Low-frequency oscillation online suppression method and device for wide-area additional damper of doubly-fed fan based on reinforcement learning algorithm

Abstract

The invention discloses a reinforcement learning algorithm-based low-frequency oscillation online suppression method for a double-fed fan wide-area additional damper, and the method comprises the following steps: obtaining the operation data of a wind power generation system, including the wind speed, the fan output power, the rotating speed and the power grid frequency; preprocessing the acquired data, and performing detrending and normalization processing by using a band-pass filter; using a depth deterministic strategy gradient algorithm to train the model to predict the operation state of the fan and the oscillation condition of the system; based on the trained depth deterministic strategy gradient algorithm model, an optimal damping control strategy is generated, and operation parameters of the draught fan are adjusted in real time; the generated damping control strategy is applied to a fan control system, the damping characteristic of the system is improved, and low-frequency oscillation is reduced.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method and apparatus for online suppression of low-frequency oscillations in a wide-area additional damper for a doubly fed wind turbine based on a reinforcement learning algorithm. Background Technology

[0002] With the rapid development of wind power globally, wind turbines, especially doubly-fed induction generators (DFIGs), have become the mainstream model in the wind power industry. DFIGs are connected to the grid via power electronic devices, and their back-to-back converter control decouples the turbine speed from the grid frequency, hiding the system's rotational kinetic energy and resulting in insufficient damping. This characteristic, coupled with the randomness and volatility of wind energy, significantly increases the risk of low-frequency oscillations in the power system. Modern power systems, due to their increased scale, large-capacity long-distance transmission, and the widespread use of fast excitation equipment, inherently possess the risk of low-frequency oscillations. The integration of wind power exacerbates this risk. Traditional synchronous generator sets, because they can be directly connected to the grid and generate damping torque through the generator's rotor and stator circuits, effectively suppress system oscillations. However, wind turbines are connected to the grid via power electronic devices, whose internal resistance is low, contributing less to oscillation suppression, posing new challenges to system stability.

[0003] To address these challenges, a wide-area damping controller for doubly-fed induction generator (DFIG) wind turbines based on reinforcement learning algorithms has emerged. Reinforcement learning algorithms are highly adaptive and capable of autonomously learning and optimizing in complex environments, making them particularly suitable for applications with high uncertainty, such as wind power systems. Through reinforcement learning, the controller can adjust the turbine's operating parameters in real time, improving system damping characteristics, reducing low-frequency oscillations, and ensuring the stable operation of the wind power system.

[0004] Application content

[0005] The purpose of this invention is to provide an online method for suppressing low-frequency oscillations in a wide-area additional damper for doubly-fed induction generator (DFIG) wind turbines based on reinforcement learning algorithms, thereby improving the stability and reliability of power systems. This invention utilizes reinforcement learning algorithms to adaptively adjust the operating parameters of the wind turbine, enhancing the damping characteristics of the system and effectively suppressing low-frequency oscillations.

[0006] To address the aforementioned technical problems, this invention provides a method for suppressing low-frequency oscillations in a wide-area additional damper for a doubly-fed wind turbine based on a reinforcement learning algorithm, comprising the following steps:

[0007] Acquire operational data of the wind power generation system, including wind speed, turbine output power, rotational speed, and grid frequency; perform detrending and normalization processing on the acquired operational data of the wind power generation system, including wind speed, turbine output power, rotational speed, and grid frequency, using a bandpass filter; train a model using DDPG (Deep Deterministic Policy Gradient Algorithm) to predict the operating status of the turbine and the oscillation of the system; generate an optimal damping control strategy based on the trained DDPG model to adjust the operating parameters of the turbine in real time; apply the generated damping control strategy to the turbine control system to improve the damping characteristics of the system and reduce low-frequency oscillations. 1. Data Acquisition: Acquire power flow data of the grid operation; collect turbine operating data during turbine operation, including wind speed, turbine output power, rotational speed, and grid frequency.

