6 It is a PDF schematic diagram of voltage data provided by the present invention.

[0028] Figure 7 It is a PDF schematic diagram of the current data provided by the present invention.

[0029] Figure 8 It is a PDF schematic diagram of power data provided by the present invention.

[0030] Fig. 9 It is a schematic diagram of power output corresponding to the number of iterations of different methods provided by the present invention.

[0031] Fig.10 It is a schematic diagram of the Q value corresponding to the number of iterations of the BRL algorithm provided by the present invention.

[0032] Fig.11 It is a schematic diagram of the simulation results of the blade angle control provided by the present invention.

[0033] Fig.12 It is a structural schematic diagram of a wind power generation system stability control device provided by the present invention.

[0034] Fig.13 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] Figure 1 FIG. 1 is a flow chart of a method for controlling the stability of a wind power generation system provided by the present invention. Figure 1 As shown, the method comprises the following steps: Step 100: Acquire state data of the wind power generation system, and obtain a probability density function based on the state data by using a kernel density estimation method.

[0037] Specifically, in the embodiment of the present invention, the state data of the wind power generation system can be first obtained, and the state data can include frequency, voltage, current, power and other data. The state data can be obtained through a phasor measurement unit (PMU), which is a high-precision measurement device used for power system monitoring and control, and can measure the phasor (i.e., amplitude and phase) of voltage and current in the power system in real time, and transmit these data to the monitoring and control system.

[0038] After obtaining the state data of the wind power generation system, the probability density function (PDF) of each parameter of the state data can be obtained through the kernel density estimation (KDE) method, such as the probability density function of frequency, the probability density function of voltage, the probability density function of current, the probability density function of power, etc.

[0039] KDE is a non-parametric statistical method that forms a continuous probability density curve by placing a smooth kernel function on sample data points to estimate the probability density function of a random variable.

[0040] Optionally, based on the state data, a probability density function is obtained by a kernel density estimation method, specifically including: Preprocessing the state data and calculating the active output power data to obtain the preprocessed state data; Based on the preprocessed state data, the probability density function is obtained through the kernel density estimation algorithm.

[0041] Specifically, after acquiring the status data, the status data may be preprocessed by identifying and removing abnormal data, and then relevant features may be extracted.

[0042] Since the status data that can generally be directly obtained only include voltage, current and frequency, the active output power can be calculated based on the current and voltage and the power factor (or the phase relationship between voltage and current) for further analysis.

[0043] For example, in the process of calculating active power, the voltage and current phase angles can be converted from degrees to radians to ensure the accuracy of subsequent power calculations. The conversion formula is as follows: In the formula, and denote the voltage and current phase angles in radians, respectively, and These phase angles are then used to calculate the active power by : In the formula, is the voltage amplitude, is the current amplitude.

[0044] After preprocessing the state data and calculating the active output power data, the preprocessed state data can be obtained. At this time, kernel density estimation can be performed on the preprocessed state data to obtain the probability density function of frequency, the probability density function of voltage, the probability density function of current, and the probability density function of power.

[0045] Step 101: Based on the probability density function, the optimal parameter value of the wind turbine pitch angle of the wind power generation system is determined by using a Bayesian reinforcement learning method.

[0046] Specifically, in a wind power generation system, the pitch angle is an important parameter for adjusting the angle of the wind turbine blades relative to the wind to optimize the conversion of wind energy into electrical energy. In an embodiment of the present invention, the optimal parameter value of the wind turbine pitch angle is determined by a Bayesian Reinforcement Learning (BRL) method.

[0047] After obtaining the probability density function corresponding to the state data through the kernel density estimation method, the prior distribution of the wind turbine pitch angle can be set according to the probability density function. Then, in each interaction, the posterior distribution is updated using the Bayesian theorem according to the observation results of the model. After multiple iterations, when the termination condition is reached, the optimal parameter value of the wind turbine pitch angle of the wind power generation system can be obtained.

