Fast frequency control method and system for new energy power system

Abstract

The invention discloses a fast frequency control method and system for a new energy power system. The fast frequency control method comprises the steps that a supervised control method based on reinforcement learning is adopted in outer layer control to generate set output power; obtaining active set output power and reactive set output power through droop control and a power strategy; performing active control to generate a control frequency # imgabs0 #; performing reactive power control to generate a control voltage # imgabs1 #; and generating a modulation voltage reference signal through inner layer control according to the control voltage # imgabs2 # and the control frequency # imgabs3 # so as to control the voltage source converter. The invention aims to solve the problem of frequency stability caused by the reduction of inertia and damping level in the current power system, and particularly the problems of low response speed, low control precision and difficulty in adapting to complex dynamic working conditions in the traditional frequency control method under the background of increasing new energy permeability.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a fast frequency control method and system for a new energy power system. Background Art

[0002] Since renewable energy generation (such as wind power, photovoltaics, etc.) is usually connected to the grid through power electronic devices, this transformation has led to a significant reduction in system inertia and damping. Traditional frequency control strategies that rely on the inertial characteristics of synchronous generators show obvious limitations in such low-inertia grids, including slow response speed, low control accuracy, and inability to adapt to rapid dynamic disturbances. This limitation poses a major challenge to grid frequency stability. In the context of grid frequency stability, frequency control strategies are mainly divided into primary frequency regulation, secondary frequency regulation, and auxiliary service control. However, traditional primary and secondary frequency regulation methods have slow response speeds and are difficult to effectively maintain frequency stability in grids with high renewable energy penetration. Especially when the grid is disturbed, the frequency response of traditional control methods often has significant overshoot and long recovery time.

[0003] In recent years, frequency control technology based on voltage source converter (VSC) has attracted wide attention. Voltage source converter can provide frequency support for low inertia power grid through its fast dynamic power regulation capability. However, how to optimize the control strategy of voltage source converter to make full use of the response speed of voltage source converter and adapt to complex dynamic conditions at the same time is still an urgent problem to be solved. Reinforcement learning (RL) does not rely on the precise model of the system, can learn in an unknown environment, and dynamically adjust the strategy according to environmental changes. Adapt to complex and high-dimensional state space. Reinforcement learning provides new ideas and methods for the control of complex power systems. For example, existing deep reinforcement learning algorithms include frequency control methods based on deep deterministic policy gradient (DDPG algorithm), and its main steps are as follows: Design of policy network and value network: DDPG algorithm uses two independent networks, one for generating control actions (policy network) and the other for evaluating action values ​​(value network). Experience replay mechanism: By storing and reusing past experience samples, the training efficiency and algorithm stability are improved. Continuous action space optimization: Deterministic policy gradients are used to directly optimize continuous control actions to achieve frequency control. However, power grid frequency control mainly relies on traditional frequency regulation methods based on droop control and frequency control methods based on model predictive control (MPC). Droop control adjusts the output power of generators or energy storage devices through the linear relationship between frequency and power to respond to frequency deviations. Traditional droop control is essentially a proportional control strategy that requires accurate setting of the droop coefficient. It is difficult to quickly respond to large-scale disturbances and difficult to adapt to nonlinear and complex new energy power systems with variable operating conditions. Model-based predictive control requires an accurate system model, has high computational complexity, and poor real-time performance, which limits its application in fast response scenarios. Therefore, the DDPG algorithm has the following shortcomings: Overestimation of action value function: Due to the evaluation error of the single value network, the frequency control strategy falls into a local optimum. Insufficient sample efficiency: The random experience replay mechanism cannot effectively utilize samples with high learning value, which affects the training efficiency of the algorithm. Summary of the invention

[0004] Technical problem to be solved by the present invention: In view of the above-mentioned problems in the prior art, a rapid frequency control method and system for a new energy power system are provided. The present invention aims to solve the frequency stability problem in the current power system caused by the reduction of inertia and damping level, especially in the context of increasing penetration of new energy. The traditional frequency control method has the problems of slow response speed, low control accuracy and difficulty in adapting to complex dynamic working conditions. It can quickly respond to large-scale disturbances in the new energy power system, adapt to the operating requirements of complex nonlinear new energy power systems, and provide efficient and accurate frequency control solutions for the power grid.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: A fast frequency control method for a new energy power system comprises the following steps: using a supervisory control method based on reinforcement learning in an outer layer control to generate a set output power ; Obtain active set output power through droop control and power strategy and reactive power setting output power ; Combined with the set output power of the voltage source converter VSC , Active Power , given active power setting balance point and frequency setting balance point Active power control to generate control frequency ; Combined reactive power , given reactive power setting balance point and voltage set point Reactive power control to generate control voltage ; According to the control voltage and control frequency Generate modulation voltage reference signal through inner layer control ; Modulate the voltage reference signal Generate control signal modulation ratio through PWM modulation To control the voltage source converter VSC.