[0008] Preferably, the DDPG algorithm improves the learning stability of the model by creating a target Actor network and a target Critic network;

[0009] Preferably, in the step of generating the optimal damping control strategy, the strategy generated by the trained DDPG model is used to adjust the operating parameters of the wind turbine in real time to improve the damping characteristics of the system.

[0010] To address the aforementioned technical problems, this invention provides a wide-area additional damper device for doubly-fed wind turbines based on the DDPG algorithm, comprising:

[0011] Acquisition module: Used to collect wind turbine operating data and power grid frequency data.

[0012] Processing module: Used to preprocess the collected data and train the DDPG model.

[0013] Control module: Used to generate and apply damping control strategies based on the trained DDPG model to improve system stability.

[0014] Preferably, the control module is used to collect data on wind speed, wind turbine output power, rotational speed, and power grid frequency.

[0015] Preferably, the processing module uses a bandpass filter to perform detrending and normalization processing on the collected data.

[0016] Preferably, the control module uses the DDPG model to generate the optimal damping control strategy and adjusts the operating parameters of the wind turbine in real time.

[0017] Beneficial effects

[0018] 1. Significantly improves low-frequency oscillation suppression capability: This invention generates the optimal damping control strategy by training the model using the DDPG algorithm, which can adjust the operating parameters of the doubly fed wind turbine in real time, effectively enhancing the damping characteristics of the power system. Compared with traditional control without additional damping or conventional parameter optimization methods, it can make the system power deviation curve reach steady state more quickly, and significantly reduce the impact of low-frequency oscillations on power grid safety.

[0019] 2. Strong adaptability and robustness: Relying on the self-learning optimization characteristics of reinforcement learning algorithms, this invention can adapt to the complex operating conditions caused by the randomness and fluctuation of wind energy in wind power systems. It does not require frequent manual parameter adjustments and can maintain a good suppression effect under different wind speeds and fault conditions, thus solving the problem of poor adaptability of traditional controllers under changing operating conditions.

[0020] 3. More comprehensive wide-area control coverage: By collecting wide-area information such as power flow data and multi-dimensional wind turbine operation data, and combining tie-line power signals to construct an optimization model for additional dampers, the system achieves accurate perception and control of the global oscillation state of the power system, avoiding the limitations of local control and further improving the overall stability and reliability of the power grid.

[0021] 4. Strong engineering practicality: The device has a simple structure and clear core module functions. The DDPG algorithm used can handle high-dimensional input and continuous action space, and is suitable for the complex operation scenarios of actual power systems. At the same time, the control strategy can be directly applied to the converter control system of existing doubly fed wind turbine units. It is easy to modify, has good compatibility, and is easy to promote and apply on a large scale, which helps the high-quality development of new energy power systems. Attached Figure Description

[0022] The accompanying drawings illustrate exemplary embodiments of the present disclosure and, together with their description, serve to explain the principles, including providing a further understanding of the disclosure. The drawings are included in and form part of this specification.

[0023] Figure 1 This is a flowchart of the online suppression method for low-frequency oscillations of a wide-area additional damper for a doubly fed wind turbine based on a reinforcement learning algorithm, as proposed in this invention.

[0024] Figure 2 This is a schematic diagram of the additional damping controller when considering the power of the tie line in this invention;

[0025] Figure 3 This is a schematic diagram of a four-machine, two-zone system used in the demonstration of this invention.

[0026] Figure 4 This is an active power curve of the generator under fault excitation according to an embodiment of the present invention. Detailed Implementation

[0027] To better understand the objectives, technical solutions, and advantages of the embodiments of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. All other embodiments that can be implemented by those skilled in the art based on the embodiments of the present invention without creative effort should be considered as included within the protection scope of the present invention. The present invention, through in-depth analysis of the structure and operating principle of doubly-fed induction generator (DFIG) wind turbines, proposes an online low-frequency oscillation suppression method based on reinforcement learning algorithms, aiming to improve the stability of power systems, especially in the context of large-scale wind power grid integration, to enhance the system's anti-oscillation capability and ensure the safe operation of the power grid.