[0048] Step 102: Based on the optimal parameter value, the pitch angle of the wind turbine of the wind power generation system is controlled to restore the stability of the wind power generation system.

[0049] Specifically, after determining the optimal parameter value of the wind turbine pitch angle of the wind power generation system according to the Bayesian reinforcement learning method, the wind turbine pitch angle of the wind power generation system can be controlled according to the optimal parameter value, thereby restoring the stability of the wind power generation system.

[0050] The wind power generation system stability control method provided by the present invention obtains the probability density function of the wind power generation system state through the kernel density estimation method, accurately quantifies the uncertainty, and then determines the optimal parameter value of the wind turbine pitch angle of the wind power generation system through the Bayesian reinforcement learning method. Then, according to the optimal parameter value, the wind turbine pitch angle control can be used to optimize the system stability, which not only reduces the calculation cost, but also improves the reliability and adaptability of the system in practical applications, and makes up for the limitations of the existing technology.

[0051] According to a wind power generation system stability control method provided by the present invention, based on a probability density function, an optimal parameter value of a wind turbine pitch angle of the wind power generation system is determined by a Bayesian reinforcement learning method, specifically comprising: Modeling is performed through the Markov decision process, taking the current output power of the wind power generation system as the state and the change of the wind turbine pitch angle as the action, and determining the action space, state transfer function and reward function; Based on the probability density function, determine the prior distribution; A deep Q network is used as an agent. Based on the prior distribution, action space, state transfer function and reward function, a Bayesian reinforcement learning model is trained until the termination condition is reached, and the optimal parameter value of the wind turbine pitch angle of the wind power generation system is obtained.

[0052] Specifically, in the process of determining the optimal parameter value of the wind turbine pitch angle of the wind power generation system through the Bayesian reinforcement learning method, an embodiment of the present invention adopts a deep Q network (Deep Q-Network, DQN) agent based on reinforcement learning, and uses an A2C (Actor-Critic) network to control the wind turbine pitch angle of the wind power generation system.

[0053] DQN is an algorithm that combines deep learning and reinforcement learning, designed to solve high-dimensional state space problems in reinforcement learning. DQN uses deep neural networks to approximate value functions, enabling agents to learn and make decisions in complex environments.

[0054] The reinforcement learning framework first needs to be modeled through the Markov Decision Process (MDP). The details of the modeling process are as follows: For state space: state Represents the agent at a specific time step Observation, state space All possible states In this embodiment of the present invention, the current output power of the wind power generation system is As a state, it is represented as .

[0055] For the action space: Action Represents the agent at a specific time step Choices made, action space Refers to the set of all possible actions that an agent can take. Since small changes in the angle of a wind turbine blade can significantly affect power output and even cause oscillations and power fluctuations, wind turbine blade control systems need to be precisely adjusted to maintain stable power output.

[0056] Optionally, the action space ranges from , expressed as , which represent five possible actions: turn left 0.1 degrees, turn left 0.05 degrees, stay still, turn right 0.05 degrees, and turn right 0.1 degrees.

[0057] For the state transfer function, it describes the relationship between the output power and the pitch angle adjustment of the wind turbine. Optionally, the state transfer function is determined according to the following formula: In the formula, The time step is The output power, The time step is The output power, is the control coefficient of the wind turbine pitch angle, The time step is The action selected.

[0058] Regarding the reward function: In the embodiment of the present invention, the reward function The difference between the output power and the normal operating power can be considered, as well as the cost of adjusting the pitch angle.

[0059] In order to achieve optimal control of the rotor angle control, it is necessary to pursue long-term minimization of the reward. The reward function aims to keep the output power close to the set value while minimizing the adjustment of the pitch angle. Optionally, the reward function is determined according to the following formula: In the formula, is the reward value, is the current output power, Indicates rated output power, variable Indicates the allowable power error range, Indicates the adjustment range of the wind turbine pitch angle, It is the maximum allowable adjustment range of the wind turbine pitch angle.

