Active defense controller for broadband oscillation of multi-infeed system of new energy power station

Abstract

The invention discloses a multi-target broadband oscillation active defense controller for a multi-infeed system of a new energy power station. The multi-target broadband oscillation active defense controller is composed of a broadband oscillation risk sensor based on an impedance network model and a multi-target controller. The risk sensor samples key time domain working point information of each new energy power station multi-infeed system through communication, and obtains oscillation risk information under the stable operation condition of the system by using an impedance network model of the new energy power station multi-infeed system; the multi-target controller is a deep neural network obtained by deep reinforcement learning and training under the guidance of perceived oscillation risk information; the multi-target controller outputs a multi-target broadband oscillation active defense control action through forward propagation; and the active defense control action is issued to the converter in each new energy power station through communication, and is used for adjusting the working state of each converter to realize active defense control on the broadband oscillation risk. According to the invention, beforehand active defense control of broadband oscillation of the multi-infeed system is realized.

Description

Technical Field

[0001] This invention relates to the field of broadband oscillation stability control for multi-infeed systems in new energy power plants, and more specifically to a multi-target broadband oscillation active defense controller for multi-infeed systems in new energy power plants. Background Technology

[0002] With the increasing penetration of new energy sources in the power system, new energy power plants, including a large number of power electronic converters, are connected to the grid at different nodes, forming multi-infeed systems. Influenced by the inherent random fluctuations of new energy sources and the diverse topological coupling of the power grid, the broadband oscillations of multi-infeed systems exhibit significant high-dimensional coupling and dynamic time-varying complex characteristics. Against this backdrop, traditional offline damping design methods typically only conduct broadband oscillation stability analysis and optimization for a few pre-set high-risk operating points, representing a "single-point" remedial oscillation suppression measure. This is insufficient to cover the diverse oscillation problems caused by time-varying factors such as dynamic changes in operating points. Online adaptive damping control relies on time-domain measurements to extract oscillation mode information of the multi-infeed system and dynamically adjusts the suppression strategy accordingly. However, it only becomes effective after broadband oscillations actually occur, making it difficult to detect potential oscillation risks in advance and quickly implement targeted prevention and control measures when the system is in a stable state. Oscillation suppression exhibits significant passivity and lag, easily missing the optimal suppression opportunity. Summary of the Invention

[0003] To overcome the shortcomings of the existing technology, this invention provides a broadband oscillation active defense controller for multi-infeed systems in new energy power plants. It has the ability to quickly acquire potential broadband oscillation risk information and can quickly coordinate and control each new energy power plant based on the broadband oscillation risk information, actively avoid high oscillation risk conditions, realize proactive prevention and control of oscillation, and ensure that the multi-infeed system of new energy power plants always has sufficient stability margin under different operating conditions.

[0004] The present invention adopts the following technical solution to solve the technical problem: The active defense controller for multi-target broadband oscillations in multi-infeed systems of new energy power plants of this invention is characterized by being composed of a risk sensor K1 and a multi-target controller K2; the risk sensor K1 is a broadband oscillation risk sensor based on an impedance network model, and the risk sensor K1 samples the key time-domain operating point information of each multi-infeed system of the new energy power plant through communication. The oscillation risk information R1 under stable operating conditions is obtained using the impedance network model of a multi-infeed system in a new energy power plant; the key time-domain operating point information... This includes the output power of each new energy power station and the controller parameters of the converters within the new energy power station; the multi-objective controller K2 is a deep reinforcement learning-based multi-objective controller, which is a deep neural network trained using deep reinforcement learning and guided by the perceived oscillation risk information R1. The input of the deep neural network is the state of the multi-feed system of the new energy power station. The output of the deep neural network is a multi-target broadband oscillatory active defense control action. The status of the multi-infeed system of the new energy power station Includes key time-domain operating point information Oscillation risk information R1, and the set oscillation gain margin Target area T1; the multi-target broadband oscillation active defense control action Includes the parameter control quantities of the converter controller in new energy power plants and the active power output regulation of each new energy power station The multi-target controller K2 outputs multi-target broadband oscillation active defense control actions through forward propagation. The multi-target broadband oscillation active defense control action The system sends data to the converters in each new energy power station via communication, which is used to adjust the operating status of each converter in order to achieve proactive defense and control against broadband oscillation risks.