[0006] Optionally, the outer layer control uses a supervisory control method based on reinforcement learning to generate a set output power The supervisory control method based on reinforcement learning is the improved TD3 algorithm. The improved TD3 algorithm improves the way of obtaining experience from the experience replay pool with the priority experience replay mechanism, and uses the current generator frequency and frequency change rate RoCoF of the new energy power system as the state and sets the output power. As the corresponding action, the preset reward function is used to iteratively obtain the optimal set output power The priority experience replay mechanism refers to extracting the temporal difference error in the TD3 algorithm as the priority of the experience in the experience replay pool, and selecting the priority size to sample experience from the experience replay pool to replace the random sampling of experience from the experience replay pool by the TD3 algorithm.

[0007] Optionally, the function expression of the reward function is: , , , in, For reward, is the cost coefficient of the control action; The action is to set the output power ; is the deviation cost term, is the deviation penalty term, and are weight factors, is the state, the state is the current generator frequency and frequency change rate RoCoF, For the ideal state.

[0008] Optionally, the active setting output power is obtained by droop control and power strategy and reactive power setting output power The function expression is: , , in, represents the filter voltage, represents the transformer current, is the 90° rotation matrix, is the transpose operation.

[0009] Optionally, the set output power of the voltage source converter VSC , Active Power , given active power setting balance point and frequency setting balance point Active power control to generate control frequency The function expression is: , , in, To set the balance point for a given frequency, is the active power droop coefficient, To set the balance point for a given active power, is the active power after low-pass filtering, For time, for About time The differential of is the cutoff frequency of the low-pass filter, is the active power.

[0010] Optionally, the combined reactive power , given reactive power setting balance point and voltage set point Reactive power control to generate control voltage The function expression is: , , in, For a given voltage set point, is the reactive power droop coefficient, To set the balance point for a given reactive power, is the reactive power after low-pass filtering, For time, for About time The differential of is the cutoff frequency of the low-pass filter, is the reactive power.

[0011] Optionally, the control voltage and control frequency Generate modulation voltage reference signal through inner layer control The function expression is: , in, and is the proportional gain, is the control voltage, is the reference voltage, To control the frequency, is the reference frequency.

[0012] In addition, the present invention also provides a rapid frequency control system for a renewable energy power system, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the rapid frequency control method for the renewable energy power system.

[0013] In addition, the present invention also provides a computer-readable storage medium, in which a computer program or instruction is stored. The computer program or instruction is programmed or configured to execute the fast frequency control method for the new energy power system through a processor.

[0014] In addition, the present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the fast frequency control method for a new energy power system through a processor.

[0015] Compared with the prior art, the present invention has the following advantages: the present invention includes using a supervisory control method based on reinforcement learning in the outer layer control to generate a set output power ; Obtain active set output power through droop control and power strategy and reactive power setting output power ; Combined with the set output power of the voltage source converter VSC , Active Power , given active power setting balance point and frequency setting balance point Active power control to generate control frequency ; Combined reactive power , given reactive power setting balance point and voltage set point Reactive power control to generate control voltage ; According to the control voltage and control frequency Generate modulation voltage reference signal through inner layer control ; Modulate the voltage reference signal Generate control signal modulation ratio through PWM modulation The voltage source converter VSC is controlled by a double-layer control strategy of outer layer control and inner layer control, and the supervisory control method based on reinforcement learning is used in the outer layer control to generate the set output power. , which can solve the frequency stability problem caused by inertia and reduced damping level in the current power system. Especially in the context of increasing penetration of new energy, traditional frequency control methods have problems such as slow response speed, low control accuracy and difficulty in adapting to complex dynamic conditions. It can give full play to the fast response characteristics of the voltage source converter, quickly adjust the power output when disturbed, reduce frequency deviation and accelerate system recovery, and improve the stability and reliability of frequency control. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of the basic flow of the method of the embodiment of the present invention.