[0060] After modeling through the Markov decision process, the prior distribution of the wind turbine pitch angle in the Bayesian framework can be set according to the determined probability density function.

[0061] After setting the prior distribution of the wind turbine pitch angle, the deep Q network can be used to select the next action based on the action space, current state and current strategy in the Markov decision process, and then the reward function is used to calculate the Q value. The strategy is then continuously updated based on the state transfer function through Bayesian inference.

[0062] During this process, through interaction with the wind power generation environment, the intelligent agent selects the pitch angle and receives rewards, continuously iterates learning, and updates strategies.

[0063] After training the Bayesian-based reinforcement learning model, the wind power system can restore the output power to the normal operating mode by optimizing the pitch angle adjustment, while dealing with the imbalance of system stability caused by small signal interference. By combining the advantages of Bayesian inference and reinforcement learning, the optimal parameter value of the wind turbine pitch angle in the wind power system can be effectively determined, thereby improving the wind energy conversion efficiency. This method can not only adapt to environmental changes, but also continuously optimize decisions based on historical data.

[0064] The wind power generation system stability control method provided by the present invention is further explained below through embodiments in specific application scenarios.

[0065] Figure 2 Schematic diagram of the flow of the probability stability and optimization control method under small signal interference provided by the present invention. Figure 2 As shown, this embodiment establishes a probabilistic analysis framework for small signal reasoning through the KDE method to quantify the uncertainty in a random environment and gives the PDFs of frequency, voltage, current and power. On this basis, this embodiment proposes a method based on Bayesian reinforcement learning to optimize the probabilistic stability of the system through wind turbine pitch angle control. Through the updated strategy, the actor network selects new actions and continues to iterate until the reward reaches the minimum value.

[0066] In this embodiment, a reinforcement learning-based DQN agent is used to control the pitch angle in the IEEE 10 generator 39-node system using an A2C network. The agent learns and optimizes its control strategy through interaction with the environment to restore system stability under low-probability disturbance conditions. The reinforcement learning framework is modeled through MDP, and its details are as follows: 1. State space: state Represents the agent at a specific time step Observation, state space Refers to all possible states The current output power of the system As a state, it is represented as .

[0067] 2. Action Space: Action Represents the agent at a specific time step Choices made, action space Refers to the set of all possible actions that the agent can take. Since small changes in the blade angle can significantly affect the power output and even cause oscillations and power fluctuations, the blade control system needs to be precisely adjusted to maintain stable power output. The action space is represented by , indicating five possible actions: turn left 0.1 or 0.05 degrees, stay still, or turn right 0.05 or 0.1 degrees. The state transfer function describes the relationship between output power and pitch angle adjustment, and is expressed as As coefficient for rotor angle control: 3. Reward function: Considering the difference between the output power and the normal operating power and the cost of adjusting the pitch angle, the reward function It is expressed as: in, is the current output power, Indicates the rated output power. represents the allowable power error range, and Indicates the adjustment range of the pitch angle. In addition, Defines the maximum permissible adjustment range.

[0068] In order to achieve optimal control of the rotor angle control, it is necessary to pursue the long-term minimization of the reward. The reward function aims to keep the output power close to the set value while minimizing the adjustment of the pitch angle.

[0069] The step-by-step framework of this program is as follows: Step 1: Data preprocessing and power calculation. First, data preprocessing is performed by identifying and removing abnormal data, and then relevant features are extracted. Since the given data only contains voltage, current, and frequency, the active output power is first calculated for further analysis.

[0070] Step 2: Plot the PDF of each variable. Based on the KDE method, the PDF of current, voltage, frequency, and power are plotted using the preprocessed data to understand the distribution of values.

[0071] Step 3: Wind turbine rotor angle control based on Bayesian reinforcement learning. After PDF analysis, the Bayesian reinforcement learning algorithm is used to control the wind turbine pitch angle. Figure 3 It is a flow chart of propeller angle control based on BRL algorithm provided by the present invention.