[0005] The feature of this invention, the multi-target broadband oscillation active defense controller for multi-infeed systems in new energy power plants, is that the risk sensor K1 obtains the oscillation risk information R1 according to the following steps: Step 1: Initialize the frequency band for assessing broadband oscillation risk : , ≥ ≥0, key time-domain operating point information of multi-infeed systems in sampling new energy power plants. ; Step 2: Based on the component connection method, the multi-feed system of the new energy power station is divided into active sub-module M1 and passive network sub-module M2 at the grid connection point of each new energy power station. The active sub-module M1 includes each active device including each new energy power station, and the passive network sub-module M2 includes each passive device including the transmission line. Step 3: Construct the frequency domain impedance model matrix of each new energy power station in the active submodule M1. ; Step 4: Construct the impedance matrix of the passive network in the passive network submodule M2. ; Step 5: with ERepresenting a unit diagonal matrix, the characteristic equation for constructing the impedance network model of a multi-infeed system in a new energy power plant is given. As shown in equation (1), and extract the characteristic equation. Open-loop transfer function matrix As shown in equation (2): ; At the current operating point, the open-loop transfer function matrix is ​​calculated. In the frequency band of the study frequency response characteristics within , ; Step 6: At the current operating point, use the generalized Nyquist criterion to sense the oscillation risk information R1 of the multi-feed system of the new energy power plant according to equation (3). The oscillation risk information R1 includes the oscillation interaction frequency. and oscillation gain margin ; ; In formula (3): This represents the eigenvalue decomposition operation in the generalized Nyquist criterion; λ To perform eigenvalue decomposition operation obtained L The eigenvalue matrix within the examined frequency band; It is the real part value corresponding to the intersection point with the minimum real part among all intersection points of the Nyquist curve of the multi-infeed system of the new energy power plant and the negative real axis; min(·) represents the operation of finding the minimum value; Re(·) and Im(·) represent the extracted real part and imaginary part, respectively; Represents positive real numbers; Indicates taking equal The corresponding oscillation frequency .

[0006] The multi-objective broadband oscillation active defense controller for multi-infeed systems in new energy power plants of this invention is also characterized by the following: the multi-objective controller K2 is constructed based on a multi-objective broadband oscillation active defense control optimization model and using a deep neural network; it includes setting the cost function, optimization variables, and constraints in the multi-objective broadband oscillation active defense control optimization model. 3.1 Set the cost function according to formula (4) : The cost function This refers to: oscillation gain margin Control objective cost function It is the oscillation gain margin of the multi-infeed system of a new energy power plant. The control is brought within the target region T1 to ensure the safe operation of the multi-infeed system of the new energy power plant with the desired stability margin; the oscillation gain margin in the target region T1. The value is ; ; In equation (4): For oscillation gain margin Distance from the center of target area T1; The distance from the boundary of the target area to the center point of the target area. ;when When, it represents the oscillation gain margin of the multi-infeed system of the new energy power plant. Within the target area T1; and The oscillation gain margin corresponding to the target region T1 are respectively Upper and lower limits, ; 3.2 Set the cost function according to formula (5) : The cost function This refers to the objective cost function that minimizes the power reduction of new energy power plants. This is to minimize the reduction of active power output from new energy power plants during the active defense control process for broadband oscillations; ; In equation (5): The first step in the active defense control process for broadband oscillations n The active power output regulation of each new energy power station; N This refers to the total number of renewable energy power plants in the multi-feed system of renewable energy power plants. n =1,2,…, N ; 3.3. Two types of optimization variables are set: The first type is the converter controller parameters in the new energy power plant. The second category is the output active power of new energy power plants. ; 3.4 Set constraints according to formula (6) The optimization variables satisfy the constraints set by equation (6) during the optimization adjustment process. ; In formula (6): It is the first n Controller parameter values ​​of the converter in a new energy power station; Indicating after regulation The value; and They are respectively The lower and upper limits; The setting of the lower and upper limits ensures that each converter has robust stability and good dynamic performance under ideal grid conditions; It is the first n The output active power value of each new energy power station; Indicating after regulation The value; for The upper limit, The upper limit depends on the maximum power output that the new energy power plant can currently generate under the existing external natural conditions; 3.5 Constructing a multi-target broadband oscillation active defense control optimization model Comprehensive cost function Cost function The optimization model for multi-objective broadband oscillation active defense control, obtained by optimizing variables and constraints, is shown in equation (7): ; In equation (7): The overall cost function is the optimization model for multi-target broadband oscillation active defense control. This represents an array consisting of the adjusted optimization variables; and These are the upper and lower bound constraints for the optimization variables; and These are the weights of the two control objectives, respectively. + =1, by adjusting the weight relationship between the two control objectives, the priority of different control objectives can be adjusted, and the control objective corresponding to the larger weight has a higher priority; The control objectives and constraints in the multi-objective broadband oscillation active defense control optimization model are converted into multi-objective reward functions, and the multi-objective controller K2 is trained by a deep reinforcement learning algorithm, thereby realizing the rapid solution of the above multi-objective broadband oscillation active defense control optimization model.

[0007] The feature of this invention, the multi-target broadband oscillation active defense controller for multi-infeed systems in new energy power plants, also lies in: based on the oscillation gain margin of the multi-infeed system in new energy power plants. The size of the warning region T2 and the prohibition region T3 are set: the oscillation gain margin in the warning region T2. The value is: When the oscillation gain margin When in warning zone T2, the multi-target broadband oscillation active defense controller generates multi-target broadband oscillation active defense control actions. Early intervention is implemented for multi-infeed systems of new energy power plants that are in the warning zone T2 but have not yet become unstable, in order to achieve ex-ante mitigation of broadband oscillation risks; the oscillation gain margin in the prohibited zone T3. The value is: This refers to a situation where the system has experienced broadband oscillation, and the multi-target broadband oscillation active defense controller prevents the system from entering the prohibited area through active defense control, thus preventing the broadband oscillation from occurring.