[0017] Figure 2 Schematic diagram of the training process of the improved TD3 algorithm in an embodiment of the present invention.

[0018] Figure 3 It is a control flow diagram of applying the improved TD3 algorithm to frequency control in an embodiment of the present invention.

[0019] Figure 4 It is a schematic diagram of the simulation structure of the IEEE node system in an embodiment of the present invention.

[0020] Figure 5 It is a schematic diagram of the frequency control result of the improved TD3 algorithm in an embodiment of the present invention.

[0021] Figure 6This is a comparison diagram of average frequency curves of the improved TD3 algorithm in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be further described in detail below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0023] like Figure 1 As shown, the fast frequency control method for a new energy power system in this embodiment includes the following steps: using a supervisory control method based on reinforcement learning in the outer layer control to generate a set output power ; Obtain active set output power through droop control and power strategy and reactive power setting output power ; Combined with the set output power of the voltage source converter VSC , Active Power , given active power setting balance point and frequency setting balance point Active power control to generate control frequency ; Combined reactive power , given reactive power setting balance point and voltage set point Reactive power control to generate control voltage ; According to the control voltage and control frequency Generate modulation voltage reference signal through inner layer control ; Modulate the voltage reference signal Generate control signal modulation ratio through PWM modulation To control the voltage source converter VSC.

[0024] like Figure 1 As shown, the method of this embodiment includes an improved TD3 algorithm and a fast frequency control model. The fast frequency control model includes a DC side circuit and a lossless switch unit connected to the power grid, through an RLC filter and transformer Connected to the grid, its mathematical model is based on the dq coordinate system and expressed in per-unit form. The DC side circuit contains DC link capacitors for energy storage , represents the constant current source of the input current of renewable energy generation and a controlled DC current source representing the current associated with the energy flexibility of the system . The DC control sets the DC link voltage The VSC control is based on a two-level scheme consisting of an inner and outer control loop. In the outer control loop of the VSC, the droop control and power measurement system output and And adjust the set values ​​of frequency and voltage control Control the output voltage amplitude and frequency. The active power P and reactive power Q controllers are designed to control the output voltage amplitude and frequency .

[0025] The fast frequency control method for the new energy power system in this embodiment can solve the frequency stability problem caused by inertia and reduced damping level in the current power system, especially in the context of increasing penetration of new energy. The traditional frequency control method has the problems of slow response speed, low control accuracy and difficulty in adapting to complex dynamic conditions. It fully utilizes the fast response characteristics of the voltage source converter, quickly adjusts the power output when disturbed, reduces frequency deviation and accelerates system recovery, and improves the stability and reliability of frequency control.

[0026] The outer layer uses a supervisory control method based on reinforcement learning to generate the set output power When the supervised control method based on reinforcement learning is used, the DDPG (Deep Deterministic Policy Gradient) algorithm and the TD3 (Twin Delayed Deep Deterministic Policy Gradient) algorithm can be used. However, the DDPG algorithm uses the evaluation error of a single value network, which has an over-estimation problem; and the random experience replay mechanism of the DDPG algorithm and the TD3 algorithm cannot effectively utilize samples with high learning value, which affects the training efficiency of the algorithm. On this basis, as an optional implementation method, the outer layer control of this embodiment uses a supervised control method based on reinforcement learning to generate a set output power When , the supervisory control method based on reinforcement learning is the improved TD3 algorithm, such as Figure 2 As shown, the improved TD3 algorithm improves the way of obtaining experience from the experience playback pool for the TD3 algorithm to a priority experience playback mechanism, and uses the current generator frequency and frequency change rate RoCoF of the new energy power system as the state and sets the output power As the corresponding action, the preset reward function is used to iteratively obtain the optimal set output power , the priority experience replay mechanism refers to extracting the temporal difference error (a key indicator in reinforcement learning, used to measure the deviation between the current estimated action value and the target action value) in the TD3 algorithm as the priority of the experience in the experience replay pool, and selecting the priority size to sample experience from the experience replay pool to replace the TD3 algorithm's random sampling of experience from the experience replay pool. On the one hand, the improved TD3 algorithm of this embodiment can effectively avoid the over-estimation problem of a single value network based on the inherent dual critic network of the TD3 algorithm, thereby improving the stability and accuracy of the algorithm; on the other hand, through the priority experience replay mechanism, according to the learning value of the experience sample, the experience with a very high success rate or a very high failure rate is replayed preferentially, thereby accelerating the learning process. Combined with these technical improvements, the improved TD3 algorithm can quickly respond to large-scale disturbances, adapt to the operating requirements of complex nonlinear new energy power systems, and provide efficient and accurate frequency control solutions for power grids.