[0072] like Figure 3As shown in Figure 1, the prior distribution is obtained before reinforcement learning training. The state consists of the output power, and the action involves adjusting the pitch angle. The reinforcement learning model restores the output power to the normal operating mode by optimizing the pitch angle adjustment, while coping with the imbalance of system stability caused by small signal interference.

[0073] Step 4: Comparison of control performance with the baseline scheme. The control method based on Bayesian reinforcement learning is compared with the A2C network and the traditional Monte Carlo method. The convergence of the output power curve shows the efficiency of BRL in pitch angle control, while the angle adjustment chart shows the accuracy of the proposed algorithm.

[0074] The technical effect of the method is described below through specific simulation results. The simulation data is provided by China Southern Power Grid. Table 1 shows the detailed simulation factors provided by the present invention.

[0075] Table 1 Simulation parameters

[0076] In addition to the parameters described in the table, the control coefficient of the relationship between the blade angle change and the power output change is set to 0.1. The allowable error range of the output power deviation from the set value is 5. In order to stabilize the training, the target network update frequency in the BRL algorithm is 250, and the total number of time steps for training the agent is 50,000.

[0077] After abnormal data cleaning and feature extraction, the final simulation results show the maximum amplitude, indicating the main characteristics of the data set. Since the data set consists of 35 data lines, frequency, amplitude, voltage and rotor angle are considered. Figure 4 It is a schematic diagram of the main features of the simulation data set provided by the present invention, such as Figure 4 The scatter plot of dominant frequency and amplitude shown indicates a positive correlation between frequency and amplitude. Most of the data points are concentrated in the low amplitude and low frequency range, while outliers are distributed in the high amplitude and high frequency range, indicating deviations or anomalies, reflecting the inherent randomness and uncertainty in the operation of wind power systems.

[0078] In order to quantify the uncertainty associated with small signal probabilistic stability, this example first uses the KDE method to analyze the PDFs of voltage, current, frequency, and power. By analyzing the data of 35 lines, the distribution function of each variable is plotted. The output power can be calculated as follows: In order to further calculate the active power, the phase angle needs to be converted from degrees to radians to ensure the accuracy of subsequent power calculations. The conversion formula is as follows: In the formula, and denote the voltage and current phase angles in radians, respectively, and These phase angles are then used to calculate the active power by : In the formula, is the voltage amplitude, is the current amplitude.

[0079] Once the calculation is done, the PDF of the power can be plotted using the KDE method.

[0080] After analyzing the probability distribution of each parameter, the system stability can be further analyzed to improve the operational stability and reliability of the wind power system under study.

[0081] Figure 5 It is a PDF diagram of the frequency data provided by the present invention. Figure 5 It can be seen that the solid blue line represents the PDF of the frequency data, the red dashed line indicates the average frequency value of 59.99, the green dashed line marks the peak position of the PDF at 60.00, and the blue dashed line shows the median frequency at 60.01. The sharp peak at 60 Hz and minimal spread indicate that the system maintains a stable frequency with very small and insignificant deviations. Considering that the PDFs of voltage and current show large fluctuations, this may indicate the presence of small perturbations within the system. Therefore, it is necessary to further analyze the PDF of the system power to gain a more complete understanding of the stability of the system.

[0082] Figure 6 It is a PDF diagram of voltage data provided by the present invention. Figure 6 As can be seen, the blue solid line represents the PDF of the voltage data, showing the probability density distribution of different voltage values ​​in the entire data set. The red dashed line indicates that the average voltage is 205.36, reflecting the average voltage level of the system. The green dashed line marks the peak of the PDF at 204.90, indicating that voltage values ​​close to this point are most frequent. The blue dashed line shows the median voltage of 205.23. The proximity of the median to the mean indicates that the data distribution is relatively symmetrical. However, the presence of small peaks and skew indicates occasional fluctuations, suggesting instability.