[0008] The multi-target broadband oscillation active defense controller for multi-infeed systems of new energy power plants is also characterized by the following: the target region T1 is adjusted as follows: the target region T1 is set by the operator of the multi-infeed system of new energy power plants according to actual operating conditions and the need for economic and safety trade-offs: when the fluctuations of new energy power plants and loads in the multi-infeed system of new energy power plants are small, and the operator expects to minimize the output reduction of new energy power plants to improve the economic efficiency of system operation, the upper limit of the target region T1 is lowered at the same time. and lower limit To reduce the oscillation gain margin of multi-infeed systems in new energy power plants To meet the needs of the system and improve its operational economy; during the adjustment of the target area T1, the upper limit of the warning area T2 changes accordingly; when the operation of the multi-infeed system of the new energy power plant is highly uncertain, in order to ensure that it has sufficient stability margin to cope with the risk of broadband oscillation instability caused by operating point fluctuations, the upper limit of the target area T1 is also increased. and lower limit To improve the oscillation gain margin of multi-infeed systems in new energy power plants To meet the requirements and ensure the safety of system operation; during the adjustment of target area T1, the upper limit of warning area T2 changes accordingly.

[0009] Compared with the prior art, the beneficial effects of the present invention are reflected in: 1. The multi-objective broadband oscillation active defense controller for multi-infeed systems of new energy power plants of the present invention consists of a risk sensor and a multi-objective controller. The risk sensor can obtain potential oscillation risk information based on the impedance network model of the multi-infeed system of the new energy power plant and the sampled key time-domain operating point information under stable system operation conditions. The multi-objective controller is constructed based on deep reinforcement learning method. It can quickly output multi-objective broadband oscillation active defense control actions under the guidance of oscillation risk information, realize the proactive defense and optimization control of oscillation in advance. This control process can be realized without relying on iterative calculations with large computational load. 2. The multi-objective broadband oscillation active defense controller for multi-feed systems of new energy power plants of the present invention simultaneously considers the gain margin control objective and the active power output reduction objective of new energy power plants. It can achieve the coordinated optimization of multiple control objectives in the process of broadband oscillation active defense control by intelligently coordinating various new energy power plants. 3. The system gain margin target area of ​​the multi-target broadband oscillation active defense controller for multi-infeed systems of new energy power plants of the present invention can be flexibly set by the system operator; the system operator of the multi-infeed system of new energy power plants can flexibly set the system gain margin target area according to the actual operating conditions and the economic and safety requirements of the system operation, thereby realizing a dynamic trade-off between the safety and economy of broadband oscillation active defense control. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the structure and control framework of the controller of the present invention; Figure 2 This is a topology diagram of a multi-infeed system for a new energy power plant in the embodiment; Figure 3 This example illustrates the convergence of reward values ​​during the training process of the multi-objective controller. Figure 4a and Figure 4b The controller of this invention regulates the output active power of the new energy power plants #2 and #3 before and after the power plant under typical operating conditions; Figure 4c and Figure 4d These are the controller parameter values ​​for new energy power plants #2 and #3 after the controller of this invention is adjusted under typical operating conditions; Figure 4e This invention relates to the gain margin of system oscillation before and after regulation under typical operating conditions using the controller. G M value; Figure 4f This refers to the control time of the controller under typical operating conditions. Figure 5 The figure shows the experimental results of the controller of the present invention under typical operating conditions; Table 1a shows the detailed parameters of the converter in the new energy power station #1 in the embodiment of the present invention; Table 1b shows the detailed parameters of the converters in new energy power plants #2 and #3 in the embodiments of the present invention; Table 2 shows the network parameters of the multi-feed system for new energy power plants in the embodiments of the present invention; Table 3 shows the initial states before adjustment for seven typical operating conditions in the embodiments of the present invention. Detailed Implementation

[0011] In this embodiment, the multi-objective broadband oscillation active defense controller for the multi-infeed system of the new energy power plant consists of a risk sensor K1 and a multi-objective controller K2. The risk sensor K1 is a broadband oscillation risk sensor based on an impedance network model. The risk sensor K1 samples the key time-domain operating point information of each multi-infeed system of the new energy power plant through communication. The oscillation risk information R1 under stable operating conditions is obtained using the impedance network model of a multi-infeed system in a new energy power plant. The oscillation risk information R1 includes the wideband oscillation interaction frequency. f osc and gain margin G M The topology diagram of the multi-infeed system of the new energy power plant in this embodiment is as follows: Figure 2 As shown, the new energy converters in new energy power plants #1 and #2 adopt grid-following control, while the new energy converters in new energy power plant #3 adopt grid-connected control. Detailed parameters of the converters in new energy power plants #1 and #2 are shown in Table 1a, and detailed parameters of the converters in new energy power plant #3 are shown in Table 1b. The network parameters of the multi-feed system are shown in Table 2. Key time-domain operating point information. This includes the output power of each new energy power station and the controller parameters of the converters within the new energy power station; during implementation, settings are... s TD ={ K FL,pllp,n , P FLS,n},in, K FL,pllp,n Indicates the first n The phase-locked loop proportional coefficient of the converter in a new energy power station P FLS,n Indicates the first n Each new energy power station outputs active power. n =1,2.