[0028] Figure 1 This is a schematic diagram of an online low-frequency oscillation suppression method for a wide-area additional damper for a doubly-fed wind turbine based on a reinforcement learning algorithm, provided in an embodiment of the present invention. Figure 1 As shown, this embodiment of the invention provides an online method for suppressing low-frequency oscillations in a doubly-fed induction generator (DFIG) with a wide-area additional damper based on a reinforcement learning algorithm. The main execution component is a DFIG with a wide-area additional damper based on a reinforcement learning algorithm. The method includes the following steps:

[0029] 1. Data Acquisition: Acquire power flow data of the power grid operation and collect wind turbine operation data during wind turbine operation, including wind speed, wind turbine output power, speed and power grid frequency.

[0030] 2. Data Preprocessing: The electrical quantity data of each synchronous machine and doubly fed wind turbine in the power system are preprocessed to obtain the change values ​​of each electrical quantity for subsequent analysis, which is used to analyze the suppression of low-frequency oscillations.

[0031] 3. DDPG, or Deep Deterministic Policy Gradient Method Training: The DDPG model is trained using collected data to identify and predict the operating status of wind turbines and the oscillation of the power system.

[0032] 4. Generate damping control strategy: Generate the optimal damping control strategy through the trained DDPG model, adjust the operating parameters of the wind turbine in real time, and improve the damping characteristics of the system.

[0033] 2.5. Real-time control application: The generated damping control strategy is applied to the wind turbine control system to adjust the wind turbine operation in real time and improve the stability of the power system.

[0034] 1. Constructing a Doubly Fed Wind Turbine Model and Control Strategy: When the stator and rotor parameters of the doubly fed generator adopt the conventional parameters for electric motors, considering the influence of transmission chain flexibility on the transient performance of the wind turbine, the equivalent lumped mass method is adopted. The wind turbine and generator are respectively represented as equivalent mass blocks. The transmission chain equations under the two mass blocks are as follows:

[0035]

[0036] Among them, H w and H g ω represents the inertial time constant of the wind turbine and generator rotors, respectively. r and ω gen ω represents the electric angular velocity of the wind turbine and generator rotors, respectively. s Let ω be the system's basic electric angular velocity. s = 2πf, where f is the power frequency, θ s D represents the angular displacement of the wind turbine relative to the generator rotor. s D is the damping coefficient between the wind turbine and the generator. w and D g K represents the damping coefficient of the wind turbine and generator rotors, respectively. s T is the stiffness coefficient of the transmission shaft system. e and T w These refer to the electromagnetic torque of the generator and the mechanical torque of the wind turbine, respectively.

[0037] The control system of a doubly-fed induction generator (DFIG) wind turbine mainly includes generator converter control and turbine pitch angle control. In the generator converter control section, the grid-side PWM converter maintains a constant DC-side voltage, ensuring the grid-side input current is sinusoidal; the rotor-side PWM converter provides excitation current with variable amplitude, phase, and frequency, thereby achieving independent regulation of active and reactive power. The turbine pitch angle control, by adjusting the rotational speed and output power, maximizes wind energy capture and maintains constant power output.

[0038] 2. Constructing an optimized model for the additional damper of a doubly-fed wind turbine: This invention adds the tie-line power signal to the reactive power loop of the rotor-side converter of the doubly-fed wind turbine. Figure 2 This is a schematic diagram of a wide-area additional damper device for a doubly-fed wind turbine based on a reinforcement learning algorithm, provided in an embodiment of the present invention. Figure 2 As shown, this embodiment of the invention provides a wide-area additional damper device for doubly fed wind turbines based on a reinforcement learning algorithm.

[0039] Considering that the most widely used type in real power systems is still the single-branch PSS, which mainly consists of three parts: a filter stage, a gain stage, and a lead-lag stage, its transfer function can be described as:

[0040]

[0041] Where, Δu pss The output signal of PSS is given by K, where K is the controller gain and T is the output signal of PSS. wf Let T1 be the time constant of the filter, T2 be the time constants of the lead-lag elements, and Δx be the input signal of the PSS. Where T1, T2, and T... wf It is usually a preset constant. In this invention, Twf =10; T1=0.02; T2=0.02.