[0027] like Figure 2 As shown, in this embodiment, an experience sampled from the experience playback pool is expressed as: , in, is the current state, is the action at the current moment, is the reward at the current moment, is the state at the next moment. The TD3 algorithm includes a policy network (Actor network ) and two value networks: Critic Network 1 and Critic Network 2, as well as a policy target network and two value target networks: Critic Target Network 1 and Critic Target Network 2. The network parameters of the Actor network are recorded as , the network parameters of Critic network 1 are recorded as The network parameters of Critic network 2 are recorded as ; The network parameters of the Actor target network are recorded as , the network parameters of the Critic target network 1 are recorded as , the network parameters of Critic target network 2 are recorded as ; The network structure of the policy network and the policy target network is the same, the network structure of the critic network 1 and the critic target network 1 is the same, and the network structure of the critic network 2 and the critic target network 2 is the same. Policy network (Actor network ) is used based on the current state Generate an action by adding noise: , in, To increase the action after the noise, the action is to set the output power ; For Actor Network Based on the current status Generated actions, states is the current generator frequency and the frequency change rate RoCoF, is noise and satisfies the distribution ,in is the standard deviation of the system state s.

[0028] In this embodiment, the calculation function expression of the timing difference error is: , in, is the time series difference error of the jth experience, For reward, is the discount factor, The value target network is based on the state of the next moment and actions The obtained Q value; The value network is based on the current state and actions As an optional implementation, in this embodiment, when extracting the time difference error in the TD3 algorithm as the priority of the experience in the experience replay pool, it includes sampling the specified k experiences from the experience replay pool according to the importance sampling weight of each experience, and calculating and updating its importance sampling weight according to the above formula; and the importance sampling weight of any j-th experience The calculation function expression is: , in, is the total amount of experience in the experience replay pool, is the normalized probability of the jth experience, and: , in, and are the priority values ​​of the j-th experience and the i-th experience respectively, For priority adjustment parameters, when When , the relative gap between high-priority samples and low-priority samples will decrease, and the influence of low-priority samples may increase. When , the power of the priority will make the value of high-priority samples grow faster and the value of low-priority samples grow slower, thereby increasing the probability of high-priority samples being selected. The initial value of the priority value can be a random value, and through the update operation, its value becomes the corresponding importance sampling weight.

[0029] In this embodiment, the function expression of the reward function is: , , , in, For reward, is the cost coefficient of the control action; The action is to set the output power ; is the deviation cost term, is the deviation penalty term, and are weight factors, is the state, the state is the current generator frequency and frequency change rate RoCoF, For the ideal state.

[0030] In this embodiment, the network parameters of Critic network 1 are updated and the network parameters of Critic Network 2 The function expression is: , In the above formula, Represents the network parameters of Critic network 1 Or the network parameters of Critic network 2 , is the target value, is the Q value obtained by Critic network 1 or Critic network 2 based on state s and action a, and N is the number of samples; , In the above formula, is the reward based on the current state s and action a, is the discount factor, The state of Critic target network 1 or Critic target network 2 based on the next moment and actions In addition, in this embodiment, the updated strategy target network and the two value target networks adopt delayed update of the target network to reduce fluctuations during strategy update, further improving the convergence efficiency and stability of the algorithm. The function expressions of the updated strategy target network and the two value target networks are: , In the above formula, Represents the network parameters of Critic target network 1 Or the network parameters of Critic target network 2 , is the soft update coefficient, Represents the network parameters of Critic network 1 Or the network parameters of Critic network 2 , is the soft update coefficient, is the network parameters of the Actor target network, Network parameters for the Actor network.

[0031] After the agent training of the improved TD3 algorithm of this embodiment is completed, it can be applied to the actual system or simulation environment to perform frequency control. Figure 3 As shown in the figure, the process includes: loading the trained agent: loading the agent parameters saved during the training process. Initializing the environment and control parameters: initializing the state of the simulation or real environment, setting the initial frequency, load state, and disturbance according to the system model. Setting the execution mode of the agent. If in the real environment, disable the exploration strategy and directly execute the deterministic action. Frequency control loop: during the operation of the system, the frequency is adjusted in real time through the agent. At each time step, the action (power adjustment) output by the agent is obtained according to the current state. Continuously monitor the frequency deviation to ensure that the control effect meets the requirements. Record the control effect: record the frequency value, power output, frequency deviation and other data at each time step. Save the data as a file or chart for subsequent analysis. Draw a curve chart of frequency change over time to evaluate the dynamic performance and stability of the control. Evaluate the control performance: compare the control effect of the agent with traditional methods (such as droop control, DDPG control, TD3 control) to show the superiority of the method.