[0012] The multi-objective controller K2 is a deep reinforcement learning-based multi-objective controller. K2 is a deep neural network trained using deep reinforcement learning and guided by the perceived oscillation risk information R1. The input to the deep neural network is the state of the multi-feed system of the new energy power plant. The output of the deep neural network is a multi-target broadband oscillatory active defense control action. Status of multi-infeed systems in new energy power plants Includes key time-domain operating point information Oscillation risk information R1, and the set oscillation gain margin Target region, oscillation gain margin The target area is denoted as target area T1; multi-target broadband oscillation active defense control action. Includes the parameter control quantities of the converter controller in new energy power plants and the active power output regulation of each new energy power station .

[0013] In this embodiment, the settings are as follows: s RPS ={ s TD , f osc , G M , G M,min , G M,max}, a RPS ={Δ K FL,pllp,n , Δ P FLS,n},in, G M,min and G M,max Δ represents the upper and lower limits of the target region, respectively. K FL,pllp,n Indicates the first n In a new energy power station, the phase-locked loop proportional coefficient control quantity and Δ of the converter are... P FLS,n Indicates the first n The output active power regulation of each new energy power station n =1,2.

[0014] The multi-target controller K2 outputs multi-target wideband oscillation active defense control actions through forward propagation. Multi-target broadband oscillation active defense control action The system sends data to the converters in each new energy power station via communication, which is used to adjust the operating status of each converter in order to achieve proactive defense and control against broadband oscillation risks.

[0015] In practice, the corresponding technical measures include: Risk sensor K1 obtains oscillation risk information R1 through the following steps: Step 1: Initialize the frequency band for assessing broadband oscillation risk : , ≥ ≥0, key time-domain operating point information of multi-infeed systems in sampling new energy power plants. Specifically, this embodiment mainly examines the subsynchronous and supersynchronous oscillation risks of multi-infeed systems in new energy power plants, therefore the examination frequency band is set as Ω=[0Hz, 100Hz].

[0016] Step 2: Based on the component connection method, the multi-feed system of the new energy power plant is divided into active sub-module M1 and passive network sub-module M2 at the grid connection points of each new energy power plant. Active sub-module M1 includes all active devices, including each new energy power plant, and passive network sub-module M2 includes all passive devices, including transmission lines. The specific division method in this embodiment is as follows: Figure 2 As shown in the image.

[0017] Step 3: Construct the frequency domain impedance model matrix of each new energy power station in the active submodule M1. .

[0018] Step 4: Construct the impedance matrix of the passive network in the passive network submodule M2 .

[0019] Step 5: with E Representing a unit diagonal matrix, the characteristic equation for constructing the impedance network model of a multi-infeed system in a new energy power plant is given. As shown in equation (1), and extract the characteristic equation. Open-loop transfer function matrix As shown in equation (2): ; At the current operating point, the open-loop transfer function matrix is ​​calculated. In the frequency band of the study frequency response characteristics within , .

[0020] Step 6: At the current operating point, use the generalized Nyquist criterion to sense the oscillation risk information R1 of the multi-feed system of the new energy power plant according to equation (3). The oscillation risk information R1 includes the oscillation interaction frequency. and oscillation gain margin ; ; In formula (3): This represents the eigenvalue decomposition operation in the generalized Nyquist criterion; λ To perform eigenvalue decomposition operation obtained L The eigenvalue matrix within the examined frequency band; It is the real part value corresponding to the intersection point with the minimum real part among all intersection points of the Nyquist curve of the multi-infeed system of the new energy power plant and the negative real axis; min(·) represents the operation of finding the minimum value; Re(·) and Im(·) represent the extracted real part and imaginary part, respectively; Represents positive real numbers; Indicates taking equal The corresponding oscillation frequency .

[0021] The multi-objective controller K2 is constructed based on a multi-objective broadband oscillation active defense control optimization model and using a deep neural network; it includes the cost function, optimization variables, and constraints of each control objective in the multi-objective broadband oscillation active defense control optimization model: 3.1 Set the cost function according to formula (4) : Cost function This refers to: oscillation gain margin Control objective cost function It is the oscillation gain margin of the multi-infeed system of a new energy power plant. The control is brought within the target region T1 to ensure the safe operation of the multi-infeed system of the new energy power plant with the desired stability margin; the oscillation gain margin in the target region T1. The value is The target area settings under different working conditions in this embodiment are shown in Table 3. ; In equation (4): For oscillation gain margin Distance from the center of target area T1; The distance from the boundary of the target area to the center point of the target area. ;when When, it represents the oscillation gain margin of the multi-infeed system of the new energy power plant. Within the target area T1; and These are the target regions T1 and their corresponding regions. Upper and lower limits, .