[0083] Figure 7 It is a PDF diagram of the current data provided by the present invention. Figure 7It can be seen that the solid blue line represents the PDF of the current data, showing the probability density distribution of different current values ​​in the entire data set. The red dashed line indicates that the average current value is 400.74, reflecting the average current level of the system. The green dashed line marks the peak of the PDF at 217.63, indicating that current values ​​near this level appear most frequently. The blue dashed line shows that the median current value is 245.29, indicating that there is a significant difference between the median and the mean, indicating a positively skewed distribution. The presence of multiple smaller peaks indicates that there may be fluctuations within the system, suggesting instability.

[0084] Figure 8 It is a PDF schematic diagram of power data provided by the present invention. Figure 8 The blue solid line in the figure represents the PDF of the power data. The red dashed line indicates the average power value of 82,295.17 watts. The green dashed line marks the peak position of the PDF, which is 44,863.57 watts. The blue dashed line marks the median power value of 50,320.64 watts. The power density distribution shown in the figure shows multiple significant peaks and fluctuations, indicating that there is a certain degree of variation in the power output. Although the average power is relatively high, the peak and median are relatively low, which may indicate the presence of power fluctuations or extreme values. The instability of power output as a function of voltage and current directly reflects the fluctuations in these variables, which means that any changes in these factors may affect the stability of power output. This suggests that there may be dynamic changes and slight disturbances in the system, which may be caused by changes in external conditions or internal control adjustments. The detailed values ​​of the four PDFs are shown in Table 2.

[0085] Table 2 PDF values

[0086] First, the PDF analysis of the given frequency data was performed using KDE, and the results showed that despite macroscopic stability, small fluctuations and instabilities still existed. Subsequently, the PDFs of the voltage and current data were depicted using KDE. By analyzing their peak, mean, and median values, it was shown that there were instabilities in the system caused by small disturbances. In order to highlight the obvious changes and their relationship trends, the phase angle relationship was introduced and the PDF of the actual power was further depicted. This further illustrates the significant impact of small disturbances on the system power output, especially through multiple fluctuation peaks and large variances in the power distribution, reflecting the impact of voltage and current fluctuations on system stability. In order to deal with the small disturbance problem, this study proposes to use a BRL-based algorithm to optimize the blade angle control, and selects A2C and Monte Carlo methods as comparison schemes.

[0087] Fig. 9 Schematic diagram of power output corresponding to the number of iterations of different methods provided by the present invention. Fig. 9It can be clearly seen that the red curve represents BRL, the green curve represents A2C, and the blue curve represents MC. Overall, the power output of BRL and A2C tends to the rated operating power value as the number of iterations increases, while MC shows a downward trend during training. It is worth noting that BRL converges and reaches the target power level at about 250 iterations, while A2C only reaches the desired output near the end of the iteration. This shows that the BRL-based method is highly efficient in adjusting the blade angle to maintain the optimal power output under small disturbance conditions.

[0088] Table 3 Output power values ​​of three methods

[0089] Fig.10 Schematic diagram of the Q value corresponding to the number of iterations of the BRL algorithm provided by the present invention. Fig.10 Overall, the Q value initially drops rapidly and then gradually stabilizes. Convergence is reached at about 250 iterations, and the Q value stabilizes at around -220. This shows that the algorithm has successfully learned a relatively stable strategy and continuously optimized it throughout the iteration process, showing a strong learning ability. The continuation of small fluctuations after convergence shows that the algorithm maintains a certain degree of exploratory nature. See Table 4 for detailed values.

[0090] Table 4 Relationship between Q value and number of iterations

[0091] Fig.11 Schematic diagram of the simulation results of the blade angle control provided by the present invention. Fig.11 As shown in Figure 5, the system continuously makes fine adjustments in the range of -0.1 to 0.1 degrees to maintain effective regulation of the output power, thereby ensuring the stability of the system in the presence of small disturbances. The detailed values ​​of the angle control are shown in Table 5, which further verifies the effectiveness of the proposed BRL method for blade control in complex environments.