[0032] In this embodiment, the active setting output power is obtained by droop control and power strategy. and reactive power setting output power The function expression is: , , in, represents the filter voltage, represents the transformer current, is the 90° rotation matrix, is the transpose operation.

[0033] In this embodiment, the output power setting of the voltage source converter VSC is combined with , Active Power , given active power setting balance point and frequency setting balance point Active power control to generate control frequency The function expression is: , , in, To set the balance point for a given frequency, is the active power droop coefficient, To set the balance point for a given active power, is the active power after low-pass filtering, For time, for About time The differential of is the cutoff frequency of the low-pass filter, is the active power.

[0034] In this embodiment, combined with reactive power , given reactive power setting balance point and voltage set point Reactive power control to generate control voltage The function expression is: , , in, For a given voltage set point, is the reactive power droop coefficient, To set the balance point for a given reactive power, is the reactive power after low-pass filtering, For time, for About time The differential of is the cutoff frequency of the low-pass filter, is the reactive power.

[0035] In this embodiment, according to the control voltage and control frequency Generate modulation voltage reference signal through inner layer control The function expression is: , in, and is the proportional gain, is the control voltage, is the reference voltage, To control the frequency, is the reference frequency.

[0036] In order to verify the performance of the fast frequency control method for the new energy power system in this embodiment in the fast frequency control task, in this embodiment, Figure 4 The IEEE 14-node test system shown in the figure is used for simulation testing. This embodiment is based on the standard IEEE 14-bus system, which introduces two VSCs connected to Bus 1 and Bus 2 respectively. The converter-based device is equipped with 850 MW of power. The disturbances used for performance evaluation control are generator losses and load losses, both of which simulate step changes in active power injection on the network bus of interest. It is assumed that when the frequency deviation exceeds ±0.5 Hz or the frequency change rate RoCoF exceeds ±1 Hz / s, underfrequency load reduction (a power system protection measure) will be triggered. In order to test and verify the performance, an 800 MW disturbance was applied to Bus 14, and the results obtained are shown in the figure. Figure 5 In addition, the performance of other existing frequency control methods is compared in this embodiment, including droop control, DDPG algorithm, TD3 algorithm and improved TD3 algorithm. By analyzing the frequency response of the system under load mutation, the method of this embodiment obtains the system average frequency control effect of each method, as shown in FIG. Figure 6 As shown. Figure 6 It can be seen that as a traditional baseline method, droop control shows obvious frequency deviation under load disturbance, delayed frequency recovery speed, and large steady-state error. Droop control cannot be effectively applied to frequency control requirements in complex environments. The DDPG algorithm improves the shortcomings of droop control, with relatively small frequency drop and the ability to enter the recovery phase faster. However, the DDPG algorithm introduces the overestimation problem of a single neural network, resulting in the frequency recovery tending to fall into the local optimum. The TD3 algorithm eliminates the overestimation deviation problem in DDPG by introducing a dual evaluation network and a delayed update mechanism, significantly improving the frequency control performance. The improved TD3 algorithm introduces a priority experience replay mechanism based on TD3, which improves the learning efficiency of the control strategy and reduces the oscillation in the control strategy through reinforcement learning of the master sample. The experimental results show that the improved TD3 algorithm has the best frequency control performance, achieving smaller frequency deviation and faster stability.

[0037] In summary, the fast frequency control method for the new energy power system in this embodiment integrates reinforcement learning control technology and VSC technology into the new energy power system, and constructs a fast frequency control model for the new energy power system based on reinforcement learning. The improved TD3 algorithm is used to interactively train the improved power system. By introducing the dual Critic network and delayed update mechanism, the Q value overestimation problem is solved, and the accuracy of the control decision is improved; at the same time, combined with the priority experience playback mechanism, the priority of the learning sample is dynamically adjusted according to the TD error, so that high-value experience is fully utilized, thereby accelerating the convergence speed of the algorithm and obtaining the optimal fast frequency control strategy.