[0022] 3.2 Set the cost function according to formula (5) : Cost function This refers to the objective cost function that minimizes the power reduction of new energy power plants. This is to minimize the reduction of active power output from new energy power plants during the broadband oscillation active defense control process. In this embodiment, since the converter in new energy power plant #3 is a grid-type control, its output power is automatically adjusted according to the power balance of the system. Therefore, in this control objective, only the output power of new energy power plants #1 and #2 is considered. ; In equation (5): The first step in the active defense control process for broadband oscillations n The output active power regulation of grid-connected new energy power stations; NThis refers to the total number of grid-connected renewable energy power plants in a multi-feedback system for renewable energy power plants. n =1,2,…, N In this embodiment N =2.

[0023] 3.3. Set two types of optimization variables: Category 1 consists of converter controller parameters in new energy power plants. ; Category 2 is the active power output of new energy power plants. ; In this embodiment, the settings are as follows: The proportional coefficient of the phase-locked loop in converters #1 and #2 of the new energy power plant. and ; Set as the output active power of new energy power plants #1 and #2 , .

[0024] 3.4 Set constraints according to formula (6) The optimization variables satisfy the constraints set by equation (6) during the optimization adjustment process. ; In formula (6): K ctrl,n It is the first n Controller parameter values ​​of the converter in a new energy power station; Indicating after regulation K ctrl,n The value; K ctrl,min,n and K ctrl,max,n They are respectively The lower and upper limits; The setting of the lower and upper limits ensures that each converter has robust stability and good dynamic performance under ideal grid conditions; in this embodiment, the phase-locked loop proportional coefficients of the converters in new energy power plants #1 and #2 are set. and The constraint range is set to 1000≤ ≤3000, n =1,2; It is the first n The output active power value of each new energy power station; Indicating after regulation The value; for The upper limit, The upper limit depends on the maximum power output that the new energy power plant can generate under the current external natural conditions.

[0025] In this embodiment, assuming that the initial values ​​of the active power output of new energy power plants #1 and #2 before regulation are the upper limits under the current conditions, then the regulation amount of the active power output of new energy power plants #1 and #2 is... and Setting the value to only negative values ​​prevents overshooting after adjustment. and The upper limit.

[0026] 3.5 Constructing a multi-target broadband oscillation active defense control optimization model Comprehensive cost function Cost function By optimizing the variables and constraints, the optimization model for multi-objective broadband oscillation active defense control can be obtained as shown in equation (7): ; In equation (7): The overall cost function is the optimization model for multi-target broadband oscillation active defense control. This represents an array consisting of the adjusted optimization variables; and These are the upper and lower bound constraints for the optimization variables; and These are the weights of the two control objectives, respectively. + =1, by adjusting the weight relationship between the two control objectives, the priority of different control objectives can be adjusted, and the control objective corresponding to the larger weight has a higher priority.

[0027] In this multi-objective broadband oscillation active defense control optimization model, the control objectives and constraints are transformed into a multi-objective reward function. A deep reinforcement learning algorithm is then used to train the multi-objective controller K2, thereby enabling a rapid solution to the aforementioned multi-objective broadband oscillation active defense control optimization model. In this embodiment, the multi-objective reward function consists of constraint reward and control objective... Rewards and control objectives The reward consists of three parts, with the constraint reward having the highest priority, and the control objective being the second highest priority. Rewards are secondary; control is the primary objective. The reward has the lowest priority, and a multi-objective controller based on deep reinforcement learning is trained based on the multi-objective reward function with the above priorities.

[0028] The target region T1 is divided according to the following method: based on the oscillation gain margin of the multi-infeed system of the new energy power plant. The size settings include the warning region T2 and the prohibition region T3; the oscillation gain margin in the warning region T2. The value is: When the oscillation gain margin When in warning zone T2, the multi-target broadband oscillation active defense controller generates multi-target broadband oscillation active defense control actions. Early intervention is implemented for multi-infeed systems of new energy power plants that are in the warning zone T2 but have not yet become unstable, in order to achieve pre-emptive mitigation of broadband oscillation risks; in this embodiment, the upper limit of the warning zone T2 is... G M,min The lower limit value of the same target region is shown in Table 3; the oscillation gain margin in the forbidden region T3. The value is: This refers to a situation where the system has experienced broadband oscillation, and the multi-target broadband oscillation active defense controller prevents the system from entering the prohibited area through active defense control, thus preventing the broadband oscillation from occurring.