[0042] The controller gain satisfies the constraint shown in the following equation:

[0043] K min ≤K≤K max

[0044] For the aforementioned PSS, system stability can be improved by optimizing the control variable K. To quantitatively evaluate the magnitude of system stability, this invention constructs a global system stability index based on eigenvalues, as shown in the following formula:

[0045]

[0046] Λ i =diag(λ i1 ,λ i2 ,…,λ ij ,…,λ in )

[0047]

[0048] Where, λ ij Let α be the j-th characteristic value of the system under the i-th operating mode. ij σ ij and ξ ij For λ ij The real part, the imaginary part, and the damping.

[0049] Minimizing the index J can make the eigenvalues ​​move toward the stable region (ξ>ξ). set , σ>σ set ) moves, where ξ set and σ set This represents the desired values ​​for the damping and the real part. J = 0 is satisfied, meaning all eigenvalues ​​in the system lie within the stability region.

[0050] 3. DDPG Algorithm Training: The DDPG algorithm can handle high-dimensional inputs and continuous action spaces, and is another classic algorithm in the field of deep reinforcement learning. A deep deterministic policy gradient algorithm model is trained using preprocessed data.

[0051] DDPG creates separate copies of the current Actor network and Critic network, namely the target Actor network and the target Critic network. After training on a batch of data, DDPG updates the current network using gradient descent, and then updates the target network using a moving average method. The moving average method ensures that the parameters of the target network change only slowly, greatly improving the stability of the learning process.

[0052] The specific steps are as follows:

[0053] Create the Actor network and the Critic network, and create the target network for each of them.

[0054] The network is trained by sampling small batches of data. During training, the Actor network is responsible for generating control policies, while the Critic network evaluates the effectiveness of these policies.

[0055] During training, the network parameters are updated step by step using an experience replay mechanism and a moving average method to ensure the stability and convergence of the model.

[0056] 4. Generation damping control strategy

[0057] Based on the trained DDPG model, an optimal damping control strategy is generated. This strategy can automatically adjust the operating parameters of the wind turbine according to the real-time status of the wind turbine and the power system, thereby improving the damping characteristics of the system and reducing low-frequency oscillations.

[0058] The generated damping control strategy is applied to the wind turbine's control system.

[0059] The specific implementation is as follows:

[0060] The generated control signals are transmitted to the wind turbine's controller to adjust the active and reactive power output of the wind turbine in real time.

[0061] Monitor the operating status of wind turbines and power systems, and further optimize control strategies based on real-time data feedback.

[0062] 5. Specific Examples

[0063] This section introduces the IEEE two-area four-machine system as the test system, connecting a doubly fed wind farm in area 1 with a wind turbine rated capacity of 200MW. For simplification, the DFIG single-machine model is used as the lumped model of the wind farm to replace the large-scale wind farm.

[0064] A four-unit, two-zone system including doubly-fed wind power, such as Figure 3 As shown.

[0065] The effectiveness of the proposed method was verified through simulation tests, particularly its suppression effect on power oscillations in tie lines under different damping signal gains.

[0066] In this invention, the annual wind speed data of a wind farm in Northwest China is selected as the historical data of the wind farm in this system. The wind speed data is sampled every 3 hours, for a total of 2800 data points.

[0067] This invention uses the DDPG algorithm to train the agent to obtain the optimal low-frequency oscillation suppression strategy for the wide-area additional damper of the doubly fed wind turbine.

[0068] To test the effectiveness of the PSS parameter settings provided by the trained agent, this invention introduces a comparative analysis of parameter setting methods for undamped control and particle swarm optimization.

[0069] The two methods mentioned above and the proposed method were tested under a wind speed of 15 m / s.

[0070] To compare the performance of parameter settings provided by different parameter tuning methods, a time-domain simulation was performed under a three-phase short-circuit fault excitation. The fault duration was 0.1 s, and the fault bus was bus 3 on the tie line between the two areas. The simulation results are as follows: Figure 4 As shown.