[0038] In addition, this embodiment also provides a rapid frequency control system for a renewable energy power system, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the rapid frequency control method for the renewable energy power system.

[0039] In addition, this embodiment also provides a computer-readable storage medium, in which a computer program or instruction is stored. The computer program or instruction is programmed or configured to execute the fast frequency control method for the new energy power system through a processor.

[0040] In addition, this embodiment also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the fast frequency control method for the new energy power system through a processor.

[0041] Those skilled in the art should understand that the technical solutions provided by the embodiments of the present invention may be in the form of methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that instructions executed by the processor of a computer or other programmable data processing device generate instructions for implementing the functions in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0042] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A fast frequency control method for a new energy power system, characterized in that: The steps include: using a supervisory control method based on reinforcement learning in the outer layer control to generate a set output power ; Obtain active set output power through droop control and power strategy and reactive power setting output power ; Combined with the set output power of the voltage source converter VSC , Active Power , given active power setting balance point and frequency setting balance point Active power control to generate control frequency ; Combined reactive power , given reactive power setting balance point and voltage set point Reactive power control to generate control voltage ; According to the control voltage and control frequency Generate modulation voltage reference signal through inner layer control ; The modulated voltage reference signal Generate control signal modulation ratio through PWM modulation To control the voltage source converter VSC.

2. The rapid frequency control method for a new energy power system according to claim 1, characterized in that: The outer layer control uses a supervisory control method based on reinforcement learning to generate the set output power The supervisory control method based on reinforcement learning is the improved TD3 algorithm. The improved TD3 algorithm improves the way of obtaining experience from the experience replay pool with the priority experience replay mechanism, and uses the current generator frequency and frequency change rate RoCoF of the new energy power system as the state and sets the output power. As the corresponding action, the preset reward function is used to iteratively obtain the optimal set output power The priority experience replay mechanism refers to extracting the temporal difference error in the TD3 algorithm as the priority of the experience in the experience replay pool, and selecting the priority size to sample experience from the experience replay pool to replace the random sampling of experience from the experience replay pool by the TD3 algorithm.

3. The rapid frequency control method for a new energy power system according to claim 2, characterized in that: The functional expression of the reward function is: ， ， ， in, For reward, is the cost coefficient of the control action; The action is to set the output power ; is the deviation cost term, is the deviation penalty term, and are weight factors, is the state, the state is the current generator frequency and frequency change rate RoCoF, For the ideal state.

4. The rapid frequency control method for a new energy power system according to claim 1, characterized in that: The active output power is obtained by droop control and power strategy. and reactive power setting output power The function expression is: ， ， in, represents the filter voltage, represents the transformer current, is the 90° rotation matrix, is the transpose operation.

5. The rapid frequency control method for a new energy power system according to claim 1, characterized in that: The set output power of the combined voltage source converter VSC , Active Power , given active power setting balance point and frequency setting balance point Active power control to generate control frequency The function expression is: ， ， in, To set the balance point for a given frequency, is the active power droop coefficient, To set the balance point for a given active power, is the active power after low-pass filtering, For time, for About time The differential of is the cutoff frequency of the low-pass filter, is the active power.

6. The rapid frequency control method for a new energy power system according to claim 1, characterized in that: The combined reactive power , given reactive power setting balance point and voltage set point Reactive power control to generate control voltage The function expression is: ， ， in, For a given voltage set point, is the reactive power droop coefficient, To set the balance point for a given reactive power, is the reactive power after low-pass filtering, For time, for About time The differential of is the cutoff frequency of the low-pass filter, is the reactive power.

7. The rapid frequency control method for a new energy power system according to claim 1, characterized in that: The control voltage and control frequency Generate modulation voltage reference signal through inner layer control The function expression is: ， in, and is the proportional gain, is the control voltage, is the reference voltage, To control the frequency, is the reference frequency.

8. A fast frequency control system for a new energy power system, comprising a microprocessor and a memory connected to each other, characterized in that: The microprocessor is programmed or configured to execute the fast frequency control method for a new energy power system as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program or instruction stored therein, characterized in that: The computer program or instruction is programmed or configured to execute the fast frequency control method for a new energy power system as described in any one of claims 1 to 7 through a processor.

10. A computer program product comprising a computer program or instructions, characterized in that The computer program or instruction is programmed or configured to execute the fast frequency control method for a new energy power system as described in any one of claims 1 to 7 through a processor.

Images