[0029] The target area T1 is set as follows: The target area T1 is set by the operator of the new energy power plant multi-infeed system based on actual operating conditions and the need for economic and safety trade-offs: When the fluctuations in the output of renewable energy power plants and loads in the system are relatively small, and the operator expects to minimize the output reduction of renewable energy power plants to improve the economic efficiency of system operation, the upper limit of T1 in the target area should be lowered. and lower limit To reduce the oscillation gain margin of multi-infeed systems in new energy power plants To meet the needs of the target area T1, thereby improving the economic efficiency of system operation; during the adjustment of the target area T1, the upper limit of the warning area T2 changes accordingly; as in this embodiment t 1 to t 2-hour operating conditions t 2 to t 3-hour operating conditions and t 6 to t Operating conditions at time 7.

[0030] When the operation of a multi-infeed system in a new energy power plant is subject to significant uncertainty, in order to ensure sufficient stability margin to cope with the risk of broadband oscillation instability caused by fluctuations in the operating point, the upper limit of the target region T1 is increased. and lower limit To improve the oscillation gain margin of multi-infeed systems in new energy power plants To meet the requirements and ensure system operational security, the upper limit of the warning area T2 changes accordingly during the adjustment of the target area T1. As in this embodiment... t 4 to t Operating conditions at time 5.

[0031] Figure 3This section describes the convergence of the reward value during the training process of the multi-objective controller K2 in this embodiment. The multi-objective controller obtained in the 2000th iteration after reward convergence is ultimately selected for subsequent verification. The initial states of seven typical operating conditions in this embodiment without the control of the multi-objective broadband oscillation active defense controller are shown in Table 3. Only the following conditions are set: t Initial operating condition settings at time 1 K FL,pllp,1 and K FL,pllp,2 , t 2 to t 7-hour working conditions K FL,pllp,1 and K FL,pllp,2 This is inherited from the multi-target broadband oscillation active defense controller in the previous time step. K FL,pllp The control results are as follows. The control results for seven typical operating conditions are as follows: Figures 4a to 4f As shown.

[0032] Depend on Figure 4a , Figure 4b , Figure 4e It can be seen that without broadband oscillation active defense control for multi-infeed systems in new energy power plants, as P FLS,1 and P FLS,2 Due to fluctuations, the multi-infeed system of the new energy power plant will operate in the warning zone, and may even enter the prohibited zone. t 1 to t Among the seven typical working conditions, t 2 and t The oscillation risk at time 3 is mainly driven by the new energy power plant #1. t 4, t 5 and t The risk of oscillation under operating conditions at time 6 is mainly driven by the new energy power plant #2, while... t At time 7, the oscillation risk under this operating condition is jointly dominated by new energy power plant #1 and new energy power plant #2. According to... Figure 4e In G M The control results show that, in cases where the broadband oscillation risk of a multi-infeed system in a new energy power plant is dominated by different new energy power plants, the multi-target broadband oscillation active defense controller of this invention can effectively mitigate the risk. G M Effectively regulate to the preset target area. And according to Figures 4a-4d The results of the regulation show that, t 3. t 6 and tUnder 7 operating conditions, the multi-target broadband oscillation active defense controller of this invention achieves lower-cost controller parameters through intelligent coordination control. K FL,pllp,1 and K FL,pllp,2 Wideband oscillation of multi-infeed systems in new energy power plants G M The control was adjusted to the target area without reducing the output power of new energy power plants #1 and #2. By coordinating the global power plants in the multi-feed system, unnecessary power reduction of each new energy power plant was avoided, thereby improving the economic efficiency of the control.

[0033] In typical operating conditions, the multi-infeed system of a new energy power plant consists of... t Operating conditions change at time 4 t During the operation at time 5, the operating point remains unchanged, only the target area is adjusted from the economy-oriented [0.1, 0.2] to the safety-oriented [0.25, 0.35]. According to... Figures 4a-4d middle t 4 and t The results of the five-time-period operating condition control show that the multi-target broadband oscillation active defense controller prioritizes... K FL,pllp,1 and K FL,pllp,2 Adjust to its upper limit, at which point the multi-infeed system of the new energy power plant... G M The target area is still a considerable distance away, therefore the multi-target broadband oscillation active defense controller further... P FLS,2 The efficiency was reduced from 0.906 pu to 0.742 pu. After adjustment, the efficiency of the multi-infeed system in the renewable energy power plant was improved. G M The value increased from 0.157 to 0.279. This indicates that adjusting the target area setting allows the overall control strategy of the multi-target broadband oscillation active defense controller to be more economical or safer.

[0034] according to Figure 4f It is known that the multi-target broadband oscillation active defense controller of the present invention does not require iterative calculation and can complete the optimal control decision through only one neural network forward propagation process. In the typical working condition of this embodiment, the single control time can be controlled within 80ms.

[0035] To further verify the control effect of the multi-objective broadband oscillation active defense controller of the present invention in actual broadband oscillation active defense control, a semi-physical experiment was conducted to verify it. The verification results are as follows: Figure 5 As shown.