[0092] Table 5 Relationship between blade angle control and iteration number

[0093] In view of the uncertainty and instability of wind power generation systems in random environments, this embodiment proposes a probabilistic analysis using KDE to depict the PDFs of frequency, voltage, current, and power in the PMU data set. The existence of small signal disturbances can be confirmed from the figure, so an optimal control strategy based on Bayesian reinforcement learning is proposed to reduce the negative impact caused by such small signal disturbances. Taking the output power as an indicator of system stability, the proposed method can converge faster and more stably compared with benchmarks such as A2C networks and traditional Monte Carlo methods. The angle adjustment diagram also shows the precise control of the proposed method in angle fine-tuning. The simulation diagrams and tables generated based on the IEEE 39-node system show the efficiency of the proposed method in terms of convergence compared to the benchmark scheme. The study confirms the importance of applying the new algorithm to the power stability analysis of wind power generation systems, and provides new insights into the field of new energy and power production.

[0094] The wind power generation system stability control device provided by the present invention is described below. The wind power generation system stability control device described below and the wind power generation system stability control method described above can be referred to each other.

[0095] Fig.12 Schematic diagram of the structure of the wind power generation system stability control device provided by the present invention. Fig.12 As shown, the device includes the following modules: The processing module 1200 is used to obtain the state data of the wind power generation system, and obtain the probability density function based on the state data by using the kernel density estimation method; A determination module 1210 is used to determine the optimal parameter value of the pitch angle of the wind turbine of the wind power generation system based on the probability density function by using a Bayesian reinforcement learning method; The control module 1220 is used to control the pitch angle of the wind turbine of the wind power generation system based on the optimal parameter value to restore the stability of the wind power generation system.

[0096] According to a wind power generation system stability control device provided by the present invention, the determination module 1410 is specifically used for: Modeling is performed through the Markov decision process, taking the current output power of the wind power generation system as the state and the change of the wind turbine pitch angle as the action, and determining the action space, state transfer function and reward function; Based on the probability density function, determine the prior distribution; A deep Q network is used as an agent. Based on the prior distribution, action space, state transfer function and reward function, a Bayesian reinforcement learning model is trained until the termination condition is reached, and the optimal parameter value of the wind turbine pitch angle of the wind power generation system is obtained.

[0097] According to a wind power generation system stability control device provided by the present invention, the reward function is determined according to the following formula: In the formula, is the reward value, is the current output power, Indicates rated output power, variable Indicates the allowable power error range, Indicates the adjustment range of the wind turbine pitch angle, It is the maximum allowable adjustment range of the wind turbine pitch angle.

[0098] According to a wind power generation system stability control device provided by the present invention, the state transfer function is determined according to the following formula: In the formula, The time step is The output power, The time step is The output power, is the control coefficient of the wind turbine pitch angle, The time step is The action selected.

[0099] According to a wind power generation system stability control device provided by the present invention, the value range of the action space is .

[0100] According to a wind power generation system stability control device provided by the present invention, the processing module 1200 is specifically used for: Preprocessing the state data and calculating the active output power data to obtain the preprocessed state data; Based on the preprocessed state data, the probability density function is obtained through the kernel density estimation algorithm.

[0101] Fig.13 A schematic diagram of the structure of an electronic device provided by the present invention, such as Fig.13 As shown, the electronic device may include: a processor 1310, a communications interface 1320, a memory 1330 and a communication bus 1340, wherein the processor 1310, the communications interface 1320 and the memory 1330 communicate with each other through the communication bus 1340. The processor 1310 may call the logic instructions in the memory 1330 to execute the wind power generation system stability control method, which includes the following steps: Acquire state data of the wind power generation system, and obtain a probability density function based on the state data by using a kernel density estimation method; Based on the probability density function, the optimal parameter value of the wind turbine pitch angle of the wind power generation system is determined by the Bayesian reinforcement learning method; Based on the optimal parameter value, the pitch angle of the wind turbine of the wind power generation system is controlled to restore the stability of the wind power generation system.