[0071] Compared to the other two parameter tuning methods, the low-frequency oscillation suppression strategy of the doubly fed wind turbine wide-area additional damper based on DDPG proposed in this invention can enable the power deviation curve of the generator in the system to reach steady state in the shortest time.

[0072] This means that the proposed method provides PSS parameter settings with better oscillation suppression capabilities under different operating conditions, and shows better robustness than parameter setting methods without additional damping control and particle swarm optimization under varying operating conditions.

[0073] Through the above steps, this invention provides a method and device for online suppression of low-frequency oscillations in a wide-area additional damper for doubly fed wind turbines based on reinforcement learning algorithms. This method can effectively improve the stability of wind power systems, reduce low-frequency oscillations, and has broad application prospects.

[0074] This invention provides a wide-area additional damper device for doubly fed wind turbines based on reinforcement learning algorithms, used to execute the methods described in the above embodiments. The specific steps of executing the methods described in the above embodiments using the device provided in this embodiment are the same as those in the above embodiments, and will not be repeated here.

[0075] The doubly fed wind turbine wide-area additional damper device based on reinforcement learning algorithm provided in this embodiment of the invention can effectively improve the stability of wind power system by using a low-frequency oscillation online suppression method based on reinforcement learning algorithm, thereby providing a guarantee for the safe and stable operation of the power grid.

[0076] The above technical solutions only embody the preferred technical solutions of the present invention. Any modifications that may be made by those skilled in the art to certain parts thereof embody the principles of the present invention and fall within the protection scope of the present invention.

Claims

1. A method for online suppression of low-frequency oscillations in a wide-area additional damper of a doubly-fed wind turbine based on a reinforcement learning algorithm, characterized in that, Includes the following steps: The system acquires operational data of the wind power generation system, including wind speed, turbine output power, rotational speed, and grid frequency; it then uses a bandpass filter to detrend and normalize the acquired operational data, including wind speed, turbine output power, rotational speed, and grid frequency; finally, it uses DDPG (Deep Deterministic Policy Gradient Algorithm) to train a model to predict the operating status of the turbines and the oscillations of the system. Based on the trained DDPG model, an optimal damping control strategy is generated to adjust the operating parameters of the wind turbine in real time. The generated damping control strategy is then applied to the wind turbine control system to improve the system's damping characteristics and reduce low-frequency oscillations.

2. The method for online suppression of low-frequency oscillations in a wide-area additional damper for a doubly-fed wind turbine based on a reinforcement learning algorithm, as described in claim 1, is characterized in that... The DDPG algorithm improves the learning stability of the model by creating a target Actor network and a target Critic network.

3. The method for online suppression of low-frequency oscillations in a doubly-fed wind turbine with a wide-area additional damper according to claim 1, characterized in that, In the step of generating the optimal damping control strategy, the operating parameters of the wind turbine are adjusted in real time using the strategy generated by the trained DDPG model to improve the damping characteristics of the system.

4. A doubly-fed induction generator (DFIG) wide-area additional damper device based on reinforcement learning algorithm, which implements the low-frequency oscillation online suppression method for a doubly-fed induction generator based on reinforcement learning algorithm as described in any one of claims 1 to 3, characterized in that, include: The acquisition module is used to collect wind turbine operating data and grid frequency data; The processing module is used to preprocess the acquired data and train the DDPG model; The control module is used to generate and apply damping control strategies based on the trained DDPG model to improve the stability of the system.

5. The wide-area additional damper device for doubly-fed wind turbines based on reinforcement learning algorithm according to claim 4, characterized in that, The control module is used to collect data on wind speed, wind turbine output power, rotational speed, and grid frequency.

6. The wide-area additional damper device for doubly-fed wind turbines based on reinforcement learning algorithm according to claim 4, characterized in that, The processing module uses a bandpass filter to perform detrending and normalization processing on the collected data.

7. The wide-area additional damper device for doubly-fed wind turbines based on reinforcement learning algorithm according to claim 4, characterized in that, The control module uses the DDPG model to generate the optimal damping control strategy and adjusts the operating parameters of the wind turbine in real time.

Images