[0036] Figure 5 middle, I FLS,1,a , I FLS,2,a , u FM,g,a The waveforms represent the a-phase output current waveforms of new energy power plants #1 and #2, respectively, and the a-phase voltage waveform of new energy power plant #3 at its PCC. Experimental results show that during the switching process under different operating conditions, the multi-target broadband oscillation active defense controller of this invention can quickly perceive and respond to broadband oscillation risks based on the real-time changes in the state of the multi-infeed system of the new energy power plant. It should be noted here that... Figure 5 middle t 3 to t 4-hour operating condition switching process t 5 to t 6-minute operating condition switching process and t 6 to t During the switching process at time 7, due to the large step change in the output power command of new energy power plants #1 and #2, their output current experienced a brief transient fluctuation. This was a normal transient process and not a broadband oscillation instability caused by a change in the operating point. This indicates that it can efficiently make control action decisions based on the real-time operating status of the multi-infeed system of the new energy power plants, achieving proactive defense control against broadband oscillation risks.

[0037] Table 1a Detailed parameters of converters in new energy power plants #1 and #2

[0038] Table 1b Detailed parameters of the converter in the #3 new energy power plant

[0039] Table 2 Network Parameters of Multi-Infeed Systems for New Energy Power Plants

[0040] Table 3 Initial states of seven typical operating conditions

[0041] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-target broadband oscillation active defense controller for a multi-infeed system in a new energy power plant, characterized in that: It consists of a risk sensor K1 and a multi-objective controller K2; the risk sensor K1 is a broadband oscillation risk sensor based on an impedance network model, and the risk sensor K1 samples the key time-domain operating point information of the multi-infeed system of each new energy power plant through communication. The oscillation risk information R1 under stable operating conditions is obtained using the impedance network model of a multi-infeed system in a new energy power plant; the key time-domain operating point information... This includes the output power of each new energy power station and the controller parameters of the converters within the new energy power station; the multi-objective controller K2 is a deep reinforcement learning-based multi-objective controller, which is a deep neural network trained using deep reinforcement learning and guided by the perceived oscillation risk information R1. The input of the deep neural network is the state of the multi-feed system of the new energy power station. The output of the deep neural network is a multi-target broadband oscillatory active defense control action. The status of the multi-infeed system of the new energy power station Includes key time-domain operating point information Oscillation risk information R1, and the set oscillation gain margin Target area T1; the multi-target broadband oscillation active defense control action Includes the parameter control quantities of the converter controller in new energy power plants and the active power output regulation of each new energy power station The multi-target controller K2 outputs multi-target broadband oscillation active defense control actions through forward propagation. ; The multi-target broadband oscillation active defense control action The system sends data to the converters in each new energy power station via communication, which is used to adjust the operating status of each converter in order to achieve proactive defense and control against broadband oscillation risks.

2. The multi-target broadband oscillation active defense controller for multi-infeed systems in new energy power plants according to claim 1, characterized in that: The risk sensor K1 obtains the oscillation risk information R1 according to the following steps: Step 1: Initialize the frequency band for assessing broadband oscillation risk : , ≥ ≥0, key time-domain operating point information of multi-infeed systems in sampling new energy power plants. ; Step 2: Based on the component connection method, the multi-feed system of the new energy power station is divided into active sub-module M1 and passive network sub-module M2 at the grid connection point of each new energy power station. The active sub-module M1 includes each active device including each new energy power station, and the passive network sub-module M2 includes each passive device including the transmission line. Step 3: Construct the frequency domain impedance model matrix of each new energy power station in the active submodule M1. ; Step 4: Construct the impedance matrix of the passive network in the passive network submodule M2. ; Step 5: with E Representing a unit diagonal matrix, the characteristic equation for constructing the impedance network model of a multi-infeed system in a new energy power plant is given. As shown in equation (1), and extract the characteristic equation. Open-loop transfer function matrix As shown in equation (2): ； At the current operating point, the open-loop transfer function matrix is ​​calculated. In the frequency band of the study frequency response characteristics within , ; Step 6: At the current operating point, use the generalized Nyquist criterion to sense the oscillation risk information R1 of the multi-feed system of the new energy power plant according to equation (3). The oscillation risk information R1 includes the oscillation interaction frequency. and oscillation gain margin ; ； In formula (3): This represents the eigenvalue decomposition operation in the generalized Nyquist criterion; λ To perform eigenvalue decomposition operation obtained L The eigenvalue matrix within the examined frequency band; It is the real part value corresponding to the intersection point with the minimum real part among all intersection points of the Nyquist curve of the multi-infeed system of the new energy power plant and the negative real axis; min(·) represents the operation of finding the minimum value; Re(·) and Im(·) represent the extracted real part and imaginary part, respectively; Represents positive real numbers; Indicates taking equal The corresponding oscillation frequency .