[0102] In addition, the logic instructions in the above-mentioned memory 1330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0103] On the other hand, the present invention further provides a computer program product, the computer program product comprising a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the wind power generation system stability control method provided by the above methods, the method comprising the following steps: Acquire state data of the wind power generation system, and obtain a probability density function based on the state data by using a kernel density estimation method; Based on the probability density function, the optimal parameter value of the wind turbine pitch angle of the wind power generation system is determined by the Bayesian reinforcement learning method; Based on the optimal parameter value, the pitch angle of the wind turbine of the wind power generation system is controlled to restore the stability of the wind power generation system.

[0104] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for controlling the stability of a wind power generation system provided by the above methods is implemented, and the method comprises the following steps: Acquire state data of the wind power generation system, and obtain a probability density function based on the state data by using a kernel density estimation method; Based on the probability density function, the optimal parameter value of the wind turbine pitch angle of the wind power generation system is determined by the Bayesian reinforcement learning method; Based on the optimal parameter value, the pitch angle of the wind turbine of the wind power generation system is controlled to restore the stability of the wind power generation system.

[0105] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0106] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wind power generation system stability control method, characterized in that: include: Acquire state data of the wind power generation system, and obtain a probability density function based on the state data by a kernel density estimation method; Based on the probability density function, determining the optimal parameter value of the wind turbine pitch angle of the wind power generation system by using a Bayesian reinforcement learning method; Based on the optimal parameter value, the pitch angle of the wind turbine of the wind power generation system is controlled to restore the stability of the wind power generation system.

2. The wind power generation system stability control method according to claim 1, characterized in that: Based on the probability density function, the optimal parameter value of the wind turbine pitch angle of the wind power generation system is determined by a Bayesian reinforcement learning method, specifically including: Modeling is performed through a Markov decision process, taking the current output power of the wind power generation system as a state, taking the change of the wind turbine pitch angle as an action, and determining an action space, a state transfer function, and a reward function; Based on the probability density function, determining a prior distribution; A deep Q network is used as an agent, and a Bayesian reinforcement learning model is trained based on the prior distribution, the action space, the state transfer function and the reward function until a termination condition is reached, thereby obtaining an optimal parameter value of a wind turbine pitch angle of the wind power generation system.

3. The wind power generation system stability control method according to claim 2, characterized in that: The reward function is determined according to the following formula: In the formula, is the reward value, is the current output power, Indicates rated output power, variable Indicates the allowable power error range, Indicates the adjustment range of the wind turbine pitch angle, It is the maximum allowable adjustment range of the wind turbine pitch angle.

4. The wind power generation system stability control method according to claim 2, characterized in that: The state transfer function is determined according to the following formula: In the formula, The time step is The output power, The time step is The output power, is the control coefficient of the wind turbine pitch angle, The time step is The action selected.

5. The wind power generation system stability control method according to claim 2, characterized in that: The value range of the action space is .

6. The wind power generation system stability control method according to claim 1, characterized in that: Based on the state data, a probability density function is obtained by a kernel density estimation method, specifically including: Preprocessing the state data and calculating active output power data to obtain preprocessed state data; Based on the preprocessed state data, a probability density function is obtained through a kernel density estimation algorithm.

7. A wind power generation system stability control device, characterized in that: include: A processing module, used for acquiring state data of the wind power generation system, and obtaining a probability density function based on the state data by a kernel density estimation method; A determination module, used to determine the optimal parameter value of the wind turbine pitch angle of the wind power generation system through a Bayesian reinforcement learning method based on the probability density function; A control module is used to control the pitch angle of the wind turbine of the wind power generation system based on the optimal parameter value to restore the stability of the wind power generation system.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the wind power generation system stability control method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the wind power generation system stability control method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the wind power generation system stability control method according to any one of claims 1 to 6 is implemented.

Images