3. The multi-target broadband oscillation active defense controller for multi-infeed systems in new energy power plants according to claim 1, characterized in that: The multi-objective controller K2 is constructed based on a multi-objective broadband oscillation active defense control optimization model and using a deep neural network; it includes setting the cost function, optimization variables, and constraints in the multi-objective broadband oscillation active defense control optimization model: 3.1 Set the cost function according to formula (4) : The cost function This refers to: oscillation gain margin Control objective cost function It is the oscillation gain margin of the multi-infeed system of a new energy power plant. The control is brought within the target region T1 to ensure the safe operation of the multi-infeed system of the new energy power plant with the desired stability margin; the oscillation gain margin in the target region T1. The value is ; ； In equation (4): For oscillation gain margin Distance from the center of target area T1; The distance from the boundary of the target area to the center point of the target area. ;when When, it represents the oscillation gain margin of the multi-infeed system of the new energy power plant. Within the target area T1; and The oscillation gain margin corresponding to the target region T1 are respectively Upper and lower limits, ; 3.2 Set the cost function according to formula (5) : The cost function This refers to the objective cost function that minimizes the power reduction of new energy power plants. This is to minimize the reduction of active power output from new energy power plants during the active defense control process for broadband oscillations; ； In equation (5): The first step in the active defense control process for broadband oscillations n The output active power regulation of each new energy power station; N This refers to the total number of renewable energy power plants in the multi-feed system of renewable energy power plants. n =1,2,…, N ; 3.3 Set two types of optimization variables: The first type is the converter controller parameters in the new energy power plant. The second category is the output active power of new energy power plants. ; 3.4 Set constraints according to formula (6) The optimization variables satisfy the constraints set by equation (6) during the optimization adjustment process. ； In formula (6): It is the first n Controller parameter values ​​of the converter in a new energy power station; Indicating after regulation The value; and They are respectively The lower and upper limits; The setting of the lower and upper limits ensures that each converter has robust stability and good dynamic performance under ideal grid conditions; It is the first n The output active power value of each new energy power station; Indicating after regulation The value; for The upper limit, The upper limit depends on the maximum power output that the new energy power plant can currently generate under the existing external natural conditions; 3.5 Constructing a multi-objective broadband oscillation active defense control optimization model Comprehensive cost function Cost function The optimization model for multi-objective broadband oscillation active defense control, obtained by optimizing variables and constraints, is shown in equation (7): ； In equation (7): The overall cost function is the optimization model for multi-target broadband oscillation active defense control. This represents an array consisting of the adjusted optimization variables; and These are the upper and lower bound constraints for the optimization variables; and These are the weights of the two control objectives, + =1, by adjusting the weight relationship between the two control objectives, the priority of different control objectives can be adjusted, and the control objective corresponding to the larger weight has a higher priority; The control objectives and constraints in the multi-objective broadband oscillation active defense control optimization model are converted into multi-objective reward functions, and the multi-objective controller K2 is trained by a deep reinforcement learning algorithm, thereby realizing the rapid solution of the above multi-objective broadband oscillation active defense control optimization model.

4. The multi-target broadband oscillation active defense controller for multi-infeed systems in new energy power plants according to claim 3, characterized in that: Based on the oscillation gain margin of the multi-infeed system of new energy power plants The size of the warning region T2 and the prohibition region T3 are set: the oscillation gain margin in the warning region T2. The value is: When the oscillation gain margin When in warning zone T2, the multi-target broadband oscillation active defense controller generates multi-target broadband oscillation active defense control actions. Early intervention is implemented for multi-infeed systems of new energy power plants that are in the warning zone T2 but have not yet become unstable, in order to achieve ex-ante mitigation of broadband oscillation risks; the oscillation gain margin in the prohibited zone T3. The value is: This refers to a situation where the system has experienced broadband oscillation, and the multi-target broadband oscillation active defense controller prevents the system from entering the prohibited area through active defense control, thus preventing the broadband oscillation from occurring.

5. The multi-target broadband oscillation active defense controller for multi-infeed systems in new energy power plants according to claim 3, characterized in that: The target area T1 is adjusted as follows: The target area T1 is set by the operator of the multi-infeed system of renewable energy power plants based on actual operating conditions and the need for economic and safety trade-offs. When the fluctuations of renewable energy power plants and loads in the multi-infeed system are small, and the operator expects to minimize the output reduction of renewable energy power plants to improve the economic efficiency of system operation, the upper limit of the target area T1 is lowered. and lower limit To reduce the oscillation gain margin of multi-infeed systems in new energy power plants This addresses the needs of [the system], thereby improving the economic efficiency of system operation; During the adjustment of the target area T1, the upper limit of the warning area T2 changes accordingly; When the operation of a multi-infeed system in a new energy power plant is subject to significant uncertainty, in order to ensure sufficient stability margin to cope with the risk of broadband oscillation instability caused by fluctuations in the operating point, the upper limit of the target region T1 is increased. and lower limit To improve the oscillation gain margin of multi-infeed systems in new energy power plants To meet the requirements and ensure the safety of system operation; during the adjustment of target area T1, the upper limit of warning area T2 changes accordingly.